Modern Power Systems Analysis
Xi-Fan Wang
l
Yonghua Song
l
Malcolm Irving
Modern Power Systems Analysis
123
X...

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Modern Power Systems Analysis

Xi-Fan Wang

l

Yonghua Song

l

Malcolm Irving

Modern Power Systems Analysis

123

Xi-Fan Wang Xi’an Jiaotong University Xi’an People’s Republic of China

Yonghua Song The University of Liverpool Liverpool United Kingdom

Malcolm Irving Brunel University Middlesex United Kingdom

ISBN 978-0-387-72852-0

e-ISBN 978-0-387-72853-7

Library of Congress Control Number: 2008924670 # 2008 Springer Science+Business Media, LLC All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. While the advice and information in this book are believed to be true and accurate at the date of going to press, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein. Printed on acid-free paper. 987654321 springer.com

Preface

The power industry, a capital and technology intensive industry, is a basic national infrastructure. Its security, reliability, and economy have enormous and far-reaching effects on a national economy. An electrical power system is a typical large-scale system. Questions such as how to reflect accurately the characteristics of modern electrical power systems, how to analyze effectively their operating features, and how to improve further the operating performance are always at the forefront of electrical power systems research. Electrical power system analysis is used as the basic and fundamental measure to study planning and operating problems. In the last century, electrical power researchers have undertaken a great deal of investigation and development in this area, have made great progress in theoretical analysis and numerical calculation, and have written excellent monographs and textbooks. Over the last 20 years, the changes in electrical power systems and other relevant technologies have had a profound influence on the techniques and methodologies of electrical power system analysis. First, the development of digital computer technology has significantly improved the performance of hardware and software. Now, we can easily deal with load flow issues with over ten thousand nodes. Optimal load flow and static security analysis, which were once considered hard problems, have attained online practical applications. Second, the applications of HVDC and AC flexible transmission technologies (FACTS) have added new control measures to electrical power systems, and have increased power transmission capacity, enhanced control capability, and improved operating characteristics. However, these technologies bring new challenges into the area of electrical power system analysis. We must build corresponding mathematical models for these new devices and develop algorithms for static and dynamic analysis of electrical power systems including these devices. In addition, the rapid development of communication technology has enabled online monitoring of electrical power systems. Therefore, the demand for online software for electrical power system analysis becomes more and more pressing. Furthermore, worldwide power industry restructuring and deregulation has separated the former vertically integrated system into various parts, and the once

v

vi

Preface

unified problem of power system dispatching is now conducted via complicated bilateral contracts and spot markets. New issues such as transmission ancillary service and transmission congestion have emerged. In recent years, several power blackouts have taken place worldwide, especially the ‘‘8.13’’ blackout on the eastern grid of USA and Canada and the blackouts that occurred successively in other countries have attracted a great deal of attention. All of these aspects require new theories, models, and algorithms for electrical power system analysis. It is within such an environment that this book has been developed. The book is written as a textbook for senior students and postgraduates as well as a reference book for power system researchers. We acknowledge the support from various research funding organizations, their colleagues, and students, especially, the special funds for Major State Basic Research Projects of China ‘‘Research on Power System Reliability under Deregulated Environment of Power Market’’ (2004CB217905). We express our special gratitude to Professor Wan-Liang Fang and Professor Zheng-Chun Du for providing the original materials of Chaps. 5 and 6, and 7 and 8, respectively. We also express our sincere gratitude to the following colleagues for their contributions to various chapters of the book: Professor Zhao-Hong Bie for Chaps. 1 and 3; Professor Xiu-Li Wang for Chaps. 2 and 4; Dr. Ze-Chun Hu for Chap. 3; Dr. Xiao-Ying Ding for Chap. 4; Dr. Lin Duan for Chaps. 5 and 6; Professor De-Chiang Gang for Chap. 7; and Professor Hai-Feng Wang for Chaps. 6 and 8. Xi’an, China Liverpool, UK London, UK

Xi-Fan Wang Yonghuna Song Malcolm Irving

Contents

1

Mathematical Model and Solution of Electric Network . . . . . . . . . . . . . . . . . 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.1 Node Equation and Loop Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.2 Equivalent Circuit of Transformer and Phase Shift Transformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.3 Nodal Admittance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.1 Basic Concept of Nodal Admittance Matrix . . . . . . . . . . . . . . . . . . 13 1.3.2 Formulation and Modification of Nodal Admittance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.4 Solution to Electric Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.4.1 Gauss Elimination Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.4.2 Triangular Decomposition and Factor Table . . . . . . . . . . . . . . . . . 27 1.4.3 Sparse Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1.4.4 Sparse Vector Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 1.4.5 Optimal Ordering Schemes of Electric Network Nodes . . . . . . 43 1.5 Nodal Impedance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 1.5.1 Basic Concept of Nodal Impedance Matrix . . . . . . . . . . . . . . . . . . 48 1.5.2 Forming Nodal Impedance Matrix Using Admittance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 1.5.3 Forming Nodal Impedance Matrix by Branch Addition Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2

Load Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Formulation of Load Flow Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Classification of Node Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Node Power Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Load Flow Solution by Newton Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Basic Concept of Newton Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Correction Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71 71 73 73 76 79 79 83

vii

viii

Contents

2.4

2.5

2.6

3

2.3.3 Solution Process of Newton Method . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2.3.4 Solution of Correction Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 Fast Decoupled Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 2.4.1 Introduction to Fast Decoupled Method . . . . . . . . . . . . . . . . . . . . . 101 2.4.2 Correction Equations of Fast Decoupled method . . . . . . . . . . . . 104 2.4.3 Flowchart of Fast Decoupled Method . . . . . . . . . . . . . . . . . . . . . . . 107 Static Security Analysis and Compensation Method . . . . . . . . . . . . . . . . 113 2.5.1 Survey of Static Security Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 2.5.2 Compensation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 DC Load Flow Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 2.6.1 Model of DC Load Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 2.6.2 Outage Analysis by DC Load Flow Method . . . . . . . . . . . . . . . . . 122 2.6.3 N-1 Checking and Contingency Ranking Method . . . . . . . . . . . 123

Stochastic Security Analysis of Electrical Power Systems . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Basic Concepts of Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Probability of Stochastic Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Random Variables and its Distribution . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Numeral Character of Random Variable . . . . . . . . . . . . . . . . . . . . 3.2.4 Convolution of Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.5 Several Usual Random Variable Distributions . . . . . . . . . . . . . . 3.2.6 Markov Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Probabilistic Model of Power Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Probabilistic Model of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Probabilistic Model of Power System Components . . . . . . . . . 3.3.3 Outage Table of Power System Components . . . . . . . . . . . . . . . . 3.4 Monte Carlo Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Fundamental Theory of Monte Carlo Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Sampling of System Operation State . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 State Evaluation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Indices of Reliability Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.5 Flowchart of Composite System Adequacy Evaluation . . . . . 3.4.6 Markov Chain Monte Carlo (MCMC) Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Probabilistic Load Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Cumulants of Random Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.2 Linearization of Load Flow Equation . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3 Computing Process of Probabilistic Load Flow . . . . . . . . . . . . . 3.6 Probabilistic Network-Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.2 Network-Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.3 Lower Boundary Points of Feasible Flow Solutions . . . . . . . . 3.6.4 Reliability of Transmission System . . . . . . . . . . . . . . . . . . . . . . . . . .

129 129 130 130 132 133 135 136 138 140 140 141 142 145 145 148 150 151 152 156 161 162 168 171 178 178 180 186 188

Contents

4

5

ix

Power Flow Analysis in Market Environment . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Transmission Owner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 Independent Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.3 Power Exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.4 Ancillary Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.5 Scheduling Coordinator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Optimal Power Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 General Formulation of OPF Problem . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Approaches to OPF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Interior Point Method (IPM) for OPF Problem . . . . . . . . . . . . . . 4.3 Application of Optimal Power Flow in Electricity Market . . . . . . . . . . 4.3.1 Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Congestion Management Method Based On OPF . . . . . . . . . . . 4.4 Power Flow Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Current Decomposition Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2 Mathematical Model of Loss Allocation . . . . . . . . . . . . . . . . . . . . 4.4.3 Usage Sharing Problem of Transmission Facilities . . . . . . . . . . 4.4.4 Methodology of Graph Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Available Transfer Capability of Transmission System . . . . . . . . . . . . . 4.5.1 Introduction To Available Transfer Capability . . . . . . . . . . . . . . 4.5.2 Application of Monte Carlo Simulation in ATC Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.3 ATC Calculation with Sensitivity Analysis Method . . . . . . . .

193 193 193 194 194 195 195 196 196 198 202 217 217 223 228 230 232 234 238 241 241

HVDC and FACTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 HVDC Basic Principles and Mathematical Models . . . . . . . . . . . . . . . . . 5.2.1 HVDC Basic Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Converter Basic Equations Neglecting Lc . . . . . . . . . . . . . . . . . . . 5.2.3 Converter Basic Equations Considering Lc . . . . . . . . . . . . . . . . . 5.2.4 Converter Equivalent Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.5 Multiple Bridge Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.6 Converter Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Power Flow Calculation of AC/DC Interconnected Systems . . . . . . . 5.3.1 Converter Basic Equations in per Unit System . . . . . . . . . . . . . . 5.3.2 Power Flow Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.3 Jacobian Matrix of Power Flow Equations . . . . . . . . . . . . . . . . . . 5.3.4 Integrated Iteration formula of AC/DC Interconnected Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.5 Alternating Iteration for AC/DC Interconnected Systems . . . 5.4 HVDC Dynamic Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Basic Principles and Mathematical Models of FACTS . . . . . . . . . . . . . 5.5.1 Basic Principle and Mathematical Model of SVC . . . . . . . . . . . 5.5.2 Basic Principle and Mathematical Model of STATCOM . . .

255 255 258 258 261 267 273 276 279 281 282 283 286

245 246

289 294 299 301 302 308

x

Contents

5.5.3 5.5.4 5.5.5 5.5.6 6

7

Basic Principle and Mathematical Model of TCSC . . . . . . . . . Basic Principle and Mathematical Model of SSSC . . . . . . . . . . Basic Principle and Mathematical Model of TCPST . . . . . . . . Basic Principle and Mathematical Model of UPFC . . . . . . . . .

313 319 322 325

Mathematical Model of Synchronous Generator and Load . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Mathematical Model of Synchronous Generator . . . . . . . . . . . . . . . . . . . . 6.2.1 Basic Mathematical Equations of Synchronous Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2 Mathematical Equations of Synchronous Generator Using Machine Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.3 Simplified Mathematical Model of Synchronous Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.4 Steady-State Equations and Phasor Diagram . . . . . . . . . . . . . . . . 6.2.5 Mathematical Equations Considering Effect of Saturation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.6 Rotor Motion Equation of Synchronous Generator . . . . . . . . . . 6.3 Mathematical Model of Generator Excitation Systems . . . . . . . . . . . . . 6.3.1 Mathematical Model of Exciter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.2 Voltage Measurement and Load Compensation Unit . . . . . . . 6.3.3 Limiters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.4 Mathematical Model of Power System Stabilizer . . . . . . . . . . . 6.3.5 Mathematical Model of Excitation Systems . . . . . . . . . . . . . . . . . 6.4 Mathematical Model of Prime Mover and Governing System . . . . . . 6.4.1 Mathematical Model of Hydro-Turbine and Governing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.2 Mathematical Model of Steam Turbine and Governing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Mathematical Model of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.1 Static Load Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.2 Dynamic Load Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

333 333 335

Power System Transient Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Numerical Methods for Transient Stability Analysis . . . . . . . . . . . . . . . . 7.2.1 Numerical Methods for Ordinary Differential Equations . . . 7.2.2 Numerical Methods for Differential-Algebraic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.3 General Procedure for Transient Stability Analysis . . . . . . . . . 7.3 Network Mathematical Model for Transient Stability Analysis . . . . . 7.3.1 The Relationship Between Network and Dynamic Devices . 7.3.2 Modeling Network Switching and Faults . . . . . . . . . . . . . . . . . . . .

336 343 351 354 357 360 363 365 375 376 377 377 381 382 389 393 395 397 405 405 407 408 425 427 430 431 439

Contents

7.4

Transient Stability Analysis with Simplified Model . . . . . . . . . . . . . . . . . 7.4.1 Computing Initial Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.2 Solving Network Equations with Direct Method . . . . . . . . . . . . 7.4.3 Solving Differential Equations by Modified Euler Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.4 Numerical Integration Methods for Transient Stability Analysis under Classical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transient Stability Analysis with FACTS Devices . . . . . . . . . . . . . . . . . . 7.5.1 Initial Values and Difference Equations of Generators . . . . . 7.5.2 Initial Values and Difference Equations of FACTS and HVDC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.3 Forming Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.4 Simultaneous Solution of Difference and Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

446 447 448

Small-Signal Stability Analysis of Power Systems . . . . . . . . . . . . . . . . . . . . . 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Linearized Equations of Power System Dynamic Components . . . . . 8.2.1 Linearized Equation of Synchronous Generator . . . . . . . . . . . . . 8.2.2 Linearized Equation of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.3 Linearized Equation of FACTS Components . . . . . . . . . . . . . . . . 8.2.4 Linearized Equation of HVDC Transmission System . . . . . . . 8.3 Steps in Small-Signal Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.1 Network Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.2 Linearized Differential Equations of Whole Power System . 8.3.3 Program Package for Small-Signal Stability Analysis . . . . . . 8.4 Eigenvalue Problem in Small-Signal Stability Analysis . . . . . . . . . . . . 8.4.1 Characteristics of State Matrix Given by Its Eigensolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.2 Modal Analysis of Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.3 Computation of Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.4 Eigensolution of Sparse Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.5 Application of Eigenvalue Sensitivity Analysis . . . . . . . . . . . . . 8.5 Oscillation Analysis of Power Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

489 489 493 493 500 502 503 506 506 508 510 519

7.5

8

xi

450 457 463 464 475 484 487

519 523 526 530 533 534

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555

Chapter 1

Mathematical Model and Solution of Electric Network

1.1

Introduction

The mathematical model of an electric network is the basis of modern power system analysis, which is to be used in studies of power flow, optimal power flow, fault analysis, and contingency analysis. The electric network is constituted by transmission lines, transformers, parallel/series capacitors, and other static elements. From the viewpoint of electrical theory, no matter how complicated the network is, we can always establish its equivalent circuit and then analyze it according to the AC circuit laws. In this chapter, the electric network is represented by the linear lumped parameter model that is suitable for studies at synchronous frequency. For electromagnetic transient analysis, the high frequency phenomena and wave processes should be considered. In that situation, it is necessary to apply equivalent circuits described by distributed parameters. Generally speaking, an electric network can be always represented by a nodal admittance matrix or a nodal impedance matrix. A modern power system usually involves thousands of nodes; therefore methods of describing and analyzing the electric network have a great influence on modern power system analysis. The nodal admittance matrix of a typical power system is large and sparse. To enhance the computational efficiency, sparsity techniques are extensively employed. The nodal admittance matrix and associated sparsity techniques will be thoroughly discussed in this chapter. The nodal impedance matrix is widely applied in the fault analysis of power systems and will be introduced in Sect. 1.5. The equivalent circuits of the transformer and phase-shifting transformer are also presented in Sect. 1.1 because they require special representation methods.

X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

1

2

1.2 1.2.1

1 Mathematical Model and Solution of Electric Network

Basic Concepts Node Equation and Loop Equation

There are two methods usually employed in analyzing AC circuits, i.e., the node voltage method and loop current method. Both methods require the solution of simultaneous equations. The difference between them is that the former applies node equations while the latter applies loop equations. At present, node equations are more widespread in analyzing power systems, and loop equations are used sometimes as an auxiliary tool. In the following, we use a simple electric network as an example to illustrate the principle and characteristics of the node equation method. As shown in Fig. 1.1, the sample system has two generators and an equivalent load, with five nodes and six branches whose admittances are y1 y6 . Assigning the ground as the reference node, we can write the nodal equations according to the Kirchoff’s current law, 9 y4 ðV_ 2 V_ 1 Þ þ y5 ðV_ 3 V_1 Þ y6 V_ 1 ¼ 0 > > > > > y1 ðV_ 4 V_ 2 Þ þ y3 ðV_ 3 V_2 Þ þ y4 ðV_ 1 V_2 Þ ¼ 0 > > = y2 ðV_ 5 V_ 3 Þ þ y3 ðV_ 2 V_3 Þ þ y5 ðV_ 1 V_ 3 Þ ¼ 0 ; > > > > y1 ðV_ 4 V_ 2 Þ ¼ I_1 > > > ; y2 ðV_ 5 V_ 3 Þ ¼ I_2

ð1:1Þ

where V_1 V_ 5 denote the node voltages. Combining the coefficients of node voltages, we obtain the following equations:

•

V4

y1

y3

•

V2

•

•

V3

•

•

I1

I2

I3 •

y4

y2

•

I4

I5

•

V1 •

I6

Fig. 1.1 Sample system for node voltage method

y6

y5

•

V5

1.2 Basic Concepts

3

9 ðy4 þ y5 þ y6 ÞV_1 y4 V_ 2 y5 V_ 3 ¼ 0 > > > > _ _ _ > y4 V1 þ ðy1 þ y3 þ y4 ÞV2 y3 V3 y1 V4 ¼ 0 > > = _ _ _ _ y5 V1 y3 V2 þ ðy2 þ y3 þ y5 ÞV:3 y2 V5 ¼ 0 : > > > > y1 V_2 þ y1 V_ 4 ¼ I_1 > > > ; _ _ _ y2 V3 þ y2 V5 ¼ I2

ð1:2Þ

In (1.2), the left-hand term is the current flowing from the node and the right-hand term is the current flowing into the node. The above equations can be rewritten in more general form as follows: 9 Y11 V_1 þ Y12 V_ 2 þ Y13 V_ 3 þ Y14 V_4 þ Y15 V_ 5 ¼ I_1 > > > > _ _ _ _ _ _ > Y21 V1 þ Y22 V2 þ Y23 V3 þ Y24 V4 þ Y25 V5 ¼ I2 > > = _ _ _ _ _ _ Y31 V1 þ Y32 V2 þ Y33 V3 þ Y34 V4 þ Y35 V5 ¼ I3 : > > > Y41 V_1 þ Y42 V_ 2 þ Y43 V_ 3 þ Y44 V_4 þ Y45 V_ 5 ¼ I_4 > > > > ; _ _ _ _ _ _ Y51 V1 þ Y52 V2 þ Y53 V3 þ Y54 V4 þ Y55 V5 ¼ I5

ð1:3Þ

Comparing (1.3) with (1.2), we can see Y11 ¼ y4 þ y5 þ y6 ; Y22 ¼ y1 þ y3 þ y4 ; Y33 ¼ y2 þ y3 þ y5 ; Y44 ¼ y1 ; Y55 ¼ y2 : These elements are known as nodal self-admittances. Y12 ¼ Y21 ¼ y4 ; Y13 ¼ Y31 ¼ y5 ; Y23 ¼ Y32 ¼ y3 ; Y24 ¼ Y42 ¼ y1 ; Y35 ¼ Y53 ¼ y2 : Similarly, the above elements are known as mutual admittances between the connected nodes. The mutual admittances of the pair of disconnected nodes are zero. Equation (1.3) is the node equation of the electric network. It reflects the relationship between node voltages and injection currents. Here I_1 I_5 are the nodal injection currents. In this example, except I_4 and I_5 , all other nodal injection currents are zero.

4

1 Mathematical Model and Solution of Electric Network

Equation (1.3) can be solved to get node voltages V_ 1 V_ 5 , then the branch currents can be obtained. Thus, we have obtained all the variables of the network. Generally, for a n node network, we can establish n linear node equations in (1.3) format. In matrix notation, we have I ¼ YV;

ð1:4Þ

where 2

3 I_1 6_ 7 6 I2 7 7 I¼6 6 .. 7; 4. 5 I_n

2

3 V_ 1 6 _ 7 6 V2 7 7 V¼6 6 .. 7: 4 . 5 V_ n

Here I is the vector of nodal injection currents and V is the vector of nodal voltages; Y is called the nodal admittance matrix 2

Y11 6 Y21 Y¼6 4 Yn1

Y12 Y22 Yn2

3 Y1n Y2n 7 7: 5 Ynn

As we have seen, its diagonal element Yii is the nodal self-admittance and the off diagonal element Yij is the mutual admittance between node i and node j. Now we introduce the incidence matrix that is very important in network representations. The incidence matrix represents the topology of an electric network. Different incidence matrices correspond to different networks configurations. The elements of the incidence matrix are only 0, þ1, or 1. They do not include the parameters of network branches. For example, there are five nodes and six branches in Fig. 1.1. Its incidence matrix is a matrix with five rows and six columns. 2

0 0 6 1 0 6 A¼6 6 0 1 4 1 0 0 1

0 1 1 0 0

1 1 0 0 0

3 1 1 0 07 7 1 07 7: 0 05 0 0

In the incidence matrix, the serial numbers of rows correspond to the node numbers and the serial numbers of columns correspond to the branch numbers. For example, the first row has three nonzero elements, which denotes node 1 is connected with three branches. These three nonzero elements are in the fourth, fifth, and sixth columns, which means the branches connected with node 1 are branches 4, 5, and 6.

1.2 Basic Concepts

5

If the branch current flows into the node, the nonzero element equals 1; if the branch current flows out of the node, the nonzero element equals 1. The positions of the nonzero elements in each column denote the two node numbers of the relevant branch. For example, in the fifth column the nonzero elements are in the first and third row, which means the fifth branch connects node 1 and 3. In the sixth column, there is only one nonzero element in the first row, which means the sixth branch is a grounded branch. From the above discussion we see that an incidence matrix can uniquely determine the topology of a network configuration. The incidence matrix has a close relationship with the network node equation. If there are n nodes and b branches in an electric network, the state equation for every branch is I_Bk ¼ yBk V_ Bk ;

ð1:5Þ

where yBk is the admittance of branch k; IBk the current flowing in branch k; and V_ Bk is the voltage difference of branch k, whose direction is determined by IBk : If branch k includes a voltage source, as shown in Fig. 1.2a, it should be transformed to the equivalent current source as shown in Fig. 1.2b. yBk ¼ 1=zBk a_ Bk ¼ e_Bk =zBk ¼ yBk e_ Bk

) :

The current source can be treated as current injecting into the electric network, thus the branch can also be represented by (1.5). In matrix notation, the equation of a b branch network is I B ¼ YB V B ;

ð1:6Þ

eBK •

zBK

a aBK •

yBK •

IBK

Fig. 1.2 Transformation from voltage source to current source

•

b

VBK

6

1 Mathematical Model and Solution of Electric Network

where I B is the vector of the currents in branches, V B the vector of the branch voltage differences, and YB is a diagonal matrix constituted by the branch admittances. According to Kirchoff’s current law, the injection current I_i of node i in an electric network can be expressed as follows I_i ¼

b X

aik I_Bk

ði ¼ 1; 2; . . . ; nÞ;

ð1:7Þ

k¼1

where aik is a coefficient. If branch current I_Bk directs toward node i, aik ¼ 1; if branch current I_Bk directs away from the node i, aik ¼ 1; and if branch k does not connect to node i, aik ¼ 0. It is easy to get the relationship between nodal current vector I_ and branch current vector I_B as follows, I ¼ AI B ;

ð1:8Þ

where A is the incidence matrix of the network. Assuming the power consumed in the whole network is S, we can obtain the following equation, S¼

b X

I^Bk V_ Bk ¼ I^B V_ B ;

i¼1

where I^Bk and I^B are the conjugate of the corresponding vector and * is the scalar product of the two vectors. From the viewpoint of the nodal input power, we have S¼

n X

^ V: _ I^i V_ i ¼ I

i¼1

Obviously, I^ V_ ¼ I^B V_ B : From (1.8), we see I^ ¼ I^B AT : Substituting it into (1.9), we obtain, I^B AT V_ ¼ I^B V_ B :

ð1:9Þ

1.2 Basic Concepts

7

Therefore, AT V_ ¼ V_ B :

ð1:10Þ

Substituting (1.6) and (1.10) into (1.8) sequentially, we can get _ I_ ¼ AYB AT V_ ¼ YV;

ð1:11Þ

where Y is the nodal admittance matrix of the electric network Y ¼ AYB AT :

ð1:12Þ

Thus the nodal equations of an electric network can be obtained from its incidence matrix. In the following, the network shown in Fig. 1.1 is used again to illustrate the basic principle of analyzing the electric network by the loop current equations. In the loop equation method, the network elements are often represented in impedance form. The equivalent circuit is shown in Fig. 1.3. There are three independent loops in the network and the loop currents are I_1 ; I_2 ; and I_3 , respectively. According to Kirchoff’s voltage law, the voltage equations of the loops are 9 V_ 4 ¼ ðz1 þ z4 þ z6 ÞI_1 þ z6 I_2 z4 I_3 > = : V_ 5 ¼ z6 I_1 þ ðz2 þ z5 þ z6 ÞI_2 þ z5 I_3 > ; _ _ _ 0 ¼ z4 I1 þ z5 I2 þ ðz3 þ z4 þ z5 ÞI3

ð1:13Þ

Rewrite the above equation into the normative form, 9 E_ 1 ¼ Z11 I_1 þ Z12 I_2 þ Z13 I_3 > = E_ 2 ¼ Z21 I_1 þ Z22 I_2 þ Z23 I_3 ; > ; E_ 3 ¼ Z31 I_1 þ Z32 I_2 þ Z33 I_3

•

V4 4

z1

ð1:14Þ

z3

2

z2

3

•

i2

•

i3

•

i1 •

•

I1

I3

z4

z5

•

i4

•

•

•

Fig. 1.3 Sample system with loop currents

1 V1 z6

i6

i5

•

I2

5

8

1 Mathematical Model and Solution of Electric Network

where E_ 1 ¼ V_4 ; E_ 2 ¼ V_ 5 ; E_ 1 ¼ 0 are voltage potentials of three loops, respectively, Z11 ¼ z1 þ z4 þ z6 ; Z22 ¼ z2 þ z5 þ z6 ; Z33 ¼ z3 þ z4 þ z5 are loop self-impedances, Z12 ¼ Z21 ¼ z6 ; Z13 ¼ Z31 ¼ z4 ; Z23 ¼ Z32 ¼ z5 are the loop mutual impedances. If we know loop voltage E_ 1 ; E_ 2 ; and E_ 3 , we can solve the loop current I_1 ; I_2 ; and I_3 from (1.14), and then obtain the branch current, i_1 ¼ I_1 ; i_2 ¼ I_2 ; i_3 ¼ I_3 ; i_4 ¼ I_1 I_3 ; i_5 ¼ I_2 þ I_3 ; i_6 ¼ I_1 þ I_2 : And the node voltages are V_ 1 ¼ z6 i_6 ;

V_ 2 ¼ V_ 4 z1 i_1 ; V_ 3 ¼ V_5 z2 i_2 :

Thus all the variables of the electric network are solved. Generally, an electric network with m independent loops can be formulated by m loop equations. In matrix notation, we have E1 ¼ Z1 I1 ;

ð1:15Þ

where 2

3 I_1 6 _ 7 6 I2 7 7 I1 ¼ 6 6 .. 7; 4 . 5 I_m

2

3 E_ 1 6 _ 7 6 E2 7 7 E1 ¼ 6 6 .. 7 4 . 5 E_ m

are vectors of the loop currents and voltage phasors, respectively; 2

Z11 6 Z21 Z1 ¼ 6 4 Zm1

Z12 Z22 Zm2

3 Z1m Z2m 7 7 5 Zmm

ð1:16Þ

is the loop impedance matrix, where Zii is the self-impedance of the loop i and equals the sum of the branch impedances in the loop; Zij is the mutual impedance between loop i and loop j, and equals the sum of the impedances of their common branches. The sign of Zij depends on the directions of loop currents of loop i and loop j. If their directions are identical, Zij is positive, and if their directions are different, Zij is negative.

1.2 Basic Concepts

9

For the example shown in Fig. 1.3 we can write the basic loop incidence matrix according to the three independent loops, 2 3 1 0 0 1 0 1 B ¼ 4 0 1 0 0 1 1 5: 0 0 1 1 1 0 The serial numbers of rows correspond to the loop numbers and the serial numbers of columns correspond to the branch numbers. For example, in the third row, there are three nonzero elements in the third, fourth, and fifth columns which means loop 3 includes branches 3, 4, and 5. If the branch current has the same direction as the basic loop current, the corresponding nonzero element equals þ1; if the directions of branch current and loop current are different the corresponding nonzero element equals 1. It should be noted that a basic loop incidence matrix cannot uniquely determine a network configuration. In other words, there may be different configurations corresponding to the same basic loop incidence matrix. Similarly to the discussion on the node incidence matrix above, we can get the basic loop equations of an electric network from its basic loop incidence matrix B, ZL ¼ BZB BT ;

ð1:17Þ

where ZB is a diagonal matrix composed of the branch impedances. The application of incidence matrices is quite extensive. If we have the above basic concepts, network analysis problems can be dealt with more flexibly. The details will be discussed in the relevant later sections.

1.2.2

Equivalent Circuit of Transformer and Phase-Shift Transformer

The equivalent circuit of an electric network is established by the equivalent circuits of its elements such as transmission lines and transformers. The AC transmission line is often described by the nominal P equivalent circuit which can be found in other textbooks. In this section, only the equivalent circuits of the transformer and the phase-shift transformer are discussed, especially the transformer with off-nominal turns ratios. Flexible AC Transmission Systems (FACTS) are increasingly involved in power systems, and we will discuss the equivalent circuit of FACTS elements in Chap. 5. When the exciting circuit is neglected or treated as a load (or an impedance), a transformer can be represented by its leakage impedance connected in series with an ideal transformer as shown in Fig. 1.4a. The relation between currents and voltages can be formulated as follows:

10

1 Mathematical Model and Solution of Electric Network 1:K •

i

Vi

a

zT •

Ii

•

Vj •

Ij

j

i I&i

•

Vj

KzT i Vi Ii KzT K 2zT K−1 1−K

•

•

j

•

Ij

b

•

Vi (K−1)yT K

c

yT K

•

Ij j •

(1−K)yT K2

Vj

Fig. 1.4 Transformer equivalent circuit

9 I_i þ K I_j ¼ 0 = V_j : V_ i zT I_i ¼ ; K Solving the above equation, we can obtain 1 1 _ I_i ¼ V_ i Vj ; zT KzT 1 _ 1 I_j ¼ Vi þ 2 V_j : KzT K zT

ð1:18Þ

9 K1 _ 1 _ > Vi þ ðVi V_ j Þ > = KzT KzT : 1K 1 _ > ; Ij ¼ 2 V_j þ ðVj V_ i Þ > K zT KzT

ð1:19Þ

Rewrite (1.18) as follows Ii ¼

According to (1.19), we can get the equivalent circuit as shown in Fig. 1.4b. If the parameters are expressed in terms of admittance, the equivalent circuit is shown in Fig. 1.4c, where yT ¼

1 : zT

It should be especially noted in Fig. 1.4a the leakage impedance zT is at the terminal where the ratio is 1. When the leakage impedance zT is at the terminal where ratio is K, we should transform it to z0T by using the following equation, so that the equivalent circuit shown in Fig. 1.4 also can be applied in this situation z0T ¼ zT =K 2 :

ð1:20Þ

The equivalent circuit of a two-winding transformer has been discussed above. A similar circuit can be used to represent a three-winding transformer. For example, Fig. 1.5 shows the equivalent circuit of a three-winding transformer that can be transformed into two two-winding transformers’ equivalent circuits.

1.2 Basic Concepts

11

1 : Ki k

Fig. 1.5 Three-winding transformer equivalent circuit

k zkh zih

i

h zjh

j 1 : Ki j

After obtaining the transformer equivalent circuit, we can establish the equivalent circuit for a multivoltage network. For example, an electric network shown in Fig. 1.6 can be represented by the equivalent circuit shown in Fig. 1.6b or c when the leakage impedances of transformer T1 and T2 have been normalized to side and side . It can be proved that the two representations have an identical ultimate equivalent circuit as shown in the Fig. 1.6d. When we analysis the operation of a power system, the per-unit system is extensively used. In this situation, all the parameters of an electric network are denoted in the per-unit system. For example, in the Fig. 1.6, if the voltage base at side is Vj1 , at sides and is Vj2 and at side is Vj4 , then the base ratio (nominal turns ratio) of transformer T1 and T2 are Kj1 ¼

Vj2 Vj2 ; Kj2 ¼ : Vj1 Vj4

ð1:21Þ

The ratios of transformer T1 and T2 on a per-unit base (off-nominal turns ratio) are K1 ¼

K1 ; Kj1

K2 ¼

K2 : Kj2

ð1:22Þ

Therefore, the ratio of the transformer should be K1 or K2 when its equivalent circuit is expressed in a per-unit system. In modern power systems, especially in the circumstances of deregulation, the power flow often needs to be controlled. Therefore the application of the phaseshifting transformer is increasing. As we know, a transformer just transforms the voltages of its two terminals and its turn ratio is a real number. The phase-shifting transformer can also change the phase angle between voltages of its two terminals. Thus its turn ratio is a complex number. When the exciting current is neglected or treated as a load (or an impedance), a phase-shifting transformer can be represented

12

1 Mathematical Model and Solution of Electric Network

T1

1

T2

l

1 : K1

2

K2 : 1

3

4

a zT1 1 : K1

K2 : 1

zl

zT2

yl 2

yl 2

b zT2

zl

zT1 1 :1 K1

yl 2

yl 2

1:

1 K2

c zl

K1zT1 K1zT1 K1−1

2

K 1 zT1 1−K1

yl 2

K2zT2 yl 2

2

K 2 z21 1−K2

K 2zT2 K2−1

d Fig. 1.6 Equivalent circuit of a multivoltage electric network

by its leakage impedance, which is connected in series with an ideal transformer having a complex turns ratio as shown in Fig. 1.7. From this figure, we can obtain the equations as follows, V_ i I_i zT ¼ V_ j0 I_i þ I_0 ¼ 0:

ð1:23Þ

j

Apparently, the two terminal voltages are related by _ V_ j0 ¼ V_ j =K: Since there is no power loss in an ideal autotransformer, V_ j0 I^j0 ¼ V_ j I^j ;

ð1:24Þ

1.3 Nodal Admittance Matrix

13

Fig. 1.7 Phase-shifting transformer representation

•

1:K •

zT V ′ j

i

Vj

•

•

Vi

•

•

Ii

Ij

j

•

Ij

where I^j0 and I^j are the conjugates of I^j0 and I^j , respectively. It follows from the above equations that ^_ I_ : I_j0 ¼ K j

ð1:25Þ

Substituting (1.24) and (1.25) into (1.23) V_ j V_i I_i ¼ ¼ Yii V_ i þ Yij V_ j _ T zT Kz V_ j V_ i I_j ¼ þ 2 ¼ Yji V_ i þ Yjj V_ j ; ^ T K zT Kz

ð1:26Þ

where Yii ¼

1 ; zT

Yij ¼

1 ; _ KzT

Yji ¼

1 ; ^ KzT

Yjj ¼

1 K 2 zT

:

Equation (1.26) is the mathematical model of the phase-shifting transformer. It is easy to be proved that (1.26) is the same as (1.18) when the turn ratio is a real number. This illustrates that the transformer is a particular case of the phaseshifting transformer. Because the ratio of a phase-shifting transformer is complex number, and Yij 6¼ Yji , it has no equivalent circuit and the admittance matrix of the electric network with the phase-shifting transformer is not symmetric.

1.3 1.3.1

Nodal Admittance Matrix Basic Concept of Nodal Admittance Matrix

As mentioned above, the node equation (1.3) is usually adopted in modern power system analysis. If the number of nodes in a network is n, we have the following general simultaneous equations:

14

1 Mathematical Model and Solution of Electric Network

9 I_1 ¼ Y11 V_ 1 þ Y12 V_ 1 þ þ Y1i V_ i þ þ Y1n V_ n > > > > _I2 ¼ Y21 V_ 1 þ Y22 V_ 2 þ þ Y2i V_ i þ þ Y2n V_ n > > > > > > > .. > = . : I_i ¼ Yi1 V_ 1 þ Yi2 V_ 2 þ þ Yii V_ i þ þ Yin V_n > > > > > > .. > > > . > > > ; I_n ¼ Yn1 V_ 1 þ Yn2 V_ 2 þ þ Yni V_ i þ þ Ynn V_ n

ð1:27Þ

The matrix constituted by the coefficients of (1.27) is the nodal admittance matrix 2Y

11

6 Y21 6 . 6 . 6 . Y¼6 6 Yi1 6 . 4 . . Yn1

Y12 Y22 .. . Yi2 .. .

Yn2

Y1i Y2i .. . Yii .. .

Yni

Yan 3 Y2n 7 .. 7 7 . 7 7: Yin 7 .. 7 5 . Ynn

ð1:28Þ

A nodal admittance matrix reflects the topology and parameters of an electric network, so it can be regarded as a mathematical abstraction of the electric network. The node equation based on the admittance matrix is a widely used mathematical model of electric networks. Next we will introduce some physical meaning of the matrix elements. If we set a unit voltage at node i and ground other nodes, i.e., V_i ¼ 1 V_j ¼ 0

ðj ¼ 1; 2; . . . ; n; j 6¼ iÞ;

then the following relationships hold according to (1.27), Ij ¼ Yji

j ¼ 1; 2; . . . ; n:

ð1:29Þ

From (1.29) we can see the physical meaning of the ith column elements in the admittance matrix: the diagonal element Yii in the ith column, the self-admittance of node i, is equal to the injection current of the node i; the off-diagonal elements Yij in the ith column, the mutual-admittance of node i and node j, is equal to the injection current of node j in this situation. We will further illustrate these concepts by a simple network shown in Fig. 1.8. The network has three nodes (plus ground), thus the dimension of its admittance matrix is 3 3,

1.3 Nodal Admittance Matrix 2

15

1

z12

3

•

I2

z13

z10

•

I12 z10

3

•

V1 = 1 I1

z12

a 2

1

2

•

I10

•

I3

•

I13

z13

b 1 •

I2

3

I1

•

I12 z12

•

I2

•

I3

•

•

I13 = 0

c

•

•

I1

•

I12 = 0 z12

z13

z10

1

2

V3 = 1 •

I31 z10

3 •

I3

z13

d 1

3

2

z12

z20

z23

e Fig. 1.8 Construction process of admittance matrix in simple electric network

2

Y 11 Y ¼ 4 Y 21 Y 31

Y 12 Y 22 Y 32

3 Y 13 Y 23 5: Y 33

According to the above discussion, we can get the elements of the first column: Y11 ; Y21 ; and Y31 , by setting a unit voltage on node 1 and grounding node 2 and node 3 as shown in Fig. 1.8b. Evidently, 1 1 1 I_1 ¼ I_12 þ I_13 þ I_10 ¼ þ þ ¼ Y11 ; z12 z10 z13 1 I_2 ¼ I_12 ¼ ¼ Y21 ; z12 1 I_3 ¼ I_13 ¼ ¼ Y31 : z13 Similarly, setting a unit voltage at node 2 and grounding node 1 and node 3 as shown in Fig. 1.8c, we can get the elements of the second column: 1 I_1 ¼ I_21 ¼ ¼ Y12 ; z12 1 I_2 ¼ I_21 ¼ ¼ Y22 ; z12 I_3 ¼ 0 ¼ Y32 :

16

1 Mathematical Model and Solution of Electric Network

For the elements of the third column we have (see Fig. 1.8d), 1 I_1 ¼ I_31 ¼ ¼ Y13 ; z31 I_2 ¼ 0 ¼ Y23 ; 1 I_3 ¼ I_31 ¼ ¼ Y33 : z13 Finally, the admittance matrix of the above simple network becomes 2

1 1 1 þ þ 6 z12 z10 z13 6 1 6 Y¼6 6 z12 4 1 z13

1 z12 1 z12

0

3 1 z13 7 7 7 0 7: 7 1 5 z13

ð1:30Þ

If we change the node numbers in Fig. 1.8a, e.g., exchange the number ordering of node 1 with node 2, as shown in Fig. 1.8e, then the admittance matrix becomes, 2

1 6 z12 6 6 1 6 0 Y ¼ 6 6 z12 6 4 0

1 z12 1 1 1 þ þ z12 z20 z23 1 z23

3 0

7 7 1 7 7 : z23 7 7 7 1 5 z23

The above matrix can be obtained through exchanging the first row with the second row, and at the same time exchanging the first column with the second column of the matrix shown in (1.30). The exchange of the rows and columns of the admittance matrix corresponds to the exchange of the sequence of node equations and their variables. The properties of the admittance matrix can be summarized as follows: 1. The admittance matrix is symmetric if there is no phase-shifting transformer in the network. From (1.30) we have Y12 ¼ Y21 ¼

1 1 ; Y13 ¼ Y31 ¼ ; Y23 ¼ Y32 ¼ 0: z12 z13

Generally, according to the reciprocity of the network, Yij ¼ Yji : Therefore, the admittance matrix is symmetric. We will discuss the networks with phase-shifting transformers later.

1.3 Nodal Admittance Matrix

17

2. The admittance matrix is sparse. From the discussion above, we know that Yij and Yji will be zero if node i does not directly connect with node j. For example, in Fig. 1.8a, node 2 does not directly connect with node 3, so both of Y23 and Y32 are zero. In general, the number of nonzero off-diagonal elements of each row is equal to the number of branches that are incident to the corresponding node. Usually, the number of branches connected to one node is 2–4, thus there are only 2–4 nonzero off-diagonal elements in each row. The property that only a few nonzero elements exist in a matrix is called sparsity. This phenomenon will be more remarkable with increase of the power system scale. For instance, for a network with 1,000 nodes, if each node directly connects three branches on average, the total number of nonzero elements for the network is 4,000, which is only 0.4% of the total elements in the admittance matrix. The symmetry and sparsity of an admittance matrix are very important features for large-scale power systems. If we make full use of these two properties, the computation speed will be accelerated and the computer memory will be saved dramatically.

1.3.2

Formulation and Modification of Nodal Admittance Matrix

Now we discuss formulation of an admittance matrix by inspection first. When an electric network is composed of only transmission lines, the principles of constructing its admittance matrix can be summarized as follows: 1. The order of the admittance matrix is equal to the number of the nodes of the electric network. 2. The number of the nonzero off-diagonal elements in each row is equal to the number of the ungrounded branches connected to the corresponding node. 3. The diagonal elements of the admittance matrix, i.e., the self-admittance of the node, is equal to the sum of all the admittances of the incident branches of the corresponding node. Thus X Yii ¼ yij ; ð1:31Þ j2i

where yij is the reciprocal of zij , which is the branch impedance between node i and node j, ‘‘j I’’ denotes that only the incident branches of node i (including the grounding branch) are included to the summation. For example, in Fig. 1.8, the self-admittance of node 1, i.e., Y11 , should be Y11 ¼

1 1 1 þ þ ¼ y12 þ y10 þ y13 : z12 z10 z13

The self-admittance of node 2, i.e., Y22 , should be Y22 ¼

1 ¼ y12 : z12

18

1 Mathematical Model and Solution of Electric Network

4. The off-diagonal element of the admittance matrix, Yij , is equal to the negative of the admittance between node i and node j Yij ¼

1 ¼ yij : zij

ð1:32Þ

For example, in Fig. 1.8a, 1 ¼ y12 ; z12 1 ¼ ¼ y13 : z13

Y12 ¼ Y13

Therefore, no matter how complicated the configuration of an electric network is, its admittance matrix can be established directly by inspection according to the parameters and the topology of the network. When the electric network involves transformers or phase-shifting transformers, they need special treatment. When branch ij is a transformer, the admittance matrix certainly can be formed following the above steps if the transformer is substituted beforehand by the P equivalent circuit as shown in Fig. 1.4a. However, in practical application the transformer is often treated directly in forming the admittance matrix. If branch ij is a transformer, as shown in Fig. 1.4a, the elements of the admittance matrix related to the branch can be obtained as follows: 1. Add two nonzero off-diagonal elements into the admittance matrix Yij ¼ Yji ¼

yT : K

ð1:33Þ

2. Add to the self-admittance of node i by, DYii ¼

K1 1 yT þ yT ¼ yT : K K

ð1:34Þ

3. Add to the self-admittance of node j by DYjj ¼

1 1K yT yT þ yT ¼ 2 : K K2 K

ð1:35Þ

When branch ij is a phase-shifting transformer, its equivalent circuit is Fig. 1.7. Then the corresponding matrix elements are obtained as follows: 1. Add two nonzero off-diagonal elements into the admittance matrix Yij ¼

1 ; _ KzT

ð1:36Þ

1.3 Nodal Admittance Matrix

19

Yji ¼

1 : ^ KzT

ð1:37Þ

1 : zT

ð1:38Þ

2. Add to the self-admittance of node i by DYii ¼ 3. Add to the self-admittance of node j by DYjj ¼

1 K2 z

:

ð1:39Þ

T

It can be seen from (1.36) and (1.37) that Yij 6¼ Yji , thus the admittance matrix is not symmetric any more although its structure is still symmetric. Studies of different system operation states, such as transformer or transmission line outages, play an important part in modern power system analysis. Because the outage of branch ij only affects the self and mutual admittance of node i and node j, we can obtain the new admittance matrix for the contingency state by modifying the original admittance matrix. The modification methods for different situations are introduced as follows: 1. To add a new node with a new branch for the original network as shown in Fig. 1.9a. Assume that i is a node of the original network and j is the new node; zij is the impedance of the new branch. The dimension of the admittance matrix becomes N þ 1 because of the new node. There is only one branch connected to node j, therefore, its self-admittance is, 1 ; zij The self-admittance of node i should be modified (added) by, Yjj ¼

DYii ¼

1 : zij

zij i

N

j

i

i

N

N

zij

j

a

b

i

N

−zij

j

c

Fig. 1.9 Four cases of modifying the electric network

j

d

−zij z′ij

20

1 Mathematical Model and Solution of Electric Network

Two off-diagonal elements should also be created 1 : zij 2. To add a new branch between node i and node j as shown in Fig. 1.9b. In this case, no new node is introduced and the dimension of the new admittance matrix is the same as the original one, while the following modifications should be made. 9 1 > > DYii ¼ > > zij > > > = 1 DYjj ¼ : ð1:40Þ zij > > > > 1> > DYij ¼ DYji ¼ > ; zij Yij ¼ Yji ¼

3. To remove a branch with impedance zij between node i and node j. In this case, it is equivalent to adding a new branch of impedance zij between node i and node j as shown in Fig. 1.9c. Therefore, the modifications of the admittance matrix are as follows: 9 > > > > > > > = : > > > > 1> > DYij ¼ DYji ¼ > ; zij 1 zij 1 DYjj ¼ zij DYii ¼

ð1:41Þ

4. To change branch impedance zij for z0ij . This case is equivalent to removing branch impedance zij first and then adding a branch of impedance z0ij between node i and node j as shown in Fig. 1.9d. Thus the modifications can be carried out according to (1.40) and (1.41). It should be noted that the above discussion is based on the assumption that the added or removed branch is a pure impedance branch. If the branch is a transformer or a phase-shifting transformer, the modifications should be carried out according to (1.33)–(1.35) or (1.36)–(1.39). [Example 1.1] Figure 1.10 shows an equivalent circuit of a simple electric network with two transformers. The branch impedance and grounding admittance in per unit are shown in the figure. Determine the nodal admittance matrix for the electric network. [Solution] According to the method introduced in Sect. 1.2.2, we can assemble the elements of the admittance matrix node by node.

1.3 Nodal Admittance Matrix

21

1:1.05

0.08 + j0.30

j 0.015

j0.25

1.05:1 j0.25

+ 04 0.

j0 .2 5

j0.25

1

0.

+

j0.03

5

.3

j0

j0.25

Fig. 1.10 Equivalent circuit for Example 1.1

In Fig. 1.10, parameters are in admittance for grounding branches and in impedance for other branches (branches in series connection). Using (1.31), we obtain the self-admittance of node 1 as follows: Y11 ¼ y10 þ y12 þ y13 ¼ j0:25 þ

1 1 þ 0:04 þ j0:25 0:1 þ j0:35

¼ 1:378742 j6:291665: The mutual admittances related to node 1 can be obtained according to (1.32), 1 ¼ 0:624025 þ j3:900156 0:04 þ j0:25 1 ¼ ¼ 0:754717 þ j2:641509: 0:1 þ j0:35

Y21 ¼ Y12 ¼ y12 ¼ Y31 ¼ Y13 ¼ y13

Because branch 2–4 is a transformer, the self-admittance of node 2 should be calculated according to (1.31) and (1.35) based on the equivalent circuit as shown in Fig. 1.4a Y22 ¼ y20 þ y12 þ y23 þ

y42 2 K42

1 1 1 1 þ þ 0:04 þ j0:25 0:08 þ j0:30 j0:015 1:052 ¼ 1:453909 j66:98082: ¼ ðj0:25 þ j0:25Þ þ

The mutual admittances related to node 2 are Y23 ¼ Y32 ¼

1 ¼ 0:829876 þ j3:112033: 0:08 þ j0:30

Using (1.33) we have Y24 ¼ Y42 ¼

y42 1 1 ¼ ¼ j63:49206: K42 j0:015 1:05

22

1 Mathematical Model and Solution of Electric Network

The other elements of the admittance matrix can be calculated in a similar way. The ultimate result is 2

1:378742 6 j6:291665 6 6 6 6 0:24024 6 6 þj3:900156 6 6 6 6 0:754717 Y¼6 6 þj2:641509 6 6 6 6 6 6 6 6 6 4

3

0:924024 0:754717 þj3:900156 þj2:641509 1:453909 j66:98082

0:829876 0:000000 þj3:112033 þj63:19206

0:929876 1:584596 þj3:112033 j35:73786 0:000000 þj63:49206

0:000000 j66:66667

0:000000 þj31:74603

7 7 7 7 7 7 7 7 7 7 0:000000 7 7; þj31:74603 7 7 7 7 7 7 7 7 7 7 0:000000 5 j33:33333

where the vacancies are zero elements.

1.4 1.4.1

Solution to Electric Network Equations Gauss Elimination Method

At present, Gauss elimination is the most popular method to solve the electric network equations. In the initial stage of computer application in power systems, iterative methods were also been used because of the limitation of computer memory. The fatal disadvantage of the iterative methods is the convergence problem. Therefore, the Gauss elimination method almost has substituted for iterative methods after successful application of the sparse techniques [1, 2]. The Gauss elimination method is introduced in this section, and the sparse technique and sparse vector method will be described successively. The Gauss elimination method in solving simultaneous linear equations consists of two steps, i.e., forward elimination and back substitution. Both forward elimination and back substitution can be carried out by either row or column orientation. Generally, the column-oriented forward elimination and row-oriented back substitution scheme are widely used. The related algorithm is introduced next, and other algorithms can be easily deduced similarly. A system of n simultaneous linear equations may be written in the matrix form as AX ¼ B in which elements in matrix A and vector B can be either real or complex numbers. For example, the coefficient matrix of (1.3) is complex, while that of the correction equation in the Newton–Raphson method (see (2.40) in Chap. 2) is real.

1.4 Solution to Electric Network Equations

23

Because the forward eliminations involve manipulations with matrix A and B, a n ðn þ 1Þ augmented matrix is formed by appending B as the ðn þ 1Þth column of A, 2

A ¼ ½ A

a11 6 a21 B ¼ 6 6 6 6 an1

a12 a22 an2

a1n a2n ann

3 2 a11 b1 6 a21 b2 7 7¼6 6 7 7 6 bn 7 6 an1

a12 a22 an2

a1n a2n ann

3 a1;nþ1 a2;nþ1 7 7: 7 7 an;nþ1 7

In the above equation, bj is substituted by aj;nþ1 ðj ¼ 1; 2; . . . ; nÞ to simplify the following representation. The process of the column-oriented forward eliminations is introduced first.

Step 1.

Eliminate the first column

First, normalize the first row of the augmented matrix A, 1

ð1Þ

ð1Þ

...

ð1Þ

a1;nþ1 ;

ð1:42Þ

a12

a13

a1j a11

ðj ¼ 2; 3; . . . ; n þ 1Þ:

where ð1Þ

a1j ¼

Then the derived row as shown in (1.42) is used to eliminate the elements and the remaining elements of the second to the nth row a21 ; a31 ; . . . ; an1 of A, can be calculated by ð1Þ

ð1Þ

aij ¼ aij ai1 a1j

ðj ¼ 2; 3; . . . ; n þ 1Þ; ði ¼ 2; 3; . . . ; nÞ;

where the superscript (1) denotes that the relative element is the result of the first manipulation. At this stage, matrix A is changed into A1 , 2 A1 ¼ ½ A1

6 6 6 B1 ¼ 6 6 6 4

1

ð1Þ

a12

ð1Þ

ð1Þ

.. .

a22 .. .

an2

ð1Þ

a1n

ð1Þ

a2n .. .

ð1Þ

ann

ð1Þ

a1;nþ1

3

7 ð1Þ a2;nþ1 7 7 7 .. 7: . 7 5 ð1Þ an;nþ1

The corresponding equation is A1 X ¼ B1 which has the same solution as the original equation. In the above matrix, the vacancies are zero elements.

24

1 Mathematical Model and Solution of Electric Network

Step 2.

Eliminate the second column

Normalize the second row of the augmented matrix A as the following ð2Þ

1 a23

0

ð2Þ

a2;nþ1 ;

...

ð1:43Þ

where ð2Þ

ð1Þ

ð1Þ

a2j ¼ a2j =a22

ðj ¼ 3; 4; . . . ; n þ 1Þ:

Then the derived row shown in (1.43) is used to eliminate the elements ð1Þ ð1Þ ð1Þ a32 ; a42 ; . . . ; a4n of A1 and the remaining elements of the third to the nth row can be calculated by, ð2Þ

ð1Þ

ð1Þ ð2Þ

aij ¼ aij ai2 a2j

ðj ¼ 3; 4; . . . ; n þ 1Þ; ði ¼ 3; 4; . . . ; nÞ;

where the superscript (2) denotes that the relative element is the result of the second manipulation. Now, matrix A1 has been transformed into A2 , 2

A2 ¼ ½ A2

ð1Þ

1 a12 6 1 6 6 6 B2 ¼ 6 6 4

ð1Þ

a13

ð2Þ

ð2Þ

a23

a33 ð2Þ an3

ð1Þ

a1n

ð2Þ

a2n

ð2Þ

a3n

að2Þ nn

ð1Þ

a1;nþ1

3

7 ð2Þ a2;nþ1 7 7 7: pgð2Þ a3;nþ1 7 7 5 ð2Þ an;nþ1

Generally, the following computation should be executed when eliminating the kth column ðkÞ

ðk1Þ

akj ¼ akj ðkÞ

ðk1Þ

aij ¼ aij

ðk1Þ ðkÞ akj

aik

ðk1Þ

=akk

ðj ¼ k þ 1; . . . ; n þ 1Þ;

ð1:44Þ

ðj ¼ k þ 1; . . . ; n þ 1Þ; ði ¼ k þ 1; . . . ; nÞ:

ð1:45Þ

After proceeding with the elimination n times in this manner, the elements below the diagonal of the matrix become zero, and the nth derived augmented matrix is obtained. 2

An ¼ ½ An

6 6 6 6 Bn ¼ 6 6 6 6 4

1 a12

ð1Þ

a13

ð1Þ

...

a1n

1

a23 1

ð2Þ

... ...

a2n

..

.

ð1Þ

ð1Þ

a1;nþ1

3

ð3Þ

7 ð2Þ a2;nþ1 7 7 7 ð3Þ a3;nþ1 7 7: .. 7 7 . 5

1

an;nþ1

ð2Þ

a3n .. .

ðnÞ

ð1:46Þ

1.4 Solution to Electric Network Equations

25

The corresponding equation becomes An X ¼ Bn , that is x1 þ

ð1Þ

ð1Þ

...

þa1n xn ¼

a23 x3 þ

ð2Þ

...

þa2n xn ¼

x3

... .. .

a12 x2 þ

a13 x3 þ

x2

þ

þ

ð1Þ

a1;nþ1

ð1Þ

ð2Þ

a2;nþ1

þa3n xn ¼ .. .

ð3Þ

a3;nþ1 .. .

xn ¼

an;nþ1

ð2Þ ð3Þ

ð1:47Þ

ðnÞ

Its solution is the same as the original equation AX ¼ B. For (1.47), back substitution is carried out in a bottom-up sequence. The value of xn is obtained directly from the nth equation, ðnÞ

xn ¼ an;nþ1 : Then substituting xn into the ðn 1Þth equation we get the solution of xn1 , ðn1Þ

ðn1Þ

xn1 ¼ an1;nþ1 an1;n xn : Substituting xn1 and xn into the ðn 2Þth equation, we obtain xn2 . Generally, xi can be obtained by substituting the solved variables xiþ1 ; xiþ2 ; . . . ; xn into the ith equation, ðiÞ

xi ¼ ai;nþ1

n X j¼iþ1

ðiÞ

aij xj

ði ¼ n; . . . ; 2; 1Þ:

ð1:48Þ

This is the general equation of the row-oriented back substitution. [Example 1.2] Solve the following simultaneous linear equations by using the Gauss elimination method. x1 þ 2x2 þ x3 þ x4 ¼ 5 2x1 þ x2 ¼ 3 x1 þ x3 ¼ 2 x1 þ x4 ¼ 2 [Solution] Write the augmented below. 2 6 ð1Þ 6 6 2 6 6 4 1 1

:

matrix according to the original equations as

2

1

1

1

0

0

0

1

0

0

0

1

.. . .. . .. . .. .

3 57 7 37 7: 7 25 2

26

1 Mathematical Model and Solution of Electric Network

As an initial step, normalize the first row of the augmented matrix according to (1.44), i.e., divide the first row by its diagonal element. 2 6 1 6 6 ð2Þ 6 6 6 ð1Þ 4

2

1

1

1

0

0

0

1

0

ð1Þ

0

0

1

.. . .. . .. . .. .

3 57 7 37 7: 7 27 5 2

Then eliminate the first column according to (1.45) 2 61 6 6 6 6 4

2

1

ð3Þ

2

2

0

2

1

3 .. . 57 . 7 2 ..7 7 : .. 7 7 1 .3 5 . 0 ..3 1

The next step is the elimination of the second column. When normalizing the second row, we divide the elements in the second row by the diagonal element –3 2 61 6 6 6 6 6 4

2

1

1

1

2 3

2 3

ð2Þ

0

1

ð2Þ 1

0

3 .. . 57 .. 7 7 . 37 7: 7 .. .3 7 5 .. .3

Then eliminate the second column in terms of (1.45) to obtain 2 61 6 6 6 6 6 4

3 . 1 .. 5 7 .. 7 7 2 1 23 . 37 3 7 4 1 .. 5 7: . 37 3 3 5 .. 5 1 4 .

2

1

3

3

3

Repeat the procedure for the third column. Normalize the third row through dividing the third row by the diagonal element 4/3. 2 61 6 6 6 6 6 4

3 .. 2 1 1 . 57 .. 7 7 2 1 23 . 37 3 7 .. 5 7: 1 1 4 . 47 5 1 4 .. 5 . 3 3 3

1.4 Solution to Electric Network Equations

27

Then eliminate the third column in terms of (1.45) to obtain 2 61 6 6 6 6 6 4

3 .. . 57 .. 7 7 2 . 37 3 7 .. 5 7: 1 . 47 4 5 5 .. 5 .4 4

2 1 1

1

2 3

1

The last step is normalizing the fourth row according to (1.44), that is, dividing the fourth row by the diagonal element 5/4. 2 61 6 6 6 6 6 4

2

1

1

1

2 3

2 3

1

1 4

1

3 .. . 57 .. 7 7 . 37 7 .. 5 7: . 47 5 .. .1

The transformed equations after elimination become x1 þ

2x2 þ x2 þ

x3 þ 2 3x 3 þ x3 þ

x4 ¼ 2 3x 4 ¼ 1 4x 4 ¼

5 7 3 5 4

x4 ¼

:

1

x4 ; x3 ; x2 ; x1 can be obtained through the back substitution according to (1.48). x4 ¼ 1 x3 ¼ 54 14x4 ¼ 1 x2 ¼ 73 23x3 23x4 ¼ 1 x1 ¼ 5 2x2 x3 x4 ¼ 1

1.4.2

:

Triangular Decomposition and Factor Table

In practical applications, the simultaneous equations often need to be solved repeatedly when only right-hand vector B changes while coefficient matrix A is a constant matrix. In such cases, the factor table method is often used to improve computation efficiency. The factor table records all the operations on right-hand vector B in the Gauss elimination process. As the discussion above, The Gauss elimination method involves forward elimination and back substitution. Back substitution is determined

28

1 Mathematical Model and Solution of Electric Network

by the upper triangular elements of the coefficient matrix after elimination operation as shown (1.46). In order to execute the elimination operation (forward elimination), the relative operation factors also need to be recorded in the elimination process. The forward elimination includes normalization and elimination operation. Take column-oriented elimination as an example, operations on the i th element of B (i.e., bi;nþ1 ) in the forward elimination are as follows (see (1.44) and (1.45)), ðiÞ

ði1Þ

ðkÞ

ðk1Þ

bi ¼ bi

bi ¼ bi

ði1Þ

=aii

ði ¼ 1; 2; . . . ; nÞ;

ðk1Þ ðkÞ bk

aik

ð1Þ

ð1:49Þ

ðk ¼ 1; 2; . . . ; i 1Þ:

ð1Þ

ði2Þ

ð1:50Þ

ði1Þ

The above operation factors ai1 ; ai2 ; ai2 ; . . . ; ai;i1 and aii are to be stored in the lower triangular matrix row by row and appended to the upper triangular elements of the (1.46). Thus, we obtain the factor table as the following a11

a12

ð1Þ

a13

ð1Þ

a14

ð1Þ

a1n

ð1Þ

a21

a22

ð1Þ

a23

ð2Þ

a24

ð2Þ

a2n

a31

a32

ð1Þ

a33

ð2Þ

a34

ð3Þ

a3n

a41 .. .

a42 .. .

ð1Þ

a43 .. .

ð2Þ

a44 .. .

ð3Þ

.. .

a4n .. .

an1

an2

ð1Þ

an3

ð2Þ

an4

ð3Þ

ð2Þ ð3Þ ð4Þ

:

ðn1Þ

ann

Where the lower triangular elements are used in elimination operations on B and the upper triangular elements are used in back substitution operations. The factor table also can be denoted in the following format d11 l21 l31 l41 .. . ln1

u12 d22 l32 l42 .. . ln2

u13 u23 d33 l43 .. . ln3

u14 u24 u34 d44 .. . ln4

.. .

where ði1Þ

;

ðiÞ

ði < jÞ;

dii ¼ aii uij ¼ aij lij ¼

ðj1Þ aij

ðj < iÞ:

u1n u2n u3n u4n ; .. . dnn

ð1:51Þ

1.4 Solution to Electric Network Equations

29

We can see that the lower triangular elements of the factor table are exactly the operation elements used in the elimination process. Therefore, if we retain them in the original position and take the reciprocals of the diagonal elements, the lower triangular elements of the factor table can be readily obtained. The upper triangular elements of the factor table are just the upper triangular part of the coefficient matrix after the elimination operations. If the simultaneous equations need to be solved repeatedly for different righthand vector B, we should first carry out the elimination operation on coefficient matrix A to obtain its factor table. Then the factor table can be used directly and repeatedly to solve the equations with different B. In this situation, we will carry out the elimination operation on the following equations instead of (1.49) and (1.50), ðiÞ

ði1Þ

=dii ;

ð1:52Þ

ðkÞ

ði ¼ k þ 1; . . . ; nÞ:

ð1:53Þ

bi ¼ bi ðkÞ

ðk1Þ

bi ¼ bi

lik bk

The back substitution will be carried out on the following equations instead of (1.48) xn ¼ bðnÞ n ; ðiÞ

x i ¼ bi

n X

uij xj :

ð1:54Þ

j¼iþ1

[Example 1.3] For the simultaneous linear equations of Example 1.2, find the factor table of its coefficient matrix A and solve the equation when B ¼ ½ 1 1 2 0 T . [Solution] Inspecting the solution process of Example 1.2, we can directly obtain the factor table of coefficient matrix A, 1 2 2 3 1 2 1 2

1

1

2 3 4 3 1 3

2 3 1 4 5 4

d11 l21 , l31 l41

u12 d22 l32 l42

u13 u23 d33 l43

u14 u24 : u34 d44

The lower triangular elements of the above factor table are just the operation factors in brackets which appeared in the elimination process, and the upper triangular elements are the upper triangular part of the coefficient matrix after elimination operation. Now we first use the lower triangular elements of the factor table to operate column-oriented elimination on B. Normalize b1 according to (1.52),

30

1 Mathematical Model and Solution of Electric Network ð1Þ

b1 ¼ b1 =d11 ¼ ð1Þ=1 ¼ 1: Then operations on b2 ; b3 ; b4 are carried out by using the elements of the factor table’s first column in the lower triangular part according to (1.53) ð1Þ

ð1Þ

ð1Þ

ð1Þ

ð1Þ

ð1Þ

b2 ¼ b2 l21 b1 ¼ 1 2 ð1Þ ¼ 3; b3 ¼ b3 l31 b1 ¼ 2 1 ð1Þ ¼ 3; b4 ¼ b4 l41 b1 ¼ 0 1 ð1Þ ¼ 1: Thus the elimination operation of the first column is completed, and we have, Bð1Þ ¼ ½ 1

3

3 1 T :

ð1Þ

Next, normalize b2 according to (1.52), ð2Þ

ð1Þ

b2 ¼ b2 =d22 ¼ 3=ð3Þ ¼ 1: ð1Þ

ð1Þ

The elimination operation on b3 ; b4 is followed by using the elements of the second column in the lower triangular part according to (1.53), ð2Þ

ð1Þ

ð2Þ

ð2Þ

ð1Þ

ð2Þ

b3 ¼ b3 l32 b2 ¼ 3 ð2Þ ð1Þ ¼ 1; b4 ¼ b4 l42 b2 ¼ 1 ð2Þ ð1Þ ¼ 1: Thus the elimination operation of the second column is finished, and we have Bð2Þ ¼ ½ 1 1

1 1 T :

ð2Þ

ð3Þ

Normalize b3 according to (1.52) and operate b4 according to (1.53) ð3Þ

ð2Þ

b3 ¼ b3 =d33 ¼ 1=43 ¼ 34: ð2Þ

Again, the elimination operation on b4 is followed by using the elements of the third column in the lower triangular part according to (1.53) ð3Þ

ð2Þ

ð3Þ

b4 ¼ b4 l43 b3 ¼ 1 13 34 ¼ 54: Thus the elimination operation on the third column is finished, and we have Bð3Þ ¼ 1

1

3 4

54

T

:

1.4 Solution to Electric Network Equations

31 ð3Þ

The last step of the elimination operation is to normalize b4 according to (1.52) ð4Þ

ð3Þ

b4 ¼ b4 =d44 ¼ 45=

4 5

¼ 1:

Now, all the elimination operations are fulfilled. Bð4Þ ¼ 1 1

3 4

1

T

:

Comparing with the factor table, we obtain the following identical solution equations x1 þ 2x2 þ x2 þ

x3 2 3x3 x3

þ þ þ

x4 ¼ 1 2 x ¼ 1 4 3 : 1 3 4x 4 ¼ 4 x4 ¼ 1

Now, the unknowns could be solved using the upper triangular part of the factor table according to (1.54). ð4Þ

x4 ¼ b4 ¼ 1 ð3Þ

x3 ¼ b3 u34 x4 ¼ 34 14 ð1Þ ¼ 1 ð2Þ

x2 ¼ b2 u23 x3 u24 x4 ¼ 1 23 1 23 ð1Þ ¼ 1 ð1Þ

x1 ¼ b1 u12 x2 u13 x3 u14 x4 ¼ 1 2 ðÞ 1 1 1 ð1Þ ¼ 1: It should be pointed out that the factor table as shown in (1.50) can be established not only by the Gauss elimination method but also by the triangular decomposition method. From the above example, we can verify that the following relationship between the factor table and its coefficient matrix holds, A ¼ L0 U;

ð1:55Þ

where 2

1 6 2 L0 ¼ 6 41 1

3 0 0 0 3 0 0 7 7 2 43 0 5 2

1 3

5 4

2

1 60 U¼6 40 0

2 1 0 0

1 1 2 3

2 3 1 4

1 0 1

3 7 7: 5

L0 can be decomposed further, L0 ¼ LD:

ð1:56Þ

32

1 Mathematical Model and Solution of Electric Network

In the above example, L can be obtained through dividing off-diagonal elements in each column of L0 by the corresponding diagonal element, 2

1 62 L¼6 41 1

0 1 2 3 2 3

3 0 0 0 07 7 1 05 1 1 4

2

1 0 6 0 3 D¼6 40 0 0 0

3 0 0 0 07 7: 4 05 3 0 54

Therefore the original coefficient matrix can be generally represented as follows A ¼ LDU:

ð1:57Þ

From the example, we can also see the following relationship LT ¼ U or U ¼ LT :

ð1:58Þ

This phenomenon is not specific to this example. The relationship in (1.58) can be proved when the coefficient matrix is symmetric. In the following, we deduce the recursion formulae of the triangular decomposition. Expand (1.55) 2

a11 6 a21 6 6 a31 6 6 .. 4 . an1

a12 a22 a32 .. . an2

a13 a23 a33 .. .

an3

.. .

3 2 l0 11 a1n 6 l0 a2n 7 21 7 6 6 l0 a3n 7 7¼6 6 31 .. 7 6 .. 5 4 . . ann l0n1 2

3 l022 l032 .. . 0 ln2

1 u12 6 1 6 6 6 6 4

l033 .. .

l0n3 u13 u23 1

7 7 7 7 7 7 5

..

.

l0nn

.. .

3 u1n u2n 7 7 u3n 7 7: .. 7 . 5 1

ð1:59Þ

Comparing two sides of the above equation, the diagonal element of the first row can be found l011 ¼ a11 : Comparing the first element of the second row and the first two elements of the second column in both sides, we can obtain l021 ¼ a21 ; l011 u12 ¼ a12 ; l021 u12 þ l022 ¼ a22 :

1.4 Solution to Electric Network Equations

33

Hence the recursion formulae are l021 ¼ a21 ; u12 ¼ a12 =l011 ; l022 ¼ a12 l021 u12 : The following decomposition equation can be obtained

a11 a21

a12 a22

¼

l011 l021

l022

u12 : 1

1

Similarly, if the first k 1 rows of L0 and the first k 1 columns of U have been obtained, the equation becomes 2

a11 a21 a31

6 6 6 6 6 4 ak1;1 2 6 6 6 6 ¼6 6 6 4

a12 a22 a32 .. .

ak1;2 l011 l021 l031

a13 a23 a33 .. .

.. .

ak1;3

3

a1;k1 a2;k1 a3;k1 .. .

7 7 7 7 7 5

ak1;k1

.. .

l022 l032 .. .

l033 .. .

l0k1;1

l0k1;2

l0k1;3

3

..

.

l0k1;k1

2 1 7 6 7 6 7 6 7 6 76 7 6 7 4 5

u12 1

u13 u23 1

.. .

3 u1;k1 u2;k1 7 7 u3;k1 7 7: 7 .. 7 . 5 1

All the elements of the two matrices in the right hand of the above equation have been solved. Comparing the first k 1 elements in the kth row and the first k elements in the kth column of the two sides element by element, we can get the corresponding elements by the following formulae 1 uik ¼ 0 lii

aik

l0kj ¼ akj

i1 X p¼1

j1 X p¼1

! l0ip upk

l0kp upj

ði ¼ 1; 2; . . . ; k 1Þ ð1:60Þ

ðj ¼ 1; 2; . . . ; kÞ:

The above are recursion formulae. Taking k from 1 to n in sequence, the triangular decomposition, A ¼ L0 U, will be achieved by using these formulae. Furthermore, dividing the off-diagonal elements by the corresponding diagonal element, L can be obtained: 1 lkj ¼ ljj

akj

j1 X p¼1

! l0kp upj

ðk ¼ j þ 1; . . . ; nÞ:

ð1:61Þ

34

1 Mathematical Model and Solution of Electric Network

The diagonal elements of L0 constitute D, i.e., dii ¼ l0ii ði ¼ 1; 2; . . . ; nÞ. Now, the coefficient matrix is decomposed into the format A ¼ LDU. It should be particularly noted that (1.58) will always be true if the coefficient matrix is symmetric.

1.4.3

Sparse Techniques

From the discussion of the above section, we know that the solution process of the electric network equation is the process of operating the right-hand constant vector successively using the elements of its factor table. In Example 1.3, there are 16 elements in its factor table: four diagonal elements, six lower triangular elements, and six upper triangular elements. Therefore the solution involves 16 multiplication operations. According to (1.53) and (1.54), if elements in the factor table are zero, the corresponding multiplication operations can be avoided (since the product will be zero) and significant computational effort can be saved. Based on this idea, socalled sparse technique is widely used in power system analysis to improve solution efficiency. The concept of the sparse technique is illustrated by an example in the following. [Example 1.4] Solve the simultaneous linear equations in Example 1.2 by using the sparse method. [Solution] In Example 1.2, the simultaneous linear equations are x1 2x1 x1 x1

2x2 þx2

x3 þx3

þx4 þx4

¼5 ¼3 : ¼2 ¼2

ð1:62Þ

In order to make full use of the sparsity advantages of the equations, the following transformation should be made first, x1 ¼ y4 ; x2 ¼ y2 ; x3 ¼ y3 ; x4 ¼ y1 :

ð1:63Þ

Then, the original equations are transformed into y1 y1

y2 þ2y2

y3 þy3

þy4 þ2y4 þy4 þy4

¼2 ¼3 : ¼2 ¼5

ð1:64Þ

We will solve the equations by using its factor table. The coefficient matrix is 2

ð1Þ 6 0 6 4 0 ð1Þ

3 0 0 1 1 0 27 7: 0 1 15 2 1 1

1.4 Solution to Electric Network Equations

35

First, we normalize the first row and eliminate the first column. There are only two operations: one normalization operation and one elimination operation in this step. The elements in brackets are the computing factors. For a 4 4 coefficient matrix, the elimination of the first column should include one normalization operation and three elimination operations. However, because both a21 and a31 are zero, two corresponding operations are avoided. After the above operations, we obtain 2

1 0 6 0 ð1Þ 6 40 0 0 ð2Þ

0 0 1 1

3 1 27 7: 15 0

The next step is the normalization of the second row and elimination of the second column. There are also only two operations, one normalization operation and one elimination operation in this step. The figures in the brackets of the above matrix are the computing factors. For a 4 4 coefficient matrix, the elimination of the second column should include one normalization operation and two elimination operations. ð1Þ Because a32 is zero, the corresponding operation is avoided. After these operations, we obtain 2

1 60 6 40 0

0 0 1 0 0 ð1Þ 0 ð1Þ

3 1 2 7 7: 1 5 4

To normalize the third row and eliminate the third column, we also need two operations, one normalization operation and one elimination operation. The computing factors are the elements in the brackets of the above matrix. After these operations, we obtain 2

1 60 6 40 0

0 1 0 0

0 0 1 0

3 1 2 7 7: 1 5 ð5Þ

Here, the factor table of the coefficient matrix can be readily written, 1 0 0 1

0 1 0 2

0 1 0 2 : 1 1 1 5

The above factor table can also be found using (1.60) and (1.61). Because there are only six zero off-diagonal elements in the above factor table, six multiply–add

36

1 Mathematical Model and Solution of Electric Network

operations are avoided. In the following, we will use this factor table to obtain the solution to the constant vector: B ¼ ½2 3

2

5 T :

First, eliminating B column by column is executed by using the lower triangular part of the factor table. According to (1.52), b1 is normalized, ð1Þ

b1 ¼ b1 =d11 ¼ 2=1 ¼ 2: Then the operations on b2 ; b3 ; b4 are continued by using the elements of the first column in the lower triangular part according to (1.53). Because l21 and l31 are zero, we have ð1Þ

ð1Þ

ð1Þ

ð1Þ

b2 ¼ b2 l21 b1 ¼ b2 ¼ 3; b3 ¼ b3 l31 b1 ¼ b3 ¼ 2: The above two steps should be avoided and only the following operation is needed ð1Þ

ð1Þ

b4 ¼ b4 l41 b1 ¼ 5 1 2 ¼ 3: After the elimination operation of the first column, we obtain Bð1Þ ¼ ½ 2

3

2

3 T :

ð1Þ

Then normalize b2 according to (1.52) ð2Þ

ð1Þ

b2 ¼ b2 =d22 ¼ 3=1 ¼ 3: ð1Þ

ð1Þ

Now, the operation on b3 ; b4 should use the elements of the second column in the lower triangular part according to (1.53). Because l32 is zero, only the operation related to l42 will be performed. Thus, ð2Þ

ð1Þ

ð2Þ

b4 ¼ b4 l42 b2 ¼ 3 2 3 ¼ 3: After finishing elimination operation of the second column, we have Bð2Þ ¼ ½ 2

3

2

3 T :

ð2Þ

Next, we normalize b3 according to (1.52) ð3Þ

ð2Þ

b3 ¼ b3 =d33 ¼ 2=1 ¼ 2:

1.4 Solution to Electric Network Equations

37

ð3Þ

And then compute b4 according to (1.53) ð3Þ

ð2Þ

ð3Þ

b4 ¼ b4 l43 b3 ¼ 3 1 2 ¼ 5: After finishing the elimination operation of the third column, we obtain Bð3Þ ¼ ½ 2

3

2

5 T : ð3Þ

The last step of the elimination operation is to normalize b4 according to (1.52) ð4Þ

ð3Þ

b4 ¼ b4 =d44 ¼ 5=ð5Þ ¼ 1: At this stage, all of the elimination operation have been completed, the right-hand vector becomes Bð4Þ ¼ ½ 2

3

2

1 T :

Comparing with the factor table, we obtain the following identical solution equations of (1.64) y1

y2

y3

þy4 ¼ þ2y4 ¼ þy4 ¼ y4 ¼

2 3 : 2 1

Now, the unknowns can be solved using the upper triangular part of the factor table according to (1.54). Because u12 ; u13 ; and u23 are zero, corresponding operations are avoided in back substitution. ð4Þ

y4 ¼ b4 ¼ 1 ð3Þ

y3 ¼ b3 u34 y4 ¼ 2 1 1 ¼ 1 ð2Þ

y2 ¼ b2 u24 y4 ¼ 3 2 1 ¼ 1

:

ð1Þ

y1 ¼ b1 u14 y4 ¼ 2 1 1 ¼ 1 Substituting the above results into (1.63), the solutions to original equation (1.62) can be obtained. From the above example, we can see that the computation effort can be saved not only in the formation of the factor table but also in the forward and back substitution. The amount of computation saved by the sparse technique depends on the number of zero elements in the factor table. Therefore, the key point of improving computation efficiency is to keep the number of zero elements in the factor table as high as possible.

38

1.4.4

1 Mathematical Model and Solution of Electric Network

Sparse Vector Method

Nowadays, the sparse matrix techniques are adopted to solve almost all large-scale power network problems. In this section, the sparse vector method, which can further improve the computation efficiency, will be introduced [3]. Sparse vector methods are useful for solving a system of simultaneous linear equations when the independent (right-hand) vector is sparse, or only few elements in the unknown vector are wanted. To take advantage of vector sparsity is relatively simple, but the results of improving computational efficiency and saving memory can be quite dramatic. Therefore sparse vector methods are often used in the compensation method, fault analysis, optimal power flow problem and contingency analysis. In principle, the sparse vector method can be applied to both full- and sparsematrix equations. This section focuses only on the implementation of sparse vector methods in the sparse-matrix situation. According to the above discussion, the admittance matrix Y of an electric network without phase-shifting transformers is symmetric. If there are phase-shifting transformers in the network the sparse admittance matrix is only symmetric in its structure. Nodal voltage equations can be written as YV ¼ I:

ð1:65Þ

For generality, we assume Y is an incidence-symmetric square matrix of order n and can be factorized as Y ¼ LDU;

ð1:66Þ

where L and U are lower and upper triangular matrices with unity diagonals, respectively, and D is a diagonal matrix. It is easy to solve the nodal equations using the above expressions. For example, the simultaneous equations can be written in the following form LDUV ¼ I:

ð1:67Þ

The above formulae can be decomposed as LX ¼ I;

ð1:68Þ

DW ¼ X;

ð1:69Þ

UV ¼ W:

ð1:70Þ

V can be obtained when (1.68)–(1.70) are solved in sequence. If Y is symmetric, matrix U is the transpose of L. If Y is incidence symmetric, matrix U is not the transpose of L, but they are identical in the sparsity structure.

1.4 Solution to Electric Network Equations

39

The forward substitution operations can be expressed as W ¼ D1 L1 I:

ð1:71Þ

The back substitution operations can be expressed as V ¼ U1 W:

ð1:72Þ

Generally, these operations can be performed either by rows or by columns. However, for the sparse vector method, the forward elimination (1.71) must be performed by columns, while the back substitution (1.72) by rows. Many different schemes can be used for storing and accessing L and U. For the sparse vector method, the lowest-numbered, nonzero, off-diagonal element in each column of L or in each row of U must be directly accessed without search. This requirement is satisfied by most storage schemes for L and U. The independent vector I is sparse in many applications. However, the solution vector V is not sparse in general. The term ‘‘sparse vector’’ in the following refers to either a sparse vector I or a subset of vector V containing the elements of interest. The exact meaning is always clear from the context. If the vector I is sparse, only a subset of the columns of L is needed for the forward elimination. This is called the fast forward (FF) process. If only certain elements of vector V are actually wanted, only a subset of the rows of U is needed for the backward substitution. This is called the fast backward (FB) process. [Example 1.5] Solve the following simultaneous linear equations V1

V1

V2 þ2V2

V3 þV3

þV4 þ2V4 þV4 þV4

¼0 ¼1 : ¼0 ¼0

[Solution] The coefficient matrix of above simultaneous linear equations is the same as in (1.64) of Example 1.4. The only difference is that the right-hand vector is sparse. I ¼ B ¼ ½0

1 0

0 T :

Therefore, the factor table of these simultaneous linear equations is the same as that of (1.64). 1 0 0 1

0 1 0 2

0 1 0 2 : 1 1 1 5

40

1 Mathematical Model and Solution of Electric Network

Decomposing the factor table, we obtain 2

1 60 L¼6 40 1

1 0 2

3

2

7 7; 5 1 1 1

6 D¼6 4

1 1 1

3

2

7 7; 5

6 U¼6 4

1

5

From (1.53), we can see that all the operations related with lik ðkÞ can be avoided if bk is equal to zero: ðkÞ

ðk1Þ

bi ¼ bi

ðkÞ

lik bk

3 0 0 1 1 0 27 7: 1 15 1

ði ¼ k þ 1; . . . ; nÞ

ði ¼ k þ 1; . . . ; nÞ:

In other words, the kth column in the lower triangular matrix can be ignored. In this example, b1 is equal to zero, so we can skip the first column of L. For this sparse vector, the elimination should begin from the second column. The elimination also includes the normalization and elimination operations. After this, the right-hand sparse vector is transformed into B0 ¼ ½ 0

1 T :

1 0

The next step is elimination of the third column. Because b03 is zero, the operations related to the third column of L are skipped, thus the elimination of the fourth column is performed directly. Here, we use d44 to normalize b04 , and the ultimate result vector after the elimination operation is B00 ¼ 0 1

0

1 T : 5

As we know, the backward substitution operations must be performed by rows. If only V3 is wanted, the operations with the first and second rows of U can be neglected. If only V2 is wanted, the operations with the first row of U can be avoided. Furthermore, the operations with the third row of U also can be omitted because b03 ¼ 0. Therefore, the back substitution is only needed to perform on the second row of U. Therefore, we have V2 ¼ b002 u24 b004 ¼ 1 2 15 ¼ 35: From the above example, we can see that the key task of sparse vector methods is to identify the active subsets of L and U for FF and FB operations. The active subset of columns for FF depends on the sparsity structure of L and I while the active subset of rows for FB depends on the sparsity structures of U and V. In order to find the active subset of FF and improve the computation efficiency, the following simple algorithm can be summarized according to the above example

1.4 Solution to Electric Network Equations

41

1. Zero all locations in I, and enter the given nonzero elements in I. 2. Search the nonzero elements in I and let k be the location number of the lowestnumbered nonzero element. 3. Perform the forward eliminations defined by column k of L on I. 4. If k ¼ n, exit. Else, return to Step 2. This algorithm ensures that only the necessary nonzero operations of FF are performed, but it is wasteful because of zeroing and searching. A similar algorithm can be used to FB, but it is even more wasteful. In the following we introduce a more efficient algorithm based on the concept of the factorization path. A factorization path for a sparse vector is represented by an ordered column list of L for FF operations. A path is executed in forward order for FF and in reverse order for FB. The same or different paths may be used for FF and FB depending on the application. The path for a singleton is basic to the path concept. A singleton is a vector with only one nonzero element. Assume that the nonzero element is in location k. The following algorithm determines the path of the singleton: 1. Let k be the first number in the path. 2. Get the number of the lowest-numbered nonzero element in column k of L (or row k of U). Replace k with this number, and list it in the path. 3. If k ¼ n, exit. Else, return to Step 2. The path for a singleton can be determined directly from the indexing arrays without searching or testing. A general sparse vector is the sum of singleton vectors, and its path is the union of the paths of its composite singleton vectors. For any sparse system, a path can be always associated with a given sparse vector. [Example 1.6] Find the factorization path of the electric network shown in Fig. 1.11. [Solution] Figure 1.12 shows the sparsity structure of the incidence symmetric admittance matrix of the network as shown in Fig. 1.11 (only the lower triangular part of the matrix is labeled). Because there are 21 branches in the network, 21 l 12

13 3

1

7

9

8

11 4

2 14

10

6

Fig. 1.11 Example electric network

15

5

42

1 Mathematical Model and Solution of Electric Network

represent the off-diagonal elements of the matrix. After triangular factorization, 10 fill-in elements (labeled as *) are added. Therefore there are altogether 31 nonzero elements in the factor table. The factorization path of any singleton can be directly obtained from the structure of the factor table. For example When k ¼ 1, the singleton path is 1 ! 2 ! 7 ! 12 ! 13 ! 14 ! 15 When k ¼ 5, the singleton path is 5 ! 11 ! 13 ! 14 ! 15 When k ¼ 6, the singleton path is 6 ! 9 ! 10 ! 12 ! 13 ! 14 ! 15 When a sparse vector is not a singleton, its path is the union of the paths of its composite singletons. For a sparse vector as follows I ¼ ½1 0

0

0

1 0

0

0

0

0 0

0

0

0 0 T

we have its path as the union of the paths of its composite singletons when k ¼ 1 and k ¼ 5, 1 ! 2 ! 7 ! 12 ! 5 ! 11 ! 13 ! 14 ! 15: In Table 1.1 we list the entire factorization paths for the network shown in Fig. 1.12. A pictorial view of the path table is provided by the path graph shown in Fig. 1.13. Utilizing this path graph, highly efficient algorithms for the sparse vector can be obtained. For example, assume the injected current at node 5 is I5 while the injected currents of other nodes are zero, and the voltage at node 1 is wanted. To do so, we carry out FF operations according to the following active column sequence: 5 ! 11 ! 13 ! 14 ! 15: And then carry out FB operations according to the following active row sequence: 15 ! 14 ! 13 ! 12 ! 7 ! 2 ! 1 In the above solution process, only the elements of five columns in lower triangular and seven rows in upper triangular elements are employed, the computation efficiency is improved dramatically. For sparse vector methods, the above path graph Table 1.1 Path table Node Next node 1 2 2 7 3 4 4 8 5 11 6 9 7 12

Node 8 9 10 11 12 13 14

Next node 10 10 12 13 13 14 15

1.4 Solution to Electric Network Equations

43

I = [1 0 0 0 1 0 0 0 0 0 0 0 0 0 1

2

3

4

5

6

7

8

0] 9

T

10

11

12

13

14

15

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Fig. 1.12 Sparse structure of a network’s factor table Fig. 1.13 The path graph

6 9

3 4

1

8

10 2 7

12 13

5 11

14 15

should be determined in advance and then be utilized directly, thus unnecessary zeroing and searching can be skipped.

1.4.5

Optimal Ordering Schemes of Electric Network Nodes

At present, the Gauss elimination method introduced in Sect. 1.3.1 is applied to solve the node equations I ¼ YV in most power system analysis programs. In order to solve the network equation repeatedly, the admittance matrix is usually

44

1 Mathematical Model and Solution of Electric Network

Fig. 1.14 Relationship between Gauss elimination and Y-D transformation

i l

1 j

factorized first, and then the factor table can be directly used to solve the equations with different right-hand vectors. As we know, the admittance matrix is sparse and the triangular matrices after factorization are also sparse. Generally, the distributions of nonzero elements in the admittance matrix are different from those in the factorized triangular matrix, because some new nonzero elements, i.e., the fill-in elements, may occur in the elimination or LU factorization process. The addition of fill-ins in the elimination process can be explained intuitively by Y-D transformation. As shown in Fig. 1.14, node l does not directly connect with nodes i and j in the initial network, thus corresponding elements Yil and Ylj in its admittance matrix are zero while Yij is nonzero. It can be proved that eliminating the first column of the admittance matrix in Gauss elimination is equivalent to eliminating node 1 by Y-D transformation as shown in Fig. 1.14. New branches connecting node pairs ij, il, and lj are created. Therefore, in the new admittance matrix, Yil ; Ylj ; and Yij are all nonzero elements, thus two fill-ins occur in eliminating the first column. Generally, eliminating node k which is the central point of a star network will create a mesh network whose vertexes are nodes connecting directly with node k. If the number of nodes connecting directly with k is Jk , the branches in the mesh network should be combinations of any two nodes of Jk nodes, which is equal to ð1=2ÞJk ðJk 1Þ. Assuming that there already exist Dk branches connecting these Jk nodes, the number of new branches (the number of fill-ins) after the elimination of node k is 1 Dbk ¼ JK ðJK 1Þ Dk : 2

ð1:73Þ

The number of fill-ins highly depends on the elimination sequence or the ordering number of the nodes. In Fig. 1.15, four number ordering schemes and the corresponding fill-ins in the triangular matrix are denoted. Apparently, different number ordering schemes will result in different fill-ins. An optimal ordering minimizes the fill-ins in the factor table during the LU factorization process. Different number ordering schemes should be compared according to the number of fill-ins. At present, several effective schemes have been developed. Among them the following three ordering schemes are widely employed: 1. Static ordering scheme: This scheme numbers the nodes according to the number of branches connected to them. It means that the nodes are ordered

1.4 Solution to Electric Network Equations Different ordering schemes

2

1

3

4 3 2

1

2

4

5 2

3

3

1

4

5 2 5

1

4

4

•

Admittance matrix

• • • • • • •

3

Lower triangular matrix

• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •

• • • • • •

•

•

•

1

5

45

•

•

•

•

• • • • • • • • • • •

Non-zero element

×

6

•

• • • • × • • × × •

3

• • • • • • • × •

1

•

•

•

× • × × • × × ×

Fillins

•

•

•

• • • • • •

0

Non-zero fill-in

Fig. 1.15 Illustration of number ordering

from the node with fewest branches to the node with most branches. If the numbers of connected branches for more than one node are the same, any one of them can be ordered first. Before ordering, the number of the branches connected to each node needs be counted. The scheme can be explained intuitively as follows: in the admittance matrix, the node with the fewest connected branches corresponds to the row which has the fewest nonzero elements, so the fill-ins will be generated with less possibility in the elimination operation. This scheme is very simple and suitable to be applied to small networks with fewer loops. 2. Semidynamic ordering scheme: In the above scheme, the number of branches connected to each node is counted based on the initial network and is constant in the ordering process. In fact, in the process of node elimination, the number of branches connected to each node will change according to D Y transformation. Therefore, the number of branches of the remaining nodes should be updated

46

1 Mathematical Model and Solution of Electric Network

after each elimination and then they should be ordered according to new data. This ordering scheme might be expected to result in better fill-in reduction, because it considers the changing number of incident branches during the elimination process. 3. Dynamic ordering scheme: The above two schemes are only suboptimal, which cannot guarantee minimizing fill-in number. The more rigorous scheme numbers the node according to the principle that introduces the fewest new branches. The ordering process is as follows l

l

l

According to D Y transformation, count the number of new branches (the number of fill-ins) added after the elimination on each node, and the node with the fewest branches (including fill-ins) is numbered first. Update the new number of incident branches connected to the remaining nodes. It is clear that the computation complexity of this scheme is much more than the other two.

[Example 1.7] Ordering the nodes of the network as shown in Fig. 1.16. [Solution] The above three ordering schemes are performed and compared as follows for the network as shown in Fig. 1.16. 1. Static optimal ordering scheme: There are eight nodes and 14 branches in this network. The number of incident branches on each node is listed in Table 1.2. The ordering results according to the static ordering scheme are shown in Fig. 1.17a. There are four new branches added in the process of node eliminations. When node 1 is eliminated, branch 2–7 and branch 2–8 are generated and when node 2 is eliminated, branch 3–7 and branch 4–7 are added. Factorizing the corresponding admittance matrix, we get the structure of the lower triangular matrix as shown in Fig. 1.17b. Four fill-ins, l72 ; l73 ; l74 ; and l82 correspond with the four new added branches. 2. Semidynamic ordering scheme: The process of numbering is shown in Table 1.3 and the result in Fig. 1.18a.

O

P Q

N M

T S

Fig. 1.16 Example of the node ordering

R

1.4 Solution to Electric Network Equations

47

Table 1.2 Number of branches at each node for network shown in Fig. 1.16 Node M N O P Q R Number of incident branches 4 3 3 3 3 3

3

2

4

×

3 8

7 5

4

× × ×

5 6

8

× ×

b

1

7 6

a

T 6

1 2

1

S 3

o

×

× × × × × × × × o

o

2 3

4

o

5

6

7

Fig. 1.17 Results of static optimal ordering Table 1.3 Process of semidynamic ordering scheme Node Process of numbering

M 4 4 4 4 (3)

N (3)

O 3 4 3 (2)

P 3 (3)

Q 3 3 (2)

R 3 3 3 3 3 (2)

S 3 3 3 3 3 2 (1)

T 6 6 5 4 3 2 1 (0)

Node ordered N P Q O M R S T

Node number 1 2 3 4 5 6 7 8

In this scheme, two new branches are introduced in the elimination process, that is, when node 1 is eliminated, branch 4–5 and branch 4–8 are added. 3. Dynamic ordering scheme: In order to number the nodes, we need to count the number of new branches (the number of fill-ins) added after eliminating each node. The result is listed in Table 1.4. From this table, we can see that node R or S should be numbered first. Suppose that node R is selected as node 1. After this node is eliminated we count the new branch numbers when eliminating other nodes. The results are shown in Table 1.5. From Table 1.5, node S should be numbered as node 2. The computation is repeated until the last node has been numbered. The results are shown in Fig. 1.18b. Only one new branch is added by this scheme. Therefore, for complex networks, the dynamic ordering scheme can obtain more satisfactory results.

48

1 Mathematical Model and Solution of Electric Network

Fig. 1.18 Result of semidynamic and dynamic optimal ordering

4

5

2

7

3

1

4 8

5

Table 1.4 First step of dynamic ordering scheme Node eliminated M N O Number of new branches 2 2 2

Table 1.5 Second step of dynamic ordering scheme Node eliminated M N O Number of new branches 1 2 2

1.5 1.5.1

8

3 1

6

a

6

7

P 1

Q 1

P 1

Q 1

b

2

R 0

S 0

T 10

S 0

T 7

Nodal Impedance Matrix Basic Concept of Nodal Impedance Matrix

As described above, the nodal equation of electric network can be generally represented as I ¼ YV; where I is the column vector of the nodal injection currents. Generally, it is the known variable in power system analysis; V is the column vector of the nodal voltages. Generally, it is unknown variable in power system analysis; and Y is the nodal admittance matrix. The above linear simultaneous equations can be solved by various methods, such as the direct method by inverting the admittance matrix. Suppose Z ¼ Y1 :

ð1:74Þ

Then, the above nodal equation can be written as V ¼ ZI

ð1:75Þ

1.5 Nodal Impedance Matrix

49

or in the expansion 9 V_ 1 ¼ Z11 I_1 þ Z12 I_2 þ þ Z1i I_i þ þ Z1n I_n > > > > _ _ _ _ _ > V2 ¼ Z21 I1 þ Z22 I2 þ þ Z2i Ii þ þ Z2n In > > = : V_ i ¼ Zi1 I_1 þ Zi2 I_2 þ þ Zii I_i þ þ Zin I_n > > > > > > > ; V_ n ¼ Zn1 I_1 þ Zn2 I_2 þ þ Zni I_i þ þ Znn I_n

ð1:76Þ

Comparing (1.75) with (1.76), we can see that 2

Z11 6 Z21 6 6 6 Z¼6 6 Zi1 6 4

Z12 Z22

Zn1

Zn2

Zi2

Z1i Z2i .. .

Zii

Zni

.. .

3 Z1n Z2n 7 7 7 7 7: Zin 7 7 5

ð1:77Þ

Znn

This is the nodal impedance matrix corresponding to the nodal admittance matrix Y, and they have the same order. The diagonal element Zii is called the self-impedance or the input impedance, and the off-diagonal element Zij is called the mutual impedance or the transfer impedance between the node i and node j. When the injection currents are known, the nodal voltages of the network can be solved directly through (1.75) or (1.76). The physical meaning of the elements in the nodal impedance matrix can be explained as follows: If a unit current is injected into node i, and all other nodes are open, i.e., I_i ¼ 1 I_j ¼ 0 ðj ¼ 1; 2; . . . ; n; j 6¼ iÞ: Then from (1.76), we can get V_ 1 ¼ Zi1 V_ 2 ¼ Zi2 V_ i ¼ Zii _ Vn ¼ Zin : Thus, we know that the elements in the ith column of the impedance matrix have the following physical meaning:

50

1 Mathematical Model and Solution of Electric Network

1. The diagonal element Zii of the impedance matrix is equal in value to the voltage of node i, when a unit current is injected into node i and all the other nodes are open. Therefore, Zii can be also regarded as the equivalent impedance between node i and the ground when all other nodes are open. If the network has some grounding branches and node i is connected to the network, Zii must be a nonzero element. 2. The off-diagonal element Zij is the mutual impedance between node i and j. When a unit current is injected into node i and all the other nodes are open, Zij is equal in value to the voltage of node j. Because there are always some electromagnetic connections (including indirect connections) among the nodes of a power network, the voltage of every node should be nonzero when node i is injected with a unit current and the other nodes are open. That is to say, all the mutual impedance elements Zij are nonzero elements. Therefore, the impedance matrix is a full matrix without zero elements. The impedance matrix method for directly solving network voltage used to be very popular in the early stages of power system analysis by computer. But the impedance matrix is a full matrix, more memory and operations are required, which limits its applications especially for large-scale networks. Nevertheless, it is conceptually very useful in many aspects of power system analysis. This will be introduced in later chapters.

1.5.2

Forming Nodal Impedance Matrix by Using Nodal Admittance Matrix

Comparing with the admittance matrix, it is more difficult to formulae the nodal impedance matrix of an electric network. Two general methods of constructing the impedance matrix will be introduced in the next sections. According to the discussion in Sect. 1.2.2, the admittance matrix of an electric network can be obtained directly from its configuration and parameters. So we can get the impedance matrix by inverting the admittance matrix. Several methods can be used to invert a matrix. In the following, we will illustrate one of them – inversion of a matrix through solving linear equations. Consider an admittance matrix Y and its corresponding impedance matrix Z. Solving the linear equation YZj ¼ Bj

ð1:78Þ

we can get the element Zj of the column j in the impedance matrix, where Bj is a column vector: Bj ¼ ½ 0

0 1 0 j

0 t :

1.5 Nodal Impedance Matrix

51

Solving (1.78) successively for j ¼ 1; 2; . . . ; n, we can obtain all elements of the impedance matrix. When the elements are solved column by column, only the righthand vector Bj is changed in (1.78). Therefore, the triangular factorization algorithm is very efficient to solve (1.78) (refer to Sect. 1.3.2 for details). Since the admittance matrix is symmetric, it can be factorized as: Y ¼ LDLT : The elements of the unit lower triangular matrix L and the diagonal matrix D can be obtained from (1.61). Therefore, (1.78) can be rewritten as: LDLT Zj ¼ Bj :

ð1:79Þ

LT Zj ¼ W j ;

ð1:80Þ

DW j ¼ X j :

ð1:81Þ

LX j ¼ Bj :

ð1:82Þ

Let

Then according to (1.79), we have

Thus the whole process of solving (1.79) can be decomposed into three steps: 1. Solve Xj from (1.82) Expand (1.82) as 2

1 6 l21 6 6 l31 6 6 .. 6 . 6 6 lj1 6 6 .. 4 . ln1

32

1 l32 .. . lj2 .. .

ln2

1 .. .

..

. lj;j1 .. .

lnj

1

..

. ln;n1

3 2 3 x1 0 7 6 x2 7 6 0 7 76 7 6 . 7 7 6 x3 7 6 . 7 76 7 6 . 7 76 .. 7 6 . 7 76 . 7 ¼ 6 .. 7: 76 7 6 7 7 6 xj 7 6 1 7 76 7 6 7 76 .. 7 6 0 7 54 . 5 4 5 .. 1 . xn

ð1:83Þ

Then we can get x1 , x2 , . . ., xn sequentially from the above equation. This is the forward substitution. 2. Obtain W j from (1.81) Expand (1.81) as

52

1 Mathematical Model and Solution of Electric Network

2 6 6 6 6 6 6 6 6 6 6 4

d1

32

d2

d3

..

.

dj

3 2 3 w1 x1 76 w2 7 6 x2 7 76 7 6 7 76 w3 7 6 x3 7 76 7 6 7 76 .. 7 6 .. 7 76 . 7 ¼ 6 . 7: 76 7 6 7 76 wj 7 6 xj 7 76 7 6 7 76 .. 7 6 .. 7 .. 54 . 5 4 . 5 . dn wn xn

ð1:84Þ

Then we can get w1 ; w2 ; . . . ; wn sequentially from the above equation. This is the normalization. wi ¼ xi =di

i ¼ 1; 2; . . . ; n:

ð1:85Þ

3. Obtain Zj from (1.80) Expand (1.80) as 2

1 l21 6 1 6 6 6 6 6 6 6 6 6 4

l31 l32 .. .

1

lj1 lj2

.. . 1

ln1 ln2 .. .

32

Z1j Z2j .. .

3

2

W1 W2 .. .

3

76 7 6 7 76 7 6 7 76 7 6 7 76 7 6 7 76 7 6 7 6 Zjj 7 ¼ 6 Wj 7: lnj 7 76 7 6 7 6 . 7 6 .. 7 .. 7 . 7 6 7 6 . 76 . 7 6 . 7 7 ln;n1 54 Zn1;j 5 4 Wn1 5 Wn Znj 1

ð1:86Þ

Then we can solve Znj ; Zn1;j ; . . . ; Zjj ; . . . ; Z2j ; Z1j one by one from bottom to top sequence. This is the backward substitution. [Example 1.8] Form the impedance matrix of the electric network shown in the Example 1.1 from its admittance matrix by applying the factorization algorithm. [Solution] The admittance matrix can be factorized by using (1.61), d1 ¼ Y11 ¼ 1:378742 j6:291665 Y21 0:6242024 þ j3:900156 l21 ¼ ¼ ¼ 0:612227 þ j0:034979 d1 1:378742 j64:57121 d2 ¼ Y22 l221 d1 ¼ ð1:453909 j66:98082Þ ð0:61227 þ j0:031979Þ2 ð1:378742 j6:291665Þ ¼ 1:208288 j64:57121 Similarly, other elements can be found through using the recursion formulae. Then the admittance matrix is factorized as

:

1.5 Nodal Impedance Matrix

2 6 6 6 6 6 6 6 L¼6 6 6 6 6 6 4

1 0:612227 þj0:034979 0:425687 j0:026671

53

3

0:073971 j0:017193 0:982943 j0:018393

1 0:137743 j0:027718 0:924654 þj0:027559

1 1:189287 þj0:048151

2

1:378742 6 j6:291665 6 6 6 6 6 6 D¼6 6 6 6 6 6 4

7 7 7 7 7 7 7 7; 7 7 7 7 7 5

1

1 3

1:208288 j64:57121 1:022377 j34:30237 0:887283 j3:640902

7 7 7 7 7 7 7 7: 7 7 7 7 7 0:038964 5 þj1:263678

The first step is to get the first column elements Z1 of the impedance matrix. In this situation, (1.83) should be written as 2 6 6 6 6 6 6 6 6 6 6 6 6 6 4

1 0:612227 þj0:034979 0:425687 j0:026671

3 7 72 3 2 3 7 1 7 x1 76 7 6 7 76 x2 7 6 0 7 76 7 6 7 76 x3 7 ¼ 6 0 7: 74 5 4 5 0 7 x4 7 0 7 x5 7 5

1 0:073971 j0:017193 0:982943 j0:018393

1 0:137743 j0:027718 0:924654 þj0:027559

1 1:189287 þj0:048151

1

Therefore, x1 ¼ 1 x2 ¼ 0 l21 x1 ¼ 0:612227 j0:034979 x3 ¼ 0 l31 x1 l32 x2 ¼ 0:471576 þ j0:034609 x4 ¼ 0 l42 x2 l43 x3 ¼ 0:665138 j0:027805 x5 ¼ 0 l53 x3 l54 x4 ¼ 1:226700 j0:046890:

54

1 Mathematical Model and Solution of Electric Network

From (1.85), we obtain x1 d1 x2 ¼ d2 x4 ¼ d4 x3 ¼ d3 x5 ¼ d5

w1 ¼ w2 w4 w3 w5

1 ¼ 0:033234 þ j0:151658 1:378742 j63291665 0:612227 j0:034979 ¼ ¼ 0:000719 þ j0:009468; 1:208288 j64:57121 0:665138 j0:027805 ¼ ¼ 0:049233 þ j0:170687 0:887283 j3:640902 0:471576 þ j0:034609 ¼ ¼ 0:000599 þ j0:013765; 1:022377 j34:30237 1:226700 j0:046890 ¼ ¼ 0:006535 j0:970940: 0:03894 þ j1:263678 ¼

Back substitution is executed using the following equation 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4

1

3

0:612227 þj0:034979

0:425687 j0:026671

1

0:073971 j0:017193

0:982943 j0:018393

1

0:137743 j0:027718 1

3 0:033234 7 6 þj0:151658 7 7 6 7 72 7 3 6 7 Z11 6 7 7 6 0:000719 7 76 7 6 7 76 7 6 7 76 Z21 7 6 þj0:009468 7 76 7 6 7 76 7 6 7 76 7 6 0:000599 7 7 0:924654 76 Z31 7 ¼ 6 76 7 6 þj0:013765 7: 7 6 7 þj0:027559 76 76 7 6 7 76 Z41 7 6 7 76 7 6 0:049233 7 4 5 6 7 1:189287 7 7 6 þj0:170687 7 7 6 7 þj0:048151 7 Z51 6 7 7 6 7 5 4 0:006535 5 1

2

j0:970940

Then the first column elements of the impedance matrix are Z51 ¼ 0:006535 j0:970940 Z41 ¼ 0:005290 j0:983725 Z31 ¼ 0:006862 j1:019487 Z21 ¼ 0:005555 j1:032911 Z11 ¼ 0:017972 j0:914690: The computation can be performed in a similar way and the whole impedance matrix can be obtained column by column. Thus we finally have

1.5 Nodal Impedance Matrix

2 6 6 6 6 6 6 6 6 Z¼6 6 6 6 6 6 6 4

0:017972 j0:914690 0:0055555 j1:032911 0:006862 j1:019487 0:005290 j0:983725 0:006535 j0:970940

0:005555 j1:032911 0:007781 j0:961291 0:010007 j1:037907 0:007410 j0:918658 0:009530 j0:988482

55

0:006862 j1:019487 0:010007 j1:037907 0:026875 j0:90470 0:007410 j0:918658 0:009530 j0:988482

0:005290 j0:983725 0:007410 j0:918658 0:009530 j0:988482 0:007057 j0:859912 0:009076 j0:941412

0:006535 j0:970940 0:009530 j0:988482 0:025596 j0:861619 0:009076 j0:941412 0:024377 j0:790589

3 7 7 7 7 7 7 7 7 7: 7 7 7 7 7 7 5

As described above, the elements of the jth column in the impedance matrix are equal to the nodal voltages in value when a per-unit current is injected into node j and other nodes are open. Therefore, finding the elements of the jth column from (1.78) is equivalent to solving the following nodal equation YV ¼ I j ;

ð1:87Þ

where all the elements of the current column vector I j are zero except the jth element equals 1. Obviously, V obtained from this equation is equal to Zj in value. It is worth noting that the computation burden of this method is a little too heavy in some situations; for example, if we want to derive the impedance matrix of a network with n nodes, n linear equations must be solved n times. Hence this method is only suitable for the case in which only a few elements are of interest. In power flow and short circuit analysis, the input impedance of one pair of nodes and the transfer impedance between two node pairs are often calculated using the above method. In Fig. 1.19, in order to get the input impedance of node i and j and the

•

V1 •

V2 •

•

N

Vi Ii = 1 •

•

Vj Ij =−1 •

Vk Vl •

Fig. 1.19 Solving node pair’s input and transfer impedance

56

1 Mathematical Model and Solution of Electric Network

transfer impedance between ij and kl, a unit current is injected between node i and node j, while other nodes are open. That is I_i ¼ 1;

I_j ¼ 1:

In this case, solve the network equation YV ¼ Fij ;

ð1:88Þ

where 2

3 0 6 .. 7 6 . 7 6 7 6 1 7 i 6 7 6 0 7 6 . 7 . 7 Fij ¼ 6 : 6 . 7 6 1 7 j 6 7 6 0 7 6 7 6 . 7 4 .. 5 0 The nodal voltage can be obtained and the input impedance of node pair ij is Zijij ¼ V_i V_ j :

ð1:89Þ

The transfer impedance between ij and kl is Zklij ¼ V_ k V_ l :

1.5.3

ð1:90Þ

Forming Nodal Impedance Matrix by Branch Addition Method

In the above section, we have described a method of forming the impedance matrix by using the admittance matrix. An alternative method is to form the impedance matrix directly by the branch addition method. The method is straightforward in computation and allows easy impedance matrix modification for changes in the network. Therefore it is applied widely. The forming process is illustrated by Fig. 1.20. We start to form the impedance matrix from a grounded branch and a matrix of order 1 is formed. In Fig. 1.20, z10 first is used to form this matrix. Then branch z12 is added and the new branch creates a new node . We call it adding a tree branch if a new node is generated when adding a branch. At the same time, the order of the

1.5 Nodal Impedance Matrix

57 3

z23

z13 4

z14

z12

1

z10

z25

2

5

z20

Fig. 1.20 Process of branch addition method in forming impedance matrix

corresponding matrix increases by 1. Thus after adding tree branch z12 , we obtain a 2 2 impedance matrix. We next add branch z20 . In this situation, there is no new node generated. The order of the impedance matrix does not change. This is called adding a link branch. All the elements of the impedance matrix must be updated when a link branch is added. Repeat the operations in a similar way: after adding tree branch z13 , node 3 is created. Then the order of the impedance matrix becomes three. After adding tree branch z14 , node 4 is created and the order of the impedance matrix becomes four. After adding tree branch z25 , node 5 is generated. The order of the impedance matrix becomes five. When adding link z25 , no new node is generated and the order of the impedance matrix is still five. The impedance matrix is formed after all the branches have been added to the electric network. It should be noted that the sequence of adding the branches is not unique. An alternative sequence is as follows: Tree branch z10 ! tree branch z20 ! link z12 ! tree branch z13 ! link z23 ! tree branch z14 ! tree branch z25 . Of course, there are some other schemes besides these two schemes. And it can be proved that whatever the branch adding sequence is, the impedance matrix is the same when the node number ordering is the same. However, the computation efforts under the different adding sequences are quite different. The effects of adding a tree branch or a link branch on the impedance matrix will be discussed in the following: 1. Adding a tree branch Assume that the m m impedance matrix of an electric network has been formed for the first m nodes. 2

Z11 6 Z21 6 6 ZN ¼ 6 6 Zi1 6 4 Zm1

Z12 Z22 Zi2 Zm2

Z1i Z2i Zii Zmi

3 Z1m Z2m 7 7 7 7: Zim 7 7 5 Zmm

ð1:91Þ

58

1 Mathematical Model and Solution of Electric Network

When a tree branch Zij is added at node i, a new node j is created and the order of the impedance matrix becomes m þ 1 (see Fig. 1.21). Suppose the new impedance matrix is 2 0 Z11

6 6 6 0 6 Z21 6 6 6 6 6 0 6Z 0 ZN ¼ 6 i1 6 6 6 6 0 6 Zm1 6 6 6 4 0 Zj1

0 Z12 0 Z22

0 Zi2

0 Z1i

0 Z2i

Zii0

0 Z1m

0 Z2m

0 Zim

0 Zm2

0 Zmi

0 Zmm

0 Zj2

Zji0

.. . .. . .. . .. . .. . .. . .. . .. .

0 Zjm

3 0 Z1j

7 7 7 7 7 7 7 7 0 7 Zmj 7 7: 7 7 7 0 7 7 Zmj 7 7 7 5 Zjj0 0 Z2j

ð1:92Þ

We first solve the m m matrix inside the dashed lines of (1.92). In order to obtain 0 0 0 0 the values of the first column Z11 Z21 Zi1 Zm1 ; a unit current is injected in node 1 and the other nodes are open as shown in Fig. 1.21a. In this case, voltages of the node 1, 2, . . ., m have nothing to do with branch zij , therefore, 0 0 0 0 Z11 ¼ Z11 ; Z21 ¼ Z21 ; . . . ; Zi1 ¼ Zi1 ; . . . ; Zm1 ¼ Zm1 :

It means that the first column of Z0N is the same as the first column of ZN . Similarly, the second column of Z0N is the same as the second column of ZN . Therefore we can deduce that the m m matrix inside the dashed lines of (1.92) is the original impedance matrix before adding the branch zij .

•

V1

•

V1 V2

•

I2 = 1

•

•

N

V2

Vi zij •

N

j

•

Vm

a Fig. 1.21 Adding tree branch

Vi zij •

•

Vm

b

j Ij = 1 •

1.5 Nodal Impedance Matrix

59

We now solve the jth column of Z0N . Similarly, j and other nodes are open as shown in Fig. 1.21b. In this situation, voltages of the node 1; 2; . . . ; i; . . . ; m are the same as those when a unit current is injected in node i, so we have, 0 0 0 Z1j ¼ Z1i ; Z2j ¼ Z2i ; . . . ; Zij0 ¼ Zii ; . . . ; Zmj ¼ Zmi :

ð1:93Þ

The voltage of node j is V_ j ¼ V_i þ zij 1: According to the physical meaning of the impedance matrix, we obtain Zjj ¼ Zii þ zij :

ð1:94Þ

Due to the symmetry of the impedance matrix, the off-diagonal elements of the jth row in Z0N can be obtained as follows, 0 0 0 Zj1 ¼ Z1j ; Zj2 ¼ Z2j ; . . . ; Zji0 ¼ Zij ; . . . ; Zjm ¼ Zmj :

ð1:95Þ

Hence all the elements in the impedance matrix after adding tree branch zij are found. Additionally, although the order of the new impedance matrix increases by 1, the computation to form it is relatively simple. 1. Adding a link The impedance matrix of the initial network is denoted as ZN . When link zij is added between nodes i and j, the impedance matrix becomes Z0N . The orders of these two matrices are the same because no new node is generated in the network. We now consider how to calculate the elements of new impedance matrix Z0N . As shown in Fig. 1.22, suppose the injection current vector of the new network is I, I ¼ I_1

I_2

I_i

I_j

V_i

V_ j

I_m

t

and the nodal voltage vector is V V ¼ V_ 1

V_2

t V_ m :

Thus the following relationship holds V ¼ Z0N I:

ð1:96Þ

60

1 Mathematical Model and Solution of Electric Network

Fig. 1.22 Adding a link V1

•

I1

•

•

I2

•

V2 •

N

Vi •

Iij

zij

•

Vj •

Vm

•

Ij •

Ij •

Im

From Fig. 1.22, the nodal current injected into the initial network is 2

3 I_1 6 I_2 7 6 7 6 . 7 6 .. 7 6 7 6 I_i I_ij 7 6 7 7 I0 ¼ 6 6 ... 7 ¼ I AM I ij ; 6 7 6 I_ þ I_ 7 6 j ij 7 6 . 7 6 . 7 4 . 5 I_m

ð1:97Þ

where AM is a column vector related to the added link branch, 3 0 6 .. 7 6 . 7 6 7 6 1 7 6 7 6 0 7 6 . 7 . 7 AM ¼ 6 6 . 7 6 1 7 6 7 6 0 7 6 7 6 . 7 4 .. 5 0 2

i ð1:98Þ j

According to the nodal equation of the original network, V ¼ ZN I0 ¼ ZN I ZN AM I_ij :

ð1:99Þ

ZN AM ¼ ZL :

ð1:100Þ

Assume

1.5 Nodal Impedance Matrix

61

We know that ZL is a column vector 2

Z1i Z1j Z2i Z2j .. .

3

6 7 6 7 6 7 6 7 6 7 6 Zii Zij 7 6 7 ZL ¼ 6 7: .. 6 7 . 6 7 6 Zji Zjj 7 6 7 6 7 .. 4 5 . Zmi Zmj

ð1:101Þ

Rewrite (1.99) as, V ¼ ZN I ZL I_ij :

ð1:102Þ

The voltage difference between nodes i and j is equal to V_ i V_ j ¼ zij I_ij ¼ ATM V;

ð1:103Þ

where ATM is the transpose of AM . Substituting (1.102) into (1.103), we obtain zij I_ij ¼ ATM ZN I ATM ZL I_ij : I_ij can be solved as follows 1 T I_ij ¼ Z I; ZLL L

ð1:104Þ

ZLL ¼ ATM ZL þ zij ¼ Zii þ Zjj 2Zij þ zij ;

ð1:105Þ

where

ZTL ¼ ATM ZN ¼ ðZN AM ÞT : Substituting (1.104) into (1.102), we have

1 T V ¼ ZN ZL ZL I: ZLL

ð1:106Þ

Comparing (1.96) with (1.106), we obtain the new impedance matrix Z0N , Z0N ¼ ZN

1 ZL ZTL : ZLL

ð1:107Þ

62

1 Mathematical Model and Solution of Electric Network

Expanding the above equation, we have the following formulae of the elements in Z0N Zkl0

ZLk ZLl ¼ Zkl ZLL

k ¼ 1; 2; . . . ; m : l ¼ 1; 2; . . . ; m

ð1:108Þ

In contrast with adding a tree branch, the computation of adding a link is quite heavy and complicated in which each element of the impedance matrix must be recalculated according to (1.108). The speed of forming the impedance matrix mainly depends on computations for adding links. Therefore the sequence of adding branches affects the computation speed dramatically. For example of the network in Fig. 1.20, the computations of adding link z23 according to the first sequence are performed on a 5 5 matrix, but the recalculations are just executed on a 3 3 matrix according to the second sequence. Hence the more reasonable sequence of adding branches is to add links as early as possible. If the transformer branch is involved, the P equivalent circuit as shown in Fig. 1.4 can be used in forming the impedance matrix. Comparing with a transmission line, two more branches must be added for each transformer and in most circumstances both of them are links. Therefore the computation burden increases notably. Now a direct method of adding a transformer branch is introduced in the following, which need not use the P equivalent circuit. First, we discuss the situation that the transformer added is a tree branch. In Fig. 1.23a, the leakage impedance is put at the nominal turn ratio side of the transformer. If the leakage impedance is put at the off nominal side, the formulae can be derived in a similar way. The impedance matrix of the original network is denoted as ZN (see (1.91)). When the transformer is added as a tree branch, the order of the new impedance matrix Z0N increases by 1 (see (1.92)). It can be proved that the m m block matrix in the top-left of Z0N is just ZN . As shown in Fig. 1.23b, the transformer is substituted for its equivalent circuit. When node j is open, the transformer’s P equivalent circuit is also opened as viewed from node i. This can be explained as follows. •

V1

1 1:K Ij = 1 2 Ii = K •

•

•

V2

•

N

2

N

i z ij

Fig. 1.23 Adding a transformer

i Kzij

Kzij

m

a

•

I1 I2

1

K−1

b

N

j

•

Iij =K

•

Vi

•

zi j

Ii

1:K Ij Im m

K 2zij

•

•

−Vj

1−K

c

•

1.5 Nodal Impedance Matrix

63

The impedance of the loop constituted by nodes i, j and the ground is zij0 ¼ Kzij þ

K2 K zij ¼ zij : 1K 1K

And the impedance between node i and the ground is zi0 ¼ ðK=ð1 KÞÞzij0 . The value of the parallel impedance of zi0 and zij0 becomes infinity. When a unit current is injected at each node of the original network, the current distribution of the original network is unchanged after adding a transformer as a tree branch. Hence the nodal voltages are also unchanged. The issue now is how to solve the new elements of Z0N . Focus on this question, a unit current is injected at node j and the other nodes are open as shown in Fig. 1.23b. This is just like the injecting current K into the original network at node i. Thus the nodal voltages are V_ 1 ¼ KZ1i ; V_ 2 ¼ KZ2i ; . . . ; V_ i ¼ KZii ; V_m ¼ KZmi : The voltage of node j is V_ j ¼ KðV_i þ Kzij Þ ¼ K 2 ðZii þ zij Þ: Thus, we obtain 0 0 0 Z1j ¼ Kz1i ; Z2j ¼ Kz2i ; . . . ; Zij0 ¼ KZii ; . . . ; Zmj ¼ KZmi ;

Zjj0 ¼ K 2 ðZii þ zij Þ:

ð1:109Þ ð1:110Þ

Obviously, (1.109) and (1.110) will be changed into (1.93) and (1.94) when the turn ratio K ¼ 1. The situation when the transformer added is a link branch is shown in Fig. 1.23c. Assume that the current injected into the network after adding the transformer branch is a column vector I, thus the current injected into the original network 2

3 I_1 6 I_ 7 2 6 7 6 7 .. 6 7 . 6 7 6 7 6 I_i K I_ij 7 6 7 I0 ¼ 6 7 ¼ I A0M Iij ; .. 6 7 . 6 7 6 _ 7 6 Ij þ I_ij 7 6 7 6 7 .. 5 4 . I_m

ð1:111Þ

64

1 Mathematical Model and Solution of Electric Network

where A0M is a column vector. 3 0 6 .. 7 6 . 7 6 7 6 K 7 6 7 6 0 7 6 7 . . 7 A0M ¼ 6 6 . 7 6 1 7 6 7 6 0 7 6 7 6 . 7 4 .. 5 0 2

i j

:

The following steps are similar to that of a simple impedance link branches (see (1.99)–(1.108)). The only difference is to substitute the original AM for A0M . Therefore (1.101) should be changed as follows: 2

KZ1i Z1j KZ2i Z2j .. . KZii Zij .. .

3

6 7 6 7 6 7 6 7 6 7 6 7 6 7 ZL ¼ 6 7: 6 7 6 7 6 KZji Zjj 7 6 7 6 7 .. 4 5 . KZmi Zmj

ð1:112Þ

Equation (1.103) should be rewritten as K V_ i V_j ¼ K 2 zij I_ij ¼ A0T M V:

ð1:113Þ

Accordingly, (1.105) is changed as ZLL ¼ KZLi ZLj þ K 2 zij :

ð1:114Þ

After calculating ZL and ZLL , the elements of Z0N can be calculated according to (1.108). Briefly, the process of forming an impedance matrix by using the branch addition method is a process of adding branches one by one. If the configuration of a network is changed or a branch needs to be added, the impedance matrix can be modified directly according to the above formulae. For instance, if a branch zij needs to be removed, the equivalent operation is to add a branch zij into the network.

1.5 Nodal Impedance Matrix

65

[Example 1.9] Form the impedance matrix of the electric network shown in Fig. 1.10 by using the branch addition method. [Solution] For convenience of the computation, line-to-ground capacitances at both ends of the transmission lines are lumped to the corresponding node and denoted in the format of line-to-ground reactance. The equivalent circuit is shown in Fig. 1.24. According to the node ordering, we can make the sequence table of branch adding as follows.

Sequence of branches added

Terminal nodes of branch

(1)

0L 1

- j4

(2)

0L 2

- j2

(3)

1L 2

0.04 + j 0.25

(4)

0L 3

- j4

(5)

1L 3

0.1 + j 0.35

(6)

2L 3

0.08 + j 0.30

(7)

2L 4

j 0.015

(8)

3L 5

j 0.03

Impedance of branch

iL j

Then label the branch adding sequence on the figure as shown in Fig. 1.24. 1:1.05

1.05:1 0.08 + j 0.30

2 4

3

j 0.015 (7)

j 0.03

(6)

(2) −j 2

−j4

(3)

(5)

1 (1)

− j4

Fig. 1.24 Impedance matrix formed by using branch addition method

(4) (8)

5

66

1 Mathematical Model and Solution of Electric Network

The procedure for forming the impedance matrix is shown as follows: 1. Start from the grounded branch z01 to form a 1 1 matrix. Its element is j4 2. Add branch (2): z02 is a tree branch, i ¼ 0; j ¼ 2. According to (1.93) and (1.94), the new elements are Z12 ¼ Z21 ¼ Z10 ; Z22 ¼ Z00 þ Z02 : According to the physical meaning of the impedance matrix element, we have Z10 ¼ Z00 ¼ 0: Then Z12 ¼ Z21 ¼ 0; Z22 ¼ Z02 ¼ j2 and the 2 2 matrix is

1 1 2

2

− j4 − j2

3. Add branch (3): z12 is a link branch. The elements of ZL can be obtained according to (1.101) and (1.105), ZL1 ¼ Z11 Z12 ¼ j4 ZL2 ¼ Z12 Z22 ¼ j2: From (1.105) we know, ZLL ¼ ZL1 ZL2 þ z12 ¼ j4 j2 þ 0:04 þ j0:25 ¼ 0:04 j5:75: Modify the elements of the 2 2 matrix according to (1.108) ZL1 ZL1 ðj4Þ2 ¼ j4 ¼ 0:019356 j1:217526 ZLL 0:04 j5:75 ZL2 ZL1 j2 ðj4Þ ¼ Z12 ¼0 ¼ 0:096782 j1:301237 ZLL 0:04 j5:75

0 Z11 ¼ Z11 0 0 Z12 ¼ Z21

0 Z22 ¼ Z22

ZL2 ZL2 ðj2Þ2 ¼ j2 ¼ 0:004839 j1:304381: ZLL 0:04 j5:75

1.5 Nodal Impedance Matrix

67

Thus we obtain the impedance matrix constituted by branches 1, 2, and 3

1

2

1

2

0.019356

− 0.096282

− j1.1217526

− j1.391237

− 0.096282

0.004839

− j 1.391237

− j1.304381

4. Add branch (4): z03 is a grounded tree branch. The computation process is the same as that in Step 2. The augmented matrix 3 3 is

1

2

1

2

3

0.019356

−0.096282

− j1.1217526

−j 1.391237

− 0.096282

0.004839

− j 1.391237

− j 1.304381

3

−j 4

5. Add branch (5) z13 and branch (6) z23 . Because both of these are links, the matrix order is unchanged. The computation process is the same as that in Step 3. The augmented matrix 3 3 is

1

2

3

1

2

3

0.017972

− 0.005555

− 0.006862

− j0.914690

− j 1.032911

− j1.019487

− 0.005555

0.007781

− 0.010007

− j 1.0329111

− j 0.964591

− j1.037907

− 0.006862

− 0. 010007

0.026875

− j1.019487

− j1.037907

− j0.904700

6. Add branch (7): z24 is a transformer tree branch. In this network, the off normal turns ratio of the transformer is at node i. The computation cannot be performed

68

1 Mathematical Model and Solution of Electric Network

1:1.05

1 :1 1.05

2

j 0.015×(1.05)2

j 0.015

4

2

4

Fig. 1.25 Equivalent circuit of the transformer

directly using (1.109) and (1.110). We should transfer the off normal turns ratio to the other terminal node of the transformer. It is illustrated in the Fig. 1.25. Then the elements of the fourth row and column can be calculated according to (1.109) and (1.110). 1 ð0:005555 j1:032911Þ ¼ 0:005290 j0:983725 1:05 1 ¼ Z24 ¼ K 0 Z22 ¼ ð0:007781 j0:964591Þ ¼ 0:007410 j0:918658 1:05 1 ¼ Z34 ¼ K 0 Z23 ¼ ð0:010007 j1:037907Þ ¼ 0:009530 j0:988482 1:05 1 ¼ K 02 ðZ22 þ z024 Þ ¼ ð0:007781 j0:964591Þ þ j0:015 1:052 ¼ 0:007057 j0:859912:

Z41 ¼ Z14 ¼ K 0 Z21 ¼ Z42 Z43 Z44

We now have a 4 4 matrix

1

1

2

3

4

0.017972

− 0.005555

− 0.006862

− 0.005290

− j0.914690

−j

− j1.019487

− j0.983725

1.0329111 2

3

4

− 0.005555

0.007781

− 0.010007

0.007410

−j

−j

− j1.037907

− j0.918658

1.0329111

0.964591

− 0.006862

− 0.010007

0.026875

− 0.009530

− j1.019487

− j1.037907

− j0.904700

− j0.988482

− 0.005290

0.007410

− 0.009530

0.007057

− j0.983725

− j0.918658

− j0.988482

− j0.859912

Thinking and Problem Solving

69

7. Add branch (8): z35 is also a transformer tree branch. Its off normal turns ratio is also at node i. The computation process is the same as that in Step 6. The final impedance matrix is 2 6 6 6 6 6 6 6 6 Z¼6 6 6 6 6 6 6 4

0:017972 j0:914690 0:0055555 j1:032911 0:006862 j1:019487 0:005290 j0:983725 0:006535 j0:970940

0:0055555 j1:032911 0:007781 j0:964591 0:010007 j1:037907 0:007410 j0:918658 0:009530 j0:988482

0:006862 j1:019487 0:010007 j1:037907 0:026875 j0:904700 0:009530 j988482 0:025596 j0:861619

0:005290 j0:983725 0:007410 j0:918658 0:009530 j988482 0:007057 j0:859912 0:009076 j0:941412

0:006535 j0:970940 0:009530 j0:988482 0:025596 j0:861619 0:009076 j0:941412 0:024377 j0:790589

3 7 7 7 7 7 7 7 7 7: 7 7 7 7 7 7 5

Thinking and Problem Solving 1. Prove that the incidence matrix of an electrical power network is a singular matrix 2. Is the admittance matrix generally a singular matrix? In what condition can the admittance matrix be a singular matrix? 3. What simplifications can be made to the equivalent circuit of the transformer in Fig. 1.4? 4. Why is the admittance matrix including phase shifter(s) not a symmetric matrix? 5. How many elements are there in the admittance matrix of an electrical power network with N nodes and B branches? 6. What changes will occur in the admittance matrix when the turn ratio of a transformer varies? 7. What changes will occur in the admittance matrix when a line is out of service? 8. What characteristics does the electrical power network equation have? And what requirements are there for its solution algorithm? 9. Why is the method of Gauss elimination often adopted to solve network equations? 10. How is the factor table formed? Compare the features between two methods of forming the factor tables. 11. What is the key idea behind sparse technique? 12. What fields can the sparse vector method be applied to? 13. Compare the features and application areas of three kinds of node optimal ordering methods.

70

1 Mathematical Model and Solution of Electric Network

14. State the significance of self-impedance, input impedance, mutual impedance, and transfer impedance. 15. How can an admittance matrix be used to find self-impedance Zii and mutual impedance Zij ? Give a detailed program flowchart. 16. Describe the storage scheme of a sparse admittance matrix.

Chapter 2

Load Flow Analysis

2.1

Introduction

Load flow analysis is the most important and essential approach to investigating problems in power system operating and planning. Based on a specified generating state and transmission network structure, load flow analysis solves the steady operation state with node voltages and branch power flow in the power system. Load flow analysis can provide a balanced steady operation state of the power system, without considering system transient processes. Hence, the mathematic model of load flow problem is a nonlinear algebraic equation system without differential equations. Power system dynamic analysis (see Chaps. 5 and 6) investigates system stability under some given disturbances. Its mathematic model includes differential equations. It should be pointed out that dynamic analysis is based on load flow analysis and the algorithm of load flow analysis is also the base for dynamic analysis methods. Therefore, familiarity with the theory and algorithms of load flow analysis is essential to understanding the methodology of modern power system analysis. Using digital computers to calculate load flow started from the middle of the 1950s. Since then, a variety of methods has been used in load flow calculation. The development of these methods is mainly led by the basic requirements of load flow calculation, which can be summed up as: 1. The convergence properties 2. The computing efficiency and memory requirements 3. The convenience and flexibility of the implementation Mathematically, the load flow problem is a problem of solving a system of nonlinear algebraic equations. Its solution usually cannot avoid some iteration process. Thus reliable convergence becomes the prime criterion for a load flow calculation method. With the scale of power system continually expanding, the dimension of load flow equations now becomes very high (several thousands to tens of thousands). For the equations with such high dimensions, we cannot ensure that any mathematical method can converge to a correct solution. This situation requires the researchers and scholars in the power system analysis field to seek more reliable methods. X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

71

72

2 Load Flow Analysis

In the early stages of using digital computers to solve power system load flow problems, the widely used method was the Gauss–Seidel iterative method based on a nodal admittance matrix (it will be simply called the admittance method below) [4]. The principle of this method is rather simple and its memory requirement is relatively small. These properties made it suit the level of computer and power system theory at that time. However, its convergence is not satisfactory. When the system scale becomes larger, the number of iteration increases sharply, and sometimes the iteration process cannot converge. This problem led to the use of the sequential substitution method based on the nodal impedance matrix (also called the impedance method). At the beginning of the 1960s, the digital computer had developed to the second generation. The memory and computing speed of computers were improved significantly, providing suitable conditions for the application of the impedance method. As mentioned in Chap. 1, the impedance matrix is a full matrix. The impedance method requires the computer to store the impedance matrix that represents the topology and parameters of the power network. Thus it needs a great amount of computer memory. Furthermore, in each iteration, every element in the impedance matrix must be operated with, so the computing burden is very heavy. The impedance method improved convergence and solved some load flow problems that the admittance method could not solve. Therefore, the impedance method was widely applied from then on and made a great contribution to power system design, operation, and research. The main disadvantage of the impedance method is its high memory requirement and computing burden. The larger the system is, the more serious these defects are. To overcome the disadvantage, the piecewise solution method based on impedance matrix was developed [5]. This method divides a large system up into several small local systems and only the impedance matrixes of local systems and the impedances of tie lines between these local systems are to be stored in the computer. In this way, the memory requirement and computing burden are greatly alleviated. The other approach to overcoming the disadvantages of the impedance method is to apply the Newton–Raphson method (also called the Newton method) [6]. The Newton method is a typical method used to solve nonlinear equations in mathematics with very favorable convergence. As long as the sparsity of the Jacobean matrix is utilized in the iterative process, the computing efficiency of the Newton method can be greatly improved. Since the optimal order eliminating method [7] began to be employed in the middle of the 1960s, the Newton method has surpassed the impedance method in the aspects of convergence, memory demand, and computing speed. It is still the favored method, and is widely used in load flow calculation today. Since the 1970s, the load flow calculating method continues to develop in various ways. Among them the most successful is the fast decoupled method, also called the P Q decoupled method [8]. Comparing with the Newton method, this method is much simpler and more efficient algorithmically, and therefore more popular in many applications.

2.2 Formulation of Load Flow Problem

73

In the recent 20 years, research on load flow calculation is still very active. Many contributions seek to improve the convergence characteristics of the Newton method and the P Q decoupled method [9–15]. Along with the development of artificial intelligent theory, the genetic algorithm, artificial neural network algorithm, and fuzzy algorithm have also been introduced to load flow analysis [16–19]. However, until now these new models and new algorithms still cannot replace the Newton method and P Q decoupled method. Because the scales of power systems continue to expand and the requirements for online calculation become more and more urgent, the parallel computing algorithms are also studied intensively now and may become an important research field [20]. This chapter mainly discusses the currently widely used Newton method and P Q decoupled method. The degree of flexibility and convenience of load flow calculation are also very important to computer application. In practice, load flow analysis is usually part of an interactive environment, rather than a pure calculation problem. Therefore, the human–computer interface should be friendly, allowing users to monitor and control the calculation process. To obtain an ideal operation scheme, it is usually necessary to modify the original data according to the computing results. Thus, the computing method should be flexible, permitting users to readily modify and adjust their operation scheme. Input and output processes should also receive careful attention. Power system steady state analysis includes load flow analysis and static security analysis. Load flow analysis is mainly used in analyzing the normal operation state, while static security analysis is used when some elements are out of service. Its purpose is to check whether the system can operate safely, i.e., if there are equipment overloads, or some node voltages are too low or too high. In principle, static security analysis can be replaced by a series of load flow analyses. However, usually there are very many contingency states to be checked and the computation burden is quite large if a rigorous load flow calculation method is used. Hence special methods have to be developed to meet the requirement of efficient calculation. In the first part of this chapter, the models and algorithms of load flow calculation are introduced. In the second part, the problems related to static security analysis are discussed.

2.2 2.2.1

Formulation of Load Flow Problem Classification of Node Types

An electric power system is composed of generators, transformers, transmission lines and loads, etc. A simple power system is illustrated in Fig. 2.1. In the process of power system analysis, the static components, such as transformers, transmission lines, shunt capacitors and reactors, are represented by their equivalent circuits

74

2 Load Flow Analysis PH2 + jQH2

PF 2 + jQF 2

6

PF1 + jQF1

5

3

4

2

1

PH1 + jQH1

(a) −PH1 + jQH1

1

V1

•

I1

•

•

•

2

0

V2

I2

•

3

0

V3

I3

V4

•

I4

•

I5

PF1 + jQF1

4

−PH2 − jQH2

5

PF2 − jQF 2

6

•

Linear Network

•

•

V5 •

•

V6

I6

(It can be described by admittance matrix or impedance matrix)

(b) Fig. 2.1 Simple power system

consisting of R, L, C elements. Therefore, the network formed by these static components can be considered as a linear network and represented by the corresponding admittance matrix or impedance matrix. In load flow calculation, the generators and loads are treated as nonlinear components. They cannot be embodied in the linear network, see Fig. 2.1b. The connecting nodes with zero injected power also represent boundary conditions on the network. In Fig. 2.1b, the relationship between node current and voltage in the linear network can be described by the following node equation: I ¼ YV

ð2:1Þ

or I_i ¼

n X

Yij V_ j

ði ¼ 1; 2; . . . ; nÞ

ð2:2Þ

j¼1

where I_i and V_ j are the injected current at bus i and voltage at bus j, respectively, Yij is an element of the admittance matrix, n is the total number of nodes in the system.

2.2 Formulation of Load Flow Problem

75

To solve the load flow equation, the relation of node power with current should be used Pi jQi I_i ¼ V^i

ði ¼ 1; 2; . . . ; nÞ

ð2:3Þ

where Pi , Qi are the injected active and reactive power at node i, respectively. If node i is a load node, then Pi and Qi should take negative values. In (2.3), V^i is the conjugate of the voltage vector at node i. Substituting (2.3) to (2.2), we have, n Pi jQi X ¼ Yij V_ j V^i j¼1

ði ¼ 1; 2; . . . ; nÞ

n Pi þ jQi X ¼ Y^ij V^j V_ i

ði ¼ 1; 2; . . . ; nÞ

or ð2:4Þ

j¼1

There are n nonlinear complex equations in (2.4). They are the principal equations in load flow calculation. Based on different methods to solve (2.4), various load flow algorithms can be formed. In the power system load flow problem, the variables are nodal complex voltages and complex powers: V, y, P, Q. If there are n nodes in a power system, then the total number of variables is 4 n. As mentioned above, there are n complex equations or 2n real equations defined in principal by (2.4), thus only 2n variables can be solved from these equations, while the other 2n variables should be specified as original data. Usually, two variables at each node are assumed known, while the other two variables are treated as state variables to be resolved. According to the original data, the nodes in power systems can be classified into three types: 1. PQ Nodes: For PQ nodes, the active and reactive power (P; Q) are specified as known parameters, and the complex voltage (V; y) is to be resolved. Usually, substation nodes are taken as PQ nodes where the load powers are given constants. When output P and Q are fixed in some power plants, these nodes can also be taken as PQ node. Most nodes in power systems belong to the PQ type in load flow calculation. 2. PV Nodes: For PV nodes, active power P and voltage magnitude V are specified as known variables, while reactive power Q and voltage angle y are to be resolved. Usually, PV nodes should have some controllable reactive power resources and can thus maintain node voltage magnitude at a desirable value. Generally speaking, the buses of power plants can be taken as PV nodes, because voltages at these buses can be controlled with reactive power capacity of their generators. Some substations can also be considered as PV nodes when they have enough reactive power compensation devices to control the voltage.

76

2 Load Flow Analysis

3. Slack Node: In load flow studies, there should be one and only one slack node specified in the power system, which is specified by a voltage, constant in magnitude and phase angle. Therefore, V and y are given as known variables at the slack node, while the active power P and reactive power Q are the variables to be solved. The effective generator at this node supplies the losses to the network. This is necessary because the magnitude of losses will not be known until the calculation of currents is complete, and this cannot be achieved unless one node has no power constraint and can feed the required losses into the system. The location of the slack node can influence the complexity of the calculations; the node most closely approaching a large AGC power station should be used. We will employ different methods to treat the above three kinds of nodes in power flow calculations.

2.2.2

Node Power Equations

As described above, power system load flow calculations can be roughly considered as the problem of solving the node voltage phasor for each node when the injecting complex power is specified. If the complex power can be represented by equations of complex voltages, then a nonlinear equation solving method, such as the Newton–Raphson method, can be used to solve the node voltage phasors. In this section, node power equations are deduced first. The complex node voltage has two representation forms – the polar form and the rectangular form. Accordingly, the node power equations also have two forms. From (2.4), the node power equations can be expressed as Pi þ jQi ¼ V_ i

X

Y^ij V^j

ði ¼ 1; 2; . . . ; nÞ

ð2:5Þ

j2i

where j 2 i means the node j should be directly connected with node i, including j ¼ i. As discussed in Chap.1, the admittance matrix is a sparse matrix, and the terms in S are correspondingly few. If the voltage vector of (2.5) adopts polar form, V_i ¼ Vi ejyi

ð2:6Þ

where Vi ,yi are the magnitude and phase angle of voltage at node i. The elements of admittance matrix can be expressed as Yij ¼ Gij þ jBij

2.2 Formulation of Load Flow Problem

77

Hence (2.5) can be rewritten as Pi þ jQi ¼ Vi ejyi

X

ðGij jBij ÞVj ejyj

ði ¼ 1; 2; . . . ; nÞ

ð2:7Þ

j2i

Combining the exponential items of above equation and using the relationship ejy ¼ cos y þ j sin y we have, Pi þ jQi ¼ Vi

X

Vj ðGij jBij Þðcos yij þ j sin yij Þ

ði ¼ 1; 2; . . . ; nÞ

ð2:8Þ

j2i

where yij ¼ yi yj , is the voltage phase angle difference between node i and j. Dividing above equations into real and imaginary parts, Pi ¼ Vi

X j2i

Qi ¼ V i

X j2i

9 Vj ðGij cos yij þ Bij sin yij Þ > > = Vj ðGij sin yij Bij cos yij Þ > > ;

ði ¼ 1; 2; ; nÞ

ð2:9Þ

This is the polar form of the nodal power equations. It is not only very important in the Newton–Raphson calculation process, but also essential to establish the fast decoupled method. When the voltage vector is expressed in rectangular form, V_i ¼ ei þ jfi where ei ¼ Vi cos yi

fi ¼ Vi sin yi

We can obtain from (2.5), P i ¼ ei

X

ðGij ej Bij fj Þ þ fi

j2i

Qi ¼ fi

X j2i

X j2i

ðGij ej Bij fj Þ ei

X j2i

9 ðGij fj þ Bij ej Þ > > = ðGij fj þ Bij ej Þ > > ;

ði ¼ 1; 2; . . . ; nÞ

ð2:10Þ

78

2 Load Flow Analysis

Let X j2i

X j2i

9 ðGij ej Bij fj Þ ¼ ai > > =

ð2:11Þ

ðGij fj þ Bij ej Þ ¼ bi > > ;

Obviously, ai and bi are the real and imaginary parts of injected current at node i and (2.10) can be simplified as, Pi ¼ ei ai þ fi bi Q i ¼ f i ai e i bi

) ði ¼ 1; 2; . . . ; nÞ

ð2:12Þ

This is the rectangular form of the nodal power equations. Both (2.9) and (2.10) are the simultaneous nonlinear equations of node voltage phasors. They are usually expressed as the following forms as mathematical models of the load flow problem: X

9 Vj ðGij cos yij þ Bij sin yij Þ ¼ 0 > > = j2i X DQi ¼ Qis Vi Vj ðGij sin yij Bij cos yij Þ ¼ 0 > > ; DPi ¼ Pis Vi

ði ¼ 1; 2; . . . ; nÞ ð2:13Þ

j2i

and X

X

9 ðGij fj þ Bij ej Þ ¼ 0 > > = j2i j2i X X DQi ¼ Qis fi ðGij ej Bij fj Þ þ ei ðGij fj þ Bij ej Þ ¼ 0 > > ; DPi ¼ Pis ei

j2i

ðGij ej Bij fj Þ fi

ð2:14Þ

j2i

ði ¼ 1; 2; . . . ; nÞ where Pis , Qis are the specified active and reactive powers at node i. Based on the above two simultaneous equations, the load flow problem can be roughly summarized as: for specified Pis ,Qis ði ¼ 1; 2; . . . ; nÞ, find voltage vector Vi , yi or ei , fi ði ¼ 1; 2; . . . ; nÞ, such that the magnitudes of the power errors DPi , DQi , ði ¼ 1; 2; . . . ; nÞ of (2.13) or (2.14) are less then an acceptable tolerance.

2.3 2.3.1

Load Flow Solution by Newton Method Basic Concept of Newton Method

The Newton–Raphson method is an efficient algorithm to solve nonlinear equations. It transforms the procedure of solving nonlinear equations into the procedure

2.3 Load Flow Solution by Newton Method

79

of repeatedly solving linear equations. This sequential linearization process is the core of the Newton–Raphson method. We now introduce the Newton–Raphson method by the following nonlinear equation example, f ðxÞ ¼ 0

ð2:15Þ

Let xð0Þ be the initial guess value of the above equation solution. Assume the real solution x is close to xð0Þ , x ¼ xð0Þ Dxð0Þ

ð2:16Þ

where Dxð0Þ is a modification value of xð0Þ . The following equation holds, f ðxð0Þ Dxð0Þ Þ ¼ 0

ð2:17Þ

When Dxð0Þ is known, the solution x can be calculated by (2.16). Expanding this function in a Taylor series expansion about point xð0Þ yields: ð0Þ 2

f ðxð0Þ Dxð0Þ Þ ¼ f ðxð0Þ Þ f 0 ðxð0Þ ÞDxð0Þ þ f 00 ðxð0Þ Þ ðDx2! ð0Þ n

þ ð1Þn f ðnÞ ðxð0Þ Þ ðDxn!

Þ

þ ¼ 0

Þ

ð2:18Þ

where f 0 ðxð0Þ Þ,. . ., f ðnÞ ðxð0Þ Þ are the different order partial derivatives of f ðxÞ at xð0Þ . If the initial guess is sufficiently close to the actual solution, the higher order terms of the Taylor series expansion could be neglected. Equation (2.18) becomes, f ðxð0Þ Þ f 0 ðxð0Þ ÞDxð0Þ ¼ 0

ð2:19Þ

This is a linear equation in Dxð0Þ and can be easily solved. Using Dxð0Þ to modify xð0Þ , we can get xð1Þ : xð1Þ ¼ xð0Þ Dxð0Þ

ð2:20Þ

xð1Þ may be more close to the actual solution. Then using xð1Þ as the new guess value, we solve the following equation similar to (2.19), f ðxð1Þ Þ f 0 ðxð1Þ ÞDxð1Þ ¼ 0 Thus xð2Þ is obtained: xð2Þ ¼ xð1Þ Dxð1Þ

ð2:21Þ

Repeating this procedure, we establish the correction equation in the tth iteration: f ðxðtÞ Þ f 0 ðxðtÞ ÞDxðtÞ ¼ 0

ð2:22Þ

80

2 Load Flow Analysis y

Fig. 2.2 Geometric interpretation of Newton method

y = f (x)

f (x(t))

f (x(t+1)) α(t)

x

0

x(t+1)

x(t)

x

Δx(t+1) Δx(t)

or f ðxðtÞ Þ ¼ f 0 ðxðtÞ ÞDxðtÞ

ð2:23Þ

The left hand of the above equation can be considered as the error produced by approximate solution xðtÞ . When f ðxðtÞ Þ ! 0, (2.15) is satisfied, so xðtÞ is the solution of the equation. In (2.22), f 0 ðxðtÞ Þ is the first-order partial derivative of function f ðxÞ at point xðtÞ . It is also the slope of the curve at point xðtÞ , as shown in Fig. 2.2, tan aðtÞ ¼ f 0 ðxðtÞ Þ

ð2:24Þ

The correction value DxðtÞ is determined by the intersection of the tangent line at xðtÞ with the abscissa. We can comprehend the iterative process more intuitively from Fig. 2.2. Now we will extend the Newton method to simultaneous nonlinear equations. Assume the nonlinear equations with variables x1 ; x2 ; . . . ; xn ; 9 f1 ðx1 ; x2 ; . . . ; xn Þ ¼ 0 > > > > = f2 ðx1 ; x2 ; . . . ; xn Þ ¼ 0 > .. > > . > > > ; fn ðx1 ; x2 ; . . . ; xn Þ ¼ 0

ð2:25Þ

ð0Þ

ð0Þ

ð0Þ

ð0Þ

Specify the initial guess values of all variables x1 ; x2 ; . . . ; xn , let Dx1 ; ð0Þ

ð0Þ

Dx2 ; . . . ; Dxn be the correction values to satisfy the following equations, 9 ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ f1 ðx1 Dx1 ; x2 Dx2 ; . . . ; xð0Þ > n Dxn Þ ¼ 0 > > > > ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ = f ðx Dx ; x Dx ; . . . ; x Dx Þ ¼ 0 > 2

1

1

2

2

n

n

.. . ð0Þ

ð0Þ

ð0Þ

ð0Þ

ð0Þ fn ðx1 Dx1 ; x2 Dx2 ; . . . ; xð0Þ n Dxn Þ ¼ 0

> > > > > > ;

ð2:26Þ

2.3 Load Flow Solution by Newton Method

81

Expanding the above n equations via the multivariate Taylor series and neglecting the higher order terms, we have the following equations, 9 @f1 @f1 @f1 ð0Þ ð0Þ > ð0Þ > Dx ¼ 0 0 Dx1 þ 0 Dx2 þ; . . . ; þ > 0 n > @x1 @x2 @xn > > > > > @f @f @f > 2 2 2 ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ = f2 ðx1 ; x2 ; . . . ; xn Þ ¼ 0> 0 Dx1 þ 0 Dx2 þ; . . . ; þ 0 Dxn @x1 @x2 @xn ð2:27Þ > .. > > > . > > > > > @fn @fn @fn > ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ > ; fn ðx1 ; x2 ; . . . ; xn Þ Dx ¼ 0 0 Dx1 þ 0 Dx2 þ; . . . ; þ 0 n @x1 @x2 @xn ð0Þ

ð0Þ

f1 ðx1 ; x2 ; . . . ; xð0Þ n Þ

here

@fi @xj j0

is the partial derivative of function fi ðx1 ; x2 ; . . . ; xn Þ over independent ð0Þ

ð0Þ

ð0Þ

variable xj at point ðx1 ; x2 ; . . . ; xn Þ. Rewrite the above equation in matrix form, 3 2 3 ð0Þ Dx1 7 76 6 7 6 @f2 @f2 . . . @f2 76 ð0Þ 7 6 @x1 0 @x2 0 @xn 0 76 Dx2 7 6 7¼6 76 7 6. 7 6 76 . 7 6.. 7 6 .. 76 .. 7 4 5 6. 74 5 4 5 ð0Þ ð0Þ ð0Þ ð0Þ @fn @fn @fn fn ðx1 ; x2 ; . . . ; xn Þ Dx n @x1 0 @x2 0 . . . @xn 0 2

2

@f1 @f1 @f 0 0 . . . @x1n 0 6 @x1 @x2

3

ð0Þ ð0Þ ð0Þ f1 ðx1 ; x2 ; . . . ; xn Þ 6 7 6 ð0Þ ð0Þ ð0Þ 7 6 f2 ðx1 ; x2 ; . . . ; xn Þ 7

ð2:28Þ

ð0Þ

ð0Þ

This is a set of simultaneous linear equations in the variables Dx1 ; Dx2 ; . . . ; ð0Þ Dxn ;

usually called the correction equations of the Newton–Raphson method. ð0Þ

ð0Þ

ð0Þ

After solving Dx1 ; Dx2 ; . . . ; Dxn ; we can get, 9 ð1Þ ð0Þ ð0Þ > x1 ¼ x1 Dx1 > > > ð1Þ ð0Þ ð0Þ > > x2 ¼ x2 Dx2 = .. .. .. > > > . . . > > > ð1Þ ð0Þ ð0Þ ; xn ¼ xn Dxn ð1Þ

ð1Þ

ð2:29Þ

ð1Þ

x1 ; x2 ; . . . ; xn will approach the actual solution more closely. The updated values are used as the new guess to solve the correction equation (2.28) and to further correct the variables. In this way the iterative process of the Newton– Raphson method is formed. Generally, the correction equation in the tth iteration can be written as, 2

ðtÞ

ðtÞ

3

2 @f

32

3 ðtÞ Dx 1 7 6 7 6 7 @f2 @f2 @f2 76 ðtÞ 7 6 f ðxðtÞ ; xðtÞ ; . . . ; xðtÞ 7 6 . . . j j j 6 76 t t t Þ n 2 @x @x @x 6 7 6 1 6 Dx2 7 2 n 1 2 7 6 7¼6 6 7 6 .. 7 6 .. 7 .. .. 7 .. .. .. 76 6 . 7 6 7 . . . 74 . . . 5 4 5 6 4 5 ðtÞ ðtÞ ðtÞ @fn @fn @fn ðtÞ fn ðx1 ; x2 ; . . . ; xn Þ Dxn @x1 jt @x2 jt . . . @xn jt ðtÞ

f1 ðx1 ; x2 ; . . . ; xn Þ

@f1 @f1 @x1 jt @x2 jt . . . @xn jt 1

ð2:30Þ

82

2 Load Flow Analysis

or expressed in matrix form, FðXðtÞ Þ ¼ JðtÞ DXðtÞ

ð2:31Þ

where 2

ðtÞ

ðtÞ

ðtÞ

f1 ðx1 ; x2 ; . . . ; xn Þ

3

6 7 6 f2 ðxðtÞ ; xðtÞ ; . . . ; xðtÞ 7 n Þ7 6 1 2 7 FðXðtÞ Þ ¼ 6 6 7 .. 6 7 . 4 5 ðtÞ

ðtÞ

ð2:32Þ

ðtÞ

fn ðx1 ; x2 ; . . . ; xn Þ is the error vector in the tth iteration; 2 @f1

@f1 @f1 @x1 jt @x2 jt ::: @xn jt

JðtÞ

3

6 @f @f 7 6 2 jt 2 jt ::: @f2 jt 7 6 @x1 @x2 @xn 7 7 ¼6 7 6 .. 6 7 . 4 5

ð2:33Þ

@fn @fn @fn @x1 jt @x2 jt ::: @xn jt

is the Jacobian matrix of tth iteration; 2 DXðtÞ

ðtÞ

Dx1

3

6 7 6 DxðtÞ 7 6 2 7 7 ¼6 6 . 7 6 .. 7 4 5

ð2:34Þ

ðtÞ

Dxn

is the correction value vector in the tth iteration. We also have the equation similar to (2.29), Xðtþ1Þ ¼ XðtÞ DXðtÞ

ð2:35Þ

With (2.31) and (2.35) solved alternately in each iteration, Xðtþ1Þ gradually approaches the actual solution. Convergence can be evaluated by the norm of the correction value, ðtÞ DX < e1

ð2:36Þ

FðXðtÞ Þ < e2

ð2:37Þ

or by the norm of the function,

Here e1 and e2 are very small positive numbers specified beforehand.

2.3 Load Flow Solution by Newton Method

2.3.2

83

Correction Equations

In Section 2.3.1, we derived two forms of the nodal power equations. Either can be applied in the load flow calculation model. When the polar form (2.13) is used, the node voltage magnitudes and angles Vi ,yi ði ¼ 1; 2; . . . ; nÞ are the variables to be solved. For a PV node, the magnitude of the voltage is specified. At the same time, its reactive power Qis cannot be fixed beforehand as a constraint. Therefore, the reactive equations relative to PV nodes should not be considered in the iterative process. These equations will be used only to calculate the reactive power of each PV node after the iterative process is over and all node voltages have been calculated. Similarly, the voltage magnitude and angle of the slack node are specified, hence the related power equations do not appear in the iterative process. When the iteration has converged, the active and reactive power of the slack node can be calculated by using these power equations. Assume that total number of system nodes is n, the number of PV nodes is r. For convenience, let the slack bus be the last node, i.e., node n.Therefore, we have n 1 active power equations, DP1 ¼ P1s V1

X

9 > > > > > > > > > > > =

Vj ðG1j cos y1j þ B1j sin y1j Þ ¼ 0

j21

DP2 ¼ P2s V2

X

Vj ðG2j cos y2j þ B2j sin y2j Þ ¼ 0

j22

.. . DPn1 ¼ Pn1;s Vn1

> > > > > > X > > > Vj ðGn1;j cos yn1; j þ Bn1; j sin yn1; j Þ ¼ 0 > > ;

j2ðn1Þ

ð2:38Þ and n r 1 reactive power equations. DQ1 ¼ Q1s V1

X

Vj ðG1j sin y1j B1j cos y1j Þ ¼ 0

j21

DQ2 ¼ Q2s V2

X

Vj ðG2j sin y2j B2j cos y2j Þ ¼ 0

j22

.. . DQn1 ¼ Qn1;s Vn1

9 > > > > > > > > > > > =

> > > > > > X > > > Vj ðGn1; j sin yn1; j Bn1; j cos yn1; j Þ ¼ 0 > > ;

ð2:39Þ

j2ðn1Þ

In the above equations, node voltage angle yi and magnitude Vi are the variables to be resolved. Here the number of yi is n 1 and the number of Vi is n r 1. There

84

2 Load Flow Analysis

are 2n r 2 unknown variables in total and they can be solved by the above 2n r 2 equations. Expanding (2.38) and (2.39) in a Taylor series, neglecting the high-order terms, the correction equation can be written as, 2

DP1 DP2 .. .

3 2

H12 6 7 6 H11 6 7 6 H21 H22 6 7 6 6 7 6 6 7 6 6 DPn1 7 6 H 6 7 6 n1;1 Hn1;2 6 7 ¼ 6 ......... ......... 6 7 6 6 DQ1 7 6 J12 6 7 6 J 6 DQ2 7 6 11 6 7 6 J J22 6 .. 7 6 21 4 . 5 4 DQn1 Jn1;1 Jn1;2

.. . N11 N12 .. . N21 N22 .. .. Hn1;n1 .. Nn1;1 Nn1;2 ......... .. ......... ......... .. ... J1;n1 .. L11 L12 .. ... J2;n1 . L21 L22 .. ... .. ... Jn1;n1 .. Jn1;1 Jn1;2 ... ... ... ... ...

H1;n1 H2;n1

3 2 3 Dy1 N1;n1 7 6 7 Dy2 6 7 N2;n1 7 .. 7 6 7 7 6 7 . 7 6 7 6 Dyn1 7 Nn1;n1 7 7 6 7 6 7 ......... 7 7 6 ......... 7 ð2:40Þ 7 6 DV1 =V1 7 ... L1;n1 7 6 7 7 6 DV2 =V2 7 6 7 ... L2;n1 7 7 6 7 .. 5 4 5 . ... DV =V n1 n1 ... Jn1;n1 ... ... ... ... ...

The form of the voltage magnitude correction values represented here, DV1 =V1 ; DV2 =V2 ; . . . ; DVn1 =Vn1 ; allow the elements in the Jacobian matrix to have similar expressions. Taking partial derivations of (2.38), or (2.39), and noting that both Pis , Qis are constants, we can obtain the elements of the Jacobian matrix as, Hij ¼

@DPi ¼ Vi Vj ðGij sin yij Bij cos yij Þ @yj

Hii ¼

X @DPi ¼ Vi Vj ðGij sin yij Bij cos yij Þ @yi j2i

j 6¼ i

ð2:41Þ ð2:42Þ

j6¼i

or Hii ¼ Vi2 Bii þ Qi Nij ¼ Nii ¼

ð2:43Þ

@DPi Vj ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ @Vj

j 6¼ i

ð2:44Þ

X @DPi Vi ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ 2Vi2 Gii ¼ Vi2 Gii Pi ð2:45Þ @Vi j2i j6¼i

Jij ¼ Jii ¼

@DPi ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ @yj

j 6¼ i

X @DPi ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ ¼ Vi2 Gii Pi @yj j2i j6¼i

ð2:46Þ ð2:47Þ

2.3 Load Flow Solution by Newton Method

85

Lij ¼

@DQi Vj ¼ Vi Vj ðGij sin yij Bij cos yij Þ @Vj

Lii ¼

X @DQi Vi ¼ Vi Vj ðGij sin yij Bij cos yij Þ þ 2Vi2 Bii ¼ Vi2 Bii Qi ð2:49Þ @Vi j2i

j 6¼ i

ð2:48Þ

j6¼i

The concise form of (2.40) is

DP H ¼ DQ J

N L

Dy DV=V

ð2:50Þ

Comparing (2.50) with (2.40), the meaning of elements is obvious. The correction equation can be rearranged into the following form for convenience, 2

3 2 H11 DP1 6 DQ1 7 6 J11 6 7 6 6 DP2 7 6 H21 6 7 6 6 DQ2 7 6 J21 6 7¼6 6 .. 7 6 .. 6 . 7 6 . 6 7 6 4 DPn1 5 4 Hn1;1 DQn1 Jn1;1

N11 L11 N21 L21 .. .

Nn1;1 Ln1;1

H12 J12 H22 J22 .. .

Hn1;2 Jn1;2

N12 L12 N22 L22 .. .

Nn1;2 Ln1;2

... ... ... ... .. .

H1;n1 J1;n1 H2;n1 J2;n1 .. .

. . . Hn1;n1 . . . Jn1;n1

3 Dy1 76 DV1 =V1 7 76 7 76 7 Dy2 76 7 76 DV2 =V2 7 76 7 ð2:51Þ 76 7 .. 76 7 . 76 7 5 4 Nn1;n1 Dyn1 5 Ln1;n1 DVn1 =Vn1 N1;n1 L1;n1 N2;n1 L2;n1 .. .

32

When the rectangular form is adopted in the load flow model, the state variables to be solved are the real and imaginary parts of voltages, i.e., e1 ; f1 ; e2; f2 ; . . . ; en ; fn . Since the voltage phasor of the slack node is specified, the number of state variables is 2ðn 1Þ. We need 2ðn 1Þ equations to solve these variables. In fact, every node has two equations except the slack bus. For PQ nodes, Pis , Qis are given, so the equations are X

X

9 ðGij fj þ Bij ej Þ ¼ 0 > > = j2i j2i X X DQi ¼ Qis fi ðGij ej Bij fj Þ þ ei ðGij fj þ Bij ej Þ ¼ 0 > > ; DPi ¼ Pis ei

ðGij ej Bij fj Þ fi

j2i

ð2:52Þ

j2i

For PV nodes, Pis , Vis are given, so the equations are DPi ¼ Pis ei

X

ðGij ej Bij fj Þ fi

j2i

DVi2

¼

Vis2

ðe2i

þ

X j2i

fi2 Þ

¼0

9 ðGij fj þ Bij ej Þ ¼ 0 = ;

ð2:53Þ

86

2 Load Flow Analysis

There are 2ðn 1Þ equations included in (2.52) and (2.53). Expanding them in a Taylor series expansion, neglecting the higher order terms, we can obtain the correction equation as follows, 2

DP1

3

2

@DP1 @e1

6 7 6 6 7 6 @DQ1 6 DQ1 7 6 @e1 6 7 6 6 7 6 @DP 6 7 6 @e1 2 6 DP2 7 6 6 7 6 @DQ2 6 7 6 @e1 6 7 6 6 6 DQ2 7 ¼ 6 6 7 6 .. 6 .. 7 6 . 6. 7 6 6 7 @DPi 6 DPi 7 6 @e1 6 7 6 6 6 7 6 6 7 0 6 DV 2 7 6 4 i 5 6 4 .. .. . .

@DP1 @f1

@DP1 @e2

@DP1 @f2

@DP1 @ei

@DP1 @fi

@DQ1 @f1

@DQ1 @e2

@DQ1 @f2

@DQ1 @ei

@DQ1 @fi

@DP2 @f1

@DP2 @e2

@DP2 @f2

@DP2 @ei

@DP2 @fi

@DQ2 @f1

@DQ2 @e2

@DQ2 @f2

@DQ2 @ei

@DQ2 @fi

.. .

.. .

.. .

.. .

.. .

.. .

@DPi @f1

@DPi @e2

@DPi @f2

@DPi @ei

@DPi @fi

0

0

0

@DVi2 @ei

@DVi2 @fi

.. .

.. .

.. .

.. .

.. .

.. .

32

De1

3

76 7 76 7 76 Df1 7 76 7 76 7 6 7 7 76 De2 7 76 7 6 7 7 76 7 76 Df2 7 7 6 7 ð2:54Þ .. 76 7 . 76 . 7 76 .. 7 6 7 7 76 De 7 76 i 7 76 7 6 7 7 76 7 54 Dfi 5 .. .. . .

By differentiating (2.52) and (2.53), we can obtain elements of the Jacobian matrix. The off-diagonal elements of the Jacobian matrix for j 6¼ i can be expressed as, 9 @DPi @DQi > ¼ ¼ ðGij ei þ Bij fi Þ > > > @ej @fj > > > = @DPi @DQi ¼ ¼ Bij ei Gij fi @fj @ej > > > > > @DVi2 @DVi2 > > ; ¼ ¼0 @ej @fj The diagonal elements of the Jacobian matrix for j ¼ i, X @DPi ¼ ðGij ej Bij fj Þ Gii ei Bii fi @ei j2i Using (2.11), we can rewrite the above expression as @DPi ¼ ai Gii ei Bii fi @ei and can obtain the following elements similarly,

ð2:55Þ

2.3 Load Flow Solution by Newton Method

87

9 X @DQi ¼ ðGij ej Bij fj Þ þ Gii ei þ Bii fi ¼ ai þ Gii ei þ Bii fi > > > > @fi > j2i > > > > X > @DPi > ¼ ðGij fj þ Bij ej Þ þ Bii ei Gii fi ¼ bi þ Bii ei Gii fi > > > @fi > > j2i > > = X @DQi ¼ ðGij fj þ Bij ej Þ þ Bii ei Gii fi ¼ bi þ Bii ei Gii fi > @ei > j2i > > > > 2 > @DVi > > > ¼ 2ei > > @ei > > > > 2 > @DVi > ; ¼ 2fi @fi

ð2:56Þ

The correction equations, in either polar form or rectangular form, are the basic equations that need repeatedly solving in Newton–Raphson load flow calculation. Investigating these equations, we can observe the following properties: 1. Equations (2.54) and (2.40) include 2ðn 1Þ and 2ðn 1Þ r equations respectively. 2. From the expression of the off-diagonal elements of the Jacobian matrix either in polar form or in rectangular form, i.e., (2.41), (2.44), (2.46), (2.48), and (2.55), we can see that each of them is related to only one element of the admittance matrix. Therefore, if the element Yij in the admittance matrix is zero, the corresponding element in the Jacobian matrix of the correction equation is also zero. It means the Jacobian matrix is a sparse matrix, and has the same structure as the admittance matrix. 3. From the expression of the elements of the Jacobian matrix we can see that the Jacobian matrix is not symmetrical in either coordinate form. For example, @DPi @DPj 6¼ ; @yj @yi @DPi @DPj 6¼ ; @ej @ei

@DQi @DQj 6¼ @Vj @Vi @DQi @DQj 6¼ ; etc: @fj @fi

4. The elements in the Jacobian matrix are a function of node voltage phasors. Therefore, they will vary with node voltages during the iterative process. The Jacobian matrix must not only be updated but also be triangularized in each iteration. This has a major effect on the calculation efficiency of the Newton– Raphson method. Many improvements of the Newton–Raphson method have focused on this problem. For instance, when the rectangular coordinate is adopted and the injected current (see (2.4)) is used to form the load flow equations [12], the off-diagonal elements of

88

2 Load Flow Analysis

the Jacobian matrix become constant. This property can certainly be used to improve the solution efficiency. Semlyen and de Leon [13] suggest that the Jacobian matrix elements can be updated partially to alleviate the computing burden. Both the above two forms of coordinate system are widely used in Newton– Raphson load flow algorithms. When the polar form is used, PV nodes can be conveniently treated. When the rectangular form is used, the calculation of trigonometric functions is avoided. Generally speaking, the difference is not very significant. A comparison between the two coordinate systems is carried out in [14]. The fast decoupled method is derived from the Newton–Raphson method in polar form. It will be discussed in Sect. 2.4. In the next section, we mainly introduce the Newton–Raphson method based on the correction equation of (2.54) in rectangular form.

2.3.3

Solution Process of Newton Method

In the Newton–Raphson method, the electric network is described by its admittance matrix. From (2.52), (2.53), (2.55), and (2.56) we know that all operations are relative to the admittance matrix. Therefore, forming the admittance matrix is the first step in the algorithm. The solving process of the Newton method roughly consists of the following steps. 1. Specify the initial guess values of node voltage, eð0Þ , f ð0Þ ; 2. Substituting eð0Þ , f ð0Þ into (2.52) and (2.53), obtain the left-hand term of the correction equation, DPð0Þ , DQð0Þ , and ðDV 2 Þð0Þ ; 3. Substituting eð0Þ , f ð0Þ into (2.55) and (2.56), obtain the coefficient matrix (Jacobian matrix) of the correction equation; 4. Solving (2.54), obtain the correction variables, Deð0Þ and Df ð0Þ ; 5. Modify voltages; eð1Þ ¼ eð0Þ Deð0Þ f ð1Þ ¼ f ð0Þ Df ð0Þ

) ð2:57Þ

6. Substituting eð1Þ and f ð1Þ into (2.52) and (2.53), obtain DPð1Þ , DQð1Þ , and ðDV 2 Þð1Þ ; 7. Check whether the iteration has converged. When it has converged, calculate branch load flow and output the results; otherwise take eð1Þ and f ð1Þ as the new guess value, return to step (3) and start the next iteration. The main flowchart of the Newton–Raphson method is shown in Fig. 2.3. The above steps introduce the main principles of the solution process. There are still many details to be clarified. As mentioned above, the solution procedure of the

2.3 Load Flow Solution by Newton Method

89

Newton–Raphson method is essentially the process of iteratively forming and solving the correction equations. Dealing with the correction equation has a crucial influence over the memory requirement and computing burden. This problem will be presented in the next section. First, we discuss some other important issues. The convergence characteristic of the Newton–Raphson method is excellent. Generally, it can converge in 6–7 iterations, and the number of iteration does not depend on the scale of the power system. Theoretically speaking, the Newton– Raphson method has a quadratic convergence characteristic if the initial guess values are close to the solution. If the initial guess values are not good enough, the iterative process may not converge or may converge to a solution at which the power system cannot operate. This property stems from the Newton method itself. As described above, the substance of the Newton method is sequential linearization of nonlinear equations. It is established on the assumption that De and Df are very small so that their high-order terms can be neglected. Therefore, a good initial guess value is crucial because the Newton method is very sensitive to it. Under normal operation states of power systems, the node voltage magnitudes are usually close to their nominal voltages, and the phase angle differences between the nodes of a branch are not very large. Therefore, a ‘‘flat start’’ initial guess value, i.e., ð0Þ

ei

¼ 1:0

ð0Þ

fi

¼ 0:0

ði ¼ 1; 2; . . . ; nÞ

ð2:58Þ

can give satisfactory results. In Fig. 2.3, the convergence condition is ðtÞ DP ; DQðtÞ < e

ð2:59Þ

where DPðtÞ ; DQðtÞ is a norm representing the maximal modulus elements in vectors DPðtÞ ; DQðtÞ . This convergence criterion is very intuitive, and can be used to directly control the power errors. When the calculation is based on the per unit system, we can set e ¼ 104 or 103 . If the base value is 100 MVA, the maximum error corresponds to 0.01 MVA or 0.1 MVA. From Fig. 2.3 we know that in the Newton–Raphson load flow calculation, the Jacobian matrix must be formed and triangularized in each iteration. Hence the computing burden in each iteration is quite heavy. From the expressions of Jacobian elements one can see that in the iteration procedure, especially when it is near convergence, the change of the elements caused by voltage variation is not significant (see Example 2.1). Therefore, to decrease the computing effort, once a Jacobian matrix is formed, it could be used in several successive iterations.

2.3.4

Solution of Correction Equations

The Newton–Raphson method, with Gauss elimination solving the correction equation, has been used in load flow calculation since the 1950s.

90

2 Load Flow Analysis Input data Form admittance matrix

Give voltage initial value e(0) and f (0) t=0 Calculate ΔP (t), ΔQ(t) and ΔV 2 (t) according to (2.52) and (2.53)

Is convergent?

Yes

Output results

No Solve the elements of Jacobian matrix according to (2.55) and (2.56)

Solve modified equation (2.54) to obtain Δe(t) and Δf (t) Modify voltage on each node according to (2.57)

t = t+1

Fig. 2.3 Flowchart of Newton method

In the 1960s, the sparsity of the correction equation was fully investigated and employed in the iteration procedure. In this way, the storage and operation for zero elements in the Jacobian are avoided. When the technology of optimal node ordering is adopted, it can minimize the number of the fill-in nonzero elements in factorizing the Jacobian of the correction equation. This greatly reduces memory and computing requirements to almost proportional to the node number of the power system. Based on this sparsity technology, the Newton–Raphson method has become one of the most popular methods in power system load flow calculation [7]. With a simple system as shown in Fig. 2.4, we now illustrate some algorithmic tricks in solving the correction equation of the Newton–Raphson method. In Fig. 2.4, both node 3 and node 6 are generator nodes. We set node 3 as a PV node while node 6 the slack node; other nodes are all PQ nodes. The structure of the network admittance matrix is shown in Fig. 2.5. The correction equation is given as (2.60). It does not include the equation related to node 6, the slack node.

2.3 Load Flow Solution by Newton Method Fig. 2.4 Example of simple system

91

3

2 1

4

5

6

Y11 Y12 Y13 Y14

Fig. 2.5 Structure of admittance matrix

Y21 Y22

Y26

Y31

Y33 Y34

Y41

Y43 Y44 Y45 Y54 Y55 Y56 Y62

2

3 2 DP1 H11 6 DQ1 7 6 J11 6 7 6 6 DP2 7 6 H21 6 7 6 6 DQ2 7 6 J21 6 7 6 6 DP3 7 6 H31 6 27¼6 6 DV 7 6 0 6 37 6 6 DP4 7 6 H41 6 7 6 6 DQ4 7 6 J41 6 7 6 4 DP5 5 4 DQ5

N11 L11 N21 L21 N31 0 N41 L41

H12 J12 H22 J22

N12 H13 L12 J13 N22 L22 H33 R33 H43 J43

32

N13 H14 N14 L13 J14 L14 N33 H34 S33 0 N43 H44 L43 J44 H54 J54

N34 0 N44 L44 N54 L54

Y65 Y66

H45 J45 H55 J55

3 De1 76 Df1 7 76 7 76 De2 7 76 7 76 Df2 7 76 7 76 De3 7 76 7 76 Df3 7 ð2:60Þ 76 7 6 7 N45 7 76 De4 7 7 6 L45 76 Df4 7 7 N55 54 De5 5 L55 Df5

where the constant terms DPi , DQi can be obtained by (2.52), X

X

9 ðGij fj þ Bij ej Þ > > = j2i j2i X X DQi ¼ Qis fi ðGij ej Bij fj Þ þ ei ðGij fj þ Bij ej Þ > > ; DPi ¼ Pis ei

j2i

ðGij ej Bij fj Þ fi

j2i

or can be written as DPi ¼ Pis ðei ai þ fi bi Þ DQi ¼ Qis ðfi ai ei bi Þ

) ð2:61Þ

92

2 Load Flow Analysis

From (2.56) we know the diagonal elements of the Jacobian are 9 @DPi > Hii ¼ ¼ ai ðGii ei þ Bii fi Þ > > > @ei > > > > @DPi > > Nii ¼ ¼ bi þ ðBii ei Gii fi Þ > = @fi > @DQi > Jii ¼ ¼ bi þ ðBii ei Gii fi Þ > > > @ei > > > > > @DQi ; Lii ¼ ¼ ai þ ðGii ei þ Bii fi Þ > @fi

ð2:62Þ

Both (2.61) and (2.62) include components of the injected current at node i, ai and bi . To calculate DPi , DQi , and the diagonal elements of Jacobian Hii , Nii , Jii , Lii , we must first compute ai and bi . From (2.11) we can see, the injected current components ai and bi at node i only depends on the i th row elements of the admittance matrix and voltage components of corresponding nodes. Therefore, ai and bi can be accumulated by sequentially taking the two terms and performing multiplication plus operation. After ai , bi are known, DPi and DQi can be easily obtained according to (2.61). The nondiagonal elements of the Jacobian in (2.60) can be expressed by: 9 @DPi > > ¼ ðGij ei þ Bij fi Þ > > @ej > > > > > @DPi > > Nij ¼ ¼ Bij ei Gij fi > = @fj > @DQi > Jij ¼ ¼ Bij ei Gij fi ¼ Nij > > > @ej > > > > > @DQi > Lij ¼ ¼ Gij ei þ Bij fi ¼ Hij > ; @fj Hij ¼

ð2:63Þ

Obviously, the off-diagonal elements are only related to the corresponding admittance elements and voltage components. From (2.62), the ith diagonal element consists of, besides the injecting current components at node i(ai and bi ), only the arithmetic operation results of the diagonal elements of admittance matrix Gii þ jBii and voltage components ei þ jfi . In brief, the whole correction equation can be formed by sequentially taking and arithmetically operating the elements of the admittance matrix and corresponding voltage components. If node i is PV node, the equation of DQi should be replaced by the equation of DVi2 . The constant term DVi2 on the left hand and elements Rii and Sii of the Jacobian can be easily obtained from (2.53) and (2.56), 9 @DVi2 > ¼ 2ei > = @ei > @DVi2 ; Sii ¼ ¼ 2fi > @fi Rii ¼

ð2:64Þ

2.3 Load Flow Solution by Newton Method

93

Forming the correction equation is a very important step in the Newton–Raphson method which remarkably affects the efficiency of the whole algorithm. Therefore, we should investigate the above equations carefully in coding the program. When Gauss elimination is used to solve the correction equation, we usually eliminate the correction equation row by row. The augmented matrix corresponding to (2.60) is 2

H11 6 J11 6 6 H21 6 6 J21 6 6 H31 6 6 0 6 6 H41 6 6 J41 6 4

N11 L11 N21 L21 N31 0 N41 L41

H12 J12 H22 J22

N12 L12 N22 L22

H13 J13

N13 L13

H14 J14

N14 L14

H33 R33 H43 J43

N33 S33 N43 L43

H34 0 H44 J44 H54 J54

N34 0 N44 L44 N54 L54

H45 J45 H55 J55

N45 L45 N55 L55

DP1 DQ1 DP2 DQ2 DP3 DV32 DP4 DQ4 DP5 DQ5

3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

After the equations related to node 1 and 2 are eliminated, the augmented matrix is converted as shown in Fig. 2.6. This figure tell us when the equations related to node 2 are eliminated (row 3 and row 4), all operations are independent of equations related to node 3, 4, . . ., N. Therefore, in the eliminating procedure, we can eliminate the rows related to a node immediately after forming them. 00 00 In Fig. 2.6, elements such as H23 ; N23 ; . . . ; L0024 , etc. are fill-in nonzero elements created in the elimination process. To decrease the number of injected elements, we should optimize the node number ordering before load flow calculation (see Section 1.3.5). The element with superscript (00 ) represents that it has been manipulated. We need not save memory for the fill-in element in advance using this elimination procedure and thus the algorithm is simplified. When the whole elimination procedure finished, the augmented matrix of correction equation becomes, 1

N11′ 1

H12′ J12′ 1

N12′ L12′ ′ N22 1

H31

N31

H41 J41

H13′ J13′ ′′ H23 ′′ J23

N13′ H14′ L13′ J14′ ′′ H24 ′′ N23 ′′ J24 ′′ L23

N14′ L14′ ′′ N 24 ′′ L24

ΔP1′ ΔQ1′ ΔP2′ ΔQ2′ ΔP3 ΔV32 N 45 ΔP4 L45 ΔQ4 N55 ΔP5 L55 ΔQ5

H34

N34

N41

H33 N33 R 33 S33 H43 N43

H44

N44

H45

L41

J43

J44 H54 J54

L44 N54 L54

J45 H55 J55

Fig. 2.6 Diagram of eliminating row by row

L43

94

2 Load Flow Analysis

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4

1

0 N11 1

0 H12 0 J12 1

0 N12 0 L12 0 N22 1

0 H13 0 J13 00 H23 00 J23 1

0 N13 0 L13 00 N23 00 L23 0 N33 1

0 H14 0 J14 00 H24 00 J24 0 H34 00 J34 1

0 N14 0 L14 00 N24 00 L24 0 N34 00 L34 0 N44 1

0 H45 0 J45 1

3 DP01 DQ01 7 7 DP02 7 7 DQ02 7 7 DP03 7 0 7 DV32 7 7 0 DP0 7 N45 4 7 L045 DQ04 7 7 0 DP0 5 N55 5 1 DQ05

Finally, using a normal backward substitution, one can get De1 ; Df1 ; . . . ; De5 ; Df5 from DP01 ; DQ01 ; . . . ; DQ05 . Following to the above discussion, we can summarize the algorithm via flowchart shown in Fig. 2.7, where R represents the slack node. The correction equation can be solved by the common Gauss elimination method. The above procedure adopts the strategy of eliminating the rows related to a node immediately after forming them. At the same time, the corresponding constant terms of the correction equation are also accumulated and eliminated. Thus the operation count per iteration is significantly reduced. [Example 2.1] Calculate the load flow of the power system shown in Fig. 2.8. [Solution] The load flow is calculated according to the procedures of the flowchart. The first step includes forming the admittance matrix and specifying the initial voltage values. From Example 1.1 we know the admittance matrix of this system is 2

1:37874 6 j6:29166 6 6 0:62402 6 6 þj3:90015 6 6 0:75471 Y¼6 6 þj2:64150 6 6 6 6 6 4

0:62402 þj3:90015 1:45390 j66:98082 0:82987 þj3:11203 0:00000 þj63:49206

0:75471 þj2:64150 0:82987 0:00000 þj3:11203 þj63:49206 1:58459 0:00000 j35:73786 þj31:74603 0:00000 j66:66667 0:00000 0:00000 þj31:74603 j33:33333

3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

The initial values of node voltages are given in Table 2.1. According to (2.52) and (2.53), we can establish the expression of the constant terms (mismatch terms) of the correction equations as

2.3 Load Flow Solution by Newton Method

95

Input Optimize node number Form admittance matrix Give initial value, and iterate by using successive iteration method t=1

i=1

t=t+1

>

i>n

>

Substituted backward and modify voltage

i=R

Form two-row equation relative to node i

Is convergent?

Eliminate the (2i 1)th and (2i) th equations by using the 1st to the 2(i 1)th equations

Output

No

Yes

i = i+1

Fig. 2.7 Flowchart of Newton Method 4

1:1.05

0.08+j0.30

2

j0.015

3

1.05:1 j0.25

j0.25

j0.03

P4 = 5 V4 =1.05

0.1+j0.35 0.04+j0.25

2+j1 j0.25

j0.25

3.7+j1.3

1

1.6+j0.8

Fig. 2.8 Simple power system

5

V5=1.05 θ5 = 0

96

2 Load Flow Analysis Table 2.1 Voltage initial values Node 1 2 1.00000 1.00000 eð0Þ 0.00000 0.00000 f ð0Þ

3 1.00000 0.00000

4 1.05000 0.00000

5 1.05000 0.00000

DP1 ¼ P1s e1 ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ f1 ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ DQ1 ¼ Q1s f1 ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ þ e1 ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ DP4 ¼ P4s e4 ½ðG42 e2 B42 f2 Þ þ ðG44 e4 B44 f4 Þ f4 ½ðG42 f2 þ B42 e2 Þ þ ðG44 f4 þ B44 e4 Þ 2 DV42 ¼ V4s ðe24 þ f42 Þ

Using (2.55) and (2.56), we can obtain the expressions of Jacobian matrix elements: @DP1 @e1 @DP1 @f1 @DP1 @e2 @DP1 @e3 @DQ1 @e1 @DQ1 @f1 @DQ1 @e2 @DQ1 @e3 @DP4 @e4 @DP4 @f4

¼ ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ G11 e1 B11 f1 ¼ ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ þ B11 e1 G11 f1 ¼ ðG12 e1 þ B12 f1 Þ; ¼ ðG13 e1 þ B13 f1 Þ;

@DP1 ¼ B12 e1 G12 f1 @f2 @DP1 ¼ B13 e1 G13 f1 @f3

¼ ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ þ B11 e1 G11 f1 ¼ ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ þ G11 e1 þ B11 f1 @DP1 ; @f2 @DP1 ¼ ; @f3 ¼

@DQ1 @DP1 ¼ @f2 @e2 @DQ1 @DP1 ¼ @f3 @e3

¼ ½ðG42 e2 B42 f2 Þ þ ðG44 e4 B44 f4 Þ G44 e4 B44 f4 ¼ ½ðG42 f2 þ B42 e2 Þ þ ðG44 f4 þ B44 e4 Þ þ B44 e4 G44 f4

2.3 Load Flow Solution by Newton Method

97

@DV42 ¼ 2e4 @e4 @DV42 ¼ 2f4 @f4 Thus according to (2.60), the correction equation of the first iteration can be written as 2

1:37874 6 6:04166 6 6 0:62402 6 6 3:90015 6 6 0:75471 6 6 2:64150 6 4 2

6:54166 1:37874 3:90015 0:62402 2:64150 0:75471

3 1:60000 6 0:55000 7 6 7 6 2:00000 7 6 7 6 5:69799 7 7 ¼6 6 3:70000 7 6 7 6 2:04901 7 6 7 4 5:00000 5 0:00000

0:62402 3:90015 1:45390 60:28283 0:82897 3:11203 0:00000 0:00000

3:90015 0:62402 73:67881 1:45390 3:11203 0:82897 66:66666 0:00000

0:75471 2:64150 0:82897 3:11203 1:58459 32:38884

2:64150 0:75471 3:11203 0:00000 0:82897 63:49206 39:98688 1:58459 0:00000 2:10000

32

3 De1 76 Df1 7 76 7 6 7 63:49206 7 76 De2 7 6 Df2 7 0:00000 7 76 7 76 De3 7 76 7 76 Df3 7 76 7 63:49206 54 De4 5 0:00000 Df4

the above equation, the elements in italic have maximal absolute value in each row of the Jacobian matrix. Obviously, if elements are arranged this way, the maximal elements do not appear at the diagonal positions. It should be noted that this situation is not accidental. From the above equation @DQi i we can conclude that the maximal element of each row is @DP @fi or @ei . This is because the active power is mainly related to the vertical component of voltage while the reactive power is mainly related to the horizontal component of voltage in high voltage power systems. To reduce the rounding error of the calculations, the maximal elements should be located in diagonal positions. There are two methods to satisfy this requirement: the first is to exchange positions of the equations relative to DQ and DP, i.e., to exchange odd numbered rows with even numbered rows; the second method is to exchange the variables De and Df , i.e., to exchange odd numbered columns with even numbered columns of the Jacobian matrix. We now introduce the first approach. Thus the above equation will be rearranged as,

98

2 Load Flow Analysis

2

6:04166 6 1:37874 6 6 3:90015 6 6 0:62402 6 6 2:64150 6 6 0:75471 6 4 2

1:37874 6:54166 0:62402 3:90015 0:75471 2:64150

3 0:55000 6 1:60000 7 6 7 6 5:69799 7 6 7 6 2:00000 7 7 ¼6 6 2:04901 7 6 7 6 3:70000 7 6 7 4 0:00000 5 5:00000

3:90015 0:62402 60:28283 1:45390 3:11203 0:82897

0:62402 3:90015 1:45390 73:67881 0:82897 3:11203

0:00000

66:66666

2:64150 0:75471 3:11203 0:82897 32:38884 1:58459

32 3 0:75471 De1 76 Df1 7 2:64150 76 7 6 7 0:82897 63:49206 0:00000 7 76 De2 7 6 Df2 7 3:11203 0:00000 63:49206 7 76 7 76 De3 7 1:58459 76 7 7 6 7 39:98688 76 Df3 7 2:10000 0:00000 54 De4 5 0:00000 63:49206 Df4

We can see the maximal element of each row appears in the diagonal position except for row 8. As described in Section 2.3.4, the iteration procedure adopts the strategy of immediately eliminating the rows related to a node after forming them (see Fig. 2.7). The equations related to node 1 are formed as 2

3 .. 4 6:04166 1:37874 3:90015 0:62402 2:64150 0:75471 0 0 . 0:55000 5 . 1:37874 6:54166 0:62402 3:90015 0:75471 2:64150 0 0 .. 1:60000 After the elimination operation is executed, the first and second row of the upper triangular matrix can be obtained: 2

3 .. 4 1:00000 0:22820 0:64554 0:10328 0:43721 0:12491 0 0 . 0:09103 5 . 1:00000 0:03879 0:58961 0:02215 0:41038 0 0 .. 0:21505 Then we establish the equations related to node 2, the corresponding augmented matrix is

3:90015 0:62402 60:28283 1:45390 3:11203 0:82987 63:49206 0:62402 3:90015 1:45390 73:67881 0:82987 3:11203 0:0

3 .. . 5:69799 5 . 63:49206 .. 2:0 0:0

Executing the elimination operation, the third and forth rows of the upper triangular matrix become:

2.3 Load Flow Solution by Newton Method

99

2

3 .. 4 1:00000 0:02090 0:08348 0:02090 1:09894 0:00000 . 0:09184 5 . 1:00000 0:01528 0:06609 0:01859 0:88943 .. 0:04253 Continuing this procedure until the eliminating operation procedure is finished, we have the upper triangular matrix: 2

.. . .. . .. . .. . .. . .. . .. . . 1:00000 ..

6 1:00000 0:22820 0:64554 0:10328 0:43721 0:12491 6 6 1:00000 0:03879 0:58961 0:02215 0:41038 6 6 1:00000 0:02090 0:08348 0:02090 1:09894 0:00000 6 6 6 1:00000 0:01528 0:06609 0:01850 0:88943 6 6 6 1:00000 0:03303 0:17246 0:03146 6 6 6 1:00000 0:02816 0:11194 6 6 1:00000 0:00000 4

3 0:09103 7 7 0:21505 7 7 7 0:09148 7 7 0:04253 7 7 7 0:07548 7 7 7 0:12021 7 7 7 0:00000 5 0:45748

After the backward substitution operation, the correcting increments of node voltages can be obtained, 2

3 2 3 De1 0:03356 6 Df1 7 6 0:03348 7 6 7 6 7 6 De2 7 6 0:10538 7 6 7 6 7 6 Df2 7 6 0:36070 7 6 7¼6 7 6 De3 7 6 0:05881 7 6 7 6 7 6 Df3 7 6 0:06900 7 6 7 6 7 4 De4 5 4 0:00000 5 Df4 0:45748 Modifying the node voltage, the voltage vector becomes: 2

3 2 3 e1 0:96643 6 f1 7 6 0:33481 7 6 7 6 7 6 e2 7 6 1:10533 7 6 7 6 7 6 f2 7 6 0:36070 7 6 7¼6 7 6 e3 7 6 1:05881 7 6 7 6 7 6 f3 7 6 0:66900 7 6 7 6 7 4 e4 5 4 1:05000 5 f4 0:45748 Using this voltage vector as the initial voltage value, we can repeat above operations. If the tolerance is set to e ¼ 106 , the calculation converges after five iterations. The evolution process of node voltages and power mismatches is shown in Tables 2.2 and 2.3.

100 Table 2.2 Iterating No. 1 2 3 4 5

Table 2.3 Iterating No. 1 2 3 4 5

2 Load Flow Analysis Node voltages in iterative process e1

f1

e2

f2

e3

f3

e4

f4

0.96643 0.87365 0.85947 0.85915 0.85915

0.33481 0.07006 0.07176 0.07182 0.07182

1.10538 1.03350 1.02608 1.02600 1.02600

0.36074 0.32886 0.33047 0.33047 0.33047

1.05881 1.03564 1.03355 1.03351 1.03351

0.06900 0.07694 0.07737 0.07738 0.07738

1.05000 0.97694 0.97464 0.97461 0.97461

0.45748 0.38919 0.39061 0.39067 0.39067

Node power mismatches in iterative process DQ1 0.55000 0.07263 0.02569 0.00078 0.00000

DP1 DQ2 DP2 DQ3 1.60000 5.69799# 2.00000 2.04901 0.03473 6.00881# 2.10426 0.37144 0.06011 0.41159# 0.15764 0.00924 0.00032 0.0030# 0.00054 0.00002 0.00000 0.00000 0.00000 0.00000

DP3 3.70000 0.04904 0.00329 0.00000 0.00000

DP4 5.00000 2.39001 0.16193 0.00069 0.00000

Power error

101 100 10−1 10−2 10−3 10−4

Fig. 2.9 Convergence property of Newton–Raphson method

1

2

3

4

5

6

7 Iterations

To reveal the convergence property, the maximal power mismatches (with # in Table 2.3) in the iterative process are shown in Fig. 2.9. In the iteration process, especially when it approaches convergence, the changes of the diagonal elements in the Jacobian are not very significant. To illustrate this point, the changes of the diagonal elements are given in Table 2.4. The calculation results of node voltages are shown in Table 2.5.

2.4 Fast Decoupled Method

101

Table 2.4 Diagonal elements of Jacobian matrix in iterative process @DP1 @DP2 @DP3 @DQ3 @DQ1 @DQ2 Iterating no. @f1 @f2 @f3 @e1 @e2 @e3 1 6.04166 6.54166 60.28283 73.67881 32.38884 39.08688 2 5.22590 6.84268 79.81886 69.30868 36.62734 38.83341 3 4.37415 6.42613 69.78933 69.61682 35.38612 38.39351 4 4.23077 6.38634 68.89682 69.52026 35.29706 38.33158 5 4.22720 6.38577 68.88900 69.51747 35.29572 38.33048

Table 2.5 Node voltage vectors Node Magnitude 1 0.86215 2 1.07791 3 1.03641 4 1.05000 5 1.05000

2.4 2.4.1

@DV42 @e4 1.05000 0.96259 0.97528 0.97463 0.97461

@DP4 @f4 63.49206 70.18293 65.61929 65.14834 65.14332

Angle ( ) 4.77851 17.85353 4.28193 21.84332 0.00000

Fast Decoupled Method Introduction to Fast Decoupled Method

The basic idea of the fast decoupled method is expressing the nodal power as a function of voltages in polar form; separately solving the active and reactive power equations [9] by using active power mismatch to modify voltage angle and using reactive power mismatch to modify voltage magnitude. In this way, the computing burden of load flow calculation is alleviated significantly. In the following, the derivation of the fast decoupled method from the Newton method is discussed. As described previously, the core of the Newton load flow approach is to solve the correction equation. When the nodal power equation is expressed in polar form, the correction equation is (see (2.50)),

DP H ¼ DQ J

N L

Du DV=V

ð2:65Þ

or can be written as, DP ¼ HDu þ NDV=V DQ ¼ JDu þ LDV=V

ð2:66Þ

This equation is derived strictly from the mathematical viewpoint. It does not take the characteristics of power systems into consideration. We know that in high voltage power system the active power flow is mainly related to the angle of the nodal voltage phasor while reactive power flow is mainly

102

2 Load Flow Analysis

related to its magnitude. The experiences of many load flow calculations tell us that the element values of matrix N and J in (2.66) are usually relatively small. Therefore, the first step to simplify the Newton method is to neglect N and J, and (2.66) is simplified to )

DP ¼ HDu

ð2:67Þ

DQ ¼ LDV=V

Thus a simultaneous linear equation of dimension 2n is simplified to two simultaneous linear equations of dimension n. The second important step to simplify the Newton method is to approximate the coefficient matrices of (2.67) as constant and symmetric matrices. As the phase angle difference across a transmission line usually is not very large (does not exceed 10 20 ), so the following relations hold, )

cos yij 1

ð2:68Þ

Gij sin yij Bij

Furthermore, the admittance BLi corresponding to the node reactive power is certainly far smaller than the imaginary part of the node self-admittance, i.e., BLi ¼

Qi Bii Vi2

Accordingly, Qi Vi2 Bii

ð2:69Þ

Based on the above relationships, the element expressions of coefficient matrix in (2.67) can be represented as (see (2.41), (2.42), (2.48), and (2.49)): 9 Hii ¼ Vi2 Bii > > > > Hij ¼ Vi Vj Bij = Lii ¼ Vi2 Bii Lij ¼ Vi Vj Bij

ð2:70Þ

> > > > ;

Therefore, the coefficient matrix in (2.67) can be written as 2

V12 B11 6 V2 V1 B21 6 H¼L¼6 4 Vn V1 Bn1

V1 V2 B12 V22 B22 .. .

Vn V2 Bn2

3 . . . V1 Vn B1n . . . V2 Vn B2n 7 7 7 5 ...

Vn2 Bnn

ð2:71Þ

2.4 Fast Decoupled Method

103

It can be further represented as the product of the following matrices: 2 6 6 H¼L¼6 4

V1

32

32 3 B11 B12 . . . B1n V1 76 B21 B22 . . . B2n 76 7 0 V2 0 76 76 7 76 76 7 ð2:72Þ .. .. .. .. 5 4 5 4 5 . 0 . . . Vn Bn1 Bn2 . . . Bnn Vn

V2 0

Substituting (2.72) into (2.67), we can rewrite the correction equations as follows: 2

3 2 DP1 V1 6 DP2 7 6 6 7 6 6 .. 7 ¼ 6 4 . 5 4

32

V2

0 ..

0

DPn

B11 76 B21 76 76 54 Vn Bn1

.

B12 B22 .. . Bn2

... ... .. . ...

32 3 B1n V1 Dy1 6 7 B2n 7 76 V2 Dy2 7 76 .. 7 54 . 5

ð2:73Þ

Vn Dyn

Bnn

and 2

3 2 DQ1 V1 6 DQ2 7 6 6 7 6 6 .. 7 ¼ 6 4 . 5 4

32

V2 0

DQn

0 ..

B11 76 B21 76 76 54 Vn Bn1

.

B12 B22 .. .

Bn2

32 3 B1n DV1 6 7 B2n 7 76 DV2 7 76 .. 7 54 . 5

... ... .. . ...

ð2:74Þ

DVn

Bnn

Multiplying both sides of the above equation with matrix, 2 6 6 6 4

V1

31 V2

..

.

Vn

2

6 7 6 7 7 ¼6 6 5 4

1 V1

3 1 V2

..

7 7 7 7 5

. 1 Vn

one can obtain 2

3 2 DP1 =V1 B11 6 DP2 =V2 7 6 B21 6 7 6 6 7¼6 .. 4 5 4 . DPn =Vn

Bn1

B12 B22 .. .

Bn2

... ... .. . ...

32 3 B1n V1 Dy1 6 7 B2n 7 76 V2 Dy2 7 76 .. 7 54 . 5 Vn Dyn Bnn

ð2:75Þ

and 2

3 2 DQ1 =V1 B11 6 DQ2 =V2 7 6 B21 6 7 6 6 7¼6 .. 4 5 4 . DQn =Vn

Bn1

B12 B22 .. .

Bn2

... ... .. . ...

32 3 B1n DV1 6 7 B2n 7 76 DV2 7 76 .. 7 54 . 5 Bnn

DVn

ð2:76Þ

104

2 Load Flow Analysis

The above two equations are the correction equations of the fast decoupled load flow method. The coefficient matrix is merely the imaginary part of the nodal admittance matrix of the system, and is thus a symmetric, constant matrix. Combining with the power mismatch equation (2.13), we obtain the basic equations of the fast decoupled load flow model DPi ¼ Pis Vi

X

Vj ðGij cos yij þ Bij sin yij Þ

ði ¼ 1; 2; . . . ; nÞ

ð2:77Þ

Vj ðGij sin yij Bij cos yij Þ ði ¼ 1; 2; . . . ; nÞ

ð2:78Þ

j2i

DQi ¼ Qis Vi

X j2i

The iterative process can be briefly summarized in the following steps: ð0Þ 1. Specify node voltage vector initial value yð0Þ i , Vi 2. Calculate the node active power mismatch DPi according to (2.77), and then calculate DPi =Vi 3. Solving correction equation (2.75), calculate the node voltage angle correction Dyi 4. Modify the node voltage angle yi ðtÞ

ðt1Þ

yi ¼ yi

ðt1Þ

Dyi

ð2:79Þ

5. Calculate node reactive power mismatch DQi according to (2.78), and then calculate DQi =Vi 6. Solving correction equation (2.76), calculate the node voltage magnitude correction DVi , 7. Modify the node voltage magnitude Vi ; ðtÞ

ðt1Þ

Vi ¼ Vi

ðt1Þ

DVi

ð2:80Þ

8. Back to step (2) to continue the iterative process, until all node power mismatches DPi and DQi satisfy convergence conditions.

2.4.2

Correction Equations of Fast Decoupled Method

The main difference between the fast decoupled method and the Newton method stems from their correction equations. Comparing with correction (2.40) or (2.54) of the Newton method, the two correction equations of the fast decoupled method have the following features: 1. Equations (2.75) and (2.76) are two simultaneous linear equations of dimension n instead of a simultaneous linear equation of dimension 2n

2.4 Fast Decoupled Method

105

2. In (2.75) and (2.76), all elements of the coefficient matrix remain constant during the iterative process 3. In (2.75) and (2.76), the coefficient matrix is symmetric. The benefit of the first feature for computing speed and storage is obvious. The second feature alleviates the computing burden in forming and eliminating the Jacobian within the iterative process. We can first form the factor table for the coefficient matrix of the correction equation (see (2.76)) by triangularization. Then we can carry out elimination and backward substitution operations for different constant terms DP=V and DQ=V through repeatedly using the factor table. In this way, the correction equation can be solved very quickly. The third feature can further improve efficiency in forming and storing the factor table. All the simplifications adopted by the fast decoupled method only affect the structure of the correction equation. In other words, they only affect the iteration process, but do not affect the final results. The fast decoupled method and the Newton method use the same mathematical model of (2.13), if adopting the same convergence criteria we should expect the same accuracy of results. It seems that (2.75) and (2.76) derived above have the same coefficient matrix, but in practice the coefficient matrixes of the two correction equations in the fast decoupled algorithms are different. We can simply write them as DP=V ¼ B0 VDu DQ=V ¼ B00 DV

ð2:81Þ ð2:82Þ

Here V is a diagonal matrix with the diagonal elements being the node voltage magnitudes. First, we should point out that the dimensions of B0 and B00 are different. The dimension of B0 is n 1 while the dimension of B00 is lower than n 1. This is because (2.82) dose not include the equations related to PV nodes. Hence if the system has r PV nodes, then the dimension of B00 should be n r 1. To improve the convergence, we use different methods to treat B0 and B00 , and how we treat B0 and B00 will result in different fast decoupled methods, are not merely the imaginary part of the admittance matrix. As described above, (2.81) and (2.82) are the correction equations based on a series of simplifications. Equation (2.81) modifies the voltage phase angles according to the active power mismatch; (2.82) modifies the voltage magnitudes according to the reactive power mismatch. To speed up convergence, the factors that have no or less effect on the voltage angle should be removed from B0 . Therefore, we use the imaginary part of admittance to form B0 without considering the effects of shunt capacitor and transformer’s off-nominal taps. To be specific, the off-diagonal and diagonal elements of B0 can be calculated according to following equations: B0ij ¼

xij ; rij2 þ x2ij

B0ii ¼

X j2i

X xij ¼ B0ij rij2 þ x2ij j2i

where rij and xij is the resistance and reactance of branch ij, respectively.

ð2:83Þ

106

2 Load Flow Analysis

Theoretically, the factors that have less effect on voltage magnitude should be removed from B00 . For example, the effect of line resistance to B00 should be removed. Therefore, the off-diagonal and diagonal elements of B00 can be calculated according to the following equations: B00ij ¼

X1 X 1 1 ; B00ii ¼ bio B00ii ¼ bio xij x xij j2i ij j2i

ð2:84Þ

where bio is the shunt admittance of the grounding branch of node i. If B0 and B00 are formed according to (2.83) and (2.84), the fast decoupled method is usually called the BX algorithm. Another algorithm opposite to BX method is called the XB algorithm in which B0 used in the DP Dy iteration is formed according to (2.84), while B00 used in the DQ DV iteration is formed according to (2.83). Although these two algorithms have different correction equations, their convergence rates are almost the same. Several IEEE standard test systems have been calculated to compare the convergence of these algorithms. Table 2.6 shows the number of iterations needed to converge for these test systems. Many load flow calculations indicate that BX and XB methods can converge for most load flow problems for which the Newton method can converge. The authors of [9, 10] explain the implications of the simplifications made in the fast decoupled method. Wong et al. [19] propose a robust fast decoupled algorithm to especially treat the possible convergence problem caused by high r=x networks. Bacher and Tinney [26] adopt the sparse vector technique to improve the efficiency of the fast decoupled method. From the above discussion we know that the fast decoupled method uses different correction equations to the Newton method, hence the convergence properties are also different. Mathematically speaking, the iteration method based on a fixed coefficient matrix to solve a nonlinear equation belongs to ‘‘the constant slope method.’’ Its convergence process has the characteristic of the geometric series. If the iteration procedure is plotted on a logarithmic coordinate, the convergence characteristic is nearly a straight line. In contrast, convergence of the Newton method has a quadratic property and is quite similar to a parabola. Fig. 2.10 shows the typical convergence properties of the two methods. Figure 2.10 illustrates that the Newton method converges slower at the early stages, but once converged to some degree its convergence speed becomes very fast. The fast decoupled method converges almost at the same speed throughout the iteration procedure. If the specified convergence criterion is smaller than the errors Table 2.6 Convergence comparison of BX method and XB method Systems Newton BX XB IEEE-5 bus 4 10 10 IEEE-30 bus 3 5 5 IEEE-57 bus 3 6 6 IEEE-118 bus 3 6 7

2.4 Fast Decoupled Method

107

Power error 1

Newton Method

1e-1 P Q Decoupled Method 1e-2

A 1e-3

1e-4

1e-5

5

10

15 Iterations

Fig. 2.10 Convergence properties of fast decoupled method and Newton method

at point A in Fig. 2.10, the iteration number of the fast decoupled method is larger than that of the Newton method. It can be roughly considered that a linear relation exists between the iteration number and the required precision when using the fast decoupled method. Although the iteration number of the fast decoupled method is larger, its computing requirement in each iteration is far less than that of the Newton method. So the computing speed of the fast decoupled method is much higher than the Newton method.

2.4.3

Flowchart of Fast Decoupled Method

The principle flowchart of the fast decoupled method is shown in Fig. 2.11 which illustrates the main procedure and logical structure of the load flow calculation. The symbols used in Fig. 2.11 are first introduced below: t: counter for the iteration number K01 a flag with ‘‘0’’ and ‘‘1’’ states, ‘‘0’’ indicates the active power iteration; while ‘‘1’’ the reactive power iteration. A whole iteration includes an active power iteration and a reactive power iteration.

108

2 Load Flow Analysis

Input information and original data, and deal with original data

1 2

Form admittance matrix

Calculate coefficient matrix B⬘, and form the first factor table

3 4

Calculate coefficient matrix B⬙, and form the second factor table

5 Give voltage initial value on each node

6

t = 0, k 01= 0

Calculate [ DW(K01 ) /V] ;ERM (K01)

Solve modified equation (2.81)or(2. 82) , and modify V K 01

No

false

Yes

0: PðBÞ

ð3:7Þ

Several important formulas can be deduced according to conditional probability. 1. Multiplication probability theorem. Let A1, A2, . . ., An be n arbitrary events, the probability of their intersection set is PðA1 \ A2 \ \ An Þ ¼ PðA1 ÞPðA2 jA1 ÞP½A3 jðA1 \ A2 Þ P½An jðA1 \ A2 \ \ An1 Þ:

ð3:8Þ

However, when A1, A2, . . ., An are independent, we have PðA1 \ A2 \ \ An Þ ¼ PðA1 ÞPðA2 Þ PðAn Þ:

ð3:9Þ

2. Formula of total probability. Let event A occur according to the given condition of events B1, B2, . . ., Bn. A can only occur at the same time as one of B1, B2, . . .,

132

3 Stochastic Security Analysis of Electrical Power Systems

Bn occurs, and any two of Bi are mutually exclusive, but Ptheir union sets consist of the sample space of one event, that is, Bi Bj ¼ ’ði 6¼ jÞ; ni¼1 Bi ¼ O; PðBi Þ > 0, then the total probability of event A, P(A), is PðAÞ ¼

n X

PðBi ÞPðA=Bi Þ:

ð3:10Þ

i¼1

3. Bayes’ Formula. Assume the occurring condition of event Bi (i = 1,2,. . ., n) is same as that in (2), then the probability of occurrence of event Bi after the event A occurred, is denoted by PðBi =AÞ ¼

PðBi ÞPðA=Bi Þ ði ¼ 1; 2; . . .Þ: n P PðBi ÞPðA=Bi Þ

ð3:11Þ

i¼1

Equation (3.11) is Bayes’ Formula. It means that once event A occurred in experiment, (3.11) is used to reassess the cause Bi, so the probability P(Bi/A) is called posterior probability.

3.2.2

Random Variable and its Distribution

If the outcome of a random experiment can be described by one numerical variable, and this numerical value is determined by a certain probability, then the variable is named a random variable. In mathematical terms, it can be described that the set O of all sample points e is one sample space in a random experiment, and X is a realvalued function defined on the sample space, that is, e 2 O; XðeÞ 2 R: If there exist real values a < b, such that the set of sample points satisfies feja XðeÞ bg; then this set is an event, and the function X(e) is referred to as a random variable. If a = 1, event {e| ‐ 1 X(e) b} can be described by {X b} for short. Its probability measurement, FðxÞ ¼ PðX xÞ

ð3:12Þ

is defined as the distribution function of random variable X. x can be any given real value. The general random variable X can be classified into a discrete random variable and a continuous random variable according to its different possible values.

3.2 Basic Concepts of Probability Theory

133

For continuous random variables, another function to express its probability is the probability density function f(x), which is defined by, f ðxÞ ¼ lim

Dx!0

1 Pðx < X < x þ DxÞ; Dx

ð3:13Þ

which can also written in incremental format, Pðx < X < x þ DxÞ f ðxÞDx:

ð3:14Þ

Formula (3.14) can be interpreted as the probability under the condition that random variable X is in the interval (x, x + Dx) and Dx ! 0. Obviously, the probability of random variable X between a and b is, Zb Pða < X bÞ ¼

f ðxÞdx

ð3:15Þ

a

and the relationship between (3.15) and distribution function F(x) in formula (3.12) can be written as, Zx FðxÞ ¼

f ðxÞdx

ð3:16Þ

1

and f ðxÞ ¼

dFðxÞ : dx

ð3:17Þ

For a discrete random variable (as shown in Fig. 3.1), X may be xi (i = 1, 2, . . ., n), then its probability density function is defined as pðxÞ ¼

PðX ¼ xi Þ 0

x ¼ xi x 6¼ xi

ð3:18Þ

and the distribution function is FðxÞ ¼

X

pðxi Þ:

ð3:19Þ

xi x

3.2.3

Numeral Characteristics of Random Variable

In many practical problems, we can specify the characteristics of random variables by finding the average value of random variables and the degree of value dispersion. The two most commonly used methods are introduced as follows.

134

3 Stochastic Security Analysis of Electrical Power Systems

1.0 0.25 0.8 0.6

0.15

F(x)

P(x)

0.20

0.10

0.4

0.05

0.2

0.00

0.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13

a

0 1 2 3 4 5 6 7 8 9 10 1112 13

b

x

x

Fig. 3.1 The relative function of discrete random variable (a) probability density function; (b) distribution function

3.2.3.1

Mathematical Expectation (Mean Value)

Discrete random variable X can be x1, x2, . . ., and its corresponding probability is PðX ¼ xi Þ ¼ pi

i ¼ 1; 2; . . .

Then mathematical expectation or expectation, E(X), is defined as EðXÞ ¼

1 X

xi pi :

ð3:20Þ

i¼1

For a continuous random variable X, when its density function is f(x), we have Z1 EðXÞ ¼

xf ðxÞdx:

ð3:21Þ

1

For the mathematical expectation of a set of random variables Xi (i = 1, 2, . . ., n), there are characteristics such as described as follows E

n X

! Xi

i¼1

3.2.3.2

¼

n X

EðXi Þ:

ð3:22Þ

i¼1

Variance

Discrete random variable X is denoted as s2, which is defined by, s2 ¼

n X i¼1

ðxi mÞ2 pi ;

ð3:23Þ

3.2 Basic Concepts of Probability Theory

135

where m = E(X), that is average value. Obviously, s2 represents the degree of dispersion of its value deviating from the average value m. For a continuous random variable X, we get, Z1 ðx mÞ2 f ðxÞdx:

s ¼ 2

ð3:24Þ

1

Some properties and applications related to other numerical characteristics of random variables will be discussed in Sect. 3.5.1.

3.2.4

Convolution of Random Variable

Suppose two random variables X and Y are independent, and they have probability density functions f1(x) and f2(y), respectively, then Z = X + Y is still a random variable. The probability density function of Z is Z1 fðzÞ ¼

f ðx; z xÞdx 1 Z1

¼

f1 ðxÞf2 ðz xÞdx:

ð3:25Þ

1

Its distribution function is Zz

Z1

FðzÞ ¼

f1 ðxÞf2 ðz xÞdx dz: 1

ð3:26Þ

1

If X and Y are discrete random variables, then the distribution function is FðzÞ ¼

X

PðX ¼ xi ; Y ¼ yj Þ ¼

xi þyj C0i4 yq2 ¼ C0i4 ll yq2 ; > = l

i¼

l

1; 2;

l ¼ 1; 2; . . . ; nc :

ð5:102Þ

> > ;

We can see that 2nc among the 5nc elements of C20 yq are nonzero. 4. Building the vectors A1DX in (5.98). From (5.85) we have A1 DX ¼ ¼

0

0

0

0

A21

A22 0

0

0

A21 DVd þ A22 DId

0 0

:

DVTd

DITd

DKTT

DWT

DFT

T ð5:103Þ

294

5 HVDC and FACTS

Equations (5.66) and (5.67) yield ðA21 DVd þ A22 DId Þk ¼ ðIdk DVdk þ Vdk DIdk Þ;

k ¼ 1; 2; . . . ; nc :

ð5:104Þ

We have introduced the integrated iteration power flow calculations of interconnected systems.

5.3.5

Alternating Iteration for AC/DC Interconnected Systems

The alternating iteration method is a further simplification to the P–Q decoupled method in the integrated iteration power flow calculations. Based on the basic converter (5.52) and (5.53), the impact of AC systems on DC systems relies on the primary voltages Vt of converter transformers. If the AC voltages Vt of all converters in a multiterminal DC systems are known, the DC system will have (5.59)–(5.61), (5.63), and (5.64), a total of 5nc equations and 5nc unknown variables. We can obtain the 5nc unknown DC variables by solving only the DC system equations. The power taken out of, or injected into, AC systems, Pidc þ jQidc from converter transformers, represents the impact of DC systems on AC systems. If the power withdrawn from or injected into AC systems is known, power flow calculations of AC systems are not affected by DC systems. The ideal process is to designate the primary voltages of nc converter transformers ð0Þ

Vt

h Þ ¼ Vnð0a þ1

ð0 Þ

Vna þ2

...

i ð0Þ Vna þnc :

Obtain solution of DC variables X(0). Substituting X(0) into (5.57) yields the power of all converters Pdc(0) and Qdc(0). Using converter power in AC system equations forh conventional i power flow calculations gives rise to convergent solution Vð1Þ ¼ Vða1Þ ð1Þ

ð1Þ

Vt

ð1Þ

. Ideally the calculation completes if Vt ð0Þ

ð0Þ

equals Vt .

Generally Vt is not the same as Vt . The calculation is an iteration process. Based on the above, AC and DC system equations are separately solved in alternating iterations. When solving AC system equations, we use the known power at the DC buses to represent the corresponding DC systems. While solving DC system equations, we model AC systems as constant voltages at the AC buses of converters. At each iteration, the solution of the AC systems provides the converter AC bus voltages for the next DC iteration; the solution of DC systems in turn produces the equivalent real and reactive power of converters for the further AC iteration. The iteration goes on and on until convergence is achieved. We must point out that the convergence of this method is mathematically related to the Gauss– Seidel iteration. In fact, the alternating iteration is not a complete Gauss–Seidel iteration. For the AC system equations in the alternating iteration method, we usually use the Newton algorithm or P–Q decoupled algorithm. For DC system equations, the Newton algorithm is the most common approach [114]. The Gauss– Seidel algorithm applies only to the coupling between AC and DC equations.

5.3 Power Flow Calculation of AC/DC Interconnected Systems

295

Start Input data Build factor tables of B⬘ and B ⬙ Set initial values Compute DD from Eqs. 5.59

5.61, 5.63 and 5.64; Y

|| DD || kTmax, take Vds kTmax/kTworst as new voltage control value; otherwise take Vds kTmin/kTworst as the new voltage control value. Return to Step 1. The above steps are the main steps of this method. Its basic characteristics are the simplicity in theory and in programming. Comparing with Newton’s iteration for DC equations, it saves considerable memory. When assuming the converter transformer ratio is a continuous variable and with no over-limits, DC and AC system power flow solution can be attained in one computation. With necessary amendment to the above method, it can be applied to power flow calculations for fixed control angle control, discrete transformer ratio changes, etc. In which case, iteration is required. The details can be found in [115]. In the above we have discussed the two major types of power flow calculation methods for AC/DC interconnected systems. Integrated iteration takes into consideration the complete coupling between AC and DC systems, and has good convergence for various conditions of network and system operation. The Jacobian matrix has a higher order than for pure AC systems. The approach requires more programming, uses more memory, and needs longer computation time. Alternating iteration can be accomplished by adding DC modules to the existing power flow programs, due to its separated solution of AC and DC equations. It is easier to take into consideration the constraints on DC variables and the adjustment of operation modes. However, the convergence of alternating iteration is not as good as integrated iteration. The computational practice indicates that its convergence is good when the AC system is strong. If the AC system is weak, its convergence deteriorates, requiring more iterations or even becoming nonconvergent. This is the shortcoming of the alternating iteration method. The strength of AC systems is related to the rated capacity of converters. Taking the converter rated power PdcN as the base, the reciprocal of per unit equivalent reactance of AC system, as viewed from the AC bus of converters, is called the short-circuit ratio (SCR). The larger the SCR, the higher is the system strength. A weak AC system (SCR less than 3) has a larger equivalent reactance, making the AC bus voltage of the converter very sensitive to variation of the reactive power injection. Alternating iteration separates the solution of AC and DC equations, assuming constant Vt and Qtdc at the boundary between AC and DC systems to neglect their coupling. If the AC system is weak, the variation of Qtdc can bring potential change to Vt. This results in computational oscillation between Qtdc and Vt in alternating iterations and worsening convergence [116]. There have been some improved calculation methods [117] for alternating iteration to make it applicable to weak AC systems. We are not going to discuss them here for brevity.

5.4 HVDC Dynamic Mathematical Models

5.4

299

HVDC Dynamic Mathematical Models

We have introduced the steady-state models of HVDC systems in the previous section. The HVDC transients are quite complicated. The main causes of the complexity are as follows. (1) The firing pulses of bridge valves are triggered at discrete time points. In transients, the firing angle is regulated by the controller to make the corresponding time unevenly distributed. Thus the firing angle is a discrete variable with regards to computation. (2) We assume that AC systems are symmetric in steady-state analysis. From the steady-state analysis of converter valves we know that the valve on/off states are closely related to the commutating voltages, the time of firing, and the magnitudes of commutation angles. When firing angles or commutation angles are too large, commutation may fail. In transient states, AC systems are actually unsymmetrical. Some valves could have negative valve voltage and could not been turned on when firing pulses occur if commutation voltages are severe unsymmetrical. For HVDC systems under transient states, we need to establish derivative equations to take into consideration the variations of commutation voltages and firing angles as well as other exceptional conditions. The solution of these derivative equations reveals the time of valve state changes. (3) We should consider the distributed characteristics of DC lines for long distance transmission. Under such circumstances, the variations of voltage and current on DC lines become wave processes. Due to the above conditions, we need to solve ordinary differential equations and partial differential equations with both continuous and discrete variables to calculate accurately the transients of HVDC transmission systems. From the mathematical point of view, it is not difficult to solve these equations. Many previous works [118– 120] used detailed mathematical models resulting in huge computational requirements. We should simplify the transient DC models as much as possible without losing engineering accuracy. In general we can take a simple DC model for stability analysis, if AC systems are relatively strong; otherwise a detailed DC model is required. The general assumptions that we make in deriving DC steady-state models still apply to most analysis of power system stability. Thus we can use the steadystate mathematical models of converters (5.52), (5.54) as their dynamic models. Here we are going to introduce the mathematical models of control systems. The controllers in HVDC systems consist of electronic circuits. Their basic working principles are as follows: receiving control inputs, sending outputs to phase-control circuits, and pulse generation device to set converter firing angles in order to control converter operation. Different control signals and different control strategies result in different controller structures and control characteristics, as well as the dynamics of DC systems or even the whole power system. To achieve better operation characteristics, the adjustments of rectifiers and inverters should be coordinated. As stated before, the basic control mode is fixed current or fixed power for rectifiers, and fixed voltage or fixed extinction angle for inverters. The transformer ratio adjustment is slow and is a discrete variable. The ratio is not changed

300

5 HVDC and FACTS Idref Id

1 1 + sTc3

x1

−

kc1

+

+

PI regulator

−

aref +

a

+ a2

kc2

Measuring unit

a1

sTc2

a Fixed Current Control Pref 1 − 1 + sTp2 x2

Pd

+

Idref kp1 1+ sTp1

+ x3

kc1

a1

+

+

− Id

+ a2

kc2 sTc2

−

aref +

α

b Fixed Power Control Vdref Vd

1 1 + sTv3

− x4

k v1

+

b1

+ + kv2

−

bref +

b

b2

sTv2

c Fixed Voltage Control mref μ

1 1 + sTm3

x5

−

km1

+

+ km2

b1

+

bref +

b

+ b 2

sTm2

d Fixed Extinction Angle Control

Fig. 5.18 Transfer function block diagrams of HVDC control systems

very often and is an ancillary means to optimize the converter operation point. Figure 5.18 shows the transfer function of the four basic control modes. The transfer function of fixed current control is shown in Fig. 5.18a. It compares Id, the output of DC current, and given current Idref. The difference is amplified and goes through a proportional plus integral process. Then the signal is passed to the phase-shift control circuits to change the converter firing angle and to enforce the fixed current function. The transfer function of fixed power control is shown in Fig. 5.18b. HVDC systems are usually required to transport power as planned. Fixed power is a basic control method. When the variations of AC voltages on both terminals are not large,

5.5 Basic Principles and Mathematical Models of FACTS

301

using fixed current and fixed extinction angle can actually achieve fixed power control. When taking into consideration AC voltage fluctuations, using fixed current and fixed voltage can obtain exact fixed power control. These two control methods are to determine the DC current setting based on the given power and DC operation voltage at the control terminal. However DC voltage is related to DC current, so it is very difficult to set DC current beforehand. To overcome this problem, special control devices are set up for fixed power control. Fixed current control has high response speed, is capable of quickly constraining overcurrent to prevent converter overload, and is easy to set up. Power control devices are usually based on fixed current control and receive additional inputs rather than directly acting on phase control circuits. In the diagram, the DC power is compared with its target value. The difference is amplified and sent to the input of the fixed current controller. This works by changing the current setting of the fixed current control dynamically. The transfer functions of fixed voltage and fixed extinction angle controls are shown in Fig. 5.18c, d. They share the same structure as Fig. 5.18a with different parameters. We need to point out that extinction angles cannot be directly measured. They are indirectly obtained by measuring the time interval between valve voltage and current zero-crossing points. Although we do not show the quantity limitation block in the above diagrams, attention has to be paid to the constraints on various physical variables. There are minimum firing angle constraints for rectifier fixed current control, minimum extinction angle constraints for inverter fixed voltage control, etc. We need to notice that the controllers here are all for DC internal adjustments. DC systems can be used to affect AC system operation through these DC internal adjustments. The inputs of controllers may include AC system operation parameters, line power, the velocity of some generators, system frequency, and so on. This kind of control is the integrated control of AC/DC systems, also called external adjustments. The control strategy and control signals in these cases are an important field of power system research.

5.5

Basic Principles and Mathematical Models of FACTS

After the introduction of the FACTS concept, many FACTS devices have been proposed. We can classify them into three groups based on the maturity of the technology. The first group has been applied in the power industry, such as static VAr compensators (SVR), thyristor controlled series capacitor (TCSC), and static synchronous compensators (STATCOM). The second group has industrial sample machines and is still under investigation, such as unified power flow controller (UPFC). The third group has only a theoretical design without any industrial application, such as static synchronous series compensator (SSSC), thyristor controlled phase shifting transformer (TCPST). We will introduce their basic principles and mathematical models in this section. The power flow calculation for systems having these devices will be discussed in the next section.

302

5 HVDC and FACTS

FACTS devices can be classified based on their connection types as series, shunt, and combined types. SVC and STATCOM are shunt type. TCSC and SSSC are series type. TCPST and UPFC are combined type. Designed by US Electrical Power Research Institute (EPRI), manufactured by Westinghouse, and installed at AEP power system in USA for industrial testing operation, UPFC is the most powerful FACTS device proposed as of today. Its control strategy is presently under further research.

5.5.1

Basic Principle and Mathematical Model of SVC

A common practice of system voltage adjustment is shunt reactive power compensation. The synchronous condenser was historically an important tool of shunt reactive power compensation. Since it is a rotating machine, its operation and maintenance are quite complicated. New synchronous condensers are now seldom installed. The static shunt reactive power compensation, as opposed to the rotating synchronous condenser, has wide industrial application due to its low cost and simple operation and maintenance. Conventional static shunt reactive power compensation is to install capacitors, reactors, or their combination, at the compensated buses to inject or extract reactive power from the system. Mechanical switches are used to put the shunt capacitor/reactors into or out of operation. There are three disadvantages in this type of compensation. First, their adjustment is discrete. Second, their control actions are slow and cannot meet system dynamic requirements. Third, they have negative voltage characteristics. When system voltages drop (rise), the reactive power injection of shunt capacitors decreases (increases). However, they are widely applied in power systems due to their economic advantages and easy maintenance. Modern SVR with FACTS technology integrate power electronic elements into conventional static shunt reactive power compensation devices to achieve fast and continuously smooth adjustment. Ideal SVCs can maintain nearly constant voltages at the compensated buses. The good steady and dynamic characteristics render them widely applicable. Their basic elements are thyristor controlled reactors (TCRs) and thyristor switched capacitors. It is not difficult to understand other types of SVCs if we know the working principles of these two. Figure 5.19 shows their basic diagrams. To save cost, most SVCs connect to systems through step-down transformers. The valve control of the SVC produces harmonics. Filters are installed with SVCs to reduce harmonic contamination. They are capacitive as regards to fundamental frequency and inject reactive power into systems. Figure 5.20a, b shows TCR and TSC branches. Below we will analyze the control theory of TCR and TSC. TCR branch consists of reactors connected with two back-to-back thyristors as control elements. The system voltage on the branch is sinusoidal and shown in Fig. 5.21a. The valve delayed firing angle is a 2 [p/2, p]. The firing time is ot ¼ a þ kp

k ¼ 0; 1; 2; . . . :

5.5 Basic Principles and Mathematical Models of FACTS

303

High-voltage bus Step-down transformer

L

C1

TCR

C2

C3 TSC

Filter

Fig. 5.19 SVC basic diagram iC

iL

C

L VmSin wt

VmSin wt

TCR

TSC

Fig. 5.20 TCR and TSC branches

Apparently the inductor current is zero when the two valves are off. When the valve conducts, neglecting the resistance in the reactor, the inductor current is L

diL ¼ Vm sin ot; dt

ð5:113Þ

where L is the inductance of the reactor, Vm is the magnitude of the system voltage. Its general solution is iL ¼ K

Vm cos ot; oL

ð5:114Þ

where K is the integral constant. Since the inductor current is zero at firing, the above equation yields iL ¼ K

Vm cosða þ kpÞ ¼ 0: oL

304

5 HVDC and FACTS

Substituting the solution of K into (5.114) gives rise to the inductor current. iL ¼

Vm ½cosða þ kpÞ cos ot oL

k ¼ 0; 1; 2; . . . :

ð5:115Þ

Based on the above equation, inductor current returns to zero at ot ¼ (k þ 2)p a. Thus the valve conducting period is ot 2 ½kp þ a; ðk þ 2Þp a k ¼ 0; 1; 2; . . . : The waveform of inductor current is shown in Fig. 5.21b. The width of a single ripple of inductor current is ðk þ 2Þp a ðkp þ aÞ ¼ 2ðp aÞ ¼ 2b: b ¼ p a is called the conducting angle. To make sure that there is always one valve conducting at any moment, we should have ðk þ 2Þp a ¼ ðk þ 1Þp þ a;

k ¼ 0; 1; 2; . . . :

One valve should conduct the moment and another one is turned off, so a ¼ p/2. This operation mode corresponds to connecting the shunt reactor directly to the system. From the waveforms we can see that the valve conducting period decreases from p to zero as the firing angle rises from p/2 to p. Now the two valves are turned off at all times, corresponding to reactors out of service. Besides when a is less than V VmSinωt ωt α

a iL iL π- α α

b

π

2π-α

3π-α 2π

3π 2π+α

3π+α 4π-α

Fundamental component

Fig. 5.21 (a) TCR voltage waveforms (b) TCR current waveforms

ωt

5π-α 4π

iL1

5.5 Basic Principles and Mathematical Models of FACTS

305

p/2, the moment at which the current of a conducting valve returns to zero is later than the firing moment of the off valve as ðk þ 2Þp a > ðk þ 1Þp þ a: In this case, the conducting valve has not been turned off when the other valve receives a firing pulse. The off valve cannot be triggered on due to zero valve voltage. One of the two valves is off at any moment. Thus the main component of inductor current is DC. The normal operating ranges of TCR firing angles are a 2 [p/2, p]. Based on (5.115) and the waveforms, the current passing through the reactor is irregular and no longer sinusoidal due to valve control. The adjustment of firing angles changes the current peak values and conducting periods. Applying Fourier analysis to the current yields the magnitude of the fundamental frequency component

IL1

2 ¼ p

2pa Z

a

Vm Vm ðcos a cos yÞ cos ydy ¼ ½2ða pÞ sin 2a: oL poL

And the instantaneous value of fundamental frequency component is iL1 ¼ IL1 cos ot ¼

Vm ð2b sin 2bÞ sinðot p=2Þ: poL

ð5:116Þ

The equivalent fundamental frequency reactance of the TCR branch is XL ðbÞ ¼

h pi poL b 2 0; : 2b sin 2b 2

ð5:117Þ

Thus the TCR equivalent reactance of fundamental frequency components is the function of conducting angle b or the firing angle a. The control of firing angle a can smoothly adjust the equivalent shunt reactance. The reactive power consumed by TCR is V2 2b sin 2b 2 QL ¼ V_ I_L1 ¼ ¼ V : XL ðbÞ poL

ð5:118Þ

As shown in Fig. 5.20b, the TSC branch consists of a capacitor connected in series with two thyristors connected in parallel and in opposite directions. The TSC source voltage is the same as TCR. Its waveforms are in Fig. 5.21a. The TSC creates two operating states for the capacitors through valve control: shunt capacitors in service or out of service. Stopping the firing can simply put the capacitor out of service. Note that the natural switch-off from conduction happens when the capacitor

306

5 HVDC and FACTS

current is zero and its voltage at the peak of source voltage. Neglecting the capacitor leakage current, capacitor voltage maintains the peak value if firing stops after the natural switch-off. We need to pay attention to the timing of putting the capacitor into service. The principle is to reduce the impulse current in capacitors at the moment of in-service operation. We should use the correct valve based on the sign of the capacitor initial voltage, and put the capacitor into service at the moment when source voltage equals capacitor initial voltage. So the transient component of capacitor current is zero when put into service. After capacitors are in service, we need a ¼ p/2 to keep one valve conducting at all times. Ideally the capacitor voltage is the peak of source voltage. Using a ¼ p/2 makes no transients for the inservice operation. In reality, the source voltage and the capacitor initial voltage cannot be exactly the same. There is a small inductor in the TSC branch to reduce the possible impulse current. From the above analysis, we can see that the main difference between TSC and mechanically switched capacitors (MSC) is the fast control of in-service or out-of-service operation by valves in TSC. TSC dynamic characteristics can meet system control demands. The reactive power injection of the capacitors is QC ¼ oCV 2 ;

ð5:119Þ

where C is the capacitance of the capacitor. From (5.118) and (5.119) we have the reactive power injection from the SVC is QSVC ¼ QC QL ¼

2b sin 2b 2 oC V : poL

ð5:120Þ

The SVC reactive power injection can be smoothly adjusted when b 2 [0,p/2]. To expand the regulation ranges of SVC, we can have many TSC branches in one SVC, based on the compensation requirements. Figure 5.19 shows an SVC with three TSCs. When all three TSCs are in service, the C in (5.120) is C1 þ C2 þ C3. To guarantee a continuous adjustment, the TCR capacity should be slightly larger than a group of TSCs, that is, oC1 < 1/oL. Based on (5.120), the equivalent reactance of SVC is

XSVC

2b sin 2b 1 poL ¼ oC ¼ : poL 2b sin 2b po2 LC

ð5:121Þ

The SVC equivalent voltage–current characteristics are the combination of TCR and TSC. As b increases from zero to p/2, XSVC will change from capacitive maximum to inductive maximum. Generally, the control signal of SVC is derived from the voltage of the bus to which they are connected. Figure 5.22 shows that as the voltage V changes, the SVC equivalent reactance varies with b. In Fig. 5.22, there is a straight line going through the original corresponding to every b. The slope of the straight line is XSVC. Suppose that the system voltage

5.5 Basic Principles and Mathematical Models of FACTS Fig. 5.22 Equivalent reactance variation with b as voltage changes

307 b5

b6 = 0

b4 V

b3

b2 b = p/2 1

V2

A

V4 V6

V1 V3

B

V5

C

ISVC

Fig. 5.23 Voltage–current characteristic

V XSVGmin = −1/wC Vref B

b1 =

p/2

(

XSVGma = (w L) 1 − w 2LC

)

b6 = 0

ISVC

characteristic is V1. The control scheme is to make the TCR conducting angle b1 ¼ p/2, corresponding to maximum equivalent inductive reactance. The SVC operating point is the crossover point A between system voltage characteristic V1 and the straight line b1. With system voltage characteristic V2 and TCR conduction angle b2 < b1, XSVC decreases and the SVC operating point shifts accordingly. Until system voltage characteristic is V6 and conduction angle b6 ¼ 0, SVC equivalent reactance is maximum capacitive with operating point B. Apparently, voltage at B is higher than at C. When voltage changes between V1 and V6, the adjustment of b puts voltage under control. All the operating points constitute the straight line AB. The slope of AB and the crossover point with voltage axis Vref is determined by the control scheme of b. From voltage control point of view, the slope of AB is zero at best, without steady-state error. To maintain the control stability, SVC should have a small steady-state error and the slope of AB is around 0.05. Taking into consideration the steady-state control scheme, the SVC voltage– current characteristics are shown in Fig. 5.23. When system voltage varies within the SVC control range, SVC can be seen as a synchronous condenser having source voltage of Vref and internal reactance of Xe. V ¼ Vref þ Xe ISVC ;

ð5:122Þ

where Xe is the slope of the straight line AB in Fig. 5.23, V and ISVC are the SVC terminal voltage and current. When system voltage is out of the SVC control range, SVC becomes a fixed reactor, XSVCmin or XSVCmax.

308

5 HVDC and FACTS

SVC is considered as a variable shunt reactor in system stability and control analysis. SVC controller determines its admittance. Reference [122] provided the controller block diagram. We have introduced SVC basic principles. Special attention needs to be paid in industrial applications of SVC to capacity settings of reactors and capacitors, control strategy, flexibility of adjustments, protection, elimination of harmonics, etc. For example, in practical operation of an SVC, the range of the control angle is slightly less than [p/2, p] to make sure that valves can be triggered on and turned off securely.

5.5.2

Basic Principle and Mathematical Model of STATCOM

A STATCOM is also called an advanced static Var generator (ASVG). Its function is basically same as SVC with wider operation ranges and faster responses. As stated before, the control element of SVC is a thyristor, a semi-controllable element that can only be turned off when valve current crosses zero. STATCOM is made of fully controllable elements. Gyugyi et al. [123] presented the basic principles of using gate turn off thyristors (GTOs) to build a STATCOM. As yet, there have been several samples STATCOM operated in real systems [124–126]. The basic connection of a STATCOM is shown in Fig. 5.24. Its control element is the fully controlled valve (GTO). The ideal GTO switch characteristic is that the valve is turned on under positive valve voltage with positive control current on its gate; valve is turned off with negative control current on its gate. Valve resistor is zero when it conducts, and is infinity when it is turned off. A GTO can manage the switch-off by gate control in comparison with the thyristor where switch-off is only possible at current zero-crossing. STATCOM in Fig. 5.24 is a voltage type selfcommutation full-bridge inverter according to power electronic theory. The capacitor DC voltage acts as an ideal DC voltage source to support the inverter. The regular diode connected in the opposite direction and parallel with the GTO is a path for continuous current, providing route for the feedback energy from the AC side. The inverter normal operation is to transfer the DC voltage into AC voltage having controllable magnitude and phase angle at the same frequency as the AC system. The sum of instantaneous power of a symmetric three-phase system is a

vc +

. .

−

Fig. 5.24 Circuit of STATCOM

. .

. .

ia ib V ic ASVC •

5.5 Basic Principles and Mathematical Models of FACTS

309

constant. Thus the reactive power exchanges periodically within phases instead of between source and load. There is no need to have an energy storage element on the DC side, if considering the inverter as a load. However, the interacting power among harmonics produces a small amount of reactive power exchange between the inverter and the system. The capacitor on the inverter DC side will provide both DC voltage and energy storage. The electrical energy stored in the capacitor is 1 W ¼ CVC2 : 2 If the above energy is not considered as energy support for AC systems during power system dynamic events, the value of capacitor C can be small while the reactive capacity provided by the STATCOM is much more than the stored energy. We will see later that the maximum reactive power capacity of a STATCOM depends on the inverter capacity. The STATCOM does not need large size reactors and capacitors as the SVC does. Generally, there are three output voltage control modes for voltage-type inverters: phase-shift adjustment, pulse-width modulation, and direct DC source voltage control. The DC voltage of STATCOM is the charged voltage of the capacitor, not a DC source. So phase-shift adjustment and pulse-width modulation, instead of direct DC source voltage control, are usually used in STATCOMs. For brevity, we are not going to discuss the inverter working principles in detail. The width of the output voltage square waves y is controlled by the GTO gates (the magnitude of voltage square waves is the DC voltage on the capacitor). By Fourier analysis, we have the fundamental frequency voltage on the AC side y VASVG ¼ KVC sin ; 2

ð5:123Þ

where K is a constant related to inverter structure; VC is the capacitor DC voltage; y is the control variable. STATCOM connection to the systems is shown in Fig. 5.25. It must connect to systems through reactors or transformers because the use of voltage-bridge circuits. The connection reactor is needed to link the two unequal voltage sources, STATCOM and AC system. Its other function is to suppress the high-order harmonics in the current. Its inductance does not need to be large. The reactor in the figure is the transformer equivalent leakage reactance or the connection reactor. The resistor is the equivalent copper loss of the transformer or STATCOM loss. STATCOM is represented as an ideal synchronous condenser. Using the system voltage as the

•

VASVG

•

•

Fig. 5.25 STATCOM connection to systems

C

I P

r + jx

QASVG

VS

310

5 HVDC and FACTS

reference vector, the fundamental frequency component of the inverter output pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ voltage is VASVG and lagged phase angle is d. With y ¼ 1= r2 þ x2 , a ¼ arctg r/x, we have the real power consumed by the inverter as 2 P ¼ Vs VASVG y sinðd þ aÞ VASVG y sin a:

ð5:124Þ

The reactive power injection from STATCOM is QASVG

VASVG ﬀ d Vs _ _ ¼ Im Vs I ¼ Im Vs r jx

¼ Vs VASVG y cosðd aÞ Vs2 y cos a:

ð5:125Þ

In steady state, the inverter neither consumes nor generates real power. Based on (5.124), making P zero yields VASVG ¼ Vs

sinðd þ aÞ : sin a

ð5:126Þ

Taking (5.126) into (5.125) and (5.123) yields QASVG ¼ VC ¼

Vs2 sin 2d; 2r

Vs sinðd þ aÞ : K sin a sinðy=2Þ

ð5:127Þ ð5:128Þ

From the above two equations, we know that the adjustment of phase angle d while maintaining constant pulse width y can change the output reactive power as well as the capacitor voltage. The simultaneous adjustment of y and d can maintain capacitor voltage and change the reactive power output. The vector diagram of STATCOM steady-state operation is shown in Fig. 5.26. We use the equivalent resistance r to represent the inverter real power loss so that the inverter model neither consumes nor generates real power. In the diagram, compensation current I_ is perpendicular to inverter output voltage V_ ASVG . The inverter injects reactive power into the system when I_ leads V_ASVG . Otherwise it consumes reactive power. While an SVC changes its equivalent inductance through adjusting the timing of its connection to the system, the STATCOM controls the magnitude and phase-angle of its output voltage. As shown in the vector diagram, the reactive power provided by STATCOM is QASVG ¼ IVs cos d:

ð5:129Þ

5.5 Basic Principles and Mathematical Models of FACTS

311 •

•

I

jxI

•

VASVG •

d

•

d

jxI

•

VS

rI •

•

VASVG

•

rI

a STATCOM Inject Reactive

VS

•

I

b STATCOM Consumes Re-active Power from System

Power into System

Fig. 5.26 STATCOM steady-state vector diagram

Fig. 5.27 STATCOM voltage adjustment

VASVG1

I1r I0r

jI1x

VASVG0

jI0x Vs1

I2r

Vs0

Vref

jI2x

Vs2 I1

I0

I2

I

Note that d is the angle by which vector V_ASVG lags V_ s . The positive sign corresponds to a greater than zero d; the negative sign to a less than zero d. Substituting the above into (5.127) yields the magnitude of compensation current as I¼

Vs sin d: r

ð5:130Þ

The phase angle of the compensation current is (p/2d) as shown in Fig. 5.26. The real and reactive power components of the compensation current are IP ¼ I cos

p V Vs s d ¼ sin2 d ¼ ð1 cos 2dÞ; 2 r 2r

p V s IQ ¼ I sin d ¼ sin 2d: 2 2r

ð5:131Þ ð5:132Þ

Figure 5.27 shows the system voltage adjusted by STATCOM. Voltage Vs0 is the voltage setting value Vref under STATCOM output voltage of VASVG0 and compensation current of I0. When system operating conditions vary and the bus voltage

312

5 HVDC and FACTS V

Fig. 5.28 STATCOM volt– ampere characteristics

Vref

I ICmax

ILmax

decreases, STATCOM increases d to inject more reactive power. The compensation current is I1 while the voltage is maintained as Vref. The STATCOM keeps system voltage constant through the adjustment of its control parameters. A practical STATCOM usually implements bus voltage mismatch control. From the above analysis, the operation characteristics of STATCOM are shown in Fig. 5.28, and approach rectangular. The constraints of maximum voltage and current are determined by the STATCOM capacity. Voltage setting is determined by the control scheme. Comparing with SVC inverse triangular operational characteristics, STATCOM has wider operation ranges. We must notice that only one of the two control variables of the STATCOM is independent. The adjustment of d will change both the magnitude and phase angle of the compensation current. The control variable y is constrained by (5.128). As d changes, y should vary accordingly to maintain a constant capacitor voltage. The range of d variation is very limited. When a STATCOM consumes reactive power from the system, V_ s lags V_ ASVG by d. We can see from the vector diagram in Fig. 5.26b that d is always less than a. The equivalent resistance r is much less than equivalent reactance x so that a is very small. When a STATCOM injects reactive power into the system, V_ s leads V_ ASVG by d. As seen in (5.130) a small r makes d less constrained by compensation current. Hence (5.132) indicates an approximately linear relationship between reactive compensation current and d. To neglect resistance r for approximate analysis, setting a and d to zero in (5.124) and (5.125) yields P ¼ 0;

QASVG ¼ Vs

VASVG Vs : x

Now the free control variable of STATCOM is y, and VASVG is determined by (5.123). If VASVG is greater than VS, the STATCOM injects reactive power into the system, otherwise it consumes reactive power. A STATCOM can be represented as a shunt connected, controllable current source as noted in (5.130) for power system stability and control analysis. The magnitude and phase angle are determined by the STATCOM controller.

5.5 Basic Principles and Mathematical Models of FACTS

5.5.3

313

Basic Principle and Mathematical Model of TCSC

TCSC can rapidly and continuously change the equivalent reactance of the compensated line, to maintain a constant power flow on the line within certain operating conditions. In system transients, the TCSC can increase system stability through its fast variation of line reactance. The earliest TCSC was first put into operation in USA in 1991. TCSC have many different structures. One of its basic formations is shown in Fig. 5.29. Figure 5.29 shows a fixed capacitor and a parallel connected TCR. Its control element is the thyristor. We have seen TCR utilization in the above analysis of SVC. Since SVC is shunt-connected, the voltage on TCR is considered to be sinusoidal. However, the current flowing through TCR is irregular due to valve control, as shown in Fig. 5.21b. The TCR in a TCSC operates in different conditions as compared to those in an SVC. Note that the TCSC is series connected in the transmission line. The current flowing through the TCSC, the line current, is sinusoidal, due to harmonic filtering requirements and to physical operating constraints. Hence the irregular current in the TCR due to valve control will generate a nonsinusoidal capacitor voltage. This is the main difference between the two. Below we introduce the TCSC equivalent reactance at fundamental frequency to understand its working and control principles. The reference directions of various physical variables are shown in Fig. 5.29. The line current is sinusoidal with the waveforms shown in Fig. 5.30a. i ¼ Im sin ot

ð5:133Þ

Suppose that the circuits are in steady state. When the valve conducts, we have the following equations based on circuit theory i ¼ iL þ i C ;

v¼L

diL ; dt

iC ¼ C

dv : dt

ð5:134Þ

From the above we have iL þ LC

d 2 iL ¼ Im sin ot: dt2

ð5:135Þ

This is a nonhomogeneous differential equation of the inductor current. Its particular solution is the steady-state solution of the second-order circuit as follows: i

v

iC C

Fig. 5.29 Basic structure of TCSC

iL

L

314

5 HVDC and FACTS

V

i = Imsinωt π 2

0

π

3π 2

5π 2

wt

a iL

0 α−π 2

π 2

π 2 +α

3π 2

5π − α 2 wt

3π − α 2

b Fig. 5.30 (a) TCSC line current and capacitor voltage waveforms (b) inductor current waveforms

9 > l2 ¼ D sin ot; D ¼ 2 Im = : lp ﬃﬃﬃﬃﬃﬃ1 > ; l ¼ o0 =o; o0 ¼ 1= LC

isL

ð5:136Þ

The complementary solution to homogeneous equation is ifL ¼ A cos o0 t þ B sin o0 t;

ð5:137Þ

where A and B are the undetermined coefficients. The general solution to (5.135) is iL ¼ A cos o0 t þ B sin o0 t þ D sin ot:

ð5:138Þ

Denote a as the firing angle and assume its value in [p/2, p]. A is the electrical angle from capacitor voltage crossing zero to the time of firing. Under steady state, the waveform of inductor current is symmetric to the time point of capacitor voltage crossing zero. The capacitor voltages at the moments of valve turning on and off are equal in magnitude and opposite in direction. Supposing that capacitor voltage magnitude is V0 when valves turn on and off, the corresponding electrical angles are p yk ¼ a þ kp; 2

dk ¼

3p a þ kp; 2

k ¼ 0; 1; 2; . . . :

ð5:139Þ

5.5 Basic Principles and Mathematical Models of FACTS

315

Based on the initial conditions of inductor current and capacitor voltage: inductor current of zero and capacitor voltage of V0 at turning on (refer to Fig. 5.30a), we have the following equations A cos lyk þ B sin lyk þ D sin yk ¼ 0; Lðo0 A sin lyk þ o0 B cos lyk þ oD cos yk Þ ¼ ð1Þk V0 :

ð5:140Þ ð5:141Þ

The capacitor voltage is V0 at the turning off time of the valve Lðo0 A sin ldk þ o0 B cos ldk þ oD cos dk Þ ¼ ð1Þkþ1 V0 :

ð5:142Þ

The solution to the above three equations yields sin yk 2k þ 1 cos l p ; cos lb 2

ð5:143Þ

sin yk 2k þ 1 B ¼ D sin l p ; cos lb 2

ð5:144Þ

V0 ¼ DLðo sin a þ o0 cos a tg lbÞ;

ð5:145Þ

A ¼ D

where b ¼ p a, is called the conducting angle having a value within [0, p/2]. Substituting A and B in (5.138) yields the inductor current when the valve conducts as

cos a p iL ¼ D sin ot þ ð1Þ cos l ot kp : cos lb 2 k

ð5:146Þ

We can obtain the capacitor voltage from (5.134) as cos a p v ¼ DL o cos ot ð1Þk o0 sin l ot kp : cos lb 2

ð5:147Þ

The conducting period is 3p p ot 2 a þ kp ; a þ kp ; 2 2

k ¼ 0; 1; 2; . . . :

The capacitor current iC ¼ i þ (iL). There are two components in capacitor current, one is the line current; the other has the same magnitude and opposite direction to the inductor current.

316

5 HVDC and FACTS

We have assumed that the firing angle is within [p/2, p]. The reason for such an assumption is the same as for the TCR in SVC. As seen from the waveforms of inductor current, to make one valve conduct at any time we have 3p p a þ kp ¼ þ a þ kp: 2 2 There is one valve turned on when the other is turned off, so a ¼ p/2. The inductor current, as indicated in (5.146), is iL ¼ D sin ot: This is the inductor current when the inductor connects directly with the capacitor in parallel. We usually call this bypass mode. When a increases from p/2 to p, the valve conducting period decreases from p to zero, corresponding to the turning-off of two valves at any moment. This is as if the inductor is not in operation, called off mode. Besides, if a is less than p/2, the time at which the current of the conducting valve crosses zero is later than the firing time of the other nonconducting valve. Thus 3p p a þ kp > a þ kp þ p: 2 2 In this case the nonconducting valve cannot be triggered on with a zero voltage across it at the time of firing since the conducting valve is not turned off. Thus one of the valves is always nonconducting at any time, making DC current the main component in inductor current. Under normal operation, the firing angle of TCR in TCSC has an operating range of [p/2, p]. When both valves are turned off ot 2

hp

i p a þ kp ; a þ kp ; 2 2

k ¼ 0; 1; 2; . . . :

The inductor current is zero while the capacitor current is the line current. The capacitor voltage is C

dv ¼ Im sin ot: dt

The solution is v¼K

Im cos ot: oC

ð5:148Þ

5.5 Basic Principles and Mathematical Models of FACTS

317

We know from (5.147) that the absolute value of capacitor voltage when the valve conducts is V0. Take ot ¼ a p/2 þ kp as the moment of valve turns on in the above, so ð1Þk V0 ¼ K

Im p cos a þ kp : oC 2

Obtaining integral constant K and substituting into (5.148) yields capacitor voltage as v ¼ ð1Þk

Im Im sin a þ V0 cos ot: oC oC

ð5:149Þ

Equations (5.147) and (5.149) give the capacitor voltages when valves are turned on and off, respectively. Apparently, the capacitor voltage is not sinusoidal when a 6¼ p/2. Figure 5.30a, b shows the waveforms of capacitor voltage and inductor current. The Fourier analysis of nonsinusoidal capacitor voltage provides the fundamental frequency component V1 ¼

Im Im sin a þ V0 cos y cos y dy oC oC 0

Z 3p=2a 2 cos a p þ DL o cos ot o0 sin l ot cos y dy : ð5:150Þ p ap=2 cos lb 2 Z 2 p Im Im þ V0 sin a cos y cos y dy p 3p=2a oC oC

2 p

Z

ap=2

The integral of the first item above equals the integral of the third. The sum of the two is F1 þ F3 ¼

4 Im V0 cos a ð2a p þ sin 2aÞ : p 4oC

ð5:151Þ

Taking into account (5.136), the integral of the second item is

2 o 2o0 2 F2 ¼ DL ob þ sin 2a 2 ðl tg a þ tg lbÞ cos a : p 2 l 1

ð5:152Þ

Substituting (5.145) into (5.151) and rearranging yields the fundamental frequency reactance of TCSC XTCSC ¼

V1 F1 þ F3 þ F2 ¼ ¼ Kb XC ; Im Im

ð5:153Þ

318

5 HVDC and FACTS

Fig. 5.31 Kb b curve

where XC ¼ 1=oC;

2 l2 2 cos2 b sin 2b Kb ¼ 1 þ ðltg lb tg bÞ b : p l2 1 l2 1 2

ð5:154Þ ð5:155Þ

As shown in (5.155), the adjustment of the firing angle changes the reactance XTCSC that is series connected in the line, rendering a controllable equivalent line reactance. The valve control scheme is predefined. TCSC ideal dynamic responses can allow the transmission line capacity to reach its thermal limit. TCSC usually has oL < 1/oC and l2 around 7 to reduce its cost. Figure 5.31 shows the Kb b curve at l ¼ 3. When b 2 [0, p/2l], Kb is greater than zero and TCSC is capacitive. When b 2 [p/2l, p/2], Kb is less than zero and TCSC is inductive. In the off mode, b ¼ 0, Kb ¼ 1. In by-pass mode, b ! p/2, Kb ! 1/ (1 l2). When b ! p/2l, Kb ! 1 due to tglb ! 1, corresponding to parallel LC resonance. To prevent TCSC resonance over voltage, b is prohibited from being operated near p/2l. TCSC shown in Fig. 5.29 is a single module. A practical TCSC usually consists of many modules connected in series. Each module has its independent firing angle. The firing angle combination of different modules gives the TCSC equivalent reactance a wider range of variation and smoother adjustment. To protect the TCSC from damage due to overvoltages and overcurrents, there are various protection devices installed and corresponding operation constraints [127]. For power system stability and control analysis, a TCSC can be represented as a variable reactor series connected in the transmission line. The reactance is determined by the TCSC controller.

5.5 Basic Principles and Mathematical Models of FACTS

5.5.4

319

Basic Principle and Mathematical Model of SSSC

TCSC is a series compensation device using semi-controllable power electronic elements. There are many types of series compensation with fully controllable elements. Here we are going to introduce the SSSC built with GTO voltage-type inverters. The STATCOM discussed before uses voltage-type inverters. Connected to systems in parallel via reactors or transformers; the SSSC employs voltage-type inverters connected in series in a transmission line through transformers. Neglecting the line-ground branches, the basic connection is shown in Fig. 5.32, where r þ jx is the line impedance. Note that the inverter is different to STATCOM as a DC source may be present on the DC side. With a DC source, SSSC can provide reactive power compensation as well as real power compensation to AC systems. When an SSSC only supplies or consumes reactive power, the capacity of its DC source can be small or even zero (the SSSC loss being provided by the AC system). We know (from the introduction of the STATCOM) that the magnitude and phase angle of inverter output AC voltages are controllable. Hence we can consider the voltage of an SSSC, connected in series on a line, as an approximately ideal voltage source, as shown in Fig. 5.33a. Denote VSSSC the voltage magnitude of the ideal voltage source and d the voltage leading phase angle regarding voltage at bus l. The vector diagram is shown in Fig. 5.33b, where ’ is the leading phase angle of voltage at bus l with regards to line current. Apparently r + jx

Fig. 5.32 SSSC basic connection

i

Voltage source inverter

DC power source Vl ′ l

a

I

VSSSC

Vl ′

r+jx

j b

I

Fig. 5.33 (a) SSSC equivalent circuit (b) SSSC vector diagram

VSSSC d

Vl

j

320

5 HVDC and FACTS

V_ l0 ¼ V_ l þ V_ SSSC :

ð5:156Þ

For pure reactive power compensation, the inverter vector V_ SSSC is perpendicular to line current I_ d þ ’ ¼ p=2:

ð5:157Þ

In this way, SSSC corresponds to a reactor connected in series on a transmission line, denoting XSSSC its equivalent reactance, so V_ l V_l0 ¼ jXSSSC I_ ¼ V_ SSSC ;

ð5:158Þ

XSSSC ¼ VSSSC =I:

When V_ SSSC leads I,_ it is capacitive with negative sign; otherwise it is inductive with positive sign. Note that in the above equation, VSSSC is not related to line current and is controlled by the inverter. Hence the adjustment of VSSSC can change the equivalent reactance. In system analysis, once XSSSC is given, the line current can be determined by V_ l V_ m : r þ jðx þ XSSSC Þ

ð5:159Þ

9 VSSSC ¼ IjXSSSC j = : p d ¼ ’; 2

ð5:160Þ

I_ ¼ Thus

When XSSSC is less than zero, use positive sign; otherwise use negative sign. Generally, the source branch in Fig. 5.33a can be represented as a current source and impedance connected in parallel as shown in Fig. 5.34a by Norton’s theorem. The current source is I_c ¼ V_SSSC =ðr þ jxÞ:

r+jx

l

Ic a

l

Plc+jQlc b

ð5:161Þ

r+jx

m

Pmc+jQmc

Fig. 5.34 (a) Use equivalent current source (b) use equivalent power injection

5.5 Basic Principles and Mathematical Models of FACTS

321

In power system analysis, the bus power injection is used in most cases, further simplifying Fig. 5.34a, b. As indicated in (5.161) 9 _ SSSC > V > > Plc þ jQlc ¼ ¼ V_l = r þ jx : V_ SSSC > > > Pmc þ jQmc ¼ V_ m I_c ¼ V_ m ; r þ jx V_ l I_c

Note that the phase angle of V_SSSC is yl þ d, so Plc ¼ Vl VSSSC ðb sin d g cos dÞ

)

Qlc ¼ Vl VSSSC ðg sin d þ b cos dÞ

;

Pmc ¼ Vm VSSSC ½g cosðylm þ dÞ b sinðylm þ dÞ Qmc ¼ Vm VSSSC ½b cosðylm þ dÞ þ g sinðylm þ dÞ

ð5:162Þ ) ;

ð5:163Þ

where 9 r > > > r 2 þ x2 > = x : b¼ 2 > r þ x2 > > > ylm ¼ yl ylm ; g¼

ð5:164Þ

The power generated by SSSC is PSSSC þ jQSSSC ¼ V_ SSSC I_ ¼ V_SSSC

_ VSSSC þ V_l V_m ; r þ jx

2 PSSSC ¼ gVSSSC þ gVSSSC ½Vl cos d Vm cosðylm þ dÞ þ bVSSSC ½Vl sin d Vm sinðylm þ dÞ,

ð5:165Þ 2 QSSSC ¼ bVSSSC þ gVSSSC ½Vl sin d Vm sinðylm þ dÞ

bVSSSC ½Vl cos d Vm cosðylm þ dÞ:

ð5:166Þ

Apparently PSSSC is zero for pure reactive power compensation. Neglecting line resistance, d satisfies the following for pure reactive power compensation Vl sin d ¼ Vm sinðylm þ dÞ:

ð5:167Þ

322

5 HVDC and FACTS

In (5.166), the reactive power compensated by SSSC is not related to the line current directly since the adjustment of VSSSC is not related to the line current. The power flows from bus l to bus m is: Plm þ jQlm ¼ V_ l I_ ¼ V_ l

V_SSSC þ V_l V_ m r þ jx

;

Plm ¼ gVl2 þ gVl ½VSSSC cos d Vm cos ylm bVl ½VSSSC sin d þ Vm sin ylm Qlm ¼ bVl2 gVl ½VSSSC sin d þ Vm sin ylm bVl ½VSSSC cos d Vm cos ylm

) :

ð5:168Þ So the power on the line is controlled by two parameters. For pure reactive power compensation, SSSC has only one independent control variable due to the constraint of (5.167) and has one control objective. In power system stability and control analysis, SSSC can also be represented as a voltage source connected in series in the line. The controller determines the magnitude and phase angle. The voltage vector is always perpendicular to line current for pure reactive power compensation.

5.5.5

Basic Principle and Mathematical Model of TCPST

Thyristor controlled phase shifting transformer is abridged as TCPST. Phase shifters using mechanical switches to change the tap positions have been utilized in power systems for a long time. It is also called a series voltage booster. Since the response speeds of mechanical switches are slow in tap changing, this type of shifter can only be used in power system steady-state adjustment. Furthermore, the short operational life is a major drawback of this type of shifter. Substituting mechanical switches with thyristors can provide the phase shifter with faster responses and wider application. There are many types of implementation [130, 131]. We are going to use a relative simple type to introduce the working principles and mathematical models. Figure 5.35 shows a basic connection of TCPST. Phase shifters consist of parallel transformer (ET), series transformer (BT), and switches. Parallel and series transformers are also called excitation transformer and boosting transformer. Figure 5.35 shows only phase c of the secondary side of parallel transformer and secondary and the primary side of the series transformer. The other two phases have the same structure. Switch S is made up of a pair of thyristors connected in parallel in opposite directions, having the same working principles as discussed in TCSC. S1–S5 can only have one conducting and all others are turned off under all circumstances. We can see that the ratio of the parallel transformer varies with the conducting conditions of S1–S4. When S1–S4 are all turned off, S5 must conduct to short-circuit the primary of the series transformer.

5.5 Basic Principles and Mathematical Models of FACTS

a b c

VE

323

I2a

*

I1b

I2b

*

I1c

I2c

I1a

I3b *a VEa1

x

VEb1

*

I3c

y

*

b VEc1

I3a c

z

s1 s2

IBC1

*

BT

*

IEC2

s3

ET

VP

s5 VBC1

s4 VEC2

Fig. 5.35 TCPST basic configuration

This is to prevent series connection of the series transformer excitation reactance into the transmission line. Notice the relationship between the primary voltage of the parallel transformer and the line phase voltage. Phase a, b, and c on the primary of the parallel transformer correspond to phase b, c, and a of the line voltage, respectively. Since the parallel transformer has D connection, the relationship between the primary voltages of the parallel transformer and the line phase voltages are 9 V_ Ea1 ¼ V_ Eb V_ Ec > = ð5:169Þ V_ Eb1 ¼ V_ Ec V_ Ea : > ; _ _ _ VEc1 ¼ VEa VEb Supposing that the ratios of parallel and series transformers are kE and kB, respectively, and neglecting the voltage loss of transformers, the phase voltages on the secondary of the parallel transformer have the following relationship with the line phase voltages 9 pﬃﬃﬃ V_Ea2 ¼ kE V_ Eb V_Ec = 3 ¼ jkE V_ Ea > > = pﬃﬃﬃ _ _ _ _ VEb2 ¼ kE VEc VEa = 3 ¼ jkE VEb : > pﬃﬃﬃ > V_ Ec2 ¼ kE V_ Ea V_ Eb = 3 ¼ jkE V_Ec ;

ð5:170Þ

The phase voltages on the secondary of the series transformer are 9 V_ Ba2 ¼ kB V_Ba1 ¼ kB V_ Ea2 ¼ jkB kE V_ Ea > = V_ Bb2 ¼ kB V_Bb1 ¼ kB V_ Eb2 ¼ jkB kE V_ Eb : > ; V_ Bc2 ¼ kB V_ Bc1 ¼ kB V_ Ec2 ¼ jkB kE V_Ec

ð5:171Þ

324

5 HVDC and FACTS VBa VPa

I 3a

VEa I2a

I1a

j

j

VPb VEc VBc

j VPc

j

VBb VEb

I1c I3c

j

j

I2c

I1b

I2b I3b

Fig. 5.36 Phase shifter vector diagram

Using a single phase expression to replace the above three-phase we have V_ B ¼ jkB kE V_ E ;

ð5:172Þ

where V_ E and V_B are input voltage of the parallel transformer and output voltage of the series transformer. Similarly we can obtain the expression for the currents I_3 ¼ jkB kE I_2 :

ð5:173Þ

The vector diagrams are shown in Fig. 5.36. From (5.172), (5.173), and Fig. 5.35 we can obtain V_ P ¼ V_ E þ V_B ¼ ð1 þ jkB kE ÞV_ E ;

ð5:174Þ

I_1 ¼ I_2 þ I_3 ¼ ð1 jkB kE ÞI_2 :

ð5:175Þ

Hence we can represent phase shifter as a transformer having complex ratio as follows: 9 _P > V > K_ P ¼ ¼ 1 þ jkB kE ¼ KP ﬀ ’ > > > = V_ E ’ ¼tg1 kB kE qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ KP ¼ 1 þ ðkB kE Þ2 ¼ sec ’

> > > > > ;

:

ð5:176Þ

Since the ratio of the parallel transformer, kE is related to the on/off state of switches S1–S5, we can change ’ by controlling switch states. Apparently ’ is a discrete variable. Note that the product of kE and kB is much less than 1. VP is a little larger than VE. The main function of phase shifters is to change the phase angle ’ of VE. Based on (5.174) and (5.175), we can use the phase shifter equivalent circuits shown in Fig. 5.37 for power system stability and control analysis.

5.5 Basic Principles and Mathematical Models of FACTS

325

Fig. 5.37 Phase shifter equivalent circuit

The above phase shifter is also called a quadrature boosting transformer (QBT) since its output voltage V_ B is always perpendicular to V_ E . A new type of phase shifter has been proposed to use a series voltage source V_ B , and its voltage magnitude and phase angle can both be continuously adjusted to provide easier implementation. Generally, this type of phase shifting transformer (PST) has the mathematical model shown in Fig. 5.37. The control variables are voltage V_ B magnitude and phase angle. Note that the phase shifter is an inactive element. Neglecting its loss, phase shifter output complex power equals its input complex power. So V_B I_2 ¼ V_ E I_3 :

ð5:177Þ

The power generated from the series transformer is consumed by the shunt current source. Thus V_ B I_3 ¼ ¼ k ﬀ ’; V_E I_B V_ B where k ¼ : V_E The adjustment of V_ B magnitude and phase angle can control k and ’. Both phase angle and magnitude of V_ P are controllable, distinguishing this from QBT. There are two independent control variables. This type of phase shifter is similar to a UPFC and we are not going to discuss it in detail.

5.5.6

Basic Principle and Mathematical Model of UPFC

The FACTS devices that we discussed above manipulate only one of the three parameters affecting power transmission. TCSC and SSC compensate a line parameter. SVC and STATCOM control a bus voltage magnitude. TCPST adjusts bus voltage phase angle. The UPFC [132] is a combination of the above FACTS devices and can adjust the three parameters at the same time. In June 1998, the first UPFC was put into trial operation at AEP in the United States. Its application and control strategy are still under investigation. The basic structure of UPFC is shown in Fig. 5.38.

326

5 HVDC and FACTS VEt

VBt I1

I3 = It + Iq

I2

VP

TB

TE Series Converter

Parallel Converter

IB

IE VE

VEa

VEb

Cdc VEc

Idc vdc

+

VBc

VBb

VBa

VB

Variable Reference

Parameter setting

Controller Measurements

Fig. 5.38 UPFC basic configuration

UPFC is like a combination of SATCOM and SSSC. The two GTO voltage inverters share a capacitor to couple the STATCOM and the SSSC. Nabavi-Niaki and Iravani [133] presented a dynamic model of UPFC at fundamental frequency and for symmetrical operation. The converters utilize sinusoid pulse width modulation (SPWM). Here we are going to introduce this model. The control variables of SPWM are modulation ratio and phase angle of sinusoidal control signal. As shown in Fig. 5.38, we can separate UPFC into an AC part and a DC part using transformers as the delimiters. The output voltages of the two converters are 1 V_ E ¼ pﬃﬃﬃ mE vdc ﬀ dE 2 2

mE 2 ½0; 1;

ð5:178Þ

1 V_ B ¼ pﬃﬃﬃ mB vdc ﬀ dB 2 2

mB 2 ½0; 1;

ð5:179Þ

where mE and mB are parallel and series converter modulation ratios; dE and dB are the phase angles of sinusoidal control signals; vdc is the instantaneous voltage on the DC capacitor. It is not difficult to understand that there is the following relation between the variation rate of electrical energy on the capacitor and the real power of the converter Cdc vdc

dvdc ¼ Re V_ E I_E V_ B I_B ; dt

ð5:180Þ

where I_E and I_B are AC currents on the parallel and series converters. In power system stability analysis, we use a sub-steady-state model for power networks.

5.5 Basic Principles and Mathematical Models of FACTS

327

Correspondingly, the AC currents of the converters and the voltages on AC side have the following relation using the reference direction in Fig. 5.38 ðrE þ jolE ÞI_E ¼ V_ Et V_ E ;

ð5:181Þ

ðrB þ jolB ÞI_B ¼ V_B V_Bt ;

ð5:182Þ

where impedances ZB and ZE are equivalent impedances of parallel and series transformers and the converter losses; V_Et and V_ Bt are the voltages transferred from UPFC terminal to converter side. Below we will convert (5.178)–(5.182) into the per unit system. To determine the voltage base, we generally assume that the converter output voltages are rated values when vdc reaches its rated value vdcN and modulation ratios approach 1. In the design of UPFC physical parameters, we have 0 VEN

¼ TE VEN

0 VBN

¼ TB VBN

9 pﬃﬃﬃ > 2 > ¼ TE 1 vdcN ¼ kE VN > = 4 pﬃﬃﬃ ; > 2 > > ¼ TB 1 vdcN ¼ kB VN ; 4

ð5:183Þ

where TE and TB are parallel and series transformer ratios; VN is the AC network 0 0 rated voltage; VEN and VBN are the AC side voltages transformed from rated converter output voltages; kE and kB are the two parameters of UPFC. Due to voltage static security constraints, kE and kB must not be too large, for instance, 1.2 and 0.3, respectively. As seen from the above two equations, ratios of parallel and series transformers are different. Having VN as the voltage base of the AC network, we use the following voltage bases for converters to work with the network per unit system while considering (5.183): VEB ¼

VN vdcN ¼ pﬃﬃﬃ ; TE 2 2 k E

ð5:184Þ

VBB ¼

VN vdcN ¼ pﬃﬃﬃ : TB 2 2 k B

ð5:185Þ

We now have the corresponding current base and impedance base of converters from the above voltage base. DC voltage base is vdcN. The expressions of (5.178) and (5.179) in the per unit system are V_ E ¼ kE mE vdc ﬀ dE ;

mE 2 ½0; 1;

ð5:186Þ

V_B ¼ kB mB vdc ﬀ dB ;

mB 2 ½0; 1:

ð5:187Þ

328

5 HVDC and FACTS

In the per unit system, (5.181) and (5.182) are unchanged. On dividing the two sides of (5.180) by the power base leaves, the right side is unchanged and the left side is Cdc vdc SB Cdc vdc SB

dvdc 2 1 vdc dvdc =vdcN 2 ¼ Cdc vdcN ; dt SB 2 vdcN dt dvdc 2 1 vdc dvdc =vdcN dvdc ¼ Cdc v2dcN ¼ vdc Tu dt SB 2 vdcN dt dt

where Tu is the UPFC time constant with the following value Tu ¼

2W 2 1 ¼ Cdc v2dcN : SB SB 2

ð5:188Þ

The time constant of a UPFC is related to the rated electrical energy stored in the DC capacitor. Equation (5.180) in the per unit system becomes vdc Tu

dvdc ¼ Re V_ E I_E V_B I_B : dt

ð5:189Þ

For the convenience of expression, we remove the subscripts for per unit system. In power system stability analysis, we need to use the two algebraic equations together with network equations. Substituting (5.186) and (5.187) into (5.181), (5.182), and (5.189), and separating real and imaginary parts we have

rE xE

xE rE

rB xB

xB rB Tu

IEx IEy IBx IBy

VEtx kE mE vdc cos dE ¼ ; VEty kE mE vdc sin dE

VBtx kB mB vdc cos dB ¼ ; VBty kB mB vdc sin dB

ð5:190Þ

dvdc ¼ kE mE IEx cos dE þ IEy sin dE dt kB mB IBx cos dB þ IBy sin dBy :

ð5:191Þ

ð5:192Þ

Equations (5.190)–(5.192) constitute UPFC dynamic models in the per unit system. In steady-state operation, UPFC is an inactive device and has constant capacitor voltage, so Re V_ E I_E V_B I_B ¼ 0:

ð5:193Þ

5.5 Basic Principles and Mathematical Models of FACTS VEt

I1 I3

VB

ZB I2

VP

VEt

Pc + jQc

VB

I1 It

VE

329

ZB I2

Iq

VP

Pc + jQc

PE = PB

ZE

a

b

Fig. 5.39 Equivalent circuit of UPFC

Hence, UPFC can be represented as two branches having impedance, connected in series with ideal voltage sources, as shown in Fig. 5.39. V_ B and V_ E are adjusted by GTO gate control signals from the parallel and series converters. Parallel branch current I_3 can be separated into two components I_t and I_q as shown in Fig. 5.39. V_ Et V_ E I_3 ¼ ¼ I_t þ I_q ; ZE

ð5:194Þ

where I_t and I_q components are in phase and perpendicular to bus voltage V_Et . The parallel branch power is PE ¼ Re V_Et I_3 ¼ V_ Et I_t ¼ VEt It ;

ð5:195Þ

jQE ¼ jIm V_Et I_3 ¼ V_Et I_q ¼ jVEt Iq :

ð5:196Þ

In (5.195), we use a negative sign when current is in opposite phase with voltage. In (5.196), we use a positive sign when current leads voltage; otherwise a use negative sign. For the parallel branch, the magnitude and phase of V_ E determine the magnitudes of It and Iq from (5.194). We can see from the above two equations that Iq is the reactive power component of the parallel branch to provide parallel reactive power compensation as in the STATCOM; It is the real power component to consume or inject real power into the AC system. This is to maintain constant DC voltage Vdc and make the phase of the series voltage source V_ B to be 360 controllable. The power generated from the series voltage source is SB ¼ PB þ jQB ¼ V_ B I_2 :

ð5:197Þ

If we control the phase of V_B to make it perpendicular to line current, the function of the series voltage source is like the SSSC series compensation. Generally, the phase and magnitude of V_ B are fully controllable. TCPST and SSSC do not have this kind of capability. Such a function of UPFC comes from the fact that the real power in

330

5 HVDC and FACTS

(5.197) is provided by the parallel branch. The control of two voltage sources under the conditions of (5.193) means that the real power generated or consumed by the series voltage source equals to that consumed or generated by the parallel voltage source. Apparently in this case, the electric field energy stored in the DC capacitor does not change and DC voltage is constant. This is the steady state of the UPFC. On the basis of the above analysis, the relationships for the variables in Fig. 5.39 under steady-state conditions are given:

arg I_t ¼

V_ p ¼ V_ Et þ V_B I_2 ZB ;

ð5:198Þ

I_2 ¼ I_1 I_t I_q ;

ð5:199Þ

V_ Et V_ E I_t þ I_q ¼ ; ZE

ð5:200Þ

It ¼ Re V_B I_2 =VEt ;

ð5:201Þ

(

arg V_ Et þ 0 arg V_ Et þ p

Re V_ B I_2 0 ; Re V_ B I_ > 0

arg I_q ¼ arg V_ Et p=2;

ð5:202Þ

2

ð5:203Þ

where arg represents the phase angle of the vector. Equations (5.201) and (5.202) correspond to (5.193). Although the magnitudes and phases of voltage sources V_ B and V_E can be continuously adjusted, the constraint of (5.193) reduces the number of independent variables from four to three 9 0 VB VB max > = 0 ’B 2p ; > ; 0 Iq Iq max

ð5:204Þ

where ’B is the phase angle of parallel voltage source; VBmax and Iqmax are constants related to UPFC rated capacity. The phase vector diagram for UPFC steady-state operation is shown in Fig. 5.40. For the convenience of analysis, ZB is ignored in the vector diagram. The series connected V_ B changes the bus voltage from V_ Et to V_ p . The change of V_ B makes V_ p vary within the circle centered at V_ Et to control the real and reactive power on the line directly. Note that the compensation of I_q makes the magnitude of V_ Et controllable by the UPFC. One UPFC has three independent control variables to manipulate three operation variables Pc, Qc, and VEt. The steady-state equivalent circuit of UPFC can also be represented as shown in Fig. 5.39b.

Thinking and Problem Solving

331 Vp VB VEt

ϕB

Iq VB max It

o

I1 −Iq

Iq max

I2

Reference phase vector

−It

Fig. 5.40 Phase vector diagram for UPFC

We know from the previous analysis of STATCOM and SSSC that both of them require DC voltages VC to be constant in steady-state operation. The converter AC voltage VASVG of a STATCOM is perpendicular to the AC current flowing out of the system; the VSSSC of SSSC is perpendicular to the AC current on the transmission line. VASVG is to satisfy (5.126) and VSSSC to satisfy (5.160). Their phase angles cannot be freely adjusted. Although UPFC still needs to maintain a constant DC voltage, the coupling between the two converters through the DC capacitor allows the real power consumed by the STATCOM to be sent back through the SSSC or vice versa. The magnitude and phase angle of series transformer output V_ B can then be freely adjusted. The parallel transformer can provide not only reactive power compensation but also the real power transfer between the system and the series transformer. The functional difference between UPFC and PST is due to (5.193) and (5.177). Iq in UPFC is a free variable. For PST, the real power and reactive power taken from the system by the parallel branch are injected into the system by the series branch due to the constraint (5.177). Hence UPFC has STATCOM function while PST does not.

Thinking and Problem Solving 1. What are the factors that limit power transmission distance and capacity? 2. What are the advantages and disadvantages of AC transmission? 3. Discuss the advantages and disadvantages of DC transmission and the applications for which DC transmission is more suitable. 4. What are the characteristics of other new power transmission modes being studied at present? 5. What is the free load flow?

332

5 HVDC and FACTS

6. Why should flexible electrical power systems be introduced? 7. Can we contemplate Id < 0 when V2d > V1d in (5.1)? Why? 8. Why can DC transmission lines only transmit active power, but the converters absorb reactive power from the AC system? 9. What is the physical significance of phase-shifter resistance Rg in (5.21)? 10. Distinguish trigger delay angle, phase-shifter angle, extinguish angle, trigger lead angle, and extinguish lead angle, paying attention to their operating areas. 11. Discuss the steady load flow control method of DC transmission. 12. Compare load flow calculation models with and without DC transmission lines. 13. Give an appropriate value L, C, and draw the curve of SVC equivalent reactance XSVC–b denoted in (5.121). 14. Draw V–b curve according to (5.121), in which V 2 [0.9,1.1], when per unit value Vref ¼ 1.05, Xe ¼ 0.05 in (5.122). 15. Draw the equivalent circuit diagram when S5 is also tripped with S1–S4 tripping in Fig. 5.35. 16. In steady state, UPFC can be regarded equivalently as two voltage source converters (VSC), with voltage amplitude values VB and VE, respectively, and phase angles dB and dE, respectively. Analyze why UPFC can only control three operational variables (active power Pc, reactive power Qc, and nodal VEt on a transmission line) in the steady state? 17. Discuss the capacitive and inductive value range of XTCSC when conduction angle b should avoid the resonance region according to (5.153)–(5.155). Then discuss whether line transmission active power Pc ¼ Plp controlled by TCSC can vary continuously.

Chapter 6

Mathematical Model of Synchronous Generator and Load

6.1

Introduction

The continuous increase of power system complexity and installation of more and more new equipment in power systems has demanded better methods for power system analysis, planning, and control. At present, analysis of modern power systems is generally based on digital computers. Hence, establishment of a mathematical model, describing the physical processes of a power system, is the foundation for the analysis and investigation of various power system problems. Correct and accurate computation for power system analysis requires a correct and accurate mathematical model of the power system. Transient processes of the power system are very fast. This is why power system operation heavily relies on the applications of automatic control. With the installation of many different automatic control devices, the operation of which largely depends on the application of electronic and computing technology, modern power system operation has reached a very high level of automation. For such large-scale and complex systems, the mathematical description is nonlinear and high dimensional, consisting of a large number of nonlinear equations. Hence it is both appropriate and practical that the analysis and computation of such a system ought to start from simple local devices and be completed finally for the complex overall system. Therefore, in modeling of large-scale and complex power systems, these systems are first decomposed into independent basic components, such as synchronous generators, transformers, transmission lines, governors and automatic voltage regulators (AVR), etc. Then those components are modeled separately according to circuit theory or other related principles. Models of those components are building-bricks to construct the mathematical model of whole power systems. For the study of different problems on the same system, different models are required. Mathematically, a power system is a nonlinear dynamic system. When the steady-state operation of a dynamic system is studied, the mathematical description of the system is in the form of algebraic equations. Differential equations (sometimes partial differential equations) give a mathematical description of system

X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

333

334

6 Mathematical Model of Synchronous Generator and Load

dynamics. For the study of some specific problems, model parameters could be time-variable and variables may not be continuous. In addition, to meet the requirement of different computing accuracy, different models could be used. Obviously, a mathematical model for qualitative analysis could be simpler than that for quantitative analysis. Computing accuracy and speed are always two conflicting factors which need to be considered carefully when a power system model is established. The more accurate the computation is, the more the computing work and hence the longer the computing time. On the other hand, to sacrifice some computing accuracy will be compensated with high computing speed, which has been a common practice in modeling power systems and developing computing algorithms. In this aspect, the effort has been to develop a mathematical model of a power system and the associated solution methods, such that the need for both computing accuracy and computing speed is met. Often, the result is a compromise between those two requirements based on available computing tools. There are two major issues in mathematical modeling. The first one is to describe a subject under investigation mathematically in the form of equations. There are two methods to establish those mathematical equations. The first method, the analytical method, is to derive those mathematical descriptions by using special knowledge and theory about the subject; the second is to identify them by carrying out experiments or using data obtained from its operation. That is the method of system identification in control theory. The second major issue in mathematical modeling is to obtain parameters of the mathematical description of the subject. No matter whether the plant is described by algebraic or differential equations, various parameters in those equations need to be obtained. Generally, for simple components of the subject, model parameters can be derived from design parameters according to certain physical (such as mechanical or electrical) principles. For example, four parameters of an overhead line, i.e., resistance, inductance, capacitance, and conductance to earth, can be obtained by applying electromagnetic theory to the way the line is arranged in space, the materials of the line, and the natural environment where the line is located. That is a typical analytical method. However, for complex components or systems, usually there is certain difference between the actual parameters and design parameters. A typical example is the generator parameter which could be affected by variations of power system operating conditions, saturation, and a series of complex conversion processes among mechanical, electrical, magnetic, and thermal energy. Therefore, in addition to the method of theoretical derivation, there is another important way to obtain model parameters of complex components and systems. This is the method of parameter estimation which is one of the methods in system identification. Parameter estimation and system identification is a research field which will not be discussed in this book. In Chap. 1, the mathematical model of a power network has been introduced. Mathematical models of HVDC and FACTS are discussed in Chap. 5. Hence, in this chapter the focus is the introduction to mathematical models of generator and load, including the mathematical models of synchronous generator, excitation systems, and governing systems.

6.2 Mathematical Model of Synchronous Generator

6.2

335

Mathematical Model of Synchronous Generator

The dynamics of a synchronous generator is the basis for the study of the dynamic behavior of the power system. In the history of developing the mathematical model of a synchronous generator, two milestones are the establishment of two-reaction theory in 1920s [146, 147] and the proposal of Park’s transformation [148]. Under the ideally assumed conditions and by using two-reaction principle, Park derived the basic mathematical equations of a synchronous generator in dq0 coordinate system. Since then, mathematical models of synchronous generators have been based on Park’s contribution with further major development regarding the number of equivalent windings to model the generator rotor winding, different assumptions about when a synchronous generator should be described by transient or subtransient parameters, different ways to describe magnetic saturation, etc. Details of all these points above can be found in [149–152]. In this section, we shall focus on those mathematical descriptions of the synchronous generator which have been widely used. Readers should note that in other references, different symbols, defined positive directions of physical variables, form of transformation matrix, and selection of base values may be used. From the structure of a synchronous generator we know that on the rotor, the field winding is a physical winding; while damping windings may just be electrically equivalent windings. For a salient-pole generator, damping windings represent the damping function of damping rods distributed on the rotor. While for a round rotor generator, they simulate the damping function produced by the eddy current inside the whole rotor. Since they are just equivalent windings, the damping function can be represented by a single or multiple damping windings. In theory, the more the equivalent damping windings, the more accurate the representation can be. However, if more equivalent damping windings are used, there could be two problems. The first is the increase of the order of differential equations in the mathematical model, adding computational burden for their solution. The second problem is that it is more difficult to obtain the relative electrical parameters accurately. Hence in the commonly used mathematical model of a synchronous generator, the number of equivalent damping windings is usually not more than three. Since the damping rods on the rotor of a salient-pole generator are more like real windings than the whole rotor of a round rotor generator and the magnetic circuit of the salient-pole generator is different in d and q directions, the damping function of the salient-pole generator is usually represented by two damping windings, one in the direction of direct axis (d), denoted as D damping winding and another in that of the quadrature axis (q), denoted as Q damping winding; For the round rotor generator, in addition to D and Q damping windings, one more equivalent damping winding in the quadrature direction (g winding) is used. Q and g winding represents the weaker and stronger eddy current effect, respectively. According to the theory of electric machines, the ideal assumptions about the synchronous generators are that the magnetic circuits are symmetrical, saturation is negligible, and flux waveforms have sinusoidal space distribution. In the following,

336

6 Mathematical Model of Synchronous Generator and Load

we shall first derive the mathematical model of an ideal synchronous generator with D, g, and Q damping winding, followed by introduction of a method considering magnetic saturation effects.

6.2.1

Basic Mathematical Equations of Synchronous Generator

6.2.1.1

Three-Phase Mathematical Equations

Figure 6.1a, b shows the structure of a synchronous generator and winding circuit diagram. We consider the general case of a salient-pole generator with D, g, Q three damping windings and treat a round rotor generator as a special case since it has only D, Q two damping windings. In the figures, the defined positive direction of voltage, current, and magnetic flux is related to the three-phase armature windings abc, field winding f and damping winding D, g, Q. It must be pointed out that the positive direction of magnetic flux related to the three-phase armature windings is opposite to that induced by the armature current of each winding in the positive direction; while magnetic flux associated with rotor windings is defined in the same direction as that induced by the current in each winding in the positive direction; q-axis leads d-axis by 90 in the rotational direction of the generator rotor. In addition, the positive direction of all flux axes is chosen to be those of the corresponding magnetic flux.

α

ia

L aa

b

+

D

−

ω

−

−

f b

a

−

q

c

−

if R f Lff

f ⫻

−

D

⫻

Rb

Lbb ib

c

⫻

Vf +−

R

−

z

⫻

g − Q − − ⫻

D

−

a

c

x

⫻

D

⫻ Q X g ⫻ X

Ra

−

iD R D

Lc

d

ic

LDD

⫻

vc Lgg

y

c

Rg

ig

vb va

LQQ RQ iQ

b

Fig. 6.1 Structure of synchronous generator and winding circuit. (a) Structure of synchronous generator (b) winding circuit

6.2 Mathematical Model of Synchronous Generator

337

From Fig. 6.1b, the following voltage equation for all the windings can be obtained 2

3 2 Ra va 6 vb 7 6 0 6 7 6 6 vc 7 6 0 6 7 6 6 7 6 6 7 6 6 vf 7 ¼ 6 0 6 7 6 6 0 7 6 0 6 7 6 4 0 5 4 0 0 0

0 Ra 0

0 0 Ra

0 0 0

0 0 0

0 0 0

0 0 0 0

0 0 0 0

Rf 0 0 0

0 RD 0 0

0 0 Rg 0

3 2 3 ’a ia 76 ib 7 6 ’b 7 76 7 6 7 76 ic 7 6 ’c 7 76 7 6 7 76 7 6 7 76 7 þ p6 7; 6 if 7 6 ’f 7 0 7 76 7 6 7 6 7 6 ’D 7 0 7 76 iD 7 6 7 4 ’g 5 0 54 ig 5 ’Q RQ iQ 0 0 0

32

ð6:1Þ

where p ¼ dtd denotes the differentiation operator. For an ideal synchronous generator, magnetic saturation effects can be ignored. Hence magnetic flux linkage of each winding can be written in the form of selfinductance and mutual inductance as shown by the following flux linkage equation 2 3 2 32 3 ’a Laa Mab Mac Maf MaD Mag MaQ ia 6 ’b 7 6 Mba Lbb Mbc 6 7 Mbf MbD Mbg MbQ 7 6 7 6 76 ib 7 6 ’c 7 6 Mca Mcb Lcc 76 ic 7 M M M M cf cD cg cQ 6 7 6 76 7 6 7 6 76 7 6 7¼6 76 7: ð6:2Þ 6 ’f 7 6 Mfa Mfb Mfc 6 7 Lff MfD Mfg MfQ 7 6 7 6 76 if 7 6 ’D 7 6 MDa MDb MDc 7 6 MDf LDD MDg MDQ 76 iD 7 6 7 6 7 4 ’g 5 4 Mga Mgb Mgc Mgf MgD Lgg MgQ 54 ig 5 ’Q MQa MQb MQc MQf MQD MQg LQQ iQ From circuit theory we know that the above coefficient matrix is symmetrical. From Fig. 6.1a we can see that due to the rotor rotation, the reluctance of the magnetic circuit of some windings changes periodically with the variation of rotor position. Hence the self-inductance and mutual inductance of those windings are a function of rotor position. According to the assumptions of an ideal synchronous generator, both the magnetomotive force (mmf) induced by armature current and mutual flux between armature windings and rotor windings have sinusoidal space distribution. Rotor position can be described by the angle between d-axis and flux axis of phase a armature winding y ¼ y0 þ ot. Hence the self-inductance and mutual inductance of each winding can be expressed as follows [153]. 1. Self-inductance and mutual inductance of armature windings 9 Laa ¼ l0 þ l2 cos 2y > = Lbb ¼ l0 þ l2 cos 2ðy 2p=3Þ ; > ; Lcc ¼ l0 þ l2 cos 2ðy þ 2p=3Þ

9 Mab ¼ ½m0 þ m2 cos 2ðy þ p=6Þ > = Mbc ¼ ½m0 þ m2 cos 2ðy p=2Þ : > ; Mca ¼ ½m0 þ m2 cos 2ðy þ 5p=6Þ

ð6:3Þ

ð6:4Þ

338

6 Mathematical Model of Synchronous Generator and Load

Under the assumptions of an ideal synchronous generator, it can be proved that l2 ¼ m2. Furthermore, for a round rotor generator, reluctance of magnetic circuits related to the self-inductance and mutual inductance of armature windings does not vary with rotor rotation, and we have l2 ¼ m2 ¼ 0. Hence those selfinductance and mutual inductance above are constant. 2. Mutual inductance between armature and rotor windings 9 9 Maf ¼ maf cos y > > = MaD ¼ maD cos y = Mbf ¼ maf cosðy 2p=3Þ ; MbD ¼ maD cosðy 2p=3Þ ; > > ; ; Mcf ¼ maf cosðy þ 2p=3Þ McD ¼ maD cosðy þ 2p=3Þ 9 9 Mag ¼ mag sin y > > = Mag ¼ maQ sin y = Mbg ¼ mag sinðy 2p=3Þ ; Mbg ¼ maQ sinðy 2p=3Þ : > > ; ; Mcg ¼ mag sinðy þ 2p=3Þ Mcg ¼ maQ sinðy þ 2p=3Þ

ð6:5Þ

ð6:6Þ

3. Self-inductance and mutual inductance of rotor windings Since rotor windings rotate with the generator rotor, for salient-pole or round rotor generator, reluctance of magnetic circuits does not vary with the change of rotor position. Hence self-inductance and mutual inductance of rotor windings are constant. D, f winding on direct axis (d) is vertical to g, Q winding on quadrature axis (q). Hence mutual inductance between them is zero, that is Mfg ¼ MfQ ¼ MDg ¼ MDQ ¼ 0: 6.2.1.2

ð6:7Þ

Basic Equations in dq0 Coordinate

From the discussion above we know that the self-inductance and mutual inductance of generator windings are not constant and some of them vary with the position of the generator rotor. Hence (6.1) and (6.2) are time-variant differential equations which are difficult to solve. To transfer these into time-invariant differential equations, some of methods of coordinate transformation have been proposed, among which the dq0 transformation proposed by Park [148] has been most widely used. In dq0 coordinate, flux linkage equations become time invariant. Hence the mathematical model of a synchronous generator is presented as a group of timeinvariant differential equations. In the following, we shall discuss the details of Park’s transformation. Park’s transformation converts three-phase flux linkage, armature current, and voltage into d, q, 0 components in the dq0 coordinates, through an equivalent coordinate transformation. It can be written as 2

3 2 Ad cos y 2 4 Aq 5 ¼ 4 sin y 3 A0 1=2

32 3 cosðy 2p=3Þ cosðy þ 2p=3Þ Aa sinðy 2p=3Þ sinðy þ 2p=3Þ 54 Ab 5: 1=2 1=2 Ac

ð6:8Þ

6.2 Mathematical Model of Synchronous Generator

339

For simplicity of expression, the equation above can be written in the compact form as follows Adq0 ¼ PAabc :

ð6:9Þ

The inverse Park’s transformation is 2

3 2 Aa cos y sin y 4 Ab 5 ¼ 4 cosðy 2p=3Þ sinðy 2p=3Þ Ac cosðy þ 2p=3Þ sinðy þ 2p=3Þ

32 3 1 Ad 1 54 Aq 5 1 A0

or Aabc ¼ P1 Adq0 :

ð6:10Þ ð6:11Þ

In (6.8)–(6.11), symbol A represents current, voltage, or flux linkage, i.e., idq0 ¼ Piabc ; vdq0 ¼ Pvabc ; Cdq0 ¼ PCabc ;

ð6:12Þ

iabc ¼ P1 idq0 ; vabc ¼ P1 vdq0 ; Cabc ¼ P1 Cdq0 :

ð6:13Þ

Applying the transformations of (6.12) and (6.13) as well as (6.3)–(6.7), (6.1) and (6.2) can be converted into the following equations in dq0 coordinates 2

3 2 Ra vd 6 vq 7 6 0 6 7 6 6 v0 7 6 0 6 7 6 6 vf 7 ¼ 6 0 6 7 6 607 60 6 7 6 405 40 0 0 2

0 Ra 0 0 0 0 0

0 0 Ra 0 0 0 0

0 0 0 0 0 0 Rf 0 0 RD 0 0 0 0

3 2 ’d Ld 0 6 ’q 7 6 0 L q 6 7 6 6 ’0 7 6 0 0 6 7 6 6 7 6 6 7¼6 6 ’f 7 6 3maf =2 0 6 7 6 6 ’D 7 6 3maD =2 0 6 7 6 4 ’g 5 4 0 3mag =2 ’Q 0 3maQ =2

0 0 0 0 0 Rg 0 0 0 L0 0 0 0 0

32 3 2 3 2 3 ’d 0 id o’q 6 7 6 ’q 7 6 o’d 7 0 7 76 iq 7 6 7 6 7 6 i0 7 6 ’0 7 6 0 7 0 7 76 7 6 7 6 7 6 7 6 7 6 7 0 7 76 if 7 þ p6 ’f 7 6 0 7; ð6:14Þ 6 iD 7 6 ’D 7 6 0 7 0 7 76 7 6 7 6 7 4 ’g 5 4 0 5 0 54 ig 5 ’Q RQ iQ 0 32 3 maf maD 0 0 id 6 7 0 0 mag maQ 7 76 iq 7 7 6 0 0 0 0 76 i0 7 7 76 7 76 7; 7 6 Lf mfD 0 0 76 i f 7 7 6 7 mfD LD 0 0 7 7 6 iD 7 5 4 0 0 Lg mgQ ig 5 0 0 mgQ LQ iQ

ð6:15Þ

where Ld ¼ l0 þ m0 þ 3l2 =2; Lq ¼ l0 þ m0 þ 3l2 =2; L0 ¼ l0 2m0 ; Lf ¼ Lff ; LD ¼ LDD ; Lg ¼ Lgg ; LQ ¼ LQQ ; mgQ ¼ MfD ; mgQ ¼ MgQ and o ¼

dy is the angular speed of the synchronous generator. dt

ð6:16Þ

340

6 Mathematical Model of Synchronous Generator and Load

Park’s transformation, in fact, replaces three-phase armature windings by their three structurally equivalent windings – d winding, q winding, and 0 winding. The difference is that the magnetic flux axis of three-phase armature windings is stationary in space; while that of dq0 windings rotates in space at rotor speed. The positive direction of magnetic flux axis of d winding and q winding is as same as that of d- and q-axis of generator rotor, respectively, to describe the behavior of electrical variables in the direction of d- and q-axis; while 0 winding represents the zero-sequence component in the three-phase armature current, voltage and flux linkage. Ld, Lq, and L0 in (6.16) is the self-inductance of equivalent d, q, and 0 winding, corresponding to d, q, and 0 synchronous reactance, respectively. From (6.16) we can see that the coefficient matrix in (6.15) is a constant matrix. Hence the mathematical model of (6.14) of synchronous generator has been transformed into a set of time-invariant differential equations. Equation (6.14) indicates that the phase voltage of the synchronous generator consists of three parts. The first part is the voltage drop across the resistance of armature windings; the second is the EMF induced from the variation of flux linking the armature windings, which is usually called the transformer voltage of a synchronous generator; the third part is the EMF due to the rotation of the synchronous generator which is termed speed voltage. The value of speed voltage is much greater than that of transformer voltage. The coefficient matrix in (6.15) is nonsymmetrical, i.e., the mutual inductance between windings on generator rotor and d, q, and 0 winding is not reciprocal. That is caused by the transformation. If the current in rotor windings is multiplied by 3/2, or an orthogonal transformation matrix is adopted, these mutual inductance will become reciprocal. From (6.16) we can see that for a salient-pole generator Ld > Lq and round rotor generator Ld ¼ Lq because l2 ¼ 0. This difference makes it applicable to represent round rotor generators by the mathematical model of the salient-pole generator. According to the reference direction of current and voltage given in Fig. 6.1b, the total output power from the three-phase armature windings is po ¼ va ia þ vb ib þ vc ic ¼ vTabc iabc :

ð6:17Þ

Applying Park’s transformation to the equation above, from (6.13) we can obtain the output power from armature windings in dq0 coordinates to be 3 po ¼ ðP1 vdq0 ÞT ðP1 idq0 Þ ¼ ðvd id þ vq iq þ 2v0 i0 Þ: 2

6.2.1.3

ð6:18Þ

Per Unit Equations of the Synchronous Generator

The per unit system is commonly used in power system analysis and calculation due to its many advantages. Parameters of synchronous generator are also usually given in per unit. Hence we need to convert the mathematical model of synchronous generator of (6.14) and (6.15) using actual values of various variables into the per unit equations

6.2 Mathematical Model of Synchronous Generator

341

where variables are described by per unit values. When we introduce the per unit equations used in HVDC in 4.3.1, we have mentioned the principle that, in a per unit system, the base values of different physical variables must have the same relationship that they have when using actual values. Hence in a per unit system, some base values are defined by users and others are derived from the physical relationships among variables. Obviously, the difference in defining those base values by users will lead to different per unit systems. This book will adopt a widely used per unit system – ‘‘unit excitation voltage/unit stator voltage’’ base value system. Subscript B is still used to denote base values of various physical variables and ‘‘*’’ to represent per unit variables. Firstly we define the base value for generator speed to be the synchronous angular speed os. Because ot ¼ y and y is dimensionless (without base value), oBtB ¼ 1 which can lead to the base value for time t. Hence oB ¼ os tB ¼ 1=os

) :

ð6:19Þ

On the generator stator side, we define the magnitude of armature current and voltage as their base values. From the definition, we can derive the base values for power, impedance, and flux linkage as follows VB IB 3 SB ¼ 3 pﬃﬃﬃ pﬃﬃﬃ ¼ VB IB ; 2 2 2

ð6:20Þ

ZB ¼

VB 3 VB2 ¼ ; IB 2 SB

ð6:21Þ

’B ¼

ZB IB ¼ ZB IB tB ¼ VB tB : oB

ð6:22Þ

In a per unit system, there should be only one base value for power. Hence for f, D, g, and Q, four rotor windings, we have 3 Vf B If B ¼ VDB IDB ¼ VgB IgB ¼ VQB IQB ¼ VB IB ¼ SB : 2

ð6:23Þ

Due to the constraint of above equation, we can only define one base value between current and voltage, for each rotor winding, and then derive the other. After the base values for the voltage and current of rotor windings are obtained, base values for impedance and flux linkage can be found from the following equations Zf B ¼ Vf B =If B ; ZDB ¼ VDB =IDB ; ZgB ¼ VgB =IgB ; ZQB ¼ VQB =IQB ; ’f B ¼ Vf B tB ; ’DB ¼ VDB tB ;

’gB ¼ VgB tB ; ’QB ¼ VQB tB :

ð6:24Þ

342

6 Mathematical Model of Synchronous Generator and Load

With base values introduced, we can convert the mathematical equations of (6.14) and (6.15) into the description in per unit as follows. Dividing both sides of each voltage equations of (6.14) by corresponding base voltage value and using the relationships among various base values as given in (6.19)–(6.24), we can obtain 2

3 2 Ra 0 0 vd 6 vq 7 6 0 Ra 0 6 7 6 6 v0 7 6 0 0 Ra 6 7 6 6 7 6 6 7¼6 6 vf 7 6 0 0 0 6 7 6 6 0 7 6 0 0 0 6 7 6 4 0 5 4 0 0 0 0 0 0 0

0 0 0

0 0 0

Rf 0 0 RD 0 0 0 0

3 2 3 2 3 ’d id o ’q 76 iq 7 6 ’q 7 6 o ’d 7 76 7 6 7 6 7 76 i0 7 6 ’0 7 6 0 7 76 7 6 7 6 7 76 7 6 7 6 7 76 7 þ p 6 76 7 6 if 7 6 ’f 7 6 0 7; ð6:25Þ 0 0 7 76 7 6 7 6 7 6 7 6 ’D 7 6 0 7 0 0 7 76 iD 7 6 7 6 7 4 ’g 5 4 0 5 Rg 0 54 ig 5 ’Q 0 RQ iQ 0 0 0 0

0 0 0

32

where p* is differentiation operator in per unit: p d d ¼ tB ¼ ; oB dt dt Ra ¼ Ra =ZB 2 Rf If B 2 Rf Rf ¼ ¼ Zf B 3 ZB IB 2 RD 2 RD IDB : RD ¼ ¼ ZDB 3 ZB IB 2 Rg IgB 2 Rg Rg ¼ ¼ ZgB 3 ZB IB 2 RQ 2 RQ IQB RQ ¼ ¼ ZQB 3 ZB IB p ¼

ð6:26Þ

Using the similar procedure for (6.15), we can obtain 2

3 2 ’d Xd 6 ’q 7 6 0 6 7 6 6 ’0 7 6 0 6 7 6 6 7 6 6 7 6 6 ’f 7 ¼ 6 Xaf 6 7 6 6 ’D 7 6 XaD 6 7 6 4 ’g 5 4 0 ’Q 0

0 Xq 0

0 0 X0

Xaf 0 0

XaD 0 0

0 Xag 0

0 0 Xag XaQ

0 0 0 0

Xf XfD 0 0

XfD XD 0 0

0 0 Xg XgQ

3 id 6 7 XaQ 7 76 iq 7 6 i0 7 0 7 76 7 76 7 76 7 6 if 7; ð6:27Þ 0 7 76 7 7 6 0 7 76 iD 7 XgQ 54 ig 5 XQ iQ 0

32

6.2 Mathematical Model of Synchronous Generator

343

where 9 > > > > > > > > > > > > > > > > > > > > > =

Xd ¼ oB Ld =ZB Xq ¼ oB Lq =ZB X0 ¼ oB L0 =ZB

2 oB Lf 2 oB Lf If B Xf ¼ ¼ Zf B 3 ZB IB 2 oB LD 2 oB LD IDB > XD ¼ ¼ > > > ZDB 3 ZB IB > > 2 > > > oB Lg 2 oB Lg IgB > > > Xg ¼ ¼ > > ZgB 3 ZB IB > > 2 > > > oB LQ 2 oB LQ IQB > > ; XQ ¼ ¼ ZQB 3 ZB IB Xaf Xag XfD

oB maf If B ¼ ; ZB IB oB mag IgB ¼ ; ZB IB 2 oB mfD If B IDB ¼ ; 3 ZB IB2

XaD XaQ XgQ

oB maD IDB ¼ ZB IB oB maQ IQB ¼ : ZB IB 2 oB mgQ IgB IQB ¼ 3 ZB IB2

ð6:28aÞ

ð6:28bÞ

In addition, dividing both sides of (6.18) by SB, from (6.20) we can obtain the output power from synchronous generator in per unit to be po ¼ vd id þ vq iq þ 2v0 i0 :

ð6:29Þ

We should note that the per unit equations of the synchronous generator of (6.25) have the similar form to those using actual values of (6.14). However, in the per unit equations, the coefficient matrix in flux linkage equation of (6.27) is symmetrical, i.e., the mutual inductance between stator and rotor windings becomes reciprocal. Furthermore, with proper choice of base values for inductance, we can make the per unit values of inductance to be equal to that of reactance. Hence the coefficient matrix in the flux linkage equation can also be expressed by use of per unit reactance.

6.2.2

Mathematical Equations of Synchronous Generator Using Machine Parameters

For simplicity of expression, in the following discussion we shall use per unit system and omit the subscript ‘‘*’’ to express per unit variables.

344

6 Mathematical Model of Synchronous Generator and Load

In the mathematical equations of the synchronous generator of (6.25) and (6.27), a total of 18 parameters are presented in (6.26) and (6.28). We regard those 18 parameters as basic parameters of a synchronous generator which are decided by physical design and materials used. Strictly speaking, for two generators of the same type and same model, the parameters will not be exactly the same. Usually it is extremely difficult to obtain the values of those parameters through analytical calculation. Therefore, in practice we convert those 18 basic parameters of a synchronous generator into a group of 11 steady-state, transient, and subtransient parameters. These 11 parameters are called machine parameters and can be obtained directly from machine experiments. They are resistance of stator winding (Ra), q- and d-axis synchronous reactance (Xd, Xq), transient reactance (Xd0 ; Xq0 ), and 0 0 00 00 subtransient reactance (Xd00 ; Xq00 ) as well as the four time constants (Td0 ; Tq0 ; Td0 ; Tq0 ). Because machine parameters are fewer than the basic parameters, certain assumptions are needed for the conversion between these two sets of parameters. Firstly, from the basic (6.25) and (6.27) of a synchronous generator we can see that the magnetic field in space generated by zero-sequence component, i0, is zero and hence it has no impact on any electrical variables associated with generator rotor. Therefore, the zero-sequence equation in (6.25) and (6.27) and the parameter X0 can be ignored. Equation (6.25) now becomes 2

3 2 vd Ra 4 vf 5 ¼ 4 0 0 0 2

3 2 Ra vq 405¼4 0 0 0

0 Rf 0

32 3 2 3 2 3 0 id ’d o’q 0 54 if 5 þ p4 ’f 5 4 0 5; RD iD ’D 0

ð6:30Þ

0 Rg 0

32 3 2 3 2 3 ’q 0 iq o’d 0 54 ig 5 þ p4 ’g 5 þ 4 0 5: ’Q RQ iQ 0

ð6:31Þ

Equation (6.27) can be written as 2

3 2 Xd ’d 4 ’f 5 ¼ 4 Xaf XaD ’D 2

3 2 ’q Xq 4 ’g 5 ¼ 4 Xag ’Q XaQ

Xaf Xf XfD

32 3 XaD id XfD 54 if 5; XD iD

ð6:32Þ

Xag Xg XgQ

32 3 XaQ iq XgQ 54 ig 5: XQ iQ

ð6:33Þ

We can assume that there exist relationships as shown in the following (6.34) among basic parameters in (6.32) and (6.33) [154] Xaf XD ¼ XaD XfD Xag XQ ¼ XaQ XgQ

) :

ð6:34Þ

6.2 Mathematical Model of Synchronous Generator

345

d-axis machine parameters are related to basic parameters as follows: 1. The definition of d-axis synchronous reactance Xd is that when f and D winding are open-circuited and there exists only the d-axis component of current in the armature winding, the measured armature reactance is Xd. From the definition we know that in (6.32) when if ¼ iD ¼ 0, we have ’d ¼ Xd id ; i.e., basic parameter Xd is machine parameter Xd. 2. d-axis transient reactance Xd0 is defined such that when f winding is shortcircuited, D winding open-circuited and only a d-axis component of current suddenly flows through the armature winding, the measured armature reactance is Xd0 . From the definition we know that with D winding being open-circuited, iD ¼ 0; and with f winding being short-circuited, at the moment of sudden flow of current through the armature winding, ff ¼ 0. Hence in (6.32) we have ’d ¼ Xd id þ Xaf if

)

’f ¼ Xaf id þ Xf if ¼ 0

:

Canceling if in the above equations we obtain ! 2 Xaf ’d ¼ Xd id : Xf Therefore Xd0 ¼

2 Xaf ’d ¼ Xd : id Xf

ð6:35Þ

3. The definition of d-axis subtransient reactance Xd00 is that when f and D winding are short-circuited and only a d component of current suddenly flows through the armature winding, the measured armature reactance is Xd00 . According to the definition, with ff ¼ fD ¼ 0 in (6.32), we have 9 ’d ¼ Xd id þ Xaf if þ XaD iD > = ’f ¼ Xaf id þ Xf if þ XfD iD ¼ 0 : > ; ’D ¼ Xad id þ XfD if þ XD iD ¼ 0 By canceling if and iD in the above equation, we obtain ’d ¼ Xd

2 2 XD Xaf 2Xaf XfD XaD þ Xf XaD 2 XD Xf XfD

! id :

346

6 Mathematical Model of Synchronous Generator and Load

That is Xd00 ¼

2 2 XD Xaf 2Xaf XfD XaD þ Xf XaD ’d ¼ Xd : 2 id XD Xf XfD

ð6:36Þ

From the first equation, on the previous assumption of (6.34), we can find XfD. By substituting it into (6.36) we have Xd00 ¼ Xd

2 XaD : XD

ð6:37Þ

4. The definition of d-axis open-circuit transient time constant is the decaying time constant of if when d and D winding are open-circuited. This means that in (6.30) and (6.32), we have id ¼ iD ¼ 0, fd ¼ fD ¼ 0. Hence ) vf ¼ Rf if þ p’f : ’f ¼ Xf if In per unit we have Xf ¼ Lf. From the equation above we can obtain v f ¼ R f if þ Lf

dif : dt

Hence 0 Td0 ¼ Lf =Rf ¼ Xf =Rf :

ð6:38Þ

In fact, when d and D winding are open-circuited, f winding becomes an isolated winding. Hence the decaying time constant of the winding current is the time constant of f winding itself. 00 5. d-axis open-circuit subtransient time constant Td0 is defined to be the decaying time constant of D winding when d winding is open-circuited and f winding short-circuited. From the definition we have id ¼ 0, vf ¼ 0 in (6.30) and (6.32). Hence 9 Rf if þ p’f ¼ 0 > > > > RD iD þ p’D ¼ 0 = : > ’f ¼ Xf if þ XfD iD > > > ; ’D ¼ XfD if þ XD iD That is

Xf XfD

XfD i Rf p f ¼ XD iD 0

0 RD

if : iD

6.2 Mathematical Model of Synchronous Generator

347

This is obviously a second-order electrical circuit and hence there are two time constants. Because usually Rf is very small we can assume Rf ¼ 0. By canceling if in the above equation we have ! 2 XfD XD piD ¼ RD iD : Xf Hence 00 Td0

¼

2 XfD XD Xf

!, RD :

ð6:39Þ

So far we have established the relationship between five d-axis machine parameters and basic parameters. In the similar way, from the definition of various q-axis machine parameters, q-axis voltage equation of (6.31), flux linkage equation of (6.33), and the assumption of (6.34), we can also obtain the relationship between five q-axis machine parameters and basic parameters. In total, the relationship between 11 machine parameters and 18 basic parameters can be listed as follows (on the left side of equations are the machine parameters and the right side the basic parameters). Ra ¼ Ra ; Xd ¼ Xd ; Xq ¼ Xq ; 2 Xd0 ¼ Xd Xaf =Xf 2 Xq0 ¼ Xq Xag =Xg 2 Xd00 ¼ Xd XaD =XD 2 Xq00 ¼ Xq XaQ =XQ 0 Td0 ¼ Xf =Rf 0 Tq0 ¼ Xg =Rg

ð6:40aÞ

) ;

ð6:40bÞ

) ;

ð6:40cÞ

) ;

9 00 2 = Td0 ¼ XD XfD =Xf =RD > : 00 2 ; Tq0 ¼ XQ XgQ =Xg =RQ >

ð6:40dÞ

ð6:40eÞ

Eleven machine parameters can be obtained through experiment. We should point out that the relationship between machine and basic parameters of (6.40) depends on the assumption given in (6.34). Different assumption may be made that will lead to different relationship between the machine and basic parameters such as that given in; while different relationships will result in different mathematical equations of a synchronous generator represented by using machine parameters. However, values of machine parameters are only affected by their definitions, irrelevant of the initial assumptions.

348

6 Mathematical Model of Synchronous Generator and Load

In the following, we will establish the mathematical equations of synchronous generators represented by machine parameters. To do so, we first introduce the noload voltage that is proportional to the current of various rotor windings, and transient and subtransient excitation voltages that are proportional to the flux linkage of rotor windings as follows. No-load voltage: 9 eq1 ¼ Xaf if > > > ed1 ¼ Xag ig = : ð6:41Þ eq2 ¼ XaD iD > > > ; ed2 ¼ XaQ iQ Transient and subtransient voltage: e0q ¼ ðXaf =Xf Þ’f e0d ¼ ðXag =Xg Þ’g

9 > > > > =

e00q ¼ ðXaD =XD Þ’D > > > > ; e00d ¼ ðXaQ =XQ Þ’Q

:

ð6:42Þ

In the mathematical equations of a synchronous generator represented by basic parameters of (6.30)–(6.33), we can express current and flux linkage of all rotor windings by the associated voltage defined in (6.40)–(6.42). By using the relationship between basic and machine parameters of (6.40) and the assumption of (6.34), we will obtain the following mathematical equations of a synchronous generator represented by machine parameters. Flux linkage equation of armature windings ’d ¼ Xd id þ eq1 þ eq2 ’q ¼ Xq iq ed1 ed2

) :

ð6:43Þ

Flux linkage equation of rotor windings 9 Xd Xd0 > ¼ ðXd þ eq1 þ eq2 > > > > Xd Xd00 > > > > 00 00 = eq ¼ ðXd Xd Þid þ eq1 þ eq2 : Xq Xq0 > > > e0d ¼ ðXq Xq0 Þiq þ ed1 þ e d2 > > Xq Xq00 > > > > ; 00 00 ed ¼ ðXq Xq Þiq þ ed1 þ ed2 e0q

Xd0 Þid

ð6:44Þ

Voltage equation of armature windings vd ¼ r’d o’q Ra id vq ¼ r’q þ o’d Ra iq

) :

ð6:45Þ

6.2 Mathematical Model of Synchronous Generator

349

Voltage equation of rotor windings 9 > > > > 0 00 > > X X > 00 00 d d > Td0 req ¼ e > q2 00 = Xd X 0 Td0 re0q ¼ Efq eq1

d

0 Tq0 re0d ¼ ed1

> > > > > 0 00 > X X > q q 00 00 > Tq0 red ¼ e d2 > ; 00 X X q

;

ð6:46Þ

q

where Efq ¼

Xaf uf : Rf

ð6:47Þ

Efq is the voltage across the armature winding when synchronous generator is connected to no load at the steady-state operation. In fact, vf/Rf is an imaginary field current due to vf at steady state. During the transient process, it is not equal to the actual if. From the definition of (6.41) we can see that the product of this steadystate field current and Xaf gives the no-load voltage. Hence Efq is called the imaginary voltage. To express eq1, eq2, ed1, and ed2 directly from the flux linkage equation of rotor windings of (6.44), we have

eq1 eq2 ed1 ed2

9 Xd Xd00 0 Xd Xd0 00 > > ¼ 0 e e > > Xd Xd00 q Xd0 Xd00 q > > > > 00 00 > Xd Xd 0 Xd Xd 00 > 00 > ¼ 0 eq þ 0 eq þ ðXd Xd Þid > > 00 00 = Xd Xd Xd Xd : 00 0 Xq Xq 0 Xq Xq 00 > > > ¼ 0 e e > > Xq Xq00 d Xq0 Xq00 d > > > > 00 00 > Xq Xq 0 Xq Xq 00 > 00 > ¼ 0 ed þ 0 ed ðXq X Þiq > ; 00 00 Xq Xq Xq Xq

ð6:48Þ

Substituting (6.48) into (6.43) and (6.46) we can obtain the flux linkage equation of the armature windings ’d ¼ e00q Xd00 id ’q ¼ e00d Xq00 iq and the voltage equation of rotor windings

) ð6:49Þ

350

6 Mathematical Model of Synchronous Generator and Load

9 Xd Xd00 0 Xd Xd0 00 > ¼ 0 e þ e þ Efq > > > > Xd Xd00 q Xd0 Xd00 q > > > > 00 00 0 00 0 00 = Td0 req ¼ eq eq ðXd Xd Þid 0 Td0 re0q

0 Tq0 re0d ¼

Xq Xq00 0 Xq Xq0 00 e þ e Xq0 Xq00 d Xq0 Xq00 d

> > > > > > > > > ;

00 Tq0 re00d ¼ e0d e00d þ ðXq0 Xq00 Þiq

:

ð6:50Þ

In (6.47) we still have two basic parameters Xaf and Rf. To avoid these two parameters in the expression, we need to choose proper base values such that in per unit system we have Xaf ¼ Rf and hence Efq ¼ vf. This choice of base values is usually called ‘‘unit excitation voltage/unit stator voltage’’ per unit system. The details are as follows. As we have introduced previously, SB is decided by the choice of base values on the generator stator side. For each winding on the rotor, we have to choose a base value for either voltage or current and derive the other. In the ‘‘unit excitation voltage/unit stator voltage’’ per unit system, we first choose the base value for the voltage of the field winding VfB and then derive the base value for field current IfB from (6.23). We choose VfB such that when synchronous generator operates at steady state, is subject to no load and rotates at synchronous speed, the voltage of stator winding is equal to the base value of stator voltage. Obviously, VfB can be gained by experiment. From the above definition about VfB, in (6.14) and (6.15) we only have if 6¼ 0, we have 9 vd ¼ 0 > = vq ¼ oB maf if ¼ VB : > ; vf ¼ Rf if ¼ Vf B So we can obtain Vf B ¼

Rf VB : oB maf

Because Zf B ¼ Vf B =IfB , we have Rf oB maf oB maf Rf ¼ ¼ Rf If B ¼ Zf B Rf VB ZB

If B : IB

Comparing the above equation with Xaf in (6.28), we can see Rf* ¼ Xf*. Hence in per unit Xaf Efq ¼ vf ¼ vf : ð6:51Þ Rf Up to this point, we have established the mathematical model of synchronous generator represented by 11 machine parameters that consists of the voltage

6.2 Mathematical Model of Synchronous Generator

351

equation of armature windings (6.45), flux linkage equation of armature windings (6.49), and voltage equation of rotor windings (6.50). We should point out that this model only needs the specific choice of base value for field winding. Base value for the voltage or current of damping windings can be selected according to (6.23). Besides, voltage of field winding vf is affected by excitation control and hence Efq in (6.50) will be discussed further in Sect. 6.3.

6.2.3

Simplified Mathematical Model of Synchronous Generator

In the above discussion, we established the mathematical model of synchronous generator where four rotor windings, f, g, D, and Q, are used. From (6.50) we can see that the electromagnetic transient of rotor windings is depicted by four differential equations. In a modern power system, there could be over 1,000 generators in synchronous operation. Higher-order differential equations could result in numerical difficulty in power system analysis and calculation. Therefore, in practice the mathematical model of a synchronous generator is often simplified according to requirements of computing accuracy, and only for those generators that we are particularly concerned about are higher-order models used. The simplification can be classified according to how to ignore certain rotor windings, leading to three rotor-winding model, two rotor-winding model, nondamping-winding model and constant e0q model (classical model). All these models can be derived from the full four rotor winding model of a synchronous generator. 1. Three rotor winding model (f, D, Q). For a salient-pole generator, usually we only consider one equivalent damping winding Q on q-axis and ignore the existence of g winding. This means that in the four rotor winding model, ig ¼ fg ¼ 0. Hence in (6.41), ed1 ¼ 0 and in (6.42), e0d ¼ 0 and Xq0 ¼ Xq . The voltage equations of rotor windings are reduced to an order three model 0 Td0 re0q ¼

9 Xd Xd00 0 Xd Xd0 00 > > e þ e þ E fq > q q = Xd0 Xd00 Xd0 Xd00

00 Td0 re00q ¼ e0q e00q ðXd0 Xd00 Þid 00 Tq0 re00d ¼ e00d þ ðXq0 Xq00 Þiq

> > > ;

:

ð6:52Þ

There is no change in the voltage and flux linkage equation of the armature windings. 2. Two winding model (f, g or double-axis model). We only consider one damping winding g on q-axis and ignore D, Q damping winding. This is the same as the assumption iD ¼ iQ ¼ fD ¼ fQ ¼ 0 in four winding rotor model. Hence in (6.41), we have eq2 ¼ ed2 ¼ 0 and in (6.42), e00q ¼ e00d ¼ 0. The flux linkage equation of armature windings becomes

352

6 Mathematical Model of Synchronous Generator and Load

)

’d ¼ e0q Xd0 id

:

’q ¼ e0d Xq0 iq

ð6:53Þ

Voltage equation of rotor windings is reduced to a second-order model 0 Td0 re0q ¼ e0q ðXd Xd0 Þid þ Efq

)

0 Tq0 re0d ¼ e0d þ ðXq Xq0 Þiq

:

ð6:54Þ

There is no change in the voltage equation of armature windings. 3. Nondamping winding model (f, or variable e0q model). Ignoring damping windings, we have iD ¼ iQ ¼ ig ¼ fD ¼ fQ ¼ fg ¼ 0. Hence in (6.41), ed1 ¼ eq2 ¼ ed2 ¼ 0 and in (6.42), e0d ¼ e00q ¼ e00d ¼ 0. The flux linkage equation of armature windings becomes ’d ¼ e0q Xd0 id ’q ¼ Xq iq

) :

ð6:55Þ

Voltage equation of rotor winding is reduced to a first-order model 0 Td0 re0q ¼ e0q ðXd Xd0 Þid þ Efq :

ð6:56Þ

There is no change in the voltage equation of armature windings. 4. Constant e0q model. We neglect damping windings and transient of field winding. Also we consider the right-hand side of (6.56) to be zero due to the control function of AVR, i.e., e0q ðXd0 Xd Þid þ Efq ¼ constant: Thus the mathematical model of synchronous generator is comprised of only the voltage and flux linkage equation of armature windings of (6.45) and (6.55). There is no differential equation for the rotor windings. Constant e0q model usually is used when the rotor motion equation of the synchronous generator is described by electrical torque. 5. Classical model. This is to use Xd0 ¼ Xq0 to further simplify the expression of output electrical power of a synchronous generator. The discussion above is the simplification of depicting rotor windings to reduce the order of the mathematical model of a synchronous generator. On the other hand, in the analysis of power system steady-state operation, the voltage equation of armature windings can be simplified in the following two ways:

6.2 Mathematical Model of Synchronous Generator

353

1. Ignoring the electromagnetic transient of armature windings. This is to neglect the induced voltage due to the variations of ’d and ’q in the voltage equation of armature windings. Thus the voltage equation of armature winding becomes vd ¼ o’q Ra id vq ¼ o’d Ra iq

) :

ð6:57Þ

For power system stability studies the above simplification is very important. From the flux linkage equation of armature winding (6.49) we can see that the differential of flux linkage of armature windings with respect to time will involve that of armature current with respect to time. Because the armature windings of a synchronous generator are connected to a transmission network that is formed by a certain topology of resistance, inductance, and capacitance, the differentiation of armature current with respect to time will require the description of the network by differential equations. This will greatly increase the order of the mathematical model of whole power system. In addition, if the electromagnetic transients of armature windings and network are not ignored, the armature current of the synchronous generator will contain high-frequency components. Under this circumstance, we must take very small integration time steps to achieve the required computing accuracy in the numerical solution of power system mathematical equations. For a modern large power system, increase of the order of its mathematical model and decrease of the required integration time step would add a heavy computational burden such that normal calculation would become impossible. In fact, compared to the electromechanical process of the synchronous generator, the electromagnetic transient behavior of the power network is sufficiently fast that it can be ignored as far as its influence on power system stability analysis and computation is concerned. From (6.57) we can see that when the electromagnetic transient of armature windings of synchronous generator is neglected, its voltage equations become algebraic equations, i.e., those depicting steady-state operation of the synchronous generator. 2. In the voltage equation of armature windings, we consider the rotor speed of synchronous generator o always to be the synchronous speed. This does not mean that during the transient, the rotor speed of synchronous generator does not change. It is because the range of o is small due to the existence of various control functions in generator operation. Hence in the voltage equation of the armature winding, the numerical variation caused by the small change of o is very small and hence can be ignored. This simplification does not result in great saving in computation. However, it has been shown that taking o ¼ 1 in the voltage equation of the armature winding of synchronous generator can partly correct the computational errors caused by ignoring the

354

6 Mathematical Model of Synchronous Generator and Load

electromagnetic transient [153]. Therefore, the voltage equation of armature winding becomes vd ¼ ’q Ra id vq ¼ ’d Ra iq

6.2.4

) :

ð6:58Þ

Steady-State Equations and Phasor Diagram

Mathematically, transient analysis of power systems is to solve a group of differential equations depicting power system transient behavior. Usually the steady-state operating point is the initial condition to solve the differential equations. In the following, we will derive the formula to calculate the initial conditions from the steady-state equations of the synchronous generator. In steady-state operation, the generator rotates at synchronous speed, all electrical variables are balanced and the current of damping windings is zero. Current id, iq, if, and eq1 associated with if as well as flux linkage of all windings are constant. In the following, we shall use capital letters to denote various steady-state electric variables. 1. Steady-state equations represented by synchronous reactance From (6.43) we have ) Fd ¼ Xd Id þ Eq1 : Fq ¼ Xq Iq

ð6:59Þ

At steady state Eq1 ¼ Xaf If ¼ Xaf

Vf ¼ Efq : Rf

Substituting (6.59) into (6.58) we obtain Efq ¼ Vq þ Ra Iq þ Xd Id 0 ¼ Vd þ Ra Id Xq Iq

) :

ð6:60Þ

After load flow calculation, we have had terminal voltage V_ t and current I_t of the synchronous generator in xy coordinate. To obtain Vd, Vq, Id, and Iq in dq coordinate, we need to find the connection between these two coordinate systems, i.e., to find the angle between them. For this purpose, we multiply the first equation of (6.60) by j and add it to the second equation jEfq jðXd Xq ÞId ¼ V_ t þ ðRa þ jXq ÞI_t :

6.2 Mathematical Model of Synchronous Generator

355

We can define an imaginary voltage E_ Q according to the above equation to be E_ Q ¼ V_ t þ ðRa þ jXq ÞI_t :

ð6:61Þ

Because E_ Q and jEfq are in the same phase, from phasor diagram (Fig. 6.2a), we can see that the angle between E_ Q and x, d, is that between d–q and x–y coordinate. Hence from (6.61) we can find d and obtain the transformation between two coordinate systems as follows

¼

sin d cos d cos d sin d

Ax ; Ay

ð6:62Þ

is Ax

is y Ax

It

EQ

vd

E fq

δ

vt

q

jxq I q

Ra I d

Id

R a It

It

R a I q jx′d I d

Ix

E′q Axis q E′

d

Axis q

It

t

R a Id

Ix

jxd Id

jxqIq

v

δ

Id

R a Iq

jx

vd

vq

Iq

jx

d

Iy

vq

Iq

y

It

Ad Aq

jx ′

Ra I t

It

is Ax

x

x

is Ax

Axis d

Axis d

a

b is Ax

Iq

δ

Id Ix

E q′′

vq

vt

It

Axis q

E ′′ Ed′′

It Ra

vd

y

jx ′q′ I q

jxd′′ I d

is x Ax

Axis d

c Fig. 6.2 Phasor diagram of steady-state operation of synchronous generator (a) when synchronous reactance is used (b) when transient reactance is used (c) when subtransient reactance is used

356

6 Mathematical Model of Synchronous Generator and Load

Ax Ay

sin d cos d ¼ cos d sin d

Ad ; Aq

ð6:63Þ

where A denotes current, voltage, flux linkage, and various EMF. After Vd, Vq, Id, and Iq are found, from (6.60) we can calculate the initial value of vf, Vf ¼ Efq. 2. Steady-state equations represented by transient reactance From the first and third equation in (6.44) we have E0q ¼ ðXd Xd0 ÞId þ Eq1

) :

E0d ¼ ðXq Xq0 ÞIq

Noting Eq1 ¼ Efq at steady state and canceling Xd and Xq by substituting the first and second equation in (6.60) into the first and second above equation, we have E0q ¼ Vq þ Ra Iq þ Xd0 Id

) :

E0d ¼ Vd þ Ra Id Xq0 Iq

ð6:64Þ

3. Steady-state equations represented by subtransient reactance From the second and fourth equation of (6.44), we can have E00q ¼ ðXd Xd00 ÞId þ Eq1

) :

E00d ¼ ðXq Xq00 ÞIq

Taking the similar procedure, from (6.64) we can obtain E00q ¼ Vq þ Ra Iq þ Xd00 Id E00d ¼ Vd þ Ra Id Xq00 Iq

) :

ð6:65Þ

Equations (6.60), (6.64), and (6.65) comprise steady-state equations of a synchronous generator adopting the four rotor winding model. From those three equations we can calculate the initial values of five state variables, vf, e0q ; e0d ; e00d , and e00q . Phasor diagrams related to those three equations are shown in Fig. 6.2. When a simplified model of the synchronous generator is used, we can calculate required initial values of state variables directly from the above steady-state equations of the four rotor winding model. For example, when the damping windings are ignored, we have 9 Efq ¼ Vq þ Ra Iq þ Xd Id > =

0 ¼ Vd þ Ra Id Xq Iq : > ; 0 0 Eq ¼ Vq þ Ra Iq þ Xd Id

6.2 Mathematical Model of Synchronous Generator

6.2.5

357

Mathematical Equations Considering Effect of Saturation

In the above discussion, we have established mathematical equations of the synchronous generator under the condition that the magnetic circuit of machine is unsaturated. In practice, to save materials, the design and manufacture of synchronous generator usually makes the iron core of both stator and rotor slightly saturated when operating at rated conditions. At some particular operating conditions, with the increase of flux density, saturation would become very obvious and serious. In system planning and operation analysis, errors caused by ignoring saturation are small. However, in certain applications, such as in transient stability analysis, with detailed model of AVR and its limiters included, the effect of machine saturation can greatly affect the accuracy of analysis and calculation. Study on the effect of saturation started as early as about 60 years ago. The mathematical model of a synchronous generator will become extremely complicated if machine saturation is modeled in great detail. This is because the extent of saturation of a magnetic circuit is closely related to the total mmf in the machine air gap. It is required to combine d- and q-axis mmf to air-gap total mmf and then to find the corresponding magnetic flux and linkage from the saturation curve. Even though air-gap total mmf has a strict sinusoidal distribution in space, mmf varies in different positions. Thus saturation at various positions in space is different, which will cause distortion of the flux wave in the air gap. Therefore, in practice, considering the simplicity of model used, effectiveness of parameters and accuracy of computation, proper approximation is applied to take account of the effect of machine saturation [155–158]. In the following, we shall introduce a method commonly used in stability analysis [156]. The assumptions to apply the method are: 1. The effect of saturation is simply considered on d- and q-axis separately. The difference of magnetic reluctance in d- and q-axis magnetic circuits is only caused by that of length of air gap in the direction of d- and q-axis. 2. On a same axis, the extent of saturation depends on the Potier voltage behind Potier reactance Xp. The higher the Potier voltage, the more serious the saturation. Potier voltage on d- and q-axis is given by the following equation vdp ¼ vd þ Ra id Xp iq vqp ¼ vq þ Ra iq þ Xp id

) :

ð6:66Þ

In addition, the extent of saturation of voltage and flux linkages of armature and rotor windings is approximately considered to be same on the same axis. 3. The distortion of the distribution wave of air-gap flux does not affect the selfinductance and mutual inductance of various windings and the unsaturated values of winding reactance.

358

6 Mathematical Model of Synchronous Generator and Load

Fig. 6.3 No-load saturation characteristic of synchronous generator

Vf

The unsaturated characteristics

(Vqp)

Vf 0

The no-load saturated characteristics

Vf

0

if0 if

if

The extent of saturation is described by saturation factor. For d-axis, saturation factor Sd can be calculated from the saturation characteristic of machine in no-load operation. This is because vqp is equivalent to the voltage of q winding induced from the resultant d-axis air-gap flux. From Fig. 6.3 of no-load saturation characteristic of synchronous generator, we can find the unsaturated value of vqp0 from a certain value of vqp. Hence we can define Sd to be Sd ¼ f ðvqp Þ ¼

vqp0 1: vqp

ð6:67Þ

Obviously, the bigger the value of Sd is, the more saturated is the synchronous generator. Zero Sd indicates the case of no saturation. For q-axis, the saturation characteristic is difficult to obtain through experiment. Hence from the first assumption above, the saturation factor Sq is also determined by using the no-load saturation characteristic of synchronous generator, using the following equation Sq ¼

Xq f ðvdp Þ: Xd

ð6:68Þ

To calculate the saturation factor, one commonly used method is to approximately fit the no-load saturation characteristic curve of Fig. 6.3 by an analytical function, such as if ¼ aVt þ bVtn : Obviously when b ¼ 0, we have the characteristic curve without saturation if 0 ¼ aVt : Hence according to triangle similarity of Fig. 6.3, we have Sd ¼

vqp0 vqp0 vqp if if 0 avqp þ bvnqp avqp b n1 1¼ ¼ ¼ ¼ vqp : vqp vqp if 0 avqp a

6.2 Mathematical Model of Synchronous Generator

359

That is Sd ¼ cvn1 qp ;

ð6:69Þ

where c ¼ b/a. Similarly, from (6.68) we have Sq ¼ c

Xq n1 v : Xd dp

ð6:70Þ

In the following, we shall discuss the voltage equations of the field winding, voltage and flux linkage equations of armature windings of a synchronous generator, taking account of the saturation effect. From the derivation of voltage equations of field winding of (6.50) without considering saturation effect, we can see that on the righthand side of the equations we have the voltage drop across the equivalent resistance of the field winding caused by field current and the external voltage applied on the rotor windings (i.e., the excitation voltage Vf). Hence there should no problem of saturation about this part in the equations. Hence when we consider the saturation effect, we shall still use unsaturated values for those on the right-hand side of the equations. On the other hand, on the left side of the equations, we have the induced voltage by variations of flux linkage with time. Hence when saturation effect is taken into account, for those terms on the left-hand side of the equations we should use their saturated values associated with actual flux linkage. According to the previous assumption (2) and (6.67), we know that on d-axis, the ratio of unsaturated value to the saturated of each voltage and flux linkage is (1 þ Sd). Similarly, on qaxis, the ratio is (1 þ Sq). Therefore, when we consider the saturation effect, the voltage equations of the field winding of a synchronous generator are 9 Xd Xd00 Xd Xd0 > 0 00 > ð1 þ S Þe þ ð1 þ S Þe þ E d qs d qs fq > > > Xd0 Xd00 Xd0 Xd00 > > > > 00 00 0 00 0 00 = Td0 reqs ¼ ð1 þ Sd Þeqs ð1 þ Sd Þeqs ðXd Xd Þid 0 Td0 re0qs ¼

0 Tq0 re0ds

Xq Xq00 Xq Xq0 0 ¼ 0 ð1 þ S Þe þ ð1 þ Sq Þe00ds q ds Xq Xq00 Xq0 Xq00

00 Tq0 re00ds ¼ ð1 þ Sq Þe0ds ð1 þ Sq Þe00ds þ ðXq0 Xq00 Þiq

> > > > > > > > > ;

;

ð6:71Þ

where subscript s denotes the saturated value of each voltage. Taking saturation effects into account, we have the flux linkage equations of armature windings of (6.49) becoming ð1 þ Sq Þ’qs ¼ ð1 þ Sq Þe00ds Xq00 iq ð1 þ Sd Þ’ds ¼ ð1 þ Sd Þe00qs Xd00 id

) :

ð6:72Þ

When we do not consider the saturation effect, from the voltage equations of armature windings of synchronous generator (6.58) and the definition of Potier

360

6 Mathematical Model of Synchronous Generator and Load

voltage of (6.66), we can obtain the relationship between the Potier voltage and flux linkage of armature windings to be vdp0 ¼ ’q Xp iq vqp0 ¼ ’d þ Xp id

) :

ð6:73Þ

According to the relationship between saturated and unsaturated value, we can have ð1 þ Sq Þvdp ¼ ð1 þ Sq Þ’qs Xp iq ð1 þ Sd Þvqp ¼ ð1 þ Sd Þ’ds þ Xp id

) :

Substituting (6.72) into the above equation we can establish the relationship between the Potier voltage and the EMF with saturation considered, to be vdp vqp

9 Xq00 Xp > > ¼ þ iq > = 1 þ Sq : Xd00 Xp > > 00 > ¼ eqs id ; 1 þ Sd e00ds

ð6:74Þ

Substituting the above equation into the defining equation of the Potier voltage, we have the voltage equations of armature windings with saturation being considered, to be 9 Xq00 Xp > > vd ¼ R a id þ þ Xp iq > = 1 þ Sq : 00 > Xd Xp > 00 > vq ¼ eqs Ra iq þ X p id ; 1 þ Sd e00ds

ð6:75Þ

Equations (6.66), (6.67), (6.71), and (6.75) form the mathematical model of synchronous generator with machine saturation being taken into account. From the model it would be straightforward to derive the steady-state equations of a synchronous generator. In practice, we often assume that stator leakage flux does not saturate. Hence we can use Xs as the Potier reactance Xp.

6.2.6

Rotor Motion Equation of Synchronous Generator

6.2.6.1

Rotor Motion Equation of Stiff Rotor

If we consider the prime mover and generator rotor to be a single mass, the rotor motion equation of the whole generation unit is

6.2 Mathematical Model of Synchronous Generator

9 dd = ¼ ðo 1Þos > dt ; do > TJ ¼ Tm Te ; dt

361

ð6:76Þ

where TJ ¼ 2Wk/SB, d is the electrical angle between q-axis of generator rotor and a reference axis x that rotates at synchronous speed. This angle is a dimensionless number and can be measured in radians (rad), TJ is the moment of inertia of generation unit measured in seconds (s), Wk the rotating kinetic energy of the rotor rotating at synchronous speed and measured in Joules (J), SB the base value of generation capacity in V A; Tm and Te are the output mechanical torque of prime mover and the electromagnetic torque of the generator in per unit; their base value is SB/Os measured in radian/second (rad s1), where Os is the mechanical synchronous speed of rotor. The positive direction of Tm and Te is taken to be as same as and opposite to that of rotation of the rotor, respectively. In some references, the mechanical inertia is represented by H ¼ Wk/SB. Obviously, TJ ¼ 2H. In addition, we ought to note the following two issues: 1. Since the product of torque and speed is the power and O/Os ¼ o/os ¼ o*, in per unit we can have ) Pm ¼ Tm o ; ð6:77Þ Pe ¼ Te o where Pm is the output mechanical power from the prime mover and Pe the electromagnetic power of the synchronous generator. As pointed out before, various functions of power system stability control result in a small change of o*. Hence in order to save computational time, sometimes we can just simply take o* to be 1. Thus in per unit, torque is equal to power. 2. Rotor rotation is always subject to air resistance and friction between bearing and shaft. This results in a damping torque to rotor motion. Often we assume that this damping torque is approximately proportional to rotor speed and represent it by the product of a damping coefficient D and speed o*. Considering what has been discussed above, when time is also represented in per unit, the rotor motion equation becomes 9 dd > > ¼ o 1 = dt : ð6:78Þ do > > ; TJ ¼ Do þ Pm Pe dt We would point out that the mechanical torque and power involved in the above rotor motion equation are subject to the control of the governing system of generation unit. Hence the appearance of mechanical torque and power will lead to the establishment of equations for the governing system. This will be discussed in Sect. 6.4.

362

6 Mathematical Model of Synchronous Generator and Load

In (6.78), we consider a combined rotor of generator and prime mover to be a single lumped mass. This consideration will usually bring about no obvious errors when carrying out transient stability analysis. However, when the subsynchronous resonance of power systems is studied, we cannot ignore the existence of rotor shaft elasticity, since large steam-turbine generation units often consist of multiple stage turbines and their shafts can be as long as several tens of meters. In this case, we can consider the exciter, generator rotor, and each turbine section to be separate lumped masses. Thus elasticity of the whole shaft system can be treated as torsional springs between each mass. Therefore, with elasticity being taken into account, rotation speed of each mass could be different during a transient process, resulting in difference in relative angular position of each mass. The motion equation of each mass forms the motion equation of the whole shaft system. Detailed discussion can be found in [159].

6.2.6.2

Electromagnetic Torque and Power of Synchronous Generator

In the rotor motion equation of (6.78), mechanical torque (or power) from prime mover and electromagnetic torque (or power) of synchronous generator are introduced. The former is included in the mathematical model of the prime mover and governing system of the generation unit, which will be discussed in Sect. 6.4. Here we shall introduce the computing model of electromagnetic torque and power. Electromagnetic torque represents the function of force applied on the rotor from the mutual electric and magnetic interactions between stator and rotor of the synchronous generator. Theoretical proof has been provided that electromagnetic torque is equal to the partial differentiation of total magnetic field energy stored in various windings to rotor angle [148], i.e., Te ¼

@WF ; @y

ð6:79Þ

where y is the angle between d-axis and a-axis of armature winding (see Fig. 6.1a) and WF is the total magnetic energy stored in three-phase armature windings and rotor windings, which can be represented as 1 1 WF ¼ ð’a ia þ ’b ib þ ’c ic Þ þ ð’f if þ ’D iD þ ’g ig þ ’Q iQ Þ: 2 2

ð6:80Þ

Because the reference positive direction of armature current is opposite to that of associated flux linkage, we have a negative sign in the above equation. From the base value for torque TB ¼ SB/OB and (6.2)–(6.7), we can obtain

6.3 Mathematical Model of Generator Excitation Systems

Te ¼ ’d iq ’q id :

363

ð6:81Þ

The above equation shows that electromagnetic torque is independent of zerosequence components, because they do not couple with rotor windings. In addition, although the above equation has been established from the four winding model, it is applicable to other higher or lower winding models with only slight differences in derivation. When the four rotor winding model is used, substituting the flux linkage equation of (6.49) into the above equation, we can obtain the expression of electromagnetic torque to be Te ¼ e00d id þ e00q iq ðXd00 Xq00 Þid iq :

ð6:82Þ

From the above expression and (6.77), we can directly establish the expression of electromagnetic power where state variables o*, e00d , and e00q are included. This will bring about a heavy computing burden in the solution. Hence to solve this problem, we can substitute the voltage equation of armature windings of (6.45) into (6.81) and use (6.77) to obtain pe ¼ ud id þ uq iq þ Ra i2d þ i2q id p ’d iq p ’q ;

ð6:83Þ

where Ra i2d þ i2q is the copper loss of the armature windings. When the transient of armature windings is ignored, from the comparison of the above equation with the output power expression of the synchronous generator in (6.29), we can see that the electromagnetic power of the generator is the sum of generator output power and copper loss of generator armature windings. Finally we would like to mention here that (6.83) is also applicable to cases when other types of rotor winding model is used and/or machine saturation is considered.

6.3

Mathematical Model of Generator Excitation Systems

In (6.50) we introduced variable Efq in the per unit system as ‘‘unit excitation voltage/unit stator voltage,’’ that is equal to voltage vf applied to the field winding. Hence we need to establish the mathematical model of generator excitation systems. The basic function of a generator excitation system is to provide the generator field winding with appropriate DC current to generate a magnetic field in the distributed space of the generator armature windings. In earlier times, the excitation system regulated the excitation voltage through manual control, to maintain the required terminal voltage of the generator and reactive power supply from the generator. More recently, various types of excitation and AVR were proposed

364

6 Mathematical Model of Synchronous Generator and Load

and used. In the 1960s, the proposal and application of power system stabilizers (PSS) further enhanced the role played by excitation control systems to improve power system stability. With the advancement of control theory and computer control technology, further new types of excitation regulators have been proposed. Their control tasks have been extended from simple terminal voltage regulation of the generator to multiple excitation control functions. Feedback signals used have developed from a single deviation of generator terminal voltage to the superimposition of various signals on the voltage deviation, based on factors such as electromagnetic power, electrical angular speed, system frequency, armature current, and deviation of excitation current or voltage and their combinations. The control strategy started with simple proportional control and has been enhanced by applying proportional–integral–differential (PID) control, multivariable linear system control schemes, self-tuning control, adaptive control, fuzzy control, and nonlinear control. In recent years, digital excitation controllers based on microprocessors or microcomputers have been developed and installed. In the near future, research into, and innovative applications of, excitation control will involve the development of digital excitation control systems realized by microcomputers and using modern control theory. Relatively accurate analysis of power system dynamics must be supported by mathematical models of the excitation system. Development and design of new types of excitation controller need to establish mathematical models for simulation to check if the dynamic performance is satisfactory. In this section, we shall only introduce the mathematical models of widely used excitation systems and the design principle of excitation regulators will not be discussed. Also we shall not discuss the newer type of excitation controllers, such as linear optimal excitation controller (LOEC), nonlinear optimal excitation controller (NOEC), because they are still at the stage of further theoretical research and testing. Figure 6.4 shows the construction of a general excitation system. The exciter provides field current to the field winding of the generator. The regulator controls the field current. The measurement unit for generator terminal voltage and load compensation measures generator terminal voltage V_ t and compensates for the load current of generator I_t , respectively. The auxiliary control signals are sent through the auxiliary controller. One of the most widely used

Protection and limiter Terminal voltage measurement and load compensation Reference

Regulator

Exciter

Generator Auxiliary controller

Fig. 6.4 Excitation system of generator

To power network

6.3 Mathematical Model of Generator Excitation Systems

365

auxiliary controllers is PSS. Protection and limiter are incorporated to ensure the generator’s operation within various allowed constraints. In Sect. 6.2, we have discussed the mathematical model of a synchronous generator. In the following we shall introduce the mathematical models of excitation systems of generators for power system stability analysis, as shown in Fig. 6.4, block by block. These models are applicable to power system operation when system frequency deviation is within 5% and system oscillation frequency is below 3 Hz. Generally speaking, for the study of SSR or other problems of shaft torsional oscillations, these models would not be precise enough.

6.3.1

Mathematical Model of Exciter

According to the different means of providing excitation power sources, exciters can be classified into three types: DC exciter systems, AC exciter systems, and static excitation systems. The two former types are also called rotational excitation systems. In the following, we shall introduce each of the three types of exciter.

6.3.1.1

Mathematical Model of DC Exciter

Due to the high cost of maintenance, DC exciters have not been used in recently built large generation units. However, in some power systems, we can still see DC exciters in operation. Hence it is necessary to introduce their mathematical model. We shall introduce the establishment of a mathematical model of the general case of a DC exciter that has both self-excitation and separate excitation. Figure 6.5 shows the configuration of the DC exciter. In Fig. 6.5, E represents armature of the exciter; Ref and Lef, Rsf and Lsf is the resistance and self-inductance of the self-excited and separately excited windings, respectively; ief, isf, and icf are the currents of the self-excitation, separate excitation,

Rc

isf Vsf

Rsf

Ref

Fig. 6.5 Configuration of a DC exciter

ief E

icf Lsf

if

Lef

Vf

366

6 Mathematical Model of Synchronous Generator and Load

and compound excitation, respectively; vsf is the voltage externally applied on the separately excited winding; and Rc is a variable regulating resistor. For simplicity of analysis, we assume that self-excited and separately excited windings have the same number of turns, or number of turns and parameters of the separately excited winding have been transferred to the side of the self-excited winding. Hence we can obtain the following voltage equations and flux linkage equations (without considering magnetic saturation). vf ¼ Rc ief þ Ref ðicf þ ief Þ þ p’ef

) ;

vsf ¼ Rsf isf þ p’sf ’ef ¼ Lef ðicf þ ief Þ þ Mes isf ’sf ¼ Mes ðicf þ ief Þ þ Lsf isf

ð6:84Þ

) :

ð6:85Þ

In the above flux linkage equations, we can approximately consider that the selfexcited winding and the separately excited winding are coupled completely. Hence leakage reactance of each winding is zero and unsaturated self-inductance and all mutual inductance have the same value. From (6.85) we can have ’L0 ¼ ’ef ¼ ’sf ¼ Lif S ;

ð6:86Þ

where L ¼ Lef ¼ Lsf ¼ Mes if S ¼ icf þ ief þ isf

) :

ð6:87Þ

’L0 is the flux linkage of the self-excited winding and separately excited, winding without considering saturation, ifS is the total excitation current provided by the DC exciter. If the saturation effect is considered, the relationship between the actual flux linkage ’L and the total excitation current provided by DC exciter ifS is determined according to Fig. 6.6a, which shows the saturation characteristic curve of the DC exciter. Similarly to (6.67), we define the saturation factor of the DC exciter to be SE ¼

if S ’L0 1¼ 1: ’L if S0

ð6:88Þ

As shown in Fig. 6.6, in (6.88), ifS0 is the total excitation current required to generate ’L without considering saturation. The value of SE represents the level of saturation of the DC exciter, describing the relationship between saturated flux linkage ’L and unsaturated flux linkage ’L0. It is usually obtained from the load characteristic curve of the exciter. Figure 6.6b shows that because the load of the exciter is fixed, i.e., when the influence of excitation current of generator if on the

6.3 Mathematical Model of Generator Excitation Systems

yL

vf 0

yL0 =Lif ∑ The saturated characteristics

yL

The unsaturated characteristics vf 0 = bif ∑

vf

The unsaturated characteristics

yL0

367

The saturated characteristics

vf

if ∑

if ∑ a 0

if ∑0 if ∑

b

if ∑0 if ∑

Fig. 6.6 Saturation characteristic curve of DC exciter (a) Relationship between flux linkage and excitation current (b) Load characteristic curve

armature voltage of the exciter during transients is ignored, the output voltage of exciter is approximately proportional to its internal EMF. If the variation of speed is neglected, flux linkage ’L is proportional to voltage vf. Hence the unsaturated characteristic in Fig. 6.6b can be expressed as vf 0 ¼ bif S :

ð6:89Þ

b is the slope of the unsaturated load characteristic curve of exciter, measured in Ohms. From the equation above and (6.86) we can obtain ’L0 ¼

L vf 0 : b

Because flux linkage ’L is proportional to voltage vf, the equation above can be extended to be ’L ¼

L vf : b

ð6:90Þ

Dividing both sides of the first equation, (6.84), by Rc þ Ref, the second equation by Rsf and adding these two equations, as well as using (6.86), (6.87), and (6.90) we can obtain vf vsf Rc 1 L L þ ¼ if S icf þ þ rvf : ð6:91Þ Rc þ Ref Rsf Rc þ Ref b Rc þ Ref Rsf

368

6 Mathematical Model of Synchronous Generator and Load

From (6.90), (6.88), and (6.89) we have bif S b b ’L0 b Lif S vf ¼ ’ L ¼ ¼ ¼ : L L 1 þ SE L 1 þ SE 1 þ SE Substituting the above equation into (6.91) and canceling variable ifS, we can obtain

b SE þ 1 Rc þ Ref

þ ðTef þ Tsf Þr vf ¼

b bRc vsf þ icf ; Rsf Rc þ Ref

ð6:92Þ

where ) Tef ¼ L=ðRc þ Ref Þ ; Tsf ¼ L=Rsf

ð6:93Þ

where Tef and Tsf are the time constants of self-excited and separately excited windings, respectively (measured in seconds). Equation (6.92) gives the relationship between input vsf, icf, and output vf of the exciter using physical units. In order to combine it with the mathematical model of generator in per unit, established in Sect. 6.2, we need to convert (6.92) into per unit form. Here we should use the same base value VfB that has been chosen in Sect. 6.2 for vf. To decide the base value for vsf and icf, we divide both sides of (6.92) by VfB. Then we can see that when base voltage for the excitation current and voltage of the separately excited winding of the exciter are chosen according to the following equation, the equation in per unit is in the most simple form. If SB ¼ Vf B =b Vsf B ¼ Rsf Vf B =b

) :

ð6:94Þ

Hence (6.92) in per unit becomes ðSE þ KE þ TE rÞvf ¼ vsf þ Kcf icf ;

ð6:95Þ

9 KE ¼ 1 b=ðRc þ Ref Þ > = TE ¼ T ef þ Tsf : > ; Kcf ¼ Rc =ðRc þ Ref Þ

ð6:96Þ

where

KE, TE, and Kcf are termed self-excitation factor, time constant, and gain of compound excitation, respectively. By changing variable resistance Rc, these three parameters can be adjusted properly. Equation (6.95) is the mathematical

6.3 Mathematical Model of Generator Excitation Systems Fig. 6.7 Block diagram of DC exciter

vsf icf

+

369

∑ +

Kcf

+

∑

1 sTE

vf

KE + SE

model of the exciter shown in Fig. 6.5. Figure 6.7 is its block diagram where the per unit subscript * has been omitted. Using the same method that has been adopted to consider the saturation effect in synchronous generators, in Sect. 6.2, we can obtain the relationship between the saturation factor SE and output voltage of the DC exciter. To match the saturated load characteristic of the exciter by an approximate function, we can derive the following equation, as we have done (6.69) SE ¼ aE vnf E 1 =bE :

ð6:97Þ

1. The case without separately excited winding is equivalent to Rsf ¼ 1, vsf ¼ 0. Hence from (6.93) and (6.96) we have TE ¼ Tef. 2. The case with only a separately excited winding is equivalent to Rc ¼ 1, icf ¼ 0. Hence from (6.93) and (6.96) we have TE ¼ Tsf and KE ¼ 1.

6.3.1.2

Mathematical Model of AC Exciter

An AC exciter uses a synchronous machine (alternator), usually rotating on the shaft of the generator. AC output from the armature winding of the exciter is rectified through a three-phase noncontrollable, or controllable, bridge rectifier to supply current to the field winding of the generator. There are two types of rectifiers, stationary rectifiers and rotating rectifiers, and two methods of excitation: self-excitation and separate excitation. Hence there are different combinations of types of rectifiers and means of excitation. In the following, we shall first discuss the mathematical model of the exciter and then that of the rectifiers. The majority of AC exciters use separate excitation. In this case, we can use the mathematical model of a synchronous generator, established in Sect. 6.2, to represent the AC exciter. However, the load of an AC exciter is the field winding of the generator and its operating conditions are much simpler than the generator’s. Hence to reduce the effort in analysis and calculation, we need not describe an AC exciter in such detail as we have done for a generator. There are several methods to simplify the mathematical model of a synchronous generator to derive a mathematical model of an AC exciter. Here we shall introduce one simple and commonly used method as follows.

370

6 Mathematical Model of Synchronous Generator and Load

Because the load of the exciter is the field winding of the generator, the armature current of the exciter is almost purely inductive. Hence the q component of armature current of exciter is approximately zero. In the mathematical model of a synchronous generator without damping windings considered, ignoring (6.55) and (6.58), we can obtain the voltage equation of the armature winding of the exciter to be vd ¼ 0 vq ¼ ’d ¼ e0q Xd0 id

) :

ð6:98Þ

In the above equation, we can further ignore the influence of stator current of the exciter on stator voltage. This leads to stator voltage being equal to the transient voltage. In (6.56), due to the adoption of the base value system of ‘‘unit excitation/ unit stator voltage’’, Efq is equal to field voltage. Using the same assumption, denoting the exciter’s field voltage by vR, stator voltage by vE, stator current by iE, using subscript E to denote synchronous reactance, transient reactance, and various time constants of the exciter and following a similar procedure as for deriving (6.56), we can establish the mathematical model of the exciter without considering saturation as follows TE rvE ¼ vR eqE 0 eqE ¼ vE þ ðXdE XdE ÞiE

) :

ð6:99Þ

With saturation being considered, similarly to the procedure for deriving (6.71), we can obtain 0 ð1 þ SE ÞvE ¼ eqE ðXdE XdE ÞiE ;

ð6:100Þ

where we use the same method to gain the saturation factor of exciter SE as we have done for a DC exciter, i.e., to fit the saturation curve of exciter by the approximate function SE ¼ aE vnEE 1 =bE :

ð6:101Þ

We should note that stator voltage vE and current iE only enter the field winding of the generator after rectification. The relationship between vE and vf will be established later in the mathematical model of rectifier. Here we shall derive the connection between iE and if first. When the exciter supplies the field winding of the generator through a threephase noncontrollable bridge rectifier, output current from the rectifier if is the field current of the generator that is approximately proportional to the input current of the 0 rectifier, i.e., armature current of exciter iE. Hence replacing ðXdE XdE ÞiE in (6.100) by KDif, we can describe this relationship as

6.3 Mathematical Model of Generator Excitation Systems Fig. 6.8 Block diagram of transfer function of separately excited AC exciter

vR

+

371

vE

1 sTE

∑

eqE

+

1 + SE

∑

+

TE rvE ¼ vR eqE eqE ¼ ð1 þ SE ÞvE þ KD if

KD

if

) :

ð6:102Þ

From this, the mathematical model of a separately excited AC exciter, using a threephase noncontrollable bridge rectifier, can be expressed by the block diagram of Fig. 6.8. For self-excited exciter, replacing (1 þ SE) in (6.102) and Fig. 6.8 by (KE þ SE) we can obtain its mathematical model [160], where KE is self-excitation factor and KE < 1. Because an AC exciter is connected to the field winding of the synchronous generator through a rectifier, base values of its armature and field voltage and current must not only satisfy various rules used when the mathematical model of synchronous generator is established in Sect. 6.2, but also be related to the mathematical model of the rectifier. This will be discussed in Sect. 6.3.1.3. 6.3.1.3

Mathematical Model of Power Rectifier

An AC exciter usually supplies excitation to the generator through a three-phase noncontrollable or controllable rectifying circuit. In the following, we shall introduce a mathematical model of a noncontrollable rectifier. The input to the rectifier is the stator voltage of the AC exciter vE, the output voltage and current are the field voltage and current of the synchronous generator, respectively. It is very complicated to accurately model the transient response of a rectifier. Engineering practice also suggests that a transient rectifier model is unnecessary. Consequently, a so-called quasisteady-state mathematical model is usually adopted. That is, although during a transient vE, vf, and if satisfy the transient equations of the rectifier, for their instantaneous values in numerical solutions we approximate them as satisfying a steady-state equation. In this way, the transient process is approximated as a series of continuous steady-state processes. A rectifier has three operational modes according to the value of its commutating angle, g, being less than, equal to, or greater than 60 . When g is less than 60 and harmonics are ignored, the steady-state equation of the rectifier using actual values of variables is pﬃﬃﬃ 3 2 3Xg Vf ¼ VE If ; ð6:103Þ p p

372

6 Mathematical Model of Synchronous Generator and Load

where VE is the effective value of stator line voltage of the AC exciter, Xg is the commutating reactance of the rectifier (that is often taken to be the subtransient reactance or negative-sequence reactance of the exciter). Comparing with (4.37), we can see that the equation above in fact treats the three-phase uncontrollable bridge rectifier as for the case of a six-pulse rectifier in HVDC when its firing angle a is zero. In the above equation, 3Xg If/p reflects commutating voltage drop. To connect with the mathematical model of a generator, we need to convert (6.103) into per unit form. Hence, we divide both sides of the equation by the base value of field voltage of the generator VfB Vf ¼ FEX VE ;

ð6:104Þ

pﬃﬃﬃ 3 2VE ¼ ; pVf B

ð6:105Þ

where VE

pﬃﬃﬃ FEX ¼ 1 IN = 3 IN ¼ KC If =VE pﬃﬃﬃ KC ¼ Xg =ð 3pZf B Þ

ð6:106Þ

and KC is a constant. We should point out that commutating angle g is not included in (6.106). In fact, when g is less that 60 , IN is in the range of (0–0.433). It can be proved that when g is equal to or greater that 60 , (5.104) can still be used as the mathematical model of the rectifier. However, in this case, the relationship between FEX and IN has changed. When IN is between zero and 1, FEX is given by the following equation. 8 pﬃﬃﬃ > 1 IN = 3 0 IN < 0:443 > < pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 FEX ¼ ð6:107Þ 0:75 IN 0:443 IN 0:75 : > > : pﬃﬃﬃ 3ð1 IN Þ 0:75 < IN < 1 To use the above model, IN must be nonnegative and less that 1. If for some reason, the value of IN is greater than 1, FEX should be set to zero. From (6.104), (6.106), and (6.107), we can obtain the block diagram of the transfer function of the rectifier, and the relationship of FEX and IN as shown in Fig. 6.9, where the subscript of per unit has been omitted as before. In the following, we shall discuss the base value of various variables in the mathematical model of an AC exciter. Because we have established the mathematical model of an exciter directly by using that of a synchronous generator, choice of per unit system of the exciter should be kept consistent with that of the synchronous generator. From (6.105) and noting that VE is the effective value of stator line voltage of the exciter, obviously the base value of stator line voltage of the exciter, VLEB, should be

6.3 Mathematical Model of Generator Excitation Systems

Vf FEX

0. 8

If

g = 60°

FEX

π

VE

g < 60°

1. 0

373

IN =

KcIf VE

1−IN 3 0 ≤ IN < 0.433

IN

FEX =

0.75− IN2 0.433 ≤ IN ≤ 0.75 3(1−IN) 0.75 < IN < 1

b

0. 6

0. 4

g > 60°

VE

0. 2

Vf

π FEX

a

0

0.2

0.4

0.6

IN

0.8

1.0

c

if

FEX (VE, if , IN)

Fig. 6.9 Mathematical model of power rectifier (a) FEX–IN relationship curve (b) block diagram of transfer function (c) simplified representation of block diagram of transfer function

p VLEB ¼ pﬃﬃﬃ Vf B : 3 2 From the relationship between line and phase voltage as well as effective and maximum value, we can obtain the base value of the maximum value of stator phase voltage of the exciter to be VEB

pﬃﬃﬃ 2 p ¼ pﬃﬃﬃ VLEB ¼ pﬃﬃﬃ Vf B : 3 3 3

ð6:108aÞ

From (6.20), we can give the base value of the maximum value of stator phase current of the AC exciter to be IEB

pﬃﬃﬃ 2 3SB ¼ : pVf B

ð6:108bÞ

The base value of field voltage of the exciter VRB should be decided through experiment according to the principle of ‘‘unit excitation voltage/unit stator voltage.’’ Base value of the field current of exciter is obtained by (6.23) to be IRB ¼ SB =VRB :

ð6:109Þ

In (6.102), KDif represents the load effect of the exciter. From previous derivation of (6.102) we have known that if is approximately proportional to iE, that is, if ¼ kiE

374

6 Mathematical Model of Synchronous Generator and Load

Fig. 6.10 Mathematical model of AC exciter when generator excitation is provided through controllable rectifier

VR

VRmax−KCIf Vf VRmin

where k is a coefficient of proportionality. From the second term on the right-hand side of (6.100), we have 0 ðXdE XdE ÞiE ¼

0 0 0 if If B XdE XdE iE XdE XdE XdE XdE ¼ ¼ if : ZEB IEB ZEB kIEB kZEB IEB

Hence, KD in (6.102) is given by the following equation as KD ¼

0 XdE XdE If B : kVEB

ð6:110Þ

When an AC exciter supplies generator excitation by use of a controllable rectifier, the AC exciter itself often is excited in the form of self-excitation. The voltage regulator of the exciter controls the firing angle of its rectifiers to maintain an approximately constant output voltage. In this case, the mathematical model of the exciter is simplified. In addition, in this case, terminal voltage of the AC exciter is set at a high level. This results in relatively small commutating voltage drop for the controllable commutating bridge. At normal operating conditions, commutating voltage drop can be ignored. Only under automatic field-forcing or reduction, is commutating voltage drop represented as KCIf in the upper limit of output voltage. Hence when an AC exciter provides generator excitation through controllable rectifiers, the mathematical model of the AC exciter is a unit with bidirectional limiters, as shown in Fig. 6.10. We shall discuss input–output relationships of various limiters later.

6.3.1.4

Mathematical Model of Stationary Exciter

A stationary exciter takes terminal voltage or terminal current plus voltage as the power source of excitation for the generator. The former is called a self-excited potential-source system; the latter is a self-excited compound-source system. In a self-excited potential-source system, generator terminal voltage is reduced via an exciter transformer to supply generator excitation through a controllable rectifier. Firing angle of the controllable rectifier is set by a regulator. The block diagram of the transfer function of the self-excited potential-source system is shown in Fig. 6.11. We can see that it is very similar to that of the AC exciter of Fig. 6.10. This is because the only physical difference between the two is that of the excitation power

6.3 Mathematical Model of Generator Excitation Systems Fig. 6.11 Mathematical model of stationary exciter

375 VtVRmax−KCIf

VR

Vf

VtVRmin

V·t

I·t

Vref

Vc = V·t + (Rc + jXc)It •

Vc

+ 1 VM − Σ 1+sTR

Fig. 6.12 Voltage measurement and load compensation

source. This difference is demonstrated in the output limiters of the two excitation systems. In self-excited potential-source system, the excitation power source is the generator itself. Hence upper and lower limit of output voltage is related to the terminal voltage of generator Vt to be VtVRmax KCIf and VtVRmin, respectively. VRmax and VRmin, respectively, are the maximum and minimum value of no-load voltage of the rectifier when Vt ¼ 1. In a self-excited compound-source system, the power source of the controllable rectifier is supplied by an exciter voltage transformer and current transformer. Measured voltage and current can be accumulated before or after rectification in the form of parallel or series addition. This results in many different types and here we shall not introduce their mathematical models. Details can be found in [160].

6.3.2

Voltage Measurement and Load Compensation Unit

The function of an AVR is to maintain generator terminal voltage at an ideal level. The voltage measurement unit takes generator terminal voltage V_ t to convert it into a DC signal through stages of voltage reduction, rectification and filtering, etc. The voltage measurement and conversion unit can be described by a first-order simple lag, as shown by the block diagram transfer function of Fig. 6.12. The function of load compensation is to compensate the load current of generator I_t so as to maintain the required constant voltage at a controlled voltage point in steadystate operation when the load changes. RC þ jXC represents the impedance between the controlled voltage point and generator terminal. When RC and XC are positive, the controlled voltage point is inside the generator; otherwise, outside the generator. In addition, automatic distribution of reactive load among electrically close generators is related to the voltage droop of the generator; while the voltage droop of the generator is realized by adjusting parameter RC and XC. For simplicity, often RC is ignored and set to zero. When XC is greater than zero, we have positive droop; that is, the larger the load current, the higher the terminal voltage. On the other hand, when XC is less than zero we have negative droop. Terminal voltage decreases with the increase of load current. In the case without compensation,

376

6 Mathematical Model of Synchronous Generator and Load

parameters RC, and XC are zero. Voltage measurement and load compensation units may have different time constants. For simplicity, usually we just use a single time constant TR for their description. TR is called the time constant of the measurement unit and usually is less than 60 ms. For many systems, it is very close to zero. Hence in computation, we often take it to be zero. Its output voltage VM is compared against reference voltage Vref. After amplification, the error signal is used as the control signal of the excitation system of the generator. Although the reference voltage Vref is set artificially, it reflects the ideal value for the controlled voltage point of the generator and must satisfy the initial steady-state operating conditions of the whole power system.

6.3.3

Limiters

In the mathematical model of an excitation system, due to functional limitations, physical limits, or the existence of saturation, the output of certain units is subject to limitations, which we represent via limiters. There are two types of limiters, windup limiters and nonwindup limiters. Limiters often appear in integral units, the firstorder simple lag and lead-lag units. Figure 6.13a, b show block diagrams of those two types of limiters. In the following, we shall discuss the example of an integral unit and its input–output relationship. We leave those of windup and nonwindup first-order simple lag and lead-lag units for readers to establish. Equation of an integral unit is dv/dt ¼ u. The limiting function of the two types of limiters is different. For a windup limiter, if variable v is greater than lower limit B and lower than upper limit A, output variable y is v; If v is greater or equal to upper limit A, output variable y is constrained to be upper limit A; If v is less or equal to lower limit B, output variable y is constrained to be lower limit B. We should note that variable v is not constrained and only the next output variable y is. If v is beyond limitation, output variable y is constrained to be the value of the upper or lower limit. For a nonwindup limiter, output variable y is directly constrained between upper and lower limit. If y is within the limits, input–output relationship is dv/dt ¼ u. If y is equal to upper limit and tends to increase with time, i.e., dy/dt > 0, input–output relationship is dy/dt ¼ 0 and y takes the value of upper limit A. If y is equal to lower limit and tends to decrease, i.e., dy/dt < 0, the relationship is dy/dt ¼ 0 and y takes value of lower limit B. When the output variable y takes the value of the upper or lower limit, once input variable u changes sign, y enters within

u Fig. 6.13 Limiters (a) integral unit with windup limiter (b) integral unit with nonwindup limiter

a

1 s

v

A

y

A

u

B

1 s B

b

y

6.3 Mathematical Model of Generator Excitation Systems

377 VS max

VIS

ks 1

1 1 + sT6 2

V1

sT5 1 + sT5 3

V2

1 + sT1 1 + sT2

V3

1 + sT3 1 + sT4

4

5

VS

V4 VS min

6

Fig. 6.14 Block diagram of transfer function of power system stabilizer

the limits. However, for a windup limiter, only when variable v returns within the limits, so does the output variable.

6.3.4

Mathematical Model of Power System Stabilizer

Power system stabilizer (PSS) is a widely used auxiliary regulator in excitation control. Its function is to suppress power system low-frequency oscillations or increase system damping. Its basic principle is to provide the AVR with an auxiliary control signal to make the generator produce an electrical torque in phase with the deviation of rotor speed. Details about the PSS principle, parameter setting, and installation locations can be found in [153]. There are several forms of PSS. Here we give a commonly used block diagram of a PSS transfer function as shown in Fig. 6.14. In Fig. 6.14, block is the gain of PSS; is the measurement unit with a time constant T6 (usually very small and can be ignored); is a wash-out unit or lowfrequency filter to block steady-state input signal to disable PSS at steady-state operation. T5 usually is as large as about 5 s. ; and ; are two lead-lag networks. PSS should consist of at least one lead-lag network. ; is a limiter. Input signal to PSS, VIS, usually is generator speed, terminal voltage, power, system frequency, or combination of some of them. Output signal Vs is superimposed on the AVR input signal. For PSS to play an effective role, its installing location must be selected and parameters be set properly.

6.3.5

Mathematical Model of Excitation Systems

The function of an AVR is to treat and amplify the input control signal to generate a suitable excitation control signal. The AVR usually consists of power amplifier, excitation system stabilizer, and limiters. In the following, we shall introduce mathematical models of different excitation systems. In each block diagram shown below, basic input signal VC is the output from voltage measurement and load compensation unit of Fig. 6.12 and Vs is an auxiliary regulation signal of the AVR, such as the output signal from a PSS.

378

6 Mathematical Model of Synchronous Generator and Load

6.3.5.1

Excitation System with DC Exciter

With different types of AVR being used, there are three types of DC excitation systems: controllable phase compound regulator, compound excitation plus load compensation, and thyristor-controlled regulator. The former two DC excitation systems are usually used for small generation units (100 MW or below) and have been gradually passing out of use. The block diagram of an excitation system adopting controllable phase compound regulator is shown in Fig. 6.15, where V_ t and I_t are the terminal voltage and current of generator, respectively. Block ; represents phase compound excitation, blocks ; and ; are load compensation and measurement unit, ; is composite amplifying unit, ; is limiter with input signal being compound excitation current of exciter, and are units of the DC exciter. To improve performance of the excitation system, a soft negative feedback unit is often used to provide a series adjustment to field voltage of the generator. Control parameters that can be set are KV, KI, RC, XC, KE, KA, TA, KF, and TF. For the excitation system adopting compound excitation plus load compensation, the block diagram can still be used except that block ; needs to be replaced by a simple amplifier of It. Figure 6.16 shows the block diagram of a DC excitation system using thyristorcontrolled regulator. TB and TC are time constants of the excitation regulator itself. They are usually very small and considered to be zero. Time constant and gain of composite amplifying unit is TA and KA. Due to the saturation of the amplifier and limitation of power output, the block for the amplifier has a nonwindup limiter. VF

1

+

KVVt + jK I It Vt It

Vc = Vt + (Rc + jX c )It

1 1 + sTR

2

3

VM − + +

−

Σ

KA 1 + sTR

Vs

VRmax VR min

+

Vref

Σ

5

+

Σ

−

Vf

K E + SE

8

4

6 1 sTE

7

sK F 1 + sTF

Fig. 6.15 Block diagram of transfer function of AVR of DC exciter using controllable phase compound excitation Vs VM

−

+ Vref

+

∑

− VF

1 + sTC 1 + sTB

KA 1 + sTA VRmin

VRmax VR +

∑

−

1 sTE KE + SE

sKF 1 + sTF

Fig. 6.16 Block diagram of transfer function of a DC excitation system

Vf

6.3 Mathematical Model of Generator Excitation Systems

379

is the output of the soft negative feedback unit of excitation voltage to improve the dynamic performance of whole excitation system. VR is the excitation voltage of the DC exciter. Parameters to be set for the normal operation of excitation control are RC, XC, KE, KA, TA, KF, and TF.

6.3.5.2

Excitation System with AC Exciter

Excitation system with an AC exciter is widely used for 100 MW or above generation units. Most excitation systems with AC exciter adopt uncontrollable power rectifier. They can be classified into two groups: stationary rectifier excitation systems and rotating rectifier excitation systems. Here we introduce one type of AC excitation system as shown by the block diagram of Fig. 6.17. Introduction to other types of AC excitation systems can be found in [160]. In Fig. 6.17, parameters TB, TC, KA, TA, KF, and TF describe three blocks belonging to the excitation regulator similar to that in Fig. 6.16. The input signal to the series regulation unit is the no-load voltage eqE of the AC exciter (6.102). Another kind of arrangement is to use the field voltage of the generator Vf as the feedback input signal. Field current If is also an input signal of the excitation regulator and constant KD represents the equivalent load effect of the AC exciter. In Fig. 6.17, the exciter is separately excited. When self-excitation is used, we need to replace the block 1 þ SE by kE þ SE, where kE and SE are the self-excitation coefficient and saturation factor of the AC exciter, respectively. Because the input to rectifier requires VE not to be negative, in the block of the exciter the integral unit represented by TE has a single-directional windup limiter to prevent VE from becoming negative. Parameters to be set for the normal operation of excitation control are RC, XC, KE, KA, TA, KF, and TF. The block diagram of an AC excitation system adopting a controllable rectifier to supply generator excitation is shown in Fig. 6.18. The rectifier is controlled by an independent voltage regulator and hence its output is kept approximately constant. Therefore, the mathematical model of an AC exciter and controllable rectifier is shown in Fig. 6.10. In Fig. 6.18, this has been combined with an equivalent Vs + VM − ∑ + Vref

−

VF

1+sTC 1+sTB

VRmax KA VR − ∑ 1+sTA eqE−

1 VE sTE

+

∑

Vf

0

VRmin sKF 1+sTF

π

+

1+sE KD

FEX (VE, if, IN) If

Fig. 6.17 Block diagram of transfer function of excitation system with AC exciter adopting uncontrolled power rectifier

380

6 Mathematical Model of Synchronous Generator and Load Vs VM −

Vref

VRmax − kcIf

VIm ax

+

VI

Σ

1 + sTc

VR

KA 1 + sTA

1 + sTB

+V Im in

Vf

VRm in

Fig. 6.18 Block diagram of transfer function of excitation system with AC exciter adopting controllable power rectifier Vs VM − +

Vref

+

Σ

−

VImin

VtVRmax − KcIf

VAmax

VImax KA 1+sTA

VI 1+sTc 1+sTc1 1+sTB 1+TB1

+

If

+

Σ

− ILR

Vf −

VAmin

VF

Σ VtVRmin

KLR

0

sKF 1 + sTF

Fig. 6.19 Block diagram of transfer function of self-excited potential-source system

composite amplifying unit, where time constant TA and gain KA depict the dynamic performance of the controllable rectifier and its regulator. To improve system dynamic performance, this type of excitation system usually adopts a series regulator instead of a shunt regulator. The time constants of the series regulator are TB and TC. We should point out that the load of the controllable rectifier is limited to ensure IN between 0 and 0.433 (6.107). Load effect of the excitation system is reflected in the upper limit of the bidirectional limiter. Parameters to be set for the normal operation of excitation control are RC, XC, KA, TA, TC, and TB. Here, because an independent AC exciter is used, the values of upper and lower limits of the bidirectional nonwindup limiter are not connected to the terminal voltage of the generator.

6.3.5.3

Stationary Excitation System

Figure 6.19 shows the block diagram of a self-excited potential-source system and controllable rectifier described by a bi-directional limiter. As has been introduced before, the power into stationary excitation is from the generator terminal. Hence the value of upper and lower limit is related to the terminal voltage of the generator. This type pf excitation system can provide very high automatic field forcing

6.4 Mathematical Model of Prime Mover and Governing System

381

voltages. To avoid overloading of generator field and rectifier, the field current of generator If is constrained by KLR and ILR in the diagram. Proportional unit KLR has a windup lower limit. To avoid this unit we can simply set KLR to zero. KA and TA are the system composite equivalent gain and time constant, respectively. Both series regulation and shunt regulation are displayed in the diagram. Usually only one of them is used. Hence when series regulation is used, we can set KF to zero. Or when shunt regulation is used, we just set time constants TB and TC to zero. Time constants TB1 and TC1 are for the increase of system dynamic gain. Usually we have TC1 > TB1. To simplify the model, this unit can be ignored by setting both of these time constant to zero. Here we should point out that the block diagram of Fig. 6.19 can represent the excitation system adopting full-wave controllable rectifying bridge. When a half-wave controllable rectifying bridge is used, we can simply set the lower limit of the bidirectional limiter at the system output to zero. Parameters to be set for the normal operation of excitation control are RC, XC, KA, TA, KF, TF, TC, TB, KLR, and ILR. In [160], we can find more about the mathematical model and block diagrams of other types of stationary excitation system.

6.4

Mathematical Model of Prime Mover and Governing System

Variable Pm in the rotor movement equation of the generator (6.78) is the mechanical power output from the prime mover. Pm is related to the operating condition of the prime mover and controlled by a governing system. Excluding wind, sun, and wave power generation, there are two types of prime mover used for large-scale power generation, hydraulic turbines, and steam turbines. The hydraulic turbine (or steam turbine) converts hydraulic energy (or steam thermal energy) into rotating kinetic energy of the prime mover which is then converted into electric power by the generator. Obviously, the amount of power being converted is associated with the opening position of the wicket gate of a hydraulic turbine and steam valve of a steam turbine. Because the generator rotor is driven by the prime mover and rotates on the same shaft with the prime mover, if we assume that the generator output power is fixed, when the opening position increases, the generator will accelerate; and conversely it decelerates. Therefore, regulation of the gate or valve position will change the output power from the prime mover to control generator speed. Hence it is easy to see that the main control signal to the opening position should be generator speed. From the rotor movement equation (6.78) we can see that when a power system is subject to a disturbance at steady-state operation, electric power output from the generator changes. This change destroys the balance between electric power output from the generator and mechanical power input to the generator from the prime mover, leading to variation of the generator speed. Change of generator speed results in a response of the governing system to adjust the opening position of the wicket gate (of a hydraulic turbine) or steam valve (of a

382

6 Mathematical Model of Synchronous Generator and Load

steam turbine). The disturbance causes the system to engage in a complex transient process of mechanical, magnetic, and electrical interactions. Therefore, when the function of the governing system is considered, resulting in the variable Pm, we need to establish a mathematical model of the prime mover and the governing system in order to quantitatively analyze electromechanical transients in power systems.

6.4.1

Mathematical Model of Hydroturbine and Governing System

6.4.1.1

Mathematical Model of Hydraulic Turbine

Dynamics of hydraulic turbines are closely related to those of water flow through a penstock, whereas the characteristics of water flow through a penstock are affected by many factors, such as water inertia, water compressibility, and pipe wall elasticity in the penstock. For example, due to water inertia inside a penstock, change of water flow inside a hydraulic turbine lags the opening position change of the wicket gate. When the opening position of the wicket gate increases suddenly, water volume at the wicket gate increases. However, due to the water inertia, speed of water flow at other points inside the pipe cannot increase immediately. This results in input water pressure of the hydraulic turbine decreasing instead of increasing for a short of period of time after the change, leading to a decrease of input power of the hydraulic turbine instead of an increase. On the other hand, when the opening position of the wicket gate decreases suddenly, input water pressure and input power will increase temporarily and then decrease. This phenomenon is usually called the water hammer effect. Furthermore, for the movement of a compressible fluid inside an elastic pipe, the change of water flow volume and pressure at each point inside the pipe is a wave movement, quite similar to the wave process of transmission lines with evenly distributed parameters. A detailed derivation of the mathematical model of input water pressure on the turbine with wave effects considered requires extensive application of fluid mechanics. This is only necessary for the case with a long pressure pipe. In the following, we shall establish a mathematical model of a hydraulic turbine, useful for the analysis of power system stability, with the wave effect of water flow ignored. That is, to assume that the pressure pipe is inelastic, and water is not compressible. Additionally we shall only consider an ideal hydraulic turbine, i.e., (1) neglecting the mechanical power loss caused by the resistance against water flow from the penstock wall; (2) power output of the hydraulic turbine being proportional to the product of net water head and water flow volume; and (3) speed of water flow being proportional to the product of the opening position of the wicket gate and square root of the stationary water head. Hence we can obtain the hydraulic equations as follows: pﬃﬃﬃﬃ U ¼ KU m H ; ð6:111Þ

6.4 Mathematical Model of Prime Mover and Governing System

383

Pm ¼ KP HU;

ð6:112Þ

dU g ¼ ðH H0 Þ; dt L

ð6:113Þ

where U is the water velocity; KU the proportional constant; H the net water head of hydraulic turbine; m the opening position of wicket gate; Pm the mechanical power output of hydraulic turbine; KP the proportional constant; g the gravity acceleration constant; L the length of penstock; and H0 is the steady-state value of H. Taking the initial value of various variables as their base value, the above hydraulic equations can be converted into the following per unit form (subscript * is omitted as before) pﬃﬃﬃﬃ U ¼ m H;

ð6:114Þ

Pm ¼ HU;

ð6:115Þ

dU 1 ¼ ðH 1Þ; dt Tw s

ð6:116Þ

Tw ¼ LU0 =ðgH0 Þ

ð6:117Þ

where

Tw is the time constant of equivalent water hammer effect and physically it is the time required for water head H0 to accelerate water flow in penstock from a stationary state to the flowing speed U0. We ought to point out that this time constant is affected by U0, i.e., related to the load condition of the hydraulic turbine. The heavier the load is, the higher the time constant. Usually under full load condition, Tw is set by the manufacturer between 0.5 and 4 s. Assuming that at initial steady state, the operating point of the hydraulic turbine shifts slightly due to small disturbances from the load, the above hydraulic equations can be linearized at the initial steady-state operating point and after Laplace transformation they become DU ¼

pﬃﬃﬃﬃﬃﬃ 1 m0 ﬃ DH; H0 Dm þ pﬃﬃﬃﬃﬃ 2 H0

ð6:118Þ

DP ¼ H0 DU þ U0 DH;

ð6:119Þ

Tw sDU ¼ DH:

ð6:120Þ

384

6 Mathematical Model of Synchronous Generator and Load Δm

Fig. 6.20 Transfer function of classical model of hydraulic turbine

m

Fig. 6.21 Relationships between actual and ideal opening position

AtgFL

1−TwS 1 + 0.5TwS

ΔPm

Full load

AtgNL No-load loss 0

gNL

gFL 1.0

g

Eliminating variables DH and DU in the above three equations we can obtain (per unit value of H0 is 1) DPm ¼

1 Tw s Dm: 1 þ 0:5Tw s

ð6:121Þ

The model above is called the classical model of a hydraulic turbine. Its transfer function block diagram is shown in Fig. 6.20. In the analysis of power system stability, the above classical model of a hydraulic turbine is used. From the assumptions used to derive the model we know that the classical model is applicable to cases with relatively small variations of load. When load changes over a wide range, the model may cause a large computational error [161]. In the following, we shall establish a nonlinear model of a hydraulic turbine. Basic assumptions will be the same as those for deriving the classical model except that mechanical power loss and dead zone are taken into account. Opening position of wicket gate m in (6.111) is that of an ideal wicket gate with the dead zone of the hydraulic turbine caused by factors such as friction being ignored, i.e., it is assumed that when m changes from 0 to 1, operation of the hydraulic turbine goes from no load to full load. With mechanical power loss being considered, position change of the wicket gate from closing to opening, initially has to overcome stationary friction forces in the hydraulic turbine without causing the turbine to start rotation immediately. Hence we need to replace the ideal opening position m by the actual one g. From Fig. 6.21, we can see their relationship to be m ¼ At g;

ð6:122Þ

where At ¼

1 : gFL gNL

ð6:123Þ

6.4 Mathematical Model of Prime Mover and Governing System

385

When the actual opening position is gNL, the hydraulic turbine is still at no-load. When it is gFL, the hydraulic turbine operates at full load. With power loss being considered, hydraulic equations (6.112) become Pm ¼ KP HU PL ;

ð6:124Þ

PL ¼ KP UNL H;

ð6:125Þ

where PL is no-load loss of the hydraulic turbine; UNL critical water speed when the hydraulic turbine changes from stationary to rotating. Obviously that is when the actual opening position is gNL. Taking the rated parameters of the hydraulic turbine as the corresponding base value, we can convert (6.111), (6.113), (6.124), and (6.125) into the following per unit form pﬃﬃﬃﬃ U ¼ m H;

ð6:126Þ

Pm ¼ ðU UNL ÞH;

ð6:127Þ

dU 1 ¼ ðH H0 Þ; dt TW

ð6:128Þ

where TW ¼

LUB : gHB

ð6:129Þ

TW is the time constant of the equivalent water hammer effect at rated load. From (6.117) we can see that the relationship of the time constant between any load condition and at rated load is Tw ¼

U0 HB TW : UB H0

ð6:130Þ

In (6.127), base value of power is the rated power of the hydraulic turbine. To connect it to the mathematical model of the generator, we can convert the base value of power to the rated power of generator SB Pm ¼ Pr ðU UNL ÞH;

ð6:131Þ

Pr ¼ PB =SB :

ð6:132Þ

2 U : m

ð6:133Þ

Rewriting (6.126) H¼

386

6 Mathematical Model of Synchronous Generator and Load

From (6.122) and (6.133) we can eliminate H in (6.128) and (6.131) to obtain dU 1 ¼ dt TW

! 2 U H0 ; At g

Pm ¼ Pr ðU UNL Þ

U At g

ð6:134Þ

2 :

ð6:135Þ

The two equations above are the nonlinear model of a hydraulic turbine. From the physical meaning of UNL we know that when the actual opening position of wicket gate g is gNL, acceleration of water flow is zero. From (6.134) we have dU 1 ¼ dt U¼UNL TW UNL

UNL At gNL pﬃﬃﬃﬃﬃﬃ ¼ At gNL H0

!

2 H0

¼ 0;

:

ð6:136Þ

Normally H0 is 1 and hence UNL is a constant. From (6.122), (6.133), (6.128), and (6.131) we can show the nonlinear model of the hydraulic turbine in Fig. 6.22. 6.4.1.2

Mathematical Model of Governing System of Hydraulic Turbine

Modern generation units usually use an electrical-hydraulic governing system. However, the principle of mechanical hydraulic governing system is easier to illustrate. Hence we shall take it as representative to establish the mathematical model of a governing system of a hydraulic turbine. Figure 6.23 shows the configuration of a governing system using a centrifugal pendulum (fly-ball governor). In the following we shall present the equation of motion of each component of the governing system, where variables are in per unit and their positive direction is indicated in Fig. 6.23. Compressibility of hydraulic oil will be neglected: 1. Equation of the centrifugal pendulum. The function of the centrifugal pendulum (fly balls) is to measure generator speed. Relative ring position of the fly balls is denoted by . When generator speed increases, the fly balls move away from each other due to the increase of centrifugal force and consequently decreases. On the other hand, when generator speed decreases, the fly balls come closer because of the decrease of centrifugal force and hence increases. Ignoring the γ

At

μ

π

H +

Σ H0

1 sTw

U +

Σ UNL

Fig. 6.22 Block diagram of transfer function of hydraulic turbine

π

Pr

Pm

Tm

ω

6.4 Mathematical Model of Prime Mover and Governing System

A

I

C

B

G

η

ζ

ζ2 D

F

E

ζ1

387

VI H V

μ

σ a

III

b IV

II

Fig. 6.23 Illustration of governing system of a centrifugal pendulum

mass of the fly balls and damping of motion, is approximately proportional to the deviation of generator speed with a proportional coefficient kd, that is ¼ kd ðo0 oÞ:

ð6:137Þ

2. Equation of pilot valve. If the servomotor does not function (point D in Fig. 6.23 is fixed) and the inertia of the pilot valve is ignored, the relationship between position of pilot valve, s, and that of point B, z, is s ¼ B:

ð6:138Þ

3. Equation of relay valve. Position of relay valve, m, is the integral of the position of the pilot valve, s, with respect to time, i.e., position speed of relay valve is proportional to position of pilot valve, s. The proportional constant is called the time constant of the relay. Hence Ts rm ¼ s:

ð6:139Þ

4. Feedback equation. From Fig. 6.23, we can see that when increases, s increases accordingly and m also increases. Increase of and m results in an increasing z, leading s to decrease and m to decrease accordingly. Hence z is a position variable exhibiting feedback from m. There are two parts in z, z1, and z2. z1 is a soft feedback due to the existence of a spring and dashpot; while z2 is proportional to m and hence it is a hard feedback. We have B ¼ B1 þ B2 ¼

kb Ti s m þ ka m; 1 þ Ti s

ð6:140Þ

where ka ¼ a/d, kb ¼ b/d, d ¼ 1/kd. kb and Ti are the gain and time constant of the soft feedback, respectively; ks the gain of hard feedback; d the sensitivity of

388

6 Mathematical Model of Synchronous Generator and Load

measuring component; b coefficient of soft feedback; and a is the droop coefficient. Due to the inertia, water flow cannot follow the change of opening position of the wicket gate quickly. Hence when speed deviation of the generator changes fast, the governor needs a strong negative feedback from the opening position m to slow down the change of opening position of wicket gate such that water flow and output power from the hydroturbine can follow that of m. For slow variations of generator speed in steady-state operation, the governor needs to respond promptly. Hence the gain of the negative feedback should take a small value. From (6.140) we can see that dynamic gain of the whole negative feedback unit is high. When t ¼ 0, it is the summation of kb and ka. The time constant of the soft negative feedback is large, usually between 0.5 and 5 s. At steady state, the steady-state gain of the soft negative feedback is zero and the gain of the whole negative feedback is only that of the hard feedback ka. This provides the generator with a certain droop coefficient at steady-state operation, such that a generator speed decrease will increase generator output. The droop characteristic ensures stable load sharing among multiple generation units in parallel steady-state operation, to realize the function of primary frequency control. ka and kb are often about 0.04 and 0.4, respectively. Opening position of both pilot valve and wicket gate have certain limitations imposed. In addition, due to the existence of mechanical friction and gap, there exists a certain dead zone of the governing function. Hence in the mathematical model, there are associated limiters and a nonlinear unit representing the dead zone. From (6.137) to (6.140) we can obtain the transfer function block diagram of the governing system of a hydraulic turbine as shown in Fig. 6.24. The function of an electrical-hydraulic governing system of a hydraulic turbine is quite similar to that of the mechanical hydraulic system introduced above, but more simple and flexible as far as the regulation of basic parameters is concerned. A mathematical model of an electrical-hydraulic governing system of the hydraulic turbine adopting PID control can be found in [162, 163].

ωref ω −

+

Σ

kδ

η +

εkδ

Σ

−

σ max σ

σ

1 sTs

σ min

ζ

Σ +

+

ζ2

ζ1

+

μ ref +

Σ

μ max μ

μ

μ min

kσ skβ Ti 1 +sTi

Fig. 6.24 Block diagram of governing system of centrifugal pendulum (fly ball) of a hydraulic turbine

6.4 Mathematical Model of Prime Mover and Governing System

6.4.2

Mathematical Model of Steam Turbine and Governing System

6.4.2.1

Mathematical Model of Steam Turbine

389

Dynamics of steam turbines are mainly related to the volume effect of steam. In the following, we shall first derive the time constant for the general steam volume effect. As shown in Fig. 6.25, volume of the vessel is V (m3) and input and output steam mass flow rates are Qin and Qout (kg s1), respectively. We have dW dr ¼V ¼ Qin Qout ; dt dt

ð6:141Þ

where W is the weight of steam in the vessel (kg) and r is the density of steam (kg m3). Assuming that the output steam flow is proportional to steam pressure in the vessel, we have Qout ¼

QN P; PN

ð6:142Þ

where P is steam pressure in the vessel (kPa), PN the rated steam pressure in the vessel (kPa), and QN is the rated output of steam out of the vessel (kg s1). With the steam temperature in the vessel being constant, we have dr dP @r ¼ ; dt dt @P

ð6:143Þ

where the rate of change of steam density with pressure, at a given temperature, ∂r/∂P, can be obtained from steam parameter tables and is a constant. From (6.141) to (6.143) and after Laplace transformation, we have Qout ¼

1 Qin ; 1 þ sTV

where TV ¼

ð6:144Þ

PN @r V : QN @P

ð6:145Þ

Qout

Qin V

Fig. 6.25 Steam vessel

390

6 Mathematical Model of Synchronous Generator and Load From boiler

Q0

Steam valve

Reheater

crossover

Q2

Q1 HP

shaft

Q3 IP

shaft

LP

LP

shaft

To condenser

Fig. 6.26 Illustration of a multistage steam turbine

TV is called the time constant of steam volume effect. From (6.145) we can see that the bigger the volume of the vessel, the higher is the time constant of volume effect. From (6.144) we can see that when the input steam flow increases (or decreases) suddenly, the output steam flow will not increases (or decreases) immediately because the pressure inside the vessel cannot increase (or decrease) instantly. Change of output steam flow lags that of input steam flow. This is the steam volume effect phenomena. There are many types of configuration of steam turbines. Modern steam turbine units consist of multiple-stage steam turbines to drive a single generator. According to the difference in rated operating steam pressure, multiple-stage turbines can be classified as high pressure (HP), intermediate pressure (IP), and low pressure (LP) turbines. Medium and small steam turbine units may have only a one-stage turbine. To increase thermal efficiency, modern steam turbine units usually have an intermediate reheater (RH). Figure 6.26 shows the configuration of a steam turbine with reheater. From Fig. 6.26 we can see that the high-pressure high-temperature steam from the boiler enters the HP stage through a main valve and steam chest. We should note the existence of a certain volume of steam in the pipe and chest from the main valve to the nozzle of the HP stage. Exhaust steam from the HP stage is sent into the reheater section to raise temperature before entering the IP stage. Similarly we ought to note that there exists a certain volume of steam between the output point of the HP stage and input point of IP stage. Exhaust steam from IP stage enters the LP stage through crossover that also has a certain volume. Volume effects of the three volumes mentioned above can be described by time constant TCH, TRH, and TCO, respectively. Usually TCH is between 0.2 and 0.3 s, time constant of reheater TRH is large, between 5 and 10 s, and TCO is about 0.5 s. Output mechanical torque of the steam turbine is proportional to the steam flow at the nozzle. In addition, we assume that input steam flow to the HP stage is approximately proportional to the opening position of the main steam valve m. We denote the proportionality coefficient of mechanical power of HP, IP, and LP stages to be FHP, FIP, FLP. Usually FHP, FIP, FLP is 0.3, 0.3, 0.4 and their summation is one.

6.4 Mathematical Model of Prime Mover and Governing System + +

TmH

μ Q0

1 1 + sTcH

+

Σ TmI

FHP

FIP

Q1

1 + sTRH Q 2

Tm

Σ +

TmL

FLP

1

391

π

Pm

ω

1

1 + sTco

Q3

Fig. 6.27 Block diagram of transfer function of a multi-stage steam turbine

From the analysis above and taking proper base values for per unit expressions, we can obtain the mathematical model of the steam turbine in per unit to be 9 1 > Q1 ¼ Q0 > > 1 þ TCH s > > > = 1 ð6:146Þ Q2 ¼ Q1 ; 1 þ TRH s > > > > > 1 ; Q3 ¼ Q2 > 1 þ TCO s 9 > > > > > > > > > = TmI ¼ FIP Q2 ; > TmL ¼ FLP Q3 > > > > Tm ¼ TmH þ TmI þ TmL > > > > ; F þF þF ¼1

m ¼ Q0 TmH ¼ FHP Q1

HP

IP

ð6:147Þ

LP

where TmH, TmI, TmL is the output mechanical torque of HP, IP, and LP turbines, respectively, flows Q0–Q3 are shown in Fig. 6.26. Block diagram of the transfer function of the mathematical model above is shown in Fig. 6.27. Other types of mathematical models of steam turbines can be found in [164, 165]. 6.4.2.2

Mathematical Model of Governing System of Steam Turbine

Basic functions of the governing system of a steam turbine include normal primary frequency control, secondary frequency control, over-speed control, over-speed generation shedding and generation starting and stopping control in normal operation, as well as auxiliary steam pressure control. Normal primary frequency control and secondary frequency control of the steam turbine is quite similar to those of a hydroturbine. Primary frequency control provides a droop around 4–5% to ensure stable load sharing among parallel generation units. Secondary frequency control is achieved through adjusting the load reference. In modern steam turbine units, usually there are more control valves in addition to the main valve shown in Fig. 6.26. For example, in a steam turbine unit equipped with a reheater there is a stop

392

6 Mathematical Model of Synchronous Generator and Load

valve behind the RH stage. When over-speeding generation requires an emergency reduction of output power from the steam turbine, it would not be enough to just turn off the main valve, because the steam volume of the RH stage is very large. Under this circumstance, usually the main valve and stop valve must be turned off simultaneously. Primary frequency control and secondary frequency control function only by adjusting the main valve shown in Fig. 6.26. Usually in the study of power system stability, only the control of the main valve is considered and that of other valves is ignored. However, if emergency stop and generation shedding are used as the method for stability control, the control function of other valves needs to be taken into account. In this book, we shall only introduce the control model of the main valve. The control model of other valves can be found in [164, 165]. Governing systems of steam turbines can be classified into three types, mechanical hydraulic, electrohydraulic, and power-frequency electrohydraulic. The principle of mechanical hydraulic governing system is the same as that of a centrifugal pendulum governor introduced previously, except that the governing system of a steam turbine does not need the soft feedback unit and only uses hard feedback. The coefficient of hard feedback is 1. Hence a mechanical hydraulic governing system of a steam turbine can be shown by the transfer function block diagram of Fig. 6.28, where the simple lag with time constant T1 represents the pilot valve in the governor. The value of T1 usually is small and hence this unit can be ignored. In an electrohydraulic governor, the low power output unit in the mechanical hydraulic governor, i.e., the part from speed measurement to servomotor, is realized by an electronic circuit. Compared to the mechanical hydraulic governor, an electrohydraulic governor is of better applicability and flexibility, with quicker responds speed. In order to obtain better performance and linear response, the feedback channel from steam flow (or steam pressure at the first stage in the HP turbine) and valve position of the servomotor is introduced in the electrical-hydraulic governor. The transfer function block diagram is shown in Fig. 6.29. The transfer function block diagram of the power-frequency electrohydraulic governor is shown in Fig. 6.30. By comparing frequency and power signals with the given reference, an error signal is obtained and then amplified. A PID controller conditions the amplified signal. Its output electrical signal is converted into a hydraulic signal by an electrical-hydraulic converter to drive a relay and servomotor to regulate the main valve of the steam turbine. In Fig. 6.30, kP, kI and kD are the gains of proportional, integral, and differential units, respectively; TEL the time constant of the electrical-hydraulic converter; and Ts is the time constant of the relay.

ω−

ω0 +

Σ

kδ

η +

Σ ζ

−

ε kδ

μ ref

ρ max

1 1 + sT1

ρ ρ min

1 sTs

+

+

μ max μ

Σ μ min

Fig. 6.28 Block diagram of mechanical hydraulic governing system of steam turbine

6.5 Mathematical Model of Load

393

ω ref ω−

μmax

μ open •

+

kδ

Σ

kp

+

+

Σ

−

Σ

−

μ close •

1 sTs

kp−1

μ

μ min

qHP

Fig. 6.29 Block diagram of transfer function of electrohydraulic governing system

ω−

ω0 +

Σ P0

P−

Σ

kI s kδ +

Σ

+

Σ

+

correcting signal

kp skD 1 + sTD

+

+

Σ

+

μ max

1 + 1 + sTEL

Σ

−

1 sTs

μ μ min

Fig. 6.30 Transfer function of power-frequency electro hydraulic governor

In the recent 20 years, digital governing systems for steam turbines have been developed, in which the operating unit of the main valve of the steam turbine is connected to a digital controller via a digital-analogue hybrid unit. The control function is realized by software. A digital governor provides more flexible and universal functions than an electrical-hydraulic governor. Response speed is enhanced greatly with the time constant being about 0.03 s. More details about digital governors can be found in [167]. We ought to point out that (6.147) is equivalent to ignoring the transients of the thermodynamic system. If thermodynamics is considered, obviously Q0 will be determined by a mathematical model describing the thermodynamic system. As far as the time scale for the computation of power system stability (usually for 5 s following a disturbance) is concerned, the time constant of the thermodynamic system is very large. Hence the thermodynamic system can be considered as operating in steady state. However, for long-term power system stability analysis, involving system dynamics for several minutes after a disturbance, the dynamics of the thermodynamic systems, such as the boiler, will play an important role. Mathematical modeling of the thermodynamic system is still a research subject at the moment.

6.5

Mathematical Model of Load

Load is an important part in a power system. To study power system behavior in various operational states, we need to establish a mathematical model of system load. It is not difficult to establish a mathematical model of certain power-consuming

394

6 Mathematical Model of Synchronous Generator and Load

equipment in the power system. However, it is neither necessary nor possible mathematically to describe each of hundreds and thousands of loads in detail. Hence in this section, power system load refers to all electrical equipment connected at a common node in the power system. It includes not only various end-users of power-consuming equipment but also under-load tap changing transformer, distribution network, various kinds of reactive power compensators, voltage regulation units, and even some small generators, etc. The relationship between active and reactive power absorbed, by all those mentioned above, at the node and the node voltage and system frequency constitutes the mathematical model of nodal load. Obviously, for different types of node, such as residential, commercial, industrial, and rural, the composition of load is quite different. Besides, for the same node, during different time periods, such as different seasons in a year, different days in a week, and different hours in a day, the composition of nodal load can vary. Due to the variety, randomness, and time variance of load, it is an extremely difficult problem to establish an accurate load model. A large number of studies have demonstrated that the conclusions from power system analysis are greatly affected by whether the mathematical model of the load has been established properly or not. From the point of view of system operation analysis and control, improper mathematical modeling of the load will result in analytical conclusions being poorly matched with practical results, either being too conservative – leading to inefficient utilization of the system, or too optimistic – causing hidden risks to system operation. An even more difficult problem is that at the moment there is no way to know if a certain load model is always conservative or optimistic under any disturbance. The importance and complexity of establishing mathematical models of the load has become a special research field, resulting in a large number of studies over many years [168–170]. There are many methods for the establishment of mathematical models of load, but these can be classified into two groups: ‘‘method of theoretical aggregation’’ [170] and ‘‘method of identification aggregation’’ [171]. In theoretical aggregation, nodal load is considered to be the combination of various individual users. Firstly those users are electrically categorized and average characteristics of each category are determined. Then a statistical percentage of each category of users is worked out and finally the total load model is aggregated. The method of identification aggregation uses collected field data. After a proper structure for the load model is chosen, the model parameters are identified by using field data. The two methods have their own merits and disadvantages. The former is simple and easy to use, but its accuracy is not satisfactory. The latter can produce more accurate mathematical models by treating and analyzing field data using modern identification theory. However, it is still difficult to obtain an accurate dynamic model of load because voltage and frequency of the real power system cannot vary over a large range. Therefore, power system load modeling remains a research topic to be pursued and no fully matured method is available. There are quite a few methods to classify power system load models. With regard to whether the load model can describe load dynamics, a model is categorized as either static or dynamic. Obviously, a static load model is a set of algebraic

6.5 Mathematical Model of Load

395

equations, while a dynamic model includes differential equations. Other classifications include: linear load model or nonlinear load model and voltage-related model or frequency-related model. Conventionally, we consider load models related to both voltage and frequency to be frequency-related models. According to the way that the model is established, we have derived-model or input–output model. A derived model has clear physical meaning and can easily be understood. It is usually adopted when few types of load are considered. Nonderived models only concern the mathematical relationship between load input and output. Due to the limitation of space, in this section, we shall only introduce several commonly used types of load. The simplest load model is to use a constant impedance to represent the load. That is, to assume that the equivalent impedance of the load does not change during system transients and its value is determined by the node voltage and power absorbed by the load at steady state before the occurrence of a disturbance. This load model is rather rough. However, due to its simplicity, it is still widely used when requirements on computational accuracy are not high.

6.5.1

Static Load Model

The static characteristic of load is the relationship between node voltage or frequency and power absorbed by the load, when voltage or frequency varies slowly. The usual forms of static load model are as follows. 1. Static load voltage or frequency characteristic described by a polynomial. Without considering variations of frequency, the relationship between node voltage and power absorbed by load is taken to be 9 " #

> VL 2 VL > 2 > PL ¼ PL0 aP þ bP þ cP ¼ PL0 aP VL þ bP VL þ cP > > = VL0 VL0 " #

> > VL 2 VL > QL ¼ QL0 aQ þ bQ þ cQ ¼ QL0 aQ VL2 þ bQ VL þ cQ > > ; VL0 VL0 ð6:148Þ where PL0, QL0, and VL0 are the active, reactive power absorbed by the load and node voltage before the occurrence of a disturbance. Parameters, aP, bP, cP, aQ, bQ, and cQ have different values for different nodes and satisfy ) aP þ bP þ c P ¼ 1 : aQ þ bQ þ cQ ¼ 1

ð6:149Þ

From (6.148) we can see that this model is in fact equivalent to representing the load in three parts. Coefficient a, b, and c represent the percentage of constant

396

6 Mathematical Model of Synchronous Generator and Load

impedance (Z), constant current (I), and constant power (P) in the total nodal load, respectively. Hence this type of load model is also called a ZIP model. Because system frequency does not vary much during transients, static frequency characteristics of load can be represented linearly. Without considering variation of node voltage, the relationship between node power and system frequency is 9 f f0 > > > PL ¼ PL0 1 þ kP = f0 ; ð6:150Þ f f0 > > > QL ¼ QL0 1 þ kQ ; f0 where PL0, QL0, and f0 are the active, reactive power absorbed by load and system frequency before the occurrence of a disturbance, respectively. Parameters kP and kQ have different values at different nodes and their physical meaning is the differential of node power to variation of system frequency at steady state, that is 9 f0 dPL dPL > > kP ¼ ¼ > PL0 df f ¼f0 df f ¼f0 = : ð6:151Þ > f0 dQL dQL > > kQ ¼ ¼ ; QL0 df f ¼f0 df f ¼f0 With variation of voltage and frequency being taken into account, the mathematical model of load is the product of the two per unit model expressions above, that is )

PL ¼ aP VL2 þ bP VL þ cP ð1 þ kP Df Þ ð6:152Þ

: QL ¼ aQ VL2 þ bQ VL þ cQ 1 þ kQ Df We would like to point out here that in statistical computation, various base values must be converted to maintain consistency with system base values. 2. Static load voltage characteristics expressed by exponentials. Without considering variation of frequency, static load voltage characteristics can be described by the following exponential form a 9 VL > > PL ¼ PL0 > = VL0 ð6:153Þ b : > VL > > ; QL ¼ QL0 VL0 For composite load, power a usually is between 0.5 and 1.8, b changes significantly between different nodes, typically between 1.5 and 6. With the effect of frequency change being taken into account, we have

6.5 Mathematical Model of Load

397

9 VL a f f0 > > 1 þ kP > = VL0 f0 b : QL VL f f0 > > > ¼ 1 þ kQ ; QL0 VL0 f0 PL ¼ PL0

ð6:154Þ

Although static load models are widely used in routine computation of power system stability due to their simplicity, computational errors could be very high when the magnitude of node voltage involved in the computation varies over a wide range. For example, discharge lighting load takes over 20% of commercial load. When the per unit voltage value reaches as low as 0.7 p.u., the light goes off and the load consumes zero power. When the voltage recovers, the light goes on after a short delay. Some induction motors are equipped with low voltage protection. When the voltage decreases below a certain level, the motor will be disconnected from the network. Also, due to transformer saturation at higher voltages, reactive power absorbed is very sensitive to changes in the magnitude of nodal voltage. All these factors make static load models inapplicable when nodal voltage varies over a large range. A common method to cope with this problem is to use different model parameters in different voltage ranges or to use a simple constant impedance load when the node voltage is below 0.3–0.7 p.u. Other algebraic forms of static load model can be found in [170].

6.5.2

Dynamic Load Model

When node voltage changes quickly over a large range, adoption of purely static load models will bring about excessive computational errors; especially in the study of voltage stability (or load stability) where high accuracy is required in load modeling. Many studies using different types of load model have shown that at sensitive nodes, dynamic load models should be used [172–175]. In computational practice, those nodal loads are considered to consist of two parts: static and dynamic. Although there are many different types of industrial load, induction motors takes the largest share. Hence, load dynamics are mainly determined by the transient behavior of an induction motor. In the following, we shall introduce mathematical models of induction motors of two types: a model considering only mechanical transients and a more detailed model including both electromechanical transients and mechanical transients. Induction motors of large and small capacity have obviously different dynamics. For small capacity motors, only mechanical transients need to be considered [168]. 6.5.2.1

Dynamic Load Model Considering Mechanical Transients of an Induction Motor

In this type of model, electromechanical transients of an induction motor are ignored, with only the mechanical transient being taken into consideration. From

398

6 Mathematical Model of Synchronous Generator and Load

machine theory we know that the dynamics of an induction motor can be described by the equivalent circuit of an induction motor as in Fig. 6.31, where X1 and X2 are leakage reactance of armature and field windings, respectively; Xm the mutual impedance between armature and field windings; R2/s the equivalent rotor resistance. If system frequency and motor speed are denoted by o and om, respectively, machine slip speed s ¼ (o om)/o ¼ 1 om* should satisfy the following equation of motion of the rotor TJM

ds ¼ TmM TeM ; dt

ð6:155Þ

where TJM is the equivalent moment of inertia of the machine rotor and mechanical load and TmM and TeM are the mechanical torque of load and machine electrical torque, respectively. Derivation of above equation is the same as that used to derive the rotor motion equation for a synchronous generator, noting its reference positive direction of torque is just opposite to that for the synchronous generator. From the above equation we can see that when load torque is greater than electrical torque, slip speed of the induction motor increases, i.e., motor speed decreases. Ignoring electromechanical transients, electrical torque of an induction motor can be expressed to be TeM

2TeM max VL 2 ¼ s ; scr V LN þ scr s

ð6:156Þ

where TeM max is the maximal electrical torque of the induction motor at rated voltage and scr is the critical slip speed for steady-state stability of the induction motor. For a certain induction motor, TeM max and scr are constant when change of frequency is not considered. VL and VLN are the terminal voltage and rated voltage of the induction motor, respectively. Mechanical torque of an induction motor is related to the characteristics of the mechanical load and often a function of motor speed. Traditionally it is given as TmM ¼ k½a þ ð1 aÞð1 sÞpm ;

ð6:157Þ

where a is the portion of mechanical load torque that is independent of motor speed, pm the exponent associated with the characteristic of the mechanical load, and k is R1 + jX1 VL

Fig. 6.31 Equivalent circuit of induction motor

jX2 R2 / s

Rμ + jX μ

6.5 Mathematical Model of Load

399

the percentage of motor load. To achieve better flexibility and wider applicability of computation, currently mechanical torque is expressed as the summation of polynomial and exponential forms [168] TmM om 2 om om g ¼ am þ bm þ c m þ dm ; ð6:158Þ TmM0 om0 o0 om0 where TmM0 and om0 are mechanical torque and motor speed before occurrence of disturbance. am, bm, cm, dm, and g are the characteristic parameters of mechanical torque. Parameter cm is calculated from the following equation cm ¼ 1 ðam þ bm þ dm Þ:

ð6:159Þ

From Fig. 6.31, we can obtain the equivalent impedance of an induction motor to be ZM ¼ R1 þ jX1 þ

ðRm þ jXm ÞðR2 =s þ jX2 Þ ; ðRm þ R2 =sÞ þ jðXm þ X2 Þ

ð6:160Þ

where ZM is a function of motor slip speed. Rotor motion equation of an induction motor (6.155), electrical torque ignoring electromechanical transients (6.156), load mechanical torque (6.157), (6.158), or (6.159), and equivalent impedance (6.160) form the mathematical model of an induction motor load with electromechanical transients neglected. Input variables to the model are node voltage and system frequency. Output variable is the equivalent impedance. Hence when VL and o, as functions of time, are known, s can be found by solving the above equations to obtain the equivalent impedance ZM at any time. As pointed out previously, nodal load includes all electrical equipment connected at the node. Because so many types of electrical equipment may be connected, the dynamics of nodal load are very complicated. In the following, we shall introduce a method of simplifying nodal load by use of the classical model of an induction motor. The key issue in the simplification is to obtain the equivalent impedance of nodal load at any time. Step 1. We separate the total power PL(0) and QL(0) absorbed by the nodal load, in steady-state operation, into two parts. One part is expressed by a static load model with power PLS(0) and QLS(0). The corresponding equivalent impedance is denoted 2 as ZLSð0Þ ¼ VLð0Þ ½PLSð0Þ jQLSð0Þ . Another part is modeled by an induction motor with only mechanical transients considered. The power of this part is denoted as 2 PLM(0) and QLM(0) with corresponding equivalent impedance ZLMð0Þ ¼ VLð0Þ ½PLMð0Þ jQLMð0Þ . Equivalent impedance of nodal load is ZL(0) ¼ ZLS(0) þ ZLM(0). Step 2. It is approximated that all equipment connected at the node, with their dynamics being considered, is a certain typical induction motor. Model parameters of the typical motor are s(0), TJM, TeM max, scr, R1, X1, R2, X2, Rm, Xm and k, a, pm or am, bm, dm, g. From (6.160) we can find the steady-state equivalent impedance of the typical motor ZM(0). Obviously, steady-state equivalent impedance of the typical

400

6 Mathematical Model of Synchronous Generator and Load

motor does not have to be equal to the steady-state equivalent impedance of an equivalent motor. Step 3. In a system transient, node voltage and system frequency vary with time. By using some numerical methods to solve system equations and rotor motion equation of the typical motor (details about the numerical method are introduced in Chaps. 7 and 8), we can obtain the slip speed s(t) of the typical motor, nodal voltage magnitude VL(t) and system frequency o(t) at time t. From (6.160) we then can calculate the equivalent impedance of the typical motor ZM(t) at time t. From the static load model we can find the equivalent impedance of static load ZLS(t). Step 4. We suppose that at any time, the ratio between the equivalent impedance of equivalent motor and equivalent impedance of typical motor is a constant. Hence at any time t, the equivalent impedance of the equivalent motor is ZLMðtÞ ¼ ðcr þ jci ÞZMðtÞ ;

ð6:161Þ

where the proportionality constant can be found from steady-state conditions cr þ jci ¼ ZLMð0Þ =ZMð0Þ :

ð6:162Þ

Finally we obtain the equivalent impedance of nodal load at time t to be ZLðtÞ ¼ ZLSðtÞ þ ZLMðtÞ :

6.5.2.2

ð6:163Þ

Load Dynamic Model with Electromechanical Transients of Induction Motors Considered

Compared to the model introduced above, this model considers electromechanical transients of the field winding of induction motors. Similar to the case of a synchronous generator, because the transient of the armature winding is very fast, we do not consider the electromechanical transient of armature windings for an induction motor either. Details about deriving the mathematical model of an induction motor with electromechanical transients of the field winding being taken into account can be found in [153]. In the following, we shall give a simple derivation method by use of the mathematical model of a synchronous generator established in Sect. 6.2. In fact, as far as the transient equation of the machine is concerned, an induction motor can be considered to be a synchronous generator being completely symmetrical in the two directions of d- and q-axes. Hence in some algorithms of power system transient stability analysis, modeling of induction motors and synchronous generators is treated in the same way. When an induction motor is considered individually; for simplicity, the subtransient process of a synchronous generator is ignored. In the mathematical model, the f winding has the same structure as that of

6.5 Mathematical Model of Load

401

the g winding but is short-circuited. Under these conditions, in equations of the synchronous generator ((6.43)–(6.46)), letting Xd ¼ Xq ¼ X, Xd0 ¼ Xq0 ¼ X, 0 0 eq2 ¼ ed2 ¼ e00q ¼ e00d ¼ 0, p’d ¼ p’q ¼ 0, Td0 ¼ Tq0 , o ¼ 1 s, Ra ¼ R1, we have per unit equations of an induction motor to be 9 vq ¼ ð1 sÞðe0q X0 id Þ R1 iq > > > > vd ¼ ð1 sÞðe0d þ X0 iq Þ R1 id = ; 0 > Td0 pe0q ¼ e0q ðX X0 Þid > > > ; 0 Td0 pe0d ¼ e0d þ ðX X0 Þiq

ð6:164Þ

0 where machine parameters, X, X0 , and Td0 can be derived from Fig. 6.31. Because dand q-axis are completely symmetrical and the structure of f and g windings is the same, in (6.32) and (6.33) we have

Xaf ¼ Xag ¼ Xm :

ð6:165Þ

Hence according to the definition of synchronous reactance, we have the following equation for the stator side, X ¼ Xd ¼ Xq ¼ X1 þ Xm :

ð6:166Þ

Similarly on the rotor side, we have Xf ¼ Xg ¼ X2 þ Xm :

ð6:167Þ

Substituting (6.166) and (6.167) into (6.40b), we can obtain X0 ¼ Xd0 ¼ Xq0 ¼ X1 þ

X2 Xm : X2 þ Xm

ð6:168Þ

We denote the resistance in (6.30) and (6.31) Rf ¼ Rg as R2. Substituting (6.167) into (6.40b) we have 0 0 Td0 ¼ Tq0 ¼ ðX2 þ Xm Þ=R2 :

ð6:169Þ

Equation (6.164) can be simplified by converting it in dq coordinates from (6.62) to system unified xy coordinates. Differentiation of (6.62) to per unit time can result in p

Ad Aq

¼

A sin d cos d cos d sin d Ax p x þ pd: Ay cos d sin d sin d cos d Ay

ð6:170Þ

402

6 Mathematical Model of Synchronous Generator and Load

From the geometrical meaning of a and (6.78) we know pd ¼ s. Hence in xy coordinates (6.164) becomes vx ¼ ð1 sÞe0x þ ð1 sÞX0 iy R1 ix vy ¼ ð1 sÞe0y ð1 sÞX0 ix R1 iy

)

0 0 Td0 pe0x ¼ Td0 se0y e0x þ ðX X0 Þiy 0 0 Td0 pe0y ¼ Td0 se0x e0y ðX X0 Þix

;

ð6:171Þ

) :

ð6:172Þ

At quasisteady state, multiplying the second equation in (6.171) and (6.172) by j and adding to the first equation, we have V_L ¼ ð1 sÞE_ 0M ½R1 þ jð1 sÞX0 I_M ;

ð6:173Þ

0 0 Td0 pE_ 0M ¼ ð1 þ j sTd0 ÞE_ 0M jðX X0 ÞI_M ;

ð6:174Þ

where V_L ¼ Vx þ jVy , I_M ¼ Ix þ jIy , E_ 0M ¼ E0x þ jE0y . However, with subtransient process ignored, the mathematical model of a synchronous generator cannot be converted into the form of (6.173) and (6.174) if d- and q-axis are not symmetrical. Treating an induction motor as a synchronous generator and from (6.81), (6.43), and (6.44), we can obtain the electrical torque of an induction motor to be TeM ¼ ðe0q iq þ e0d id Þ ¼ ðe0x ix þ e0y iy Þ;

ð6:175Þ

where the negative sign is because the positive reference direction of electrical torque of an induction motor is just opposite to that of a synchronous generator. Because a generator model is used, the reference direction of current is going out of, instead of into the induction motor. Therefore, the mathematical model of an induction motor considering electromagnetic transients consists of (6.155), (6.173)–(6.175) and the load mechanical torque of (6.157) or (6.158). For nodal composite load, we can use the same method adopted previously with mechanical transients being considered. For the typical motor, pE_ 0M ¼ 0 in steady-state operation, from (6.173) and (6.174) we can find I_Mð0Þ ; E_ 0Mð0Þ . Hence the equivalent impedance of the typical motor at steady state is ZMð0Þ ¼ V_ Lð0Þ I_Mð0Þ . Equivalent impedance of the equivalent motor can be calculated from nodal voltage and load power at steady state. Thus the ratio between equivalent impedance of typical and equivalent motors can be computed from (6.162). During transients, solving the combined equation describing the typical motor and system we can obtain I_MðtÞ , V_ LðtÞ , and ZM(t). Hence the equivalent impedance of equivalent motor and composite load can be calculated from (6.161) and (6.163). During transients, variation of slip speed has little effect on armature voltage,

Thinking and Problem Solving

403

numerically. It can be ignored in a simple computation and hence in the armature voltage equation of the motor, (6.173), s is taken to be a constant 0. Typical parameters of induction motors can be found in [168, 176]. There are other forms of load dynamic model. For some special loads with large capacity, such as large rolling machines, electric-arc furnaces, electric trains, large units of temperature control, and synchronous motors in pumping or energy storage power plants, etc., the model needs to be established individually. For long-term stability analysis, transformer saturation, adjustment of under-load tap changing transformers, voltage regulation arising from reactive compensators, and the action of low-frequency low-voltage load-shedding equipment, etc., ought to be represented within the load models. Overall, load modeling is still a developing subject.

Thinking and Problem Solving 1. How is the relationship between the electrical quantities in stator and rotor of a synchronous generator set up? 2. Does the mutual inductance between stator winding and rotor winding vary with time, according to whether the generator is round-rotor or salient-pole? 3. What is the function of Park conversion? 4. In the state equation of (6.1), each winding flux linkage is a state variable. Considering the motion equation of the rotor, the electrical rotational speed o of generator is also a state variable. Discuss the nonlinearity of the generator model according to this formula. 5. Discuss the physical significance of the right-hand three items in (6.14), and thereby explain the electrical mechanism of power output of a generator. 6. What are the usual applications of round-rotor generators and salient-pole generators in electrical power systems, and why? 7. The form of synchronous generator model will be influenced by such factors as the choosing of positive direction of magnetic axis, the suppositions taken during converting original parameters into rotor parameters, selection method of base values, and so on. By consulting other books, compare the common and different points of synchronous machine models with the forms that are introduced in this book. 8. By consulting books on synchronous generator experiments, find out about and describe the methods that can be used to empirically determine the parameters of a synchronous generator. 9. During a transient in an electrical power system, the electromagnetic transient process in the electrical network is much faster than the rotor flux dynamics of the generator, so in the synchronous generator model that is used to analyze the electromechanical transient process, the time derivative of stator winding flux linkage is taken to be zero. Analyze the effect of this approximation on calculation quantity.

404

6 Mathematical Model of Synchronous Generator and Load

10. There are three kinds of coordinates used to describe the electrical quantities of a generator. These are the electrical quantities in three-phase static coordinates a, b, c; three-axis rotating coordinates d, q, 0; and complex plane x – y. Discuss the relationship among these three kinds of description. 11. Given one salient-pole synchronous generator, its terminal voltage U_ t ¼ 1:0, and unit power output P þ jQ ¼ 1.0 þ j0.1. The parameters of the generator unit are Xd ¼ 1.0, Xq ¼ 0.6, Xd0 ¼ 0:3; Xq0 ¼ 0:2; Xd00 ¼ 0:15; Xq00 ¼ 0:1. If the stator resistance is neglected, calculate the emfs E0q ; E0d ; E00q , and E00q of this generator. 12. During the formulation of a synchronous generator model, in which formulae can the electrical rotational speed o be considered approximately as invariable, and in which formulae can the electrical rotational speed o not be considered approximately invariable? Why? 13. Discuss the effect of excitation current on the operating state of a synchronous generator, according to (6.50) and (6.51). 14. Discuss the working mechanism of the Washout link in PSS (in Fig. 6.14). 15. Discuss the necessity and difficulty of building steady and dynamical synthetic load models.

Chapter 7

Power System Transient Stability Analysis

7.1

Introduction

The mechanical–electrical transient of a power system that has experienced a large disturbance can evolve into two different situations. In the first situation, the relative rotor angles among generators exhibit swing (or oscillatory) behavior, but the magnitude of oscillation decays asymptotically; the relative motions among generators eventually disappear, thus the system migrates into a new stable state, and generators remain in synchronous operation. The power system is said to be transiently stable. In another situation, the relative motions of some generator rotors continue to grow during the mechanical–electrical transient, and the relative rotor angles increase, resulting in the loss of synchronism of these generators. The system is said to be transiently unstable. When a generator loses synchronization with the remaining generators in the system, its rotor speed will be above or below what is required to produce a voltage at system frequency, and the slip motion between the rotating stator magnetic field (relative to system frequency) and rotor magnetic field causes generator power output, current and voltage to oscillate with very high magnitudes, making some generators and loads trip and, in the worst case, causing the system to split or collapse. A necessary condition that a power system maintains normal operation is the synchronous operation of all generators. Therefore, analyzing the stability of a power system after a large disturbance is equivalent to analyzing the ability of generators to maintain synchronous operation after the system experiences a large disturbance, this is called power system transient stability analysis. The aforementioned power system transient stability analysis typically involves the short-term (within some 10 s) dynamic behavior of a system, nevertheless, sometimes we have to study system midterm (10 s to several minutes) and longterm (several minutes to tens of minutes) dynamic behavior, this would be termed power system midterm and long-term stability analysis. Midterm and long-term stability mainly concerns the dynamic response of a power system that experiences a severe disturbance. A severe disturbance can cause system voltages, frequency, and power flows to undergo drastic changes; therefore, it is meaningful to look into certain slow dynamics, control, and protection X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853‐7, # Springer Science þ Business Media, LLC 2008

405

406

7 Power System Transient Stability Analysis

performance that are not addressed in a short-term transient stability analysis. The response time of devices that affect voltage and frequency can be from a few seconds (such as the response time of generator control and protection devices) to several minutes (such as the response time of a prime mover system and on-load tap changing regulators, etc.) A long-term stability analysis focuses on the slow phenomenon with long duration that occur after a large disturbance has happened, and the significant mismatch between active/reactive power generation and consumption. The phenomena of concern include: boiler dynamics, water gate and water-pipe dynamics of hydraulic turbines, automatic generation control (AGC), control and protection of power plant and transmission system, transformer saturation, abnormal frequency effects of load and network, and so forth. When performing long-term stability analysis, one is often concerned about the responses of a system under extremely severe disturbances that are not taken into consideration in system design. After the occurrence of an extremely severe disturbance, a power system can undergo cascading faults and can be split into several isolated parts. The question a stability study has to answer is whether or not each isolated part can reach acceptable stable operation after any load-shedding occurs. Midterm response refers to response whose timeframe is between that of shortterm response and long-term response. Midterm stability study investigates the synchronous power oscillations among generators, including some slow phenomena and possibly large voltage and frequency deviations [177]. Large disturbances are severe threatens to power system operation, but in reality they cannot be avoided. The consequence of losing stability after a power system experiences a disturbance is in general very serious, it can even be a disaster. In fact the various large disturbances, such as short circuit, tripping or committing of large capacity generator, load, or important transmission facility, appear as probabilistic events, therefore when designing and scheduling a power system, one always ensures that the system can maintain stable operation under a set of reasonably specified credible contingencies, rather than requiring that the system can sustain the impact of any disturbance. Because every country has their own stability requirements, the selection of credible contingencies can be based on different standards. To check if a power system can maintain stable operation under credible contingencies, one needs to perform transient stability analysis. When the system under study is not stable, efficient measures that can improve system stability need to be sought. When a system experiences extremely severe stability problems, fault analysis is required to find the weak points in the system, and develop corresponding strategies. In power system stability analysis, the mathematical models of system components not only directly relate to the analysis results, but also have a significant effect on the complexity of the analysis. Therefore, if appropriate mathematical models for each system component are developed, stability analysis can be made simple and accurate. This is a crucial step in stability analysis.

7.2 Numerical Methods for Transient Stability Analysis

Differential Eqs

Rotor Circuit Eqs

Network Eqs

Excitation System Eqs

Coordinate Transformation Eqs

PSS Eqs

Rotormotion Eqs

Stator Voltage Eqs

Primer-mover & Governer Eqs

407

Other generotots Loads DC System Other dynamic devices Such as SVC, TCSC, etc

Algebraic Eqs

Fig. 7.1 Conceptual framework of mathematical models for stability studies

Figure 7.1 conceptualizes the mathematical model of all system components for power system stability studies. From the figure one can see that the mathematical model consists of the models of synchronous machines and the associated excitation systems, prime mover and speed-governing system, electrical load, and other dynamic devices and electrical network. Apparently, all the dynamic components of the system are independent; it is the electrical network that connects them with each other. Mathematically, the complete system model can be described as a set of differential-algebraic equations as follows: dx ¼ fðx; yÞ; dt

ð7:1Þ

0 ¼ gðx; yÞ:

ð7:2Þ

This chapter first introduces the composition of the component models for transient stability analysis and the numerical solution algorithms for differentialalgebraic equations, then describes the mathematical relationship between dynamic components and electrical network, followed by an exposition of how to model network switches and faults. The chapter also presents in detail the solution algorithms for simple transient stability analysis and for analysis of systems with FACTS devices represented by full mathematical models.

7.2

Numerical Methods for Transient Stability Analysis

Power system transient stability analysis can be viewed as an initial value problem of differential-algebraic equations. In this section we first introduce the numerical methods for ordinary differential equations, and then discuss the numerical methods for differential-algebraic equations. We provide a general procedure for transient stability analysis at the end of the section.

408

7 Power System Transient Stability Analysis

7.2.1

Numerical Methods for Ordinary Differential Equations

7.2.1.1

Fundamental Concept

Consider the following first-order differential equation: dx ¼ f ðt; xÞ; dt

xðt0 Þ ¼ x0 :

ð7:3Þ

In general, the function f in the above equation is a nonlinear function of x and t. In many real world situations, f is not an explicit function of time t, therefore the above equation reduces to dx ¼ f ðxÞ; dt

xðt0 Þ ¼ x0 :

ð7:4Þ

In power system stability analysis, the right-hand side of all the differential equations does not contain explicitly the time variable t. When f in (7.4) is a linear function of x, one can easily obtain the closed-form solution of the differential equation. For example, consider the following differential equation: dx ¼ x: dt

ð7:5Þ

The closed-form solution is given as x ¼ A et ;

ð7:6Þ

where A is a constant. Equation (7.6) represents a family of integral curves. Given an initial condition in the form of x(t0) ¼ x0, one can determine a solution curve. For instance, if x(0) ¼ 1, then from (7.6) the integral constant can be found as A ¼ 1, thus the solution curve is as follows: x ¼ et :

ð7:7Þ

The differential equations of real world engineering problems appear to be more complex, the right-hand sides of the equations are typically not integrable, therefore closed-form solutions, like (7.6), of such differential equations cannot be obtained. To solve these differential equations, one must rely on numerical methods. The idea of numerical methods is to employ a certain integral formula to solve for the approximate value of xn at each instant in the time series tn ¼ t0 þ nh, n ¼ 1,2, . . . (here h is the step size) in a step-by-step fashion, starting from the initial state (t ¼ t0, x ¼ x0). This method of solving differential equation is called step-by-step integration.

7.2 Numerical Methods for Transient Stability Analysis Fig. 7.2 Illustration of Euler’s method for solving differential equations

409

x

x3

true solution x = x(t)

x2 x0

x1

0

t1

t2

t3

t

In the following, we illustrate the basic idea of step-by-step integration using Euler’s method as an example. Suppose that the exact solution of the first-order differential (7.3) at t0 ¼ 0, x(t0) ¼ x0 is as follows: x ¼ xðtÞ:

ð7:8Þ

The graph of the function, that is, the integral curve of the differential (7.3) passing through the point (0, x0) is depicted in Fig. 7.2 . Euler’s method is also called the Euler’s tangent method or Euler’s polygon method. The idea of the method is to approximate the integral curve by an Euler’s polygon, the slope of each Euler’s polygon is obtained by solving for (7.3) with the initial value of the Euler’s polygon as input. Specifically, the computational steps are as follows: For the first segment, the slope of the integral curve at point (0, x0) is dx ¼ f ðx0 ; 0Þ: dt 0 Replacing the first segment with a straight line which has a slope of dx dt 0 , one can find the incremental of x at t1 ¼ h (h is the step size) as follows: dx Dx1 ¼ h: dt 0 Therefore the approximation of x at t1 ¼ h should be dx x1 ¼ x0 þ Dx1 ¼ x0 þ h: dt 0

410

7 Power System Transient Stability Analysis

For the second segment, the integral curve will be approximated by another straight line segment, the slope of which can be obtained by substituting the initial value of the segment (that is, the starting point of the segment (t1, x1)) into (7.3): dx ¼ f ðx1 ; t1 Þ: dt 1 An approximate value of x at t2 ¼ 2h can be found based on x2 ¼ x1 þ

dx h dt 1

as illustrated in Fig. 7.2. The above procedure can be repeated to find an approximate value of x3 at t3 and so forth. In general, the recursive formula for computing an approximate value of the n þ 1 point is as follows: dx xnþ1 ¼ xn þ h; dt n

n ¼ 0; 1; 2; . . . :

ð7:9Þ

Now we turn to analyzing the error introduced by this recursive formula which is used to compute (tnþ1, xnþ1) from (tn, xn). To do so, expand the integral function (7.8) at (tn, xn) using Taylor’s formula as follows: xnþ1 ¼ xn þ x0n h þ x00n

r h2 ðrÞ h þ þ x xn ; 2! r!

ð7:10Þ

where x0n ; x00n ; . . . are the first-order, second-order,. . . derivatives of the integral function with regard to variable t. The symbol xn represents a number in the interval ðrÞ [tn, tnþ1], and xxn is the residual of the Taylor’s series. When r ¼ 2, (7.10) becomes h2 n 2!

xnþ1 ¼ xn þ x0n h þ x00x0

ð7:11Þ

or in an alternative form xnþ1

dx d2 x h2 ¼ xn þ h þ 2 : dt n dt x0n 2!

ð7:12Þ

Here the symbol x0n still represents a number in interval [tn, tnþ1] and in general 6¼ xn . Obviously, Euler’s recursive (7.9) can be obtained after neglecting the residual 2 2 term ddt2x 0 h2! in (7.12). x0n

xn

7.2 Numerical Methods for Transient Stability Analysis

411

Therefore when computing the function value at point n þ 1 from that at n, the error introduced by the approximation is Enþ1

d2 x h2 ¼ 2 : dt x0n 2!

ð7:13Þ

Suppose that within the computing interval [0,tm], the maximum value 2 of ddt2x ¼ f 0 ðx; tÞ is M, then the error Enþ1 should satisfy Enþ1

M 2 h ; 2

ð7:14Þ

where M is independent of the choice of step size h. The errors in (7.13) and (7.14) are due to the approximation made when computing the function value at point n þ 1 from that at n, it is called local truncation error. The truncation error of Euler’s formula is in proportion to h2, and often expressed as of order h2 or 0(h2). It should be noted that before obtaining xnþ1, xn is solved using the same recursive formula, therefore xn itself also contains error. As a matter of fact, when computing xnþ1 based on (7.9), one should take into account the impact of the error of xn, in addition to the impact of the local truncation error associated with neglecting residual term. This error is called global truncation error or simply put truncation error. Consequently the error introduced by the inaccuracy of Euler’s formula is larger than the local truncation error expressed in (7.13) and (7.14). It can be proved that the global truncation error of Euler’s formula is in proportion to h, in other words, it is 0(h). Based on the above discussion, a smaller step size h should be selected to reduce the computational error of the Euler’s formula. But it is false to assert that the smaller the step size h is, the smaller the error would be. In the aforementioned discussion, we did not take into consideration the roundoff error of the computer. When a small step size h is used, the computational effort adversely increases; thus, the impact of rounding errors increases, as illustrated in Fig. 7.3. In the figure, hmin is the step size associated with the minimum error, therefore we cannot merely rely on reducing the step size to reduce computational error. If higher computational precision is desired, a better computational algorithm has to be used. In the above calculations, when computing the function value at tnþ1, only the function value xn at the previous point tn is required, this algorithm is called a single-step algorithm. The algorithms to be presented in this section belong to this category. There are multistep or multivalue algorithms which are more accurate. These algorithms require the information of previous steps (tn, xn), (tnþ1, xnþ1), . . . , (tnkþ1, xnkþ1) when solving for the value xnþ1 corresponding to time tnþ1.

412

7 Power System Transient Stability Analysis

error

minimum error truncation error round-off error

h

hmin

Fig. 7.3 Relationship between error and step size

7.2.1.2

Modified Euler’s Method

The large error of Euler’s method comes from the fact that the derivative dx dt n ¼ f ðxn ; tn Þ of the starting point of an Euler’s polygon is used for the entire segment [tn, tnþ1]. In other words, the slope of each Euler’s polygon is entirely determined by the starting point of the polygon. If the slope of an Euler’s segment is replaced with the average of slopes of starting point and end point, we should expect improved solution precision. This is the basic idea of the modified Euler’s method. For first-order differential equation (7.3), let the initial value is given as t0 ¼ 0, x(t0) ¼ x0, in what follows we introduce the computational steps of the modified Euler’s method. To find out the function value x1 at t1 ¼ h, first compute an approximate value of x1 using Euler’s method: ð0Þ x1

dx ¼ x0 þ h; dt 0

where dx dt 0 ¼ f ðx0 ; t0 Þ. ð0Þ

ð7:15Þ

ð0Þ

When x1 is obtained based on (7.15), substitute t1, x1 into (7.3) to solve for the derivative at the end point of the segment: dxð0Þ ð0Þ ¼ f ðx1 ; t1 Þ: dt 1

7.2 Numerical Methods for Transient Stability Analysis

413

x

Fig. 7.4 Geometrical explanation of the modified Euler’s method

x = x(t) dx dt

(0)

dx dt

1

dx dx + dt dt

x1

0

(1)

x1

2

1 (0) 1

(0)

x1 x0

dx dt

h t2

t1

0

t

dxð0Þ Now the average of dx dt 0 and dt 1 can be used to calculate an improved solution of x1 as follows:

ð1Þ

x1

dx dxð0Þ þ dt 0 dt 1 ¼ x0 þ h: 2

ð7:16Þ

ð1Þ

The solution x1 computed this way better approximates the true solution x1 than ð0Þ

does x1 which is computed using a standard Euler’s method. Figure 7.4 provides a geometrical explanation. To compute (tnþ1, xnþ1) from (tn, xn), the following general formula can be used 9 dx > > ¼ f ðx ; t Þ n n > > dt n > > > > > dx > ð0Þ > > xnþ1 ¼ xn þ h > > dt n > = ð0Þ : ð7:17Þ dx ð0Þ > ¼ f ðxnþ1 ; tnþ1 Þ > > dt nþ1 > > > ð0Þ > > > > dx dx > > þ > > dt dt ð1Þ n nþ1 > xnþ1 ¼ xnþ1 ¼ xn þ h; 2 Eliminate xn in (7.17), the fourth formula of (7.17) can be modified as xnþ1 ¼

ð0Þ xnþ1

a dx þ h; dt nþ1

ð7:18Þ

414

7 Power System Transient Stability Analysis

where a ð0Þ ! dx 1 dx dx ¼ : dt nþ1 2 dt nþ1 dt n As such, the general formula of the modified Euler’s method can be summarized as follows: 9 dx > > ¼ f ðxn ; tn Þ > > > dt n > > > > > dx ð0Þ > > xnþ1 ¼ xn þ h > = dt n a : ð7:19Þ dx 1 dx > ð0Þ > > ¼ f ðx ; t Þ > nþ1 nþ1 dt nþ1 2 dt > > > a n > > > > dx ð1Þ ð0Þ > xnþ1 ¼ xnþ1 ¼ xnþ1 þ h > ; dt nþ1 When solving for xnþ1 based on (7.19), because it takes the same form as that of ð0Þ xnþ1 , the computer code can be simplified. In addition, xn need not be saved after ð0Þ

xnþ1 is obtained, thus computer memory can be saved. In what follows we discuss the local truncation error of modified Euler’s method. To do so, recall the Taylor’s expansion formula of equation (7.10): xnþ1 ¼ xn þ x0n h þ x00n

h2 h3 þ x000 ; x00n 2! 3!

ð7:20Þ

h3 is the residual term of the Taylor’s expansion. n 3! The fourth equation of the modified Euler’s method (7.17) can be re-cast as

where x000 x00

h h ð0Þ ð1Þ xnþ1 ¼ xn þ x0n þ f ðxnþ1 ; tnþ1 Þ: 2 2 Substituting the first formula in (7.17) into the above equation, one obtains h h ð1Þ xnþ1 ¼ xn þ x0n þ f ðxn þ x0n h; tn þ hÞ: 2 2

ð7:21Þ

Expand the third term in the right-hand side of the above equation using Taylor’s formula, h h h2 @f 0 h2 @f f ðxn þ x0n h; tn þ hÞ ¼ f ðxn ; tn Þ þ x þ þ 0ðh3 Þ: 2 2 2 @xn n 2 @t n

7.2 Numerical Methods for Transient Stability Analysis

Since x00n

415

@f 0 @f ¼ xn þ ; @x n @t n

therefore h h h2 f ðxn þ x0n h; tn þ hÞ ¼ x0n þ x00n þ 0ðh3 Þ; 2 2 2 substituting the above formula into (7.21), it follows ð1Þ

xnþ1 ¼ xn þ x0n h þ x00n

h2 þ 0ðh3 Þ; 2

ð7:22Þ

subtracting the above formula from (7.20), we have ð1Þ

h3 0ðh3 Þ: n 3!

Enþ1 ¼ xnþ1 xnþ1 ¼ x000 x00

The above equation shows that the local truncation error of the modified Euler’s method is 0(h3). By the same token, it can be proved that the global truncation error of the modified Euler’s method is 0(h2). [Example 7.1] Solve the following differential equation by the modified Euler’s method dx 2t ¼ x ; dt x where the initial values are t0 ¼ 0 and x0 ¼ 1. [Solution] Taking 0.2 as step length, the computational results are summarized in the following table:

n 0 1 2 3 4 5

tn 0 0.2 0.4 0.6 0.8 1.0

xn 1 1.18667 1.34832 1.49837 2.62790 1.75430

dx dt n

xnþ1

1 0.84959 0.75499 0.69036 0.64500

1.2 1.35658 1.49932 1.63179 1.75690

ð0Þ

tnþ1 0.2 0.4 0.6 0.8 1.0

The true solution of this differential equation is x¼

pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2t þ 1:

dx0 dt nþ1 0.8667 0.7669 0.6990 0.6513 0.6185

ð0Þ dx dx þ dt n dt nþ1 2 0.9333 0.8083 0.7270 0.6708 0.6318

xn 1.18667 1.34832 1.49372 1.62788 1.75430

416

7 Power System Transient Stability Analysis

When t ¼ 1, x ¼ 1.73205, therefore the error is equal to j1:73205 1:7543j ¼ 0:02225: The modified Euler’s method can also be employed to solve first-order differential equations. For instance, for the following differential equations: 9 dx ¼ f1 ðx; y; tÞ > = dt : ð7:23Þ dy > ¼ f2 ðx; y; tÞ ; dt Let the initial values be t0, x0, y0, when step length h is determined, for the first segment, one can compute the approximate value of the true solution as follows: dx ð0Þ x1 ¼ x0 þ h dt 0 ; dy ð0Þ y1 ¼ y0 þ h dt 0 where dx ¼ f1 ðx0 ; y0 ; t0 Þ dt 0 : dy ¼ f ðx ; y ; t Þ 2 0 0 0 dt 0 ð0Þ

ð0Þ

From t1 ¼ h; x1 ; y1 , we have ð0Þ dx ð0Þ ð0Þ ¼ f1 ðx1 ; y1 ; t1 Þ dt 1 ; ð0Þ dy ð0Þ ð0Þ ¼ f2 ðx1 ; y1 ; t1 Þ dt 1 thus the true solution of the differential equation at t should be dx dxð0Þ a þ dt 0 dt 1 dx ð0Þ x1 ¼ x0 þ h ¼ x1 þ h; 2 dt 1 ð0Þ dy dy þ dt 0 dt 1 dya ð0Þ y1 ¼ y0 þ h ¼ y1 þ h; 2 dt 1

7.2 Numerical Methods for Transient Stability Analysis

417

where a ð0Þ dx 1 dx ¼ dt 1 2 dt 1 dya 1 dyð0Þ ¼ dt 1 2 dt 1

! dx dt 0 ! dy dt 0

and so forth. From (7.17), it can be concluded that the modified Euler’s method applied to one segment requires computational effort, that is, two times of that of the Euler’s method. On the other hand, if the same step length is used, the modified Euler’s method provides more accurate calculation results than the Euler’s method. As discussed before, the truncation error of the modified Euler’s method is 0(h2), while the Euler’s method is 0(h). Figure 7.5 illustrates that, when the tolerance is equal to e1, the difference between the required step length of the modified Euler’s method h01 and the Euler’s step length h1 is small. Under such circumstance, the computational effort required by the modified Euler’s method is larger than that of the Euler’s method. When the tolerance is equal to e2, the required step length of the modified Euler’s method h02 is significantly larger than the Euler’s step length h2. Obviously, if h02 > 2h2 , then the total computational effort of the modified Euler’s method is smaller than that of the Euler’s method.

7.2.1.3

Runge–Kutta Method

The modified Euler’s method is based on the observation that xnþ1 can be estimated using the derivatives or slopes of two points in the interval [tn, tnþ1], and since the Taylor’s series of the integral function is approximated by the first three terms, the local truncation error is 0(h3). This has motivated the following question: is it possible to estimate xnþ1 using the derivatives of more points in the interval [tn, tnþ1],

e

Euler Method Modified Euler Method

e1 e2 Fig. 7.5 Comparison between the modified Euler’s method and Euler’s method

h2

h2′

h1

h1′

h

418

7 Power System Transient Stability Analysis

such that more terms of the Taylor’s series can be included? The answer to this question is positive. The well-known Runge–Kutta method is built upon this idea. The most popular Runge–Kutta method is the fourth-order method. In this method, xnþ1 is estimated using the derivatives of four points in the interval [tn, tnþ1], thus the first five terms of Taylor’s series are included in the approximation: xnþ1 ¼ xn þ x0n h þ x00n

4 h2 h3 ð4Þ h þ xð3Þ þ x þ 0ðh5 Þ: n n 2! 3! 4!

The local truncation error of the method is 0(h5), and the global truncation error is 0(h4). For differential (7.3), the following Runge–Kutta formula should be used: 9 1 > xnþ1 ¼ xn þ ðk1 þ 2k2 þ 2k3 þ k4 Þ > > > 6 > > > > k1 ¼ hf ðxn ; tn Þ > > > > = k1 h k2 ¼ hf xn þ ; tn þ 2 2 > > > > > k2 h > > k3 ¼ hf xn þ ; tn þ > > > 2 2 > > ; k4 ¼ hf ðxn þ k3 ; tn þ hÞ

ð7:24Þ

to solve for x1, x2, x3, . . .. [Example 7.2] Solve the first-order differential equation in Example 7.1 using the Runge–Kutta method [Solution] Let the step length h ¼ 0.2, the computational steps are described in the following table: tn 0 0.2 0.4 0.6 0.8 1

xn 1 1.1832292 1.3416668 1.483281 1.612513 1.732141

k1 0.2 0.1698342 0.1490788 0.1348528 0.1240546

tn þ 0.1 0.3 0.5 0.7 0.9

h 2

xn þ

k1 2

1.1 1.267746 1.416026 1.550707 1.674541

k2 0.1836364 0.1588930 0.1420188 0.1295786 0.1199240

tn þ 0.1 0.3 0.5 0.7 0.9

h 2

k2 2 1.0918182 1.262676 1.412676 1.548070 1.672475 xn þ

k3 0.1817274 0.1574990 0.1409600 0.1287436 0.1192452

tn þ h

xn þ k3

0.2 0.4 0.6 0.8 1.0

1.181727 1.340728 1.482627 1.612025 1.731759

k4 0.1686478 0.1488074 0.1346506 0.1238970 0.1153728

The above table shows that, based on the Runge–Kutta method, the value of x at t ¼ 1 is x ¼ 1.732141. Comparing this result with the true solution, the error is equal to j1:73205 1:732141j ¼ 0:00009; which is a much better result in comparison with the result obtained in Example 7.1.

7.2 Numerical Methods for Transient Stability Analysis

419

The Runge–Kutta method can also be used to solve first-order differential equations. As an example, the differential equation (7.23) can be solved using the following recursive formula: 1 xnþ1 ¼ xn þ ðk1 þ 2k2 þ 2k3 þ k4 Þ; 6 1 ynþ1 ¼ yn þ ðl1 þ 2l2 þ 2l3 þ l4 Þ; 6 where 9 k1 ¼ hf1 ðxn ; yn ; tn Þ > > > > k1 l1 h > > > > k2 ¼ hf1 xn þ ; yn þ ; tn þ 2 2 2 = ; k2 l2 h > > > k3 ¼ hf1 xn þ ; yn þ ; tn þ > 2 2 2 > > > > ; k4 ¼ hf1 ðxn þ k3 ; yn þ l3 ; tn þ hÞ 9 l1 ¼ hf2 ðxn ; yn ; tn Þ > > > > k1 l1 h > > > > l2 ¼ hf2 xn þ ; yn þ ; tn þ 2 2 2 = : k2 l2 h > > > l3 ¼ hf2 xn þ ; yn þ ; tn þ > 2 2 2 > > > > ; l4 ¼ hf2 ðxn þ k3 ; yn þ l3 ; tn þ hÞ Although the Runge–Kutta method has the advantage of higher precision, it requires larger computational effort which is four times that required by the Euler’s method. The trend is that multiple-step methods, which require less computational effort, are replacing Runge–Kutta methods when higher computational accuracy is required. Runge–Kutta methods are typically used as auxiliary methods only to initiate multiple-step methods in the first few steps. 7.2.1.4

Implicit Integration Methods

Explicit and implicit methods are the major categories of solution methods for differential equations. The methods described in the previous sections belong to the category of explicit methods. From (7.9), (7.17), and (7.24), one can see that the right-hand sides of the formulas are known quantities; therefore, the value of the end point xnþ1 can be directly computed using those recursive formulas. In contrast, an implicit method does not work with recursive equations, it first converts differential equations into difference equations, then solves for the value xnþ1 using the methods of difference equations. Let us first introduce the method of the trapezoidal rule.

420

7 Power System Transient Stability Analysis

Fig. 7.6 Geometrical illustration of trapezoidal rule

dx dt

f (xn, tn)

f (xn+1, tn)

D

C

A

0

tn

B tn+1

t

When xn at tn is known, the function value xnþ1 at time tnþ1 ¼ tn þ h of the differential equation (7.3) can be solved using the following formula: Z tnþ1 xnþ1 ¼ xn þ f ðx; tÞdt: ð7:25Þ tn

The solution of the definite integral of the above equation is equal to the area of the shaded region in Fig. 7.6. Observe that if the step size h is sufficiently small, the graph of the function f(x, t) between tn and tnþ1 can be approximated by a straight line as illustrated in the figure. Apparently the area of the shaded region is equal to the area of the trapezoid ABCD. Equation (7.25) can thus be reformulated as h xnþ1 ¼ xn þ ½f ðxn ; tn Þ þ f ðxnþ1 ; tnþ1 Þ: 2

ð7:26Þ

This is the difference equation of the trapezoidal rule. Obviously, one cannot rely on certain recursive formula to compute xnþ1 because the right-hand side of (7.26) also includes unknown xnþ1. Equation (7.26) has to solve as an algebraic equation to find xnþ1. Generally speaking, the idea of implicit methods is to transform a numerical initial value problem of differential equations into a sequence of algebraic equation problems. For example, given starting point t0 and x0, according to (7.26) the difference equation for the first step should be h x1 ¼ x0 þ ½f ðx0 ; t0 Þ þ f ðx1 ; t0 þ hÞ; 2 where the only unknown variable is x1, which can be solved for using the methods for solving algebraic equation. Given t1 and x1, based on (7.26), the difference formula for the next step should be h x2 ¼ x1 þ ½f ðx1 ; t1 Þ þ f ðx2 ; t1 þ hÞ 2 from which x2 can be computed, and so forth.

7.2 Numerical Methods for Transient Stability Analysis

421

If f(xn, tn) and f(xnþ1, tnþ1) are viewed as the slopes of the integral curve at the starting point and terminating point of the interval [tn, tnþ1], then it is reasonable to term the implicit trapezoidal rule as an implicit modified Euler’s method. In other words, difference equation (7.26) can be viewed as the solution formula of the implicit modified Euler’s method. In fact, the idea of implicit methods are applicable not only to the modified Euler’s method, but also to the previously mentioned Euler’s method, Runge–Kutta method, and multistep methods. For example, the recursive formula of the Euler’s method (7.9) can be rewritten as xnþ1 ¼ xn þ x0nþ1 h ¼ xn þ f ðxnþ1 ; tnþ1 Þh:

ð7:27Þ

Changing the derivate value x0n of the starting point of the interval [tn, tnþ1] to one obtains the implicit Euler’s method. Equation (7.27) is the difference formula of the implicit Euler’s method. The difference equations (7.26) and (7.27) can be nonlinear as a result of the nonlinearity of the function f(x, t) in (7.3). Therefore the algorithms for implicit methods are more complex than those of explicit methods. It is not difficult to find out that the truncation error of implicit trapezoidal rule is introduced by the approximation of replacing the trapezoid with the shaded area (see Fig. 7.6). Using the same arguments as before, one can prove that the local truncation error of difference equation (7.26) is 0(h3). The advantage of implicit methods over explicit methods is that a larger step size can be used. This issue involves the numerical stability of numerical initial value problems; readers are referred to relevant references. Here we illustrate using a simple example. Suppose we have the following differential equation: x0nþ1 ,

dx ¼ 100x: dt

ð7:28Þ

The initial values are t ¼ 0 and x0 ¼ 1. For the above differential equation, the true solution is x ¼ e100t. This is an exponential function, as depicted in Fig. 7.7. When the step length is equal to h ¼ 0.025, the numerical solution using the Euler’s method is as follows: Steps 0 1 2 3

tn 0.000 0.025 0.050 0.075

xn 1 1.5 2.25 3.375

x0n

x0n h

100 150 225

2.5 3.75 5.625

Observe that the function value oscillates as time increases, and the magnitude of the oscillation increases, as illustrated in the dotted line in Fig. 7.7. Mathematically,

422

7 Power System Transient Stability Analysis

Fig. 7.7 Illustration of solutions obtained using different methods

x

Implicit

Euler method

Euler Method

0

0.025

0.05

0.075

t

this indicates that the numerical solution obtained using the Euler’s method is not stable. This situation can be avoided if the implicit Euler’s method is used. Let us first transform (7.28) into a difference equation as follows: xnþ1 ¼ xn þ x0nþ1 h ¼ xn 100xnþ1 h: Thus xnþ1 ¼

xn : 1 þ 100h

When h ¼ 0.025, xnþ1 ¼

xn : 3:5

One obtains the following calculation results:

Steps 0 1 2 3

tn 0.000 0.025 0.050 0.075

xn 1 1/3.5 (1/3.5)2 (1/3.5)3

The function value in the above table decays as time increases, see Fig. 7.7.

7.2 Numerical Methods for Transient Stability Analysis

423

To explain the relationship between step size and numerical solutions, we rewrite differential equation (7.28) into more general form: dx x ¼ ; dt T

ð7:29Þ

where the constant T has the unit of time, which is termed the time constant. Substituting (7.29) into the Euler’s equation (7.9), we have h xnþ1 ¼ xn 1 : T Therefore h nþ1 xnþ1 ¼ x0 1 : T

ð7:30Þ

Obviously in order for x to be a monotonically decaying function, the right-hand side of (7.30) has to meet the following condition: 0 > ¼ bx ¼ > > R2a þ Xd Xq R2a þ Xd Xq > > > > Rag sin d Xqg cos d Ra cos d Xd sin d > > > ¼ g ¼ y = 2 2 Ra þ Xd Xq Ra þ Xd Xq : 2 2 Ra ðXd Xq Þ sin d cos d Xd cos d þ Xq sin d > > > ¼ B ¼ x > > R2a þ Xd Xq R2a þ Xd Xq > > > 2 2 Xd sin d Xq cos d Ra þ ðXd Xq Þ sin d cos d > > > ; ¼ G ¼ y R2 þ Xd Xq R2 þ Xd Xq a

ð7:40Þ

a

Substituting the current injection formula derived from (7.39) into network equation (7.36), and applying some simple manipulations, one can conclude that the interconnection of a generator is equivalent to a current injection at the corresponding node: 0 Ix gx bx Ed ¼ : Iy0 by gy Eq

7.3 Network Mathematical Model for Transient Stability Analysis

433

This current is termed generator pseudocurrent. Furthermore, the corresponding block of the admittance matrix of the network should be added to by a matrix as follows: Gx Bx : By Gy It is not difficult to see that, after connecting a generator into the system, the network equations during the stability study period are still linear, however, the generator pseudocurrents and the corresponding admittance matrix are functions of the generator variables Ed , Eq , and d. Thus these linear equations are time varying. If simpler synchronous machine models are used in the study, the network equations can be simplified too. These simplified equations appear as n-order equations in the complex plane. Unless there is a fault or switch change, the network equations remain unaltered. Thus during the study period, the coefficient matrix of the network equations needs to be refactorized using triangular factorization only when there is a fault or switch change. In what follows we discuss the network model associated with two simplified machine models. If the effect of damper windings is not considered, the varying E0 q or E0 q ¼ C model for synchronous machines in Table 7.1 should be applied. In this case, (7.39) can be reformulated as 2 3 0 0 Xd0 Xq X þ X X X q q d d Ra sin 2d þ cos 2d 7 6 2 2 2 7 6 0 0 2 2 6 7 Ix Ra þ Xd Xq Ra þ Xd Xq 6 7 ¼6 0 0 0 7 Iy Xd Xq 6 Xd þ Xq þ Xd Xq cos 2d Ra þ sin 2d 7 4 5 2 2 2 2 0 2 0 Ra þ Xd Xq Ra þ Xd Xq 0 E cos d Vx q0 : ð7:41Þ Eq sin d Vy From the above, one obtains the formula of generator current into node i represented in the complex domain: I_i ¼ I_i0 Yi0 V_ i ;

ð7:42Þ

0 Rai j 12 ðXdi þ Xqi Þ ; 0 X R2ai þ Xdi qi

ð7:43Þ

where Yi0 ¼

9 1 0 Rai jXqi _ 0 2 ðXdi Xqi Þ j2di ^_ > = I_i0 ¼ 2 E j e V i 0 0 Rai þ Xdi Xqi qi R2ai þ Xdi Xqi : > ; 0 0 jd i E_ qi ¼ Eqi e

ð7:44Þ

434

7 Power System Transient Stability Analysis

Fig. 7.9 Generator equivalent circuit when damper winding is not considered

· Vi

o

I·i I·i′

i

Yi′

o

The concept underlying (7.42) can be explained using the circuit model illustrated in Fig. 7.9, where Y0 i is called generator pseudoadmittance and is dependent only on generator parameters. The generator pseudoadmittance can be incorporated into the network admittance matrix; I_i0 is the generator pseudocurrent injection which is related to generator terminal voltage. The network equations are now nonlinear, thus can only be solved using an iterative procedure. As one example, assume an initial value of voltage V_ i , compute I_i0 based on (7.44), then solve the network equations for an improved solution of V_ i , taking I_i0 as current injection. This procedure is repeated until convergence is reached. In normal computational steps, the iteration converges within 2–3 steps; while under fault or switching conditions, it may take a few more steps to obtain a converged solution [196]. If synchronous machines are represented by classical models, the effects of damper windings and salient poles are neglected; in addition, the transient voltages E0 of generators behind X0 d are assumed to be constants during the stability study period. This situation is shown in Table 7.1, where E0 ¼ Eq0 ¼ C and Xq ¼ Xd0 . Correspondingly, from (7.42)–(7.44), it follows that Yi0 ¼

1 0 ; Rai þ jXdi

9 1 = _0 > E i 0 Rai þ jXdi : > ; 0 0 jdi _ Ei ¼ Ei e I_i0 ¼

ð7:45Þ

ð7:46Þ

Obviously generator pseudocurrent I_i0 is independent of generator terminal voltage V_i ; thus, once pseudoadmittance Y0 i is incorporated into the network admittance matrix, the network equations can be solved by direct Gauss elimination since I_i0 is a known quantity. 7.3.1.2

Relationship Between Loads and Network

Depending upon the characteristics of loads, the ways loads are treated are different: 1. If loads are represented by constant impedance models, the constant impedances can be incorporated into the network admittance matrix.

7.3 Network Mathematical Model for Transient Stability Analysis

435

2. If loads are modeled as dynamic devices and only the mechanical–electrical interactions of induction motors in synthesized loads are taken into consideration, loads are still modeled as impedances. However, these impedances are not constant during the stability study period, but vary as the slip-speeds of the induction motors vary. Therefore the impedances representing induction motor loads must be updated, given the slips of the induction motors in each step of the transient stability computation. This means that the diagonal elements of the network admittance matrix are varying in the calculation. The network admittance matrix has to be refactorized in each step of the transient stability calculation when solving the network equations. 3. Again, if loads are modeled as dynamic devices and only the mechanical– electrical interactions of induction motors in synthesized loads are taken into consideration, they can be represented using the Norton equivalent circuit described in Sect. 5.5.2, as illustrated in Fig. 7.10. That is, the load impedances R þ jX and KM (r1 þ jx0 ) are incorporated into the network; thus, the loads become simple current sources. This treatment is similar to the way generators connected to the network are treated. In the above load representations, the underlying networks are linear. 4. If loads are modeled based on steady-state voltage characteristics, the corresponding node current injections are nonlinear functions of node voltage; as a result, the network equations are nonlinear. According to (6.148) and (6.153), the steady-state voltage characteristics of loads have two formulations, these are the second-order polynomial formulation and exponential formulation: " # 9 9 > Vi 2 Vi > Vi m > > > Pi ¼ Pið0Þ aP þ bP þ cP > > > = Pi ¼ Pið0Þ V = Við0Þ Við0Þ ið0Þ " # : ð7:47Þ n 2 > > Vi > > Vi Vi > > Qi ¼ Qið0Þ ; Qi ¼ Qið0Þ aQ þ bQ þ cQ > > ; Við0Þ Við0Þ Við0Þ Note that the active and reactive powers in the above equations are the loads absorbed from the network.

VM

Network

eM ′

R+jX

KM (r1 + jx′) KM (r1 + jx′)

Fig. 7.10 Load representation

436

7 Power System Transient Stability Analysis

Node voltage, current injection, and power injection are connected by the following relationship: Pi jQi ¼ V_ i^I_i ¼ ðVxi þ jVyi ÞðIxi jIyi Þ from which it is easy to find the relationships between load current injections and node voltages. If loads are represented by second-order polynomial forms, the load current injections are found to be 9 Pið0Þ aP Vxi þ Qið0Þ aQ Vyi Pið0Þ bP Vxi þ Qið0Þ bQ Vyi Pið0Þ cP Vxi þ Qið0Þ cQ Vyi > > q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ > 2 > Við0Þ Vxi2 þ Vyi2 > = Við0Þ Vxi2 þ Vyi2 ; Qið0Þ aQ Vxi Pið0Þ aP Vyi Qið0Þ bQ Vxi Pið0Þ bP Vyi Qið0Þ cQ Vxi Pið0Þ cP Vyi > > > q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ Iyi ¼ þ þ > 2 > Við0Þ Vxi2 þ Vyi2 ; Við0Þ Vxi2 þ Vyi2 Ixi ¼

ð7:48Þ where the terms proportional to the square of voltages can be incorporated into the network admittance matrix as constant admittances, thus (7.48) is reduced to the last two terms only. If loads are represented by exponential functions, the load current injections are found to be 9 Pið0Þ Vim2 Vxi Qið0Þ Vin2 Vyi > > > Ixi ¼ > m n = Vð0Þ Vð0Þ : Qið0Þ Vin2 Vxi Pið0Þ Vim2 Vyi > > > Iyi ¼ > ; n m Vð0Þ Vð0Þ

7.3.1.3

ð7:49Þ

The Relationship Between FACTS Devices and the Network

Here we will only describe the relationship between SVC/TCSC and the network; also the relationship between the other FACTS devices and the network can be derived following the same concept. 1. SVC: In general an SVC is connected to a high-voltage bus of the network through a transformer (let the index of this bus be i). Thus the shunt susceptance of the device is equal to j

BSVC : 1 XT BSVC

7.3 Network Mathematical Model for Transient Stability Analysis

437

From the relationship between nodal voltage V_i and current injection I_i it is not difficult to find the real and imaginary parts of the current injection as follows: Ixi ¼

9 > > = ; > > Vxi ;

BSVC Vyi 1 XT BSVC

Iyi ¼

BSVC 1 XT BSVC

ð7:50Þ

where XT is the impedance of the transformer, BSVC is the equivalent susceptance of the SVC, Vxi and Vyi are the real and imaginary parts of the voltage of the high-voltage bus. 2. TCSC: Regardless of the place where the TCSC is connected in series in a line, it is always possible to put two nodes around the TCSC, let the nodes be i and j. As a matter of the fact, the role a TCSC plays is equivalent to two current sources having the same magnitude but opposite directions at node i and j, the current injections are easily derived as Ixi ¼ Ixj ¼ BTCSC ðVyi Vyj Þ Iyj ¼ Iyi ¼ BTCSC ðVxi Vxj Þ

) ;

ð7:51Þ

where BTCSC is the equivalent susceptance of the TCSC, Vxi, Vyi, Vxj, and Vyj are the real and imaginary parts of the voltages of the two nodes.

7.3.1.4

The Relationship Between Two-Terminal HVDC and the Network

Let variables with subscript ‘‘d’’ denote quantities on the DC side, and subscripts ‘‘R’’ and ‘‘I’’ denote rectifier and inverter sides (they have the same meaning in subsequent text), respectively. From (5.52)–(5.54) and (5.57) (where kg 1), the steady-state equations of the rectifier are as follows: VdR ¼ kR VR cos a XcR IdR VdR ¼ kR VR cos ’R IR ¼ kR IdR PR ¼ VdR IdR

9 > > > > > > > > > > > =

> > > pﬃﬃﬃ > > ¼ 3VR IR cos ’R > > > > > > ;

QR ¼ PR tg ’R

ð7:52Þ

438

7 Power System Transient Stability Analysis

and the inverter side steady-state equations: 9 > > > > > > ¼ kI VI cos ’I > > = ¼ kI IdI : > > pﬃﬃﬃ > > ¼ VdI IdI ¼ 3VI II cos ’I > > > > ; ¼ PI tg ’I

VdI ¼ kI VI cos b þ XcI IdI VdI II PI QI

ð7:53Þ

Based on (7.52) and (7.53), the power injections into the AC system by the DC system can denoted by functions of Id, a, b, VxR, VyR, VxI, and VyI. The power injection into the AC bus from the rectifier is given by 9 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 > 2 þ V 2 cos a > PR ¼ PR ¼ VdR IdR ¼ XcR IdR kR IdR VxR > yR > > > pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 2 = kR VR VdR 2 2 2 QR ¼ QR ¼ PR ¼ IdR kR VR VdR VdR > > rﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > 2 2 2 2 2 2 2 2 > ; ¼ IdR kR ðVxR þ VyR Þ sin a þ 2kR XcR IdR VxR þ VyR cos a XcR IdR > ð7:54Þ and the power injection from the inverter is 9 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > 2 2 > PI ¼ PI ¼ VdI IdI ¼ þ kI IdI VxI þ VyI cos b > > > > pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ 2 = kI2 VI2 VdI 2 2 2 QI ¼ QI ¼ PI ¼ IdI kI VI VdI : VdI > rﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ> > > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > 2 2 > ; ¼ IdI kI2 ðVxI2 þ VyI2 Þ sin2 b 2kI XcI IdI VxI2 þ VyI2 cos b XcI IdI > 2 XcI IdI

ð7:55Þ

Thus the current injections into the AC system from the rectifier and inverter are obtained as IxR

PR VxR þ QR VyR ¼ ; 2 þ V2 VxR yR

PI VxI þ QI VyI IxI ¼ ; 2 VxI2 þ VyI

9 PR VyR QR VxR > > > IyR ¼ > 2 þ V2 = VxR yR : > PI VyI QI VxI > > > IyI ¼ ; 2 VxI2 þ VyI

ð7:56Þ

Substituting (7.54) and (7.55) into (7.56), and eliminating variables PR , QR , PI , and QI , the current injections IxR and IyR become functions of IdR, a, VxR and VyR, and IxI and IyI are functions of variables IdI, , bVxI, and VyI.

7.3 Network Mathematical Model for Transient Stability Analysis

7.3.2

439

Modeling Network Switching and Faults

When a fault or switch change is applied to a network, the network admittance matrix needs to be correctly modified. If the fault or switch is three-phase symmetrical, for example, a three-phase short circuit, the removal of three phases of a device, the forced connection of a series capacitor, the introduction or removal of a braking resistor, etc., the modification to the admittance matrix is straightforward because such a fault or switching operation results in a parameter change in a shunt branch or series branch of the network. Most of short circuits and device removals are unsymmetrical, and thus have to be analyzed using a symmetrical components method. In addition to dealing with the positive sequence network of the power system under study, one has to consider the negative sequence and zero sequence networks. On the other hand, in stability studies we are mostly interested in the quantities of the positive sequence network, paying little if any attention to the quantities of the negative and zero sequence networks. The effects of the negative and zero sequence networks can be modeled using an equivalent impedance viewed from the positive sequence network. When analyzing unsymmetrical problem using the concept of symmetrical components, phase A is often taken as the reference, the boundary conditions of various types of short circuit or open-conductor are expressed in terms of the sequence quantities of phase A. When a short circuit or an open-conductor occurs, the phase that exhibits different behavior compared with the other two phases is called the special phase. For instances, the special phase in a single-line-to-ground fault is the phase connected to ground; the special phase in a double-line-to-ground or line-to-line fault is the phase that is not faulted. The special phase in a singleline-open-conductor is the phase that is open, and the special phase in a double-lineopen-conductor is the phase that is intact. When the special phase of a short circuit or open-conductor is phase A, the three sequence networks can be directly connected to form the so-called composite sequence network according to certain boundary conditions. This is equivalent to connecting supplementary impedance to the faulted terminals of the positive sequence network. The size of the supplementary impedance depends on the type of fault, as illustrated in Tables 7.2 and 7.3. Here the ‘‘faulted terminals’’ mean, in a short circuit the terminals between faulted bus and ground and in an open-conductor fault the two nodes resulting from the open-conductor. The network admittance matrices under these circumstances can be easily formed. Table 7.2 Supplemental impedances of short circuits Type of short circuit Supplemental impedance ð2Þ ð0Þ Single-line-to-ground Z þZ Double-line-to-ground Line-to-line

S S ð2Þ ð0Þ

ð2Þ

ð0Þ

ZS ZS =ðZS þ ZS Þ ð2Þ ZS

ð2Þ

ZS is the self-impedance of the short circuit in negative sequence network, ð0Þ

ZS is the self-impedance of the short circuit in zero sequence network

440

7 Power System Transient Stability Analysis Table 7.3 Supplemental impedances of open-conductor Type of open-conductor Supplemental impedance Single-line-open-conductor Z(2)Z(0)/(Z(2) þ Z(0)) Double-line-open-conductor Z(2) þ Z(0) (2) Z is the equivalent impedance of the open-conductor terminals in the negative sequence network, Z(0) is the equivalent impedance of the openconductor terminals in the zero sequence network Table 7.4 The ratios of ideal transformers Special phase Sequence Zero Positive Negative A 1 1 1 B 1 a2 a C 1 a a2

When the special phase is not phase A, there is a complex operator a ¼ ej120 in the boundary conditions, therefore the three sequence networks cannot be directly connected to form a combined sequence network. However, we can connect the three sequence networks via three ideal transformers with ratios 1:n0 (0), 1:n0 (1), and 1:n0 (2) in the zero, positive, and negative sequence networks. The two sides of these transformers have the same voltage/current ratios thus the transformers introduce no losses. For different special phases, these transformers have different ratios in different sequence networks, as described in Table 7.4. After introducing ideal transformers, the various types of unsymmetrical short circuit and open-conductor can be classified into two categories: series and shunt (or parallel) faults based on the topology of the three sequence networks. The faults belonging to the series category include single-line-to-ground, double-line-openconductor, and single-line-to-ground of a series capacitor. The boundary conditions of these faults are as follows: the sum of three sequence voltages is zero, and the sequence currents are identical in the nonstandard ratio side of the transformer. The faults belonging to the shunt category include double-line-to-ground, single-lineopen-conductor, and double-line short circuit of capacitors. The boundary conditions of this class of faults are as follows: in the nonstandard ratio side of the transformer, the sum of sequence currents is equal to zero, and the sequence voltages are equal. When simultaneous short circuits or open-conductors occur, and they occur in different phases, the method for handling single faults can still be applied to modify the admittance matrix of the positive sequence network, but now the concept of supplementary impedance is generalized to that of synthesized impedance matrix. In what follows, we introduce the basic concept of synthesized impedance matrix using single-line-to-ground and single-line-open-conductor faults as examples. Suppose a single-line-to-ground fault occurs at bus k (let this be fault 1), and a single-line-open-conductor occurs between buses i and j (let this be fault 2), and the two faults occur in different phases. By the boundary conditions of the three sequence components at the place where a fault occurs, the combined sequence

7.3 Network Mathematical Model for Transient Stability Analysis

441

•

n 1′ (1 ) : 1

V1(1 )

k

i

o

j

1 : n ′2 (1)

I 2(1 ) •

I 1(1 ) •

k

i

o

j

Positive sequence

•

V2(1 )

Positive sequence

n 1′ ( 2 ) : 1

k

i

o

j

k

i

o

j

Negative sequence

zero sequence

a

I 2( 2 ) •

•

I1( 2 )

1 : n 2′ ( 2 )

•

V1 ( 2 )

k

i

o

j

Negative sequence

n1( 2 ) : 1

•

V2( 2 ) 1 : n 2( 2 )

•

•

I1( 0 )

I 2( 0 )

•

V1( 0 )

i

o

j

0ð1Þ

0ð1Þ

Zero sequence

n1( 0 ) : 1

b

k

•

V2( 0 ) 1 : n 2( 0 )

Fig. 7.11 Combined sequences of two simultaneous faults 0ð2Þ

0ð2Þ

network can be obtained as in Fig. 7.11a. In the figure, n1 , n2 , n1 , and n2 are the ratios of ideal transformers, the specific values of them depending on the special phases. For ease of mathematical manipulation, let us recast the combined sequence network in Fig. 7.11a as that in Fig. 7.11b. It is not difficult to see that the ratios in the two figures obey the following relationships: ð2Þ

0ð2Þ

0ð1Þ

n1 ¼ n1 =n1 ;

ð2Þ

0ð2Þ

0ð1Þ

n2 ¼ n2 =n2 ;

ð0Þ

0ð1Þ

n1 ¼ 1=n1 ;

ð0Þ

0ð1Þ

n2 ¼ 1=n2 :

In the following, we derive the impedance matrix Zf viewed from the fault buses of the positive sequence network into the negative and zero sequence network based on the combined sequence network. We call Zf the synthesized impedance matrix of simultaneous faults. In Fig. 7.11b, the single-line-to-ground part on the left forms a loop circuit, let ð1Þ the loop current be I_ , and the single-line-open-conductor part on the right forms 1

ð1Þ ð0Þ two independent loop circuits, let the currents in these circuits be I_2 and I_2 . ð2Þ ð2Þ ð0Þ ð0Þ Therefore the currents I_1 , I_2 , I_1 , and I_2 of the faulted buses in the negative and zero sequence networks can be obtained in terms of these loop currents as follows:

IS ¼ CIL ;

ð7:57Þ

where C the coincidence matrix is dependent on fault conditions. The definitions of the symbols are 2

3 ð2Þ I_1 6 _ð2Þ 7 6I 7 IS ¼ 6 2ð0Þ 7; 4 I_1 5 ð0Þ I_ 2

2

3 ð1Þ I_1 6 7 IL ¼ 4 I_2ð1Þ 5; ð0Þ I_ 2

2

3 1 0 0 6 0 1 1 7 7: C¼6 41 0 0 5 0 0 1

ð7:58Þ

442

7 Power System Transient Stability Analysis

Based on loop voltage equations, the relationship among the voltages of the faulted buses in each sequence can be obtained as VL ¼ CT VS ;

ð7:59Þ

where CT is the transpose of matrix C and 2

3 ð1Þ V_ ok 6 7 VL ¼ 4 V_ ð1Þ 5; ji 0

2

ð2Þ 3 V_1 6 _ ð2Þ 7 6V 7 VS ¼ 6 2ð0Þ 7: 4 V_1 5 ð0Þ V_

ð7:60Þ

2

From the transformer nonstandard ratio side point of view, the relationship among the currents and voltages of negative and zero sequence networks is expressed as "

"

ð2Þ V_ 1 ð2Þ V_ 2 ð0Þ V_ 1 ð0Þ V_ 2

#

" ¼

#

" ¼

Z11 ð2Þ Z21

ð2Þ

Z12 ð2Þ Z22

ð0Þ

Z12 ð0Þ Z22

Z11 ð0Þ Z21

ð2Þ

ð0Þ

#"

#"

# ð2Þ I_1 ð2Þ ; I_2

ð7:61Þ

# ð0Þ I_k ð0Þ : I_i

ð7:62Þ

Because of the existence of ideal transformers in negative and zero sequence networks, the impedance matrices in (7.61) and (7.62) are unsymmetrical, in general. The computation of the elements of these matrices will be introduced in subsequent sections. Let us incorporate (7.61) and (7.62) into a single equation: 2

3 2 ð2Þ ð2Þ V_ 1 Z11 6 _ ð2Þ 7 6 ð2Þ V 6 2 7 6 Z21 6 ð0Þ 7 ¼ 6 4 V_ 1 5 4 0 ð0Þ V_ 0 2

ð2Þ

Z12 ð2Þ Z22 0 0

0 0 ð0Þ Z11 ð0Þ Z21

32 ð2Þ 3 0 I_1 76 _ð2Þ 7 0 76 I2 7 ð0Þ 76 ð0Þ 7 Z12 54 I_1 5 ð0Þ ð0Þ Z22 I_2

ð7:63Þ

or in compact form: VS ¼ ZIS :

ð7:64Þ

Making use of the matrix Z and coincidence matrix C, the relationship between positive sequence voltage and current can be derived. To this end, substituting (7.64) and (7.57) into (7.59), the relationship between faulted loop voltage and current is found to be VL ¼ ZL IL ;

ð7:65Þ

7.3 Network Mathematical Model for Transient Stability Analysis

443

where ZL is termed the loop impedance matrix, defined by ZL ¼ CT ZC:

ð7:66Þ

In this example, 2 ð2Þ 3 Z11 1 0 1 0 6 ð2Þ 6Z ZL ¼ 4 0 1 0 0 56 21 4 0 0 1 0 1 0 2 0 3 0 0 Z11 Z12 Z13 0 0 0 5 4 ¼ Z21 Z22 Z23 : 0 0 0 Z31 Z32 Z33

ð2Þ

Z12 ð2Þ Z22 0 0

2

0 0 ð0Þ Z11 ð0Þ Z21

32 0 1 76 0 76 0 ð0Þ 76 Z12 54 1 ð0Þ 0 Z22

0 1 0 0

3 0 7 1 7 7 05 1 ð7:67Þ

ð0Þ

Eliminating current I_2 in (7.65), it follows: "

ð1Þ V_ ok ð1Þ V_ ji

#

" ¼

Z11

Z12

Z21

Z22

#"

# ð1Þ I_1 ð1Þ ; I_

ð7:68Þ

2

where the elements Zmn (m and n can be equal to 1 or 2) of the impedance matrix are computed based on: 0 Zmn ¼ Zmn

0 0 Zm3 Z3n ; 0 Z33

ð7:69Þ

(7.68) is rewritten in compact form as Vf ¼ Z f I f :

ð7:70Þ

Finally, the impedance matrix Zf, viewed from the faulted buses of the positive sequence network into the negative and zero sequence networks, is obtained. Equation (7.70) can also be expressed in the form of synthesized admittance matrix as follows: I f ¼ Yf Vf ;

ð7:71Þ

where Yf ¼ Z1 f . Once Yf is determined, the elements of the matrix can be incorporated into the correct position of the admittance matrix of the positive sequence network. In this example, notice that ð1Þ ð1Þ V_ok ¼ V_k ;

ð1Þ ð1Þ ð1Þ V_ji ¼ V_ j V_i ;

ð1Þ ð1Þ I_k ¼ I_1 ;

ð1Þ ð1Þ I_i ¼ I_2 ;

ð1Þ ð1Þ I_j ¼ I_2 :

ð7:72Þ

444

7 Power System Transient Stability Analysis

The above relationships together with (7.71) give us the relationship among the voltages and currents at node k, i and j in the positive sequence network: 2

3 2 ð1Þ Y11 I_k 6 _ð1Þ 7 6 Y 4 Ii 5 ¼ 4 21 ð1Þ Y21 I_

Y12 Y22 Y22

j

32 ð1Þ 3 Y12 V_ k 7 6 Y22 54 V_ ið1Þ 7 5: ð1Þ _ Y22 Vj

ð7:73Þ

In summary, the calculation of synthesized impedance matrix includes the following steps: 1. Form the impedance matrix of the faulted buses of the negative and zero sequence network (refer to (7.63)) 2. By use of the coincidence matrix that represents the boundary conditions of simultaneous faults, form the loop impedance matrix ZL (refer to (7.66) and (7.67)) 3. Eliminate the closed circuit from the synthesized impedance matrix Zf (refer to (7.68) and (7.69)) In what follows we describe the above steps in detail. (1) Forming the impedance matrices of the faulted buses of the negative and zero sequence networks: In a transient stability study, the admittance matrices of each sequence network should be formed first, followed by calculation of the triangular factors for these matrices. In this way the impedance matrices of the faulted buses of each sequence network can be easily obtained given the fault information. For the negative sequence network, observing Fig. 7.11b it is not difficult to see that, if one injects unity current into the nonstandard ratio node k of the ideal ð2Þ transformer with zero current injections to the other nodes, that is, I_k ¼ 1 and ð2Þ I_m ¼ 0 (m is a node other than node k), then solve the equation of the negative ð2Þ sequence network including the ideal transformer for voltages V_ and k

ð2Þ ð2Þ ð2Þ V_ ij ¼ V_ i V_ j . These quantities are the desired quantities for the first ð2Þ

ð2Þ

column Z11 and Z21 of the impedance matrix in (7.71). More specifically, injecting unity current into node k of the nonstandard transformer is equivalent to injecting into node k of negative sequence network 0ð2Þ ð2Þ a current I_k ¼ n^1 , thus after performing sparse forward substitution and backward substitution on the admittance matrix of the negative sequence 0ð2Þ 0ð2Þ 0ð2Þ 0ð2Þ network, voltages V_ k and V_ ij ¼ V_ i V_ j are obtained; in addition, we ð2Þ ð2Þ 0ð2Þ ð2Þ ð2Þ 0ð2Þ have V_ ¼ n V_ and V_ ¼ n V_ . k

1

k

ij

2

ij

By the same token, injecting into nodes i and j of the negative sequence ð2Þ ð2Þ network currents þ n^2 and n^2 , and performing sparse forward substitution

7.3 Network Mathematical Model for Transient Stability Analysis

445

0ð2Þ 0ð2Þ 0ð2Þ 0ð2Þ and backward substitution, one obtains voltages V_ k and V_ij z ¼ V_ i V_j , ð2Þ ð2Þ 0ð2Þ ð2Þ ð2Þ 0ð2Þ furthermore, V_ ¼ n V_ , V_ij ¼ n V_ ij , the quantities which we seek for k ð2Þ

1

k ð2Þ

2

the elements Z12 and Z22 of the second column in the impedance matrix in (7.61). The same principle applies to compute the elements of the impedance matrix in (7.62). (2) Forming loop impedance matrix from the coincidence matrix: As discussed before, the combined sequence network of a series fault is formed by putting the three sequence networks together in series, therefore there is only one independent loop circuit. The combined sequence network of a shunt fault is formed by putting together the three sequence networks in parallel, resulting in two independent loop circuits. Besides, a line-to-line fault is viewed as a special shunt fault. From (7.57) and (7.59), the coincidence matrix C expresses the relationship between the loop current of the boundary circuit of the combined sequence network and the current of the faulted buses of the negative and zero sequence networks. Thus the number of rows of the coincidence matrix equals the dimension of IS, that is, two times the number of simultaneous faults (when a line-to-line fault occurs, an empty faulted bus in the zero sequence network is designated). The number of columns of the coincidence matrix equals the dimension of IL. A series fault occupies one column in the coincidence matrix as illustrated below: ½ 0 0 1 0 0 0 0 1 0 . . . 0 T ; |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} corresponding to the current of the faulted bus in negative sequence network

corresponding to the current of the faulted bus in zero sequence network

where the column number of the nonzero is equal to the index number of the fault among all faults. A shunt fault occupies two columns in the coincidence matrix as illustrated below:

T 0 0 1 0 0 0 0 ; 0 0 1 0 0 0 0 1 0 0 |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} corresponding to the current of the faulted bus in negative sequence network

corresponding to the current of the faulted bus in zero sequence network

where the first column contains the information on how the negative sequence network is connected with the positive sequence network, while the second column describe the connectivity between the zero sequence and negative sequence networks. The column number of the nonzero corresponds to the index number of the fault among all faults. For a line-to-line fault, the coincidence matrix has only the first column because there is no circuit connection between negative and zero sequence networks.

446

7 Power System Transient Stability Analysis

Based on the above principle, we can easily find the coincidence matrix that represents the boundary conditions of arbitrarily complex simultaneous faults given the types of the faults. For example, if three faults simultaneously occur, and the faults are, in order, single-line-to-ground, single-line-open-conductor, and line-to-line, then the coincidence matrix is as follows: 3 9 1 0 0 0 > = 7 6 0 1 1 0 6 7 negative sequence part 6 0 7 > 0 0 1 6 7 ; 6 1 7 9 0 0 0 C¼6 : 7 6 0 0 1 0 7 = 6 7 > 6 0 7 0 0 0 5 zero sequence part 4 |{z} |ﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄ} |{z} > ; singleline singleline linetoline 2

toground openconductor

With the coincidence matrix that describes the boundary conditions of complex simultaneous faults, the loop impedance matrix of the combined sequence network can be obtained using (7.66) and (7.67). The manipulations on these matrices can be accomplished by simple addition and subtraction operations. (3) Eliminating the closed circuit of shunt faults to form synthesized impedance matrix: The order of the loop impedance matrix equals the number of independent loop currents in the combined sequence network. To seek the synthesized impedance matrix, the currents of zero and negative sequence network must be eliminated (refer to (7.68) and (7.69)).

7.4

Transient Stability Analysis with Simplified Model

For a regional power system, the duration of losing synchronous stability is very short, typically a simulation study of the first swing (1–1.5 s) after a disturbance is applied suffices to judge whether or not the system can maintain synchronous operation. In stability studies like this, the effects of speed-governing systems can be neglected, thus the output of prime movers can be assumed to be constant, the reason is that the inertias of the prime movers are sufficient to keep the outputs of the prime movers constant; besides, because the time constants of the excitor windings are relatively large, their flux linkages do not change drastically in a short range of time, as a result the effect of the excitation system can be modeled as keeping generator transient voltages Eq0 or E0 constant. In other words, the free current components of excitor windings are compensated by the regulation of excitation systems, thus the flux linkages cf of excitor windings remain constant. Correspondingly, the effects of damper windings are also ignored. The simplified models for transient stability analysis are widely used in power system operation and planning. Specific applications include feasibility studies on

7.4 Transient Stability Analysis with Simplified Model

447

system topologies and operating schedules, computation of maximum transfer capabilities, calculation of critical clearing times, and investigations into the effects of stability controls, etc. Using different models for generators, loads, and network, one can build codes for various simplified stability analyses. Which portfolio of models to use depends on the fundamental characteristics of the problem under study. To explain the principles and procedures of simplified transient stability analysis, the subsequent sections assume the following mathematical models and solution algorithms have been applied to the transient stability analysis procedure: Generators: Generator transient voltage Eq0 remains constant Loads: Small loads are modeled as constant impedances, while larger loads are modeled as motors with mechanical–electrical interactions Network: Modeled with admittance matrix The differential equations are solved by the modified Euler’s method while the network equations are solved by Gauss elimination method. The overall procedure for a transient stability calculation is still as described in Fig. 7.8. The computer implementation of the calculation is provided below.

7.4.1

Computing Initial Values

Before starting the numerical integration, the initial values of the differential equations should be calculated based on the prefault operating state obtained by performing a load flow study. In a simplified transient stability study, the calculation of initial values include prefault generator transient voltages, rotor angles, the output of prime movers, and the slips and equivalent admittances of motors representing loads, etc. These parameters do not change discontinuously at the instant immediately after the fault is applied. In what follows the initial value variables are marked with subscripts (0). First we describe how to calculate the initial values of generators. From a load flow study the generator terminal voltages before the disturbance and the generator powers are given by V_ ð0Þ ¼ Vxð0Þ þ jVyð0Þ and S(0) ¼ P(0) + jQ(0). Furthermore, the generator currents injected into the network are computed by I_ð0Þ ¼ Ixð0Þ þ jIyð0Þ ¼

S^ð0Þ : ^_ V

ð7:74Þ

ð0Þ

Thus by (6.61), one can find the pseudovoltage E_ Qð0Þ as E_ Qð0Þ ¼ EQxð0Þ þ jEQyð0Þ ¼ V_ ð0Þ þ ðRa þ jXq ÞI_ð0Þ :

ð7:75Þ

Subsequently, the generator rotor angles are calculated by dð0Þ ¼ arctgðEQyð0Þ =EQxð0Þ Þ:

ð7:76Þ

448

7 Power System Transient Stability Analysis

Under steady-state operation, generators rotate at synchronous speed, therefore: oð0Þ ¼ 1:

ð7:77Þ

Using coordinate transformation formula (6.62), the d, q components of generator stator voltages and currents are given by Vdð0Þ sin dð0Þ cos dð0Þ Vxð0Þ Idð0Þ ¼ Vqð0Þ cos dð0Þ sin dð0Þ Vyð0Þ Iqð0Þ sin dð0Þ cos dð0Þ Ixð0Þ ¼ : ð7:78Þ cos dð0Þ sin dð0Þ Iyð0Þ Now based on (6.64), the values of transient voltages are obtained as E0qð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd0 Idð0Þ :

ð7:79Þ

In addition, the electrical powers Pe(0) of generators under steady-state operation are equal to the mechanical powers of the prime movers Pm(0), that is, 2 2 Pmð0Þ ¼ Peð0Þ ¼ Pð0Þ þ ðIxð0Þ þ Iyð0Þ ÞRa :

ð7:80Þ

The calculation of the initial values of loads is simple. The prefault node voltages V_ ð0Þ and powers S(0) consumed by loads are obtained from a load flow study, therefore the equivalent admittances of loads are computed by Yð0Þ ¼

S^ð0Þ : 2 Vð0Þ

ð7:81Þ

When loads are modeled as constant impedances, the corresponding equivalent admittances remain constant in the study period, and thus can be incorporated into the network admittance matrix as discussed earlier. For loads representing motors with mechanical–electrical interactions, since the slips of motors do not jump at the instant of disturbance, the equivalent admittances of loads do not change. In other words, the equivalent admittances of loads after the disturbance are identical to those of loads under normal steady-state operation.

7.4.2

Solving Network Equations with Gauss Elimination Method

In this solution method, the network equations are represented in the domain of real numbers, as in (7.36). Before starting the simulation, the loads represented by constant impedances should be incorporated into the network to obtain the network with constant impedance loads, this set of network equations remains constant during the simulation period.

7.4 Transient Stability Analysis with Simplified Model

449

Suppose a motor load is connected at node j. In the transient period the motor slip sj is time varying, and given the sj at an instant, the actual impedance of the motor load can be calculated based on (6.160):

ZMjð0Þ ðRm þ jXm ÞðR2 =sj þ jX2 Þ ZMj ¼ R1 þ jX1 þ ; ð7:82Þ ðRm þ jXm Þ þ ðR2 =sj þ jX2 Þ ZMð0Þ where ZMj(0) and ZM(0) are the equivalent impedance of all the motors under normal operation and the equivalent impedance of a typical motor. The admittance associated with the actual impedance can be rewritten as YMj ¼

1 ¼ GMj þ jBMj : ZMj

ð7:83Þ

Now suppose a generator is located at node i of the network. When the generator is represented by a varying E0 q model, with reference to Table 7.1, in (7.39) let 0 Edi ¼ 0, Eqi ¼ E0qi , Xdi ¼ Xdi , and Xqi ¼ Xqi , the formula for generator current is as follows:

Ixi Iyi

bxi 0 Gxi ¼ E gyi qi Byi

Bxi Gyi

Vxi ; Vyi

ð7:84Þ

where the elements can be rewritten, based (7.40), as follows: Rai cos di þ Xqi sin di ; 0 R2ai þ Xdi Xqi 0 Rai ðXdi Xqi Þ sin di cos di Gxi ¼ ; 0 R2ai þ Xdi Xqi bxi ¼

Byi ¼

0 Xdi sin2 di Xqi cos2 di ; 0 X R2ai þ Xdi qi

9 > > > > > > > > = 2 0 2 Xdi cos di þ Xqi sin di : Bxi ¼ 0 X > R2ai þ Xdi qi > > > 0 > Rai þ ðXdi Xqi Þ sin di cos di > > > Gyi ¼ ; 0 X R2ai þ Xdi qi gyi ¼

Rai sin di Xqi cos di 0 R2ai þ Xdi Xqi

ð7:85Þ Substitute the generator current representations (7.84) into the network equations with constant impedance loads, and do the same for the equivalent admittance [(7.83)] of motors, we obtain the new set of network equations. Obviously, the new network equations are just modifications of the original network equations: the diagonal elements of the admittance matrix are modified, and there are nonzero pseudocurrents in elements of the current vector associated with generators, the current injections of other nodes are zero, that is: The ith diagonal block of the admittance matrix is changed to

Gxi þ Gii Byi þ Bii

Bxi Bii Gyi þ Gii

ð7:86Þ

450

7 Power System Transient Stability Analysis

and the jth diagonal block changes to

GMj þ Gjj BMj þ Bjj

BMj Bjj : GMj þ Gjj

ð7:87Þ

The pseudocurrent injections at generator nodes are given by

Ixi0 Iyi0

¼

bxi 0 E : gyi qi

ð7:88Þ

Now the linear equations obtained in each integration step can be solved by Gauss elimination or the triangular factorization method. This gives us the real and imaginary part Vx and Vy of the network voltages for this step. Finally, based on (7.84), the generator currents Ix and Iy can be found.

7.4.3

Solving Differential Equations by Modified Euler’s Method

In a transient stability analysis using simplified models, the differential equations comprise the motion (6.76) of generator rotors and the motion (6.155) of motor rotors representing loads: 9 ddi > > ¼ os ðoi 1Þ > > dt > > = doi 1 ¼ ðPmi Pei Þ : dt TJi > > > > > dsj 1 ; ¼ ðMmMj MeMj Þ > dt TJMi

ð7:89Þ

Suppose the simulation of mechanical–electrical interactions has been completed up to time t, now let us discuss how to calculate the system states for time t + Dt. Before calculating system states for the next step, whether or not there is a fault or switch operation at time t should be checked first. If the answer is no, then one proceeds to compute the states of the next step, given the states of time t; otherwise, one has to calculate the postswitch or postfault network operating parameters first, and then continue the calculation for the next step. The computational procedure for solving differential equations based on the modified Euler’s method is as follows: (1) Given generator di(t) and motor sj(t) at time t, compute system voltages Vx(t) and Vy(t), and generator currents Ixi(t) and Iyi(t) based on the method described in Sect. 7.4.2.

7.4 Transient Stability Analysis with Simplified Model

451

(2) Based on (7.89), compute the derivatives for time t: 9 ddi > > ¼ o ðo 1Þ s iðtÞ > > dt t > > > = doi 1 ¼ ðP P Þ ; mi eiðtÞ > dt t TJi > > > > > dsj 1 > ; ¼ ðM M Þ mMjðtÞ eMjðtÞ dt t TJMi

ð7:90Þ

where generator power Pei(t) is calculated by 2 2 PeiðtÞ ¼ ðVixðtÞ IixðtÞ þ ViyðtÞ IiyðtÞ Þ þ ðIixðtÞ þ IiyðtÞ ÞRai :

ð7:91Þ

The mechanical torque TmMj(t) of generators and electrical torque TeMj(t) of motors are computed based on (6.157) and (6.156) as follows: 9 MmMjðtÞ ¼ k½a þ ð1 aÞð1 sjðtÞ Þ2 > > > = 2 2 2MeM max VjxðtÞ þ VjyðtÞ ; MeMjðtÞ ¼ s > scrj V 2 þ V 2 jðtÞ > > jxð0Þ jyð0Þ ; þ scrj sjðtÞ

ð7:92Þ

in which Vjx(0) and Vjy(0) denote the real and imaginary parts of prefault node voltage at node j. (3) Compute an initial estimate of state variables for time t + Dt: ½0

9 ddi > > Dt > > dt t > > > doi = ¼ oiðtÞ þ Dt : dt t > > > > > > dsj ; ¼ sjðtÞ þ Dt > dt t

diðtþDtÞ ¼ diðtÞ þ ½0

oiðtþDtÞ ½0

sjðtþDtÞ

½0

ð7:93Þ

½0

(4) Similar to step (1), given generator diðtþDtÞ and motor sjðtþDtÞ , compute system ½0

½0

½0

½0

node voltages VxðtþDtÞ and VyðtþDtÞ , generator currents IxiðtþDtÞ and IyiðtþDtÞ based on the method of Sect. 7.4.2. (5) Similar to step compute the estimated derivatives (2), ½0 ½0 ½0 dsj ddi doi ; dt ; and dt for step t + Dt. To this end, one should replace dt tþDt

tþDt

tþDt

½0

½0

½0

oi(t), Pei(t), MmMj(t), and MeMj(t) in (7.92) with oiðtþDtÞ , PeiðtþDtÞ , MmMjðtþDtÞ , ½0

and MeMjðtþDtÞ . To compute them, one should also replace Vix(t), Viy(t), Iix(t),

452

7 Power System Transient Stability Analysis ½0

½0

½0

½0

Iiy(t), sj(t), Vjx(t), and Vjy(t) with VixðtþDtÞ , ViyðtþDtÞ , IixðtþDtÞ , Iiy(t), sj(t), VjxðtþDtÞ , ½0

and VjyðtþDtÞ . (6) Finally, compute the variable values for step t + Dt, that is: # 9 " ½0 > Dt ddi ddi > > diðtþDtÞ ¼ diðtÞ þ þ > > 2 dt t dt > > tþDt > #> " > > ½0 = Dt doi doi oiðtþDtÞ ¼ oiðtÞ þ þ : 2 dt t dt > tþDt > > > # " > > ½0 > > Dt dsj dsj > > sjðtþDtÞ ¼ sjðtÞ þ þ > ; 2 dt t dt

ð7:94Þ

tþDt

[Example 7.3] Consider the 9-bus system in Fig. 7.12 [178]. This system consists of three generators, three loads, and nine branches. The generator and branch parameters are listed in Tables 7.5 and 7.6, respectively. The system load flow under normal operation is illustrated in Table 7.7, and the system frequency is 60 Hz. [Solution] A stability analysis based on the simplified system model will be described below. The disturbances are as follows: at time zero a three-line-toground fault occurs in line 5–7 at the node 7 side, the fault is cleared five cycles (about 0.08333 s) later by the removal of line 5–7. Generators are modeled as constant Eq0 , loads are modeled as impedances, the network is modeled by admittance matrix, the differential equations are solved by the modified Euler’s method, and the network equations are solved by a direct method.

18kV

230kV

13.8kV

2

3 2

8

3

7

9 6

5

4

16.5kV

1 1

Fig. 7.12 Single-line diagram of 9-bus system

7.4 Transient Stability Analysis with Simplified Model Table 7.5 Branch data From-end To-end bus bus 4 4 5 6 7 8 1 2 3

5 6 7 9 8 9 4 7 9

Resistance (in per unit) 0.010 0.017 0.032 0.039 0.0085 0.0119 0.0 0.0 0.0

Reactance (in per unit) 0.085 0.092 0.161 0.170 0.072 0.1008 0.0576 0.0625 0.0586

453

Half of the admittance (in per unit) 0.088 0.079 0.153 0.179 0.0745 0.1045

Non-standard ratio of transformer

1.0 1.0 1.0

Table 7.6 Generator data 0 0 Generator Bus TJ Ra Xd Xd0 Xq Xq0 Td0 Tq0 D 1 1 47.28 0.0 0.1460 0.0608 0.0969 0.0969 8.96 0.0 2 2 12.80 0.0 0.8958 0.1198 0.8645 0.1969 6.00 0.535 0.0 3 3 6.02 0.0 1.3125 0.1813 1.2578 0.2500 5.89 0.600 0.0 The units for all time constants are ‘‘seconds,’’ the units of all damping coefficients D, resistances and impedances are in ‘‘per unit’’ Table 7.7 Load flow under normal system operation Bus Voltage Generator Magnitude Phase angle Active Reactive (degree) power power 1 1.040 0.0000 0.7164 0.2705 2 1.0250 9.2800 1.6300 0.0665 3 1.0250 4.6648 0.8500 0.1086 4 1.0258 2.2168 5 0.9956 3.9888 6 1.0127 3.6874 7 1.0258 3.7197 8 1.0159 0.7275 9 1.0324 1.9667

Load Active power

Reactive power

1.2500 0.9000

0.5000 0.3000

1.0000

0.3500

Based on the general procedure described in Fig. 7.8 and the method described in the previous section, the transient stability analysis can be summarized below: 1. Initial value computation: Compute the equivalent shunt admittances of loads according to (7.81), and the results are as follows: Load (node 5): 1.26099 j0.50440 Load (node 6): 0.87765 j0.29255 Load (node 8): 0.96898 j0.33914 Then compute, based on (7.74)–(7.80), generator transient voltage Eq0 , initial rotor angle d(0), and mechanical power Pm(0). The results are in Table 7.8. The

454

7 Power System Transient Stability Analysis Table 7.8 E0 q, d(0), and Pm(0) of generators Generator Neglecting the effect of With the effect of sasalient poles lient poles Eq0 d(0) Eq0 d(0) 1 1.05664 2.27165 1.05636 3.58572 2 1.05020 19.73159 0.78817 61.09844 3 1.01697 13.16641 0.76786 54.13662

Pm(0)

0.71641 1.63000 0.85000

initial values of generator rotor angles are set to o1(0) ¼ o2(0) ¼ o3(0) ¼ 1. In the calculations to be described below, the effect of generator salient poles is neglected, which is to say Eq0 ¼ C and Xq ¼ Xd0 . This is the classical model of generators. 2. The fault-on system and post-fault system model: In the fault-on system, a shunt branch with zero impedance is connected at node 7, to model this shunt branch, the diagonal element Y77 of the admittance matrix Y is set to a very high value (say 1020). The admittance matrix of the fault-on system is YF. In the postfault network, branch 5–7 is removed. Because the contribution of line 5–7 to admittance matrix is equal to: 2

Ylð57Þ

5

.. 6 . 6 1 6 þ jb 6 r þ jx 56 6 . .. ¼ 6 6 .. . 6 1 76 6 6 r þ jx 4 .. .

7 .. . .. . .. .

.. .

1 r þ jx .. .

1 þ jb r þ jx .. .

.. .

3 7 7 7 7 7 7 7; 7 7 7 7 7 5

where r ¼ 0.032, x ¼ 0.161, and b ¼ 0.153. Thus the postfault admittance matrix is YP ¼ Y Yl(5–7). 3. Integrating the differential-algebraic equations: We will only compute the transient duration from the instant the fault occurs to time equals 2 s. Thus the system for the duration 0–2 s is divided into two autonomous systems: that is, the fault-on system for duration 0–0.08333 s, and the postfault system for duration 0.08333–2 s. The step size for numerical integration is 0.001 s. Table 7.9 lists the rotor angles d(t) and relative maximum rotor angles, with and without consideration of the effect of salient pole. The later is also depicted in Fig. 7.13. From Fig. 7.13, observe that the system is stable, whether or not the salient pole effect is taken into consideration. When the salient pole effect is considered, the maximum relative rotor angle is d21 ¼ 151.48396 (t ¼ 0.80133s). When the

0.00000 0.04200 0.08333 0.13333 0.18333 0.23333 0.28333 0.33333 0.38333 0.43333 0.48333 0.53333 0.58333 0.63333 0.68333 0.73333 0.78333 0.83333 0.88333 0.93333 0.98333 1.03333 1.08333 1.13333 1.18333 1.23333

2.27165 2.28779 2.34848 2.40803 2.58251 3.19401 4.52397 6.79447 10.16415 14.73304 20.54980 27.61528 35.87927 45.23062 55.48563 66.38413 77.60323 88.79177 99.61769 109.81335 119.20630 127.73265 135.43743 142.46732 149.05619 155.50082

19.73159 22.15764 29.28237 41.21540 53.48395 65.30378 76.12258 85.62591 93.68682 100.29635 105.50168 109.36539 111.95248 113.34805 113.70536 113.31385 112.66232 112.45974 113.57983 116.92253 123.22495 132.88223 145.83696 161.57457 179.23461 197.81063

13.16641 14.63856 18.86248 25.92757 33.68320 41.94182 50.42907 58.80933 66.72892 73.85628 79.91656 84.73139 88.27281 90.72587 92.52996 94.35019 96.94304 100.94170 106.66335 114.05272 122.79557 132.52906 143.03225 154.30791 166.52345 179.84443

Table 7.9 Generator d(t) and relative maximum rotor angles Without consideration of salient pole effects d1 d2 d3 3.58572 3.69016 4.00270 4.48409 5.09318 6.11234 7.80288 10.37892 13.99684 18.75788 24.71833 31.90242 40.31440 49.94781 60.79180 72.83490 86.06717 100.48127 116.07313 132.84229 150.79173 169.92694 190.25360 211.77290 234.47339 258.31838

61.09844 63.52449 70.64922 82.69359 95.37520 108.03487 120.26523 131.92171 143.06625 153.88636 164.62086 175.50595 186.74309 198.48469 210.83172 223.83807 237.51766 251.85143 266.79269 282.27022 298.18912 314.43044 330.85111 347.28707 363.56384 379.52044

54.13662 55.31617 58.74084 64.79619 72.05804 80.45556 89.78211 99.74929 110.07065 120.53143 131.02170 141.53491 152.14627 162.98542 174.21138 185.99198 198.48689 211.83293 226.13087 241.43372 257.73701 274.97226 293.00522 311.64166 330.64456 349.76706

With consideration of salient pole effects ~d1 ~d2 ~ d3 17.45994 19.86985 26.93389 38.80737 50.90144 62.10977 71.59862 78.83143 83.52267 85.56332 84.95188 81.75011 76.07321 68.11743 58.21973 46.92972 35.05909 23.66797 13.96214 7.10918 4.01865 5.14958 10.39953 19.10725 30.17842 42.30981

d2 d1 57.51272 59.83433 66.64652 78.20951 90.28202 101.92253 112.46235 121.54280 129.06941 135.12847 139.90254 143.60353 146.42869 148.53688 150.03992 151.00317 151.45049 151.37016 150.71956 149.42793 147.39740 144.50351 140.59752 135.51417 129.09045 121.20206 (continued)

~d2 ~d1

7.4 Transient Stability Analysis with Simplified Model 455

1.28333 1.33333 1.38333 1.43333 1.48333 1.53333 1.58333 1.63333 1.68333 1.73333 1.78333 1.83333 1.88333 1.93333 1.98333

162.12683 169.25129 177.15096 186.04298 196.07909 207.34945 219.88920 233.68131 248.65233 264.66308 281.50116 298.88496 316.48551 333.96386 351.01413

216.36810 234.19924 250.87411 266.20363 280.15828 292.78226 304.12763 314.22084 323.06766 330.69730 337.23926 343.01311 348.59530 354.81755 362.66346

194.25664 209.48217 225.02567 240.30096 254.76080 267.98613 279.73658 289.98411 298.94296 307.08305 315.08928 323.73003 333.64692 345.15535 358.17917

Table 7.9 (continued) Without consideration of salient pole effects d1 d2 d3 283.22935 309.06874 335.62983 362.64368 389.80754 416.82762 443.45984 469.53546 494.97070 519.76777 544.01356 567.87632 591.59440 615.45111 639.73684

395.05209 410.17107 425.07106 440.16296 456.04799 473.42157 492.94327 515.12417 540.25876 568.39552 599.32930 632.61227 667.60021 703.55468 739.78895

368.80340 387.65294 406.38036 425.24462 444.67239 465.17298 487.22074 511.14408 537.05069 564.80111 594.03622 624.26498 655.00155 685.90126 716.82750

With consideration of salient pole effects ~d1 ~d2 ~ d3 54.24126 64.94795 73.72315 80.16065 84.07919 85.43281 84.23842 80.53953 74.41533 66.03422 55.73810 44.12814 32.10979 20.85369 11.64932

d2 d1 111.82274 101.10233 89.44123 77.51928 66.24045 56.59395 49.48343 45.58871 45.28806 48.62775 55.31574 64.73595 76.00581 88.10357 100.05211

~d2 ~d1

456 7 Power System Transient Stability Analysis

Relative Swing Angle(Degrees)

7.4 Transient Stability Analysis with Simplified Model 160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0

457

d21

d21

0 0.10.20.30.40.50.60.70.80.9 1 1.11.21.31.41.51.6 1.71.81.9 2 t(s)

Fig. 7.13 Relative rotor angles as functions of time

salient pole effect is not considered, the maximum relative rotor angle is d21 ¼ 85.65788 (t ¼ 0.44633 s), and the angle of the second swing d21 ¼ 85.43378 (t ¼ 1.53433 s), which is smaller than that of the first swing. Lastly, the critical clearing times, with and without consideration of salient pole effects, are calculated. It turns out that the critical clearing time under the former circumstance is between 0.162 and 0.163 s, and it is between 0.085 and 0.086 s under the second condition. The swing curves under these circumstances are provided in Figs. 7.14 and 7.15, respectively.

7.4.4

Numerical Integration Methods for Transient Stability Analysis Under Classical Model

In a modern energy management system (EMS), to assess system security, transient stability under various prespecified contingencies is predicted online within a limited amount of time. Because the number of contingencies is quite large, to meet the requirement of online assessment, each transient stability analysis must be completed rapidly. Obviously the traditional integration methods for transient stability analysis are no longer suitable because of the limitation of speed, fast methods customized for online applications need to be developed. Dynamic security assessment is one of the ‘‘hot’’ areas in the power system stability field. As early as 1983, IEEE established a transient stability analysis working group, the responsibility of which was to lead and review the research in

458

7 Power System Transient Stability Analysis 300 280 260

d 21 ( ct = 0.163s )

Relative Swing Angle(Degrees)

240 220 200 180 160 140 120

d 21 ( ct = 0.162 s )

100 80 60 40 20 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.81.9 2 t(s)

Fig. 7.14 Relative rotor angle near critical clearing time (salient pole effects not considered)

300 280

Relative Swing Angle(Degrees)

260

~

d21 (ct = 0.086s)

240 220 200 180 160 140 120

~

d21 (ct = 0.085s)

100 80 60 40 20 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 t/s

Fig. 7.15 Relative rotor angle near critical clearing time (salient pole effects considered)

7.4 Transient Stability Analysis with Simplified Model

459

this area. A dynamic security assessment method must be extra-rapid, especially in an online environment. Though the requirement on computational precision can be relaxed to some extent, reliability and robustness are still required. Currently, the methods for improving the speed of online dynamic security assessment include no more than: first, simplifying the mathematical model for stability analysis; and second, developing rapid algorithms for stability analysis. In what follows we introduce a rapid algorithm for transient stability analysis based on a classical model. 1. The classical model of power system stability: The ‘‘classical model’’ places the following assumptions on the mathematical model: (1) Assume that the generator mechanical power remains constant during the transient stability period, and neglect the effects of damper windings (2) Assume that generator transient voltage E0 does not change during the transient stability study period, and furthermore the phase angle of this voltage is equal to the rotor angle (3) Loads are modeled as constant impedances With the above assumptions, the motion equation of the ith generator is obtained as 9 ddi > > ¼ os ðo 1Þ = dt i ¼ 1; 2; . . . ; m: doi 1 > ; ¼ ðPmi Pei Þ > dt TJi

ð7:95Þ

From a load flow study one finds the transient voltage: 0 E_ 0i ¼ E0i ﬀdið0Þ ¼ V_ ið0Þ þ ðRai þ jXdi Þ

Pið0Þ jQið0Þ : ^_ V

ð7:96Þ

ið0Þ

And the normal operation conditions give oið0Þ ¼ 1:

ð7:97Þ

Based on (7.42), (7.45), and (7.46) in Sect. 7.3.1, which describe the generator–network relationship, we can incorporate the pseudoadmittances of generators (see (7.45)) and load equivalent admittances (see (7.81)) into the network. The diagonal elements of the admittance matrix of the network (7.35) should have the generator pseudoadmittances or load equivalent

460

7 Power System Transient Stability Analysis

admittances added. The right-hand side vector, as given by (7.46), contains nonzeros in rows corresponding to generator nodes only, the other rows are zero. The expression for electromagnetic power of generators is easily obtained as ! ^_ 0 V ^_ E i Pei ¼ Re E_ 0i i : 0 Rai jXdi

ð7:98Þ

2. Solving for the network equations: First perform triangular factorization on the admittance matrix Y (a symmetrical matrix): Y ¼ UT DU;

ð7:99Þ

where U is an upper triangular matrix, D is a matrix with nonzeros on the diagonal only. After performing the following forward substitution and backward substitution, one obtains the voltage: F ¼ D1 UT I;

ð7:100Þ

V ¼ U1 F:

ð7:101Þ

Vector I in the above equations is a sparse vector. To compute the electromagnetic powers of generators, it is only necessary to know the generator voltages. Thus the unknown vector V is also sparse. Therefore, the network equations can be solved using rapid forward and backward substitutions. The experience of medium size computations demonstrates that, some 1/3 computational effort can be saved if a sparse vector method is used to perform rapid forward and back substitution. When solving the network equations with a sparse vector method, the majority of the time is spent in factorizing the admittance matrix. In dynamic security assessment, the admittance matrices for fault-on and postfault networks under different contingency scenarios are different. If these admittance matrices are factored every time, it would take a large amount of computer time. However, in general, the fault-on and postfault network admittance matrices differ from the prefault network admittance matrix only in a few places. This allows the utilization of compensation methods for solving network equations. The idea of a compensation method is to avoid matrix refactorizations, thus computational burden can be greatly relieved. Consider network equation: ðY þ DYÞV ¼ I;

ð7:102Þ

7.4 Transient Stability Analysis with Simplified Model

461

where Y is the prefault network admittance matrix, and DY is the adjustment to Y due to network switching or a fault and can be represented as DY ¼ MdyMT ;

ð7:103Þ

where dy is a (q q) matrix, including the information for the adjustment to Y, q is in general of order 1 or 2, and M is a (n q) coincidence matrix related to the specific fault or switch. By the matrix inversion lemma, (7.102) and (7.103) become V ¼ ðY1 Y1 MCMT Y1 ÞI;

ð7:104Þ

where the (q q) matrix C is equal to C ¼ ½ðdyÞ1 þ Z1 ;

ð7:105Þ

while the (q q) matrix Z is Z ¼ MT Y1 M:

ð7:106Þ

Thus according to (7.104), taking into account (7.99), the computational steps for solving the network (7.102) using compensation are as follows: Preparatory calculation: ð1Þ

9 > > > > > > =

W ¼ UT M

~ ¼ D1 W ð2Þ W ð3Þ

~ TW Z ¼W

ð4Þ

C ¼ ½ðdyÞ1 þ Z1

> > > > > > ;

:

ð7:107Þ

Solving the network equations:

ð2Þ

9 > > > > > > T > ~ ~ DF ¼ WCF F =

ð3Þ

F ¼ F~ þ DF

ð4Þ

V ¼ U1 D1 F

ð1Þ F~ ¼ UT I

> > > > > > > ;

:

ð7:108Þ

The forward and backward substitutions in (7.107) and (7.108) are all completed using a sparse vector method.

462

7 Power System Transient Stability Analysis

3. A numerical integration algorithm for second-order conservative systems: The differential equations in (7.95) can be rewritten in the following compact form: d2 d ¼ fðdÞ; dt2

ð7:109Þ

9 > > > = T fðdÞ ¼ ½f1 ðdÞ; . . . ; fm ðdÞ : > os > > ; fi ðdÞ ¼ ðPmi Pei Þ TJi

ð7:110Þ

where d ¼ ½d1 ; . . . ; dm T

The right-hand side functions of the differential equations in (7.109) do not contain arguments with first-order derivatives, the equations are thus termed a second-order conservative system. Compared with solving two first-order equations, the equations can be solved by direct differencing which results in an efficiency one level higher. Consider the Stormer and Numerov integration formula [186]: dkþ2 ¼ 2dkþ1 dk þ h2 fðdkþ1 Þ; dkþ2 ¼ 2dkþ1 dk þ

ð7:111Þ

h2 ½fðdkþ2 Þ þ 10fðdkþ1 Þ þ fðdk Þ: 12

ð7:112Þ

Equation (7.111) is an explicit second-order method, while (7.112) is an implicit forth-order method. To solve the differential (7.109) based on (7.111) requires a smaller step size because of the poor numerical stability. To solve (7.109) based on (7.112) allows a larger step size because the method has higher order and has a larger region of absolute stability. However this method still takes a large amount of computational effort because it involves solving a set of nonlinear simultaneous ½0 equations. On the other hand, if good initial estimates dkþ2 are provided when solving (7.112), the convergence can be speeded up. This suggests the application of a predictor–corrector method for solving (7.109); specifically, the explicit method (7.111) is adopted for the predictor, while the implicit method (7.112) is adopted for the corrector. Let P and C represent the application of one predictor and one corrector, E represent computing function f(d) once, the pair of predictor–corrector is formed as ½0 PECE. More concretely, one computes dkþ2 based on the predictor, and calculates ½0

½0

½1

f kþ2 ¼ fðdkþ2 Þ, substitute the result into the corrector to obtain dkþ2 , and finally ½1

½1

compute f kþ2 ¼ fðdkþ2 Þ.

7.5 Transient Stability Analysis with FACTS Devices

463

The above method falls into the category of multistep methods, the procedure can be started using the following special fourth-order Runge–Kutta formula [185]: 9 h2 > dkþ1 ¼ dk þ þ ðk1 þ 2k2 Þ > > > > 6 > > > h > 0 0 > dkþ1 ¼ dk þ ðk1 þ 4k2 þ k3 Þ > > > 6 > = k1 ¼ fðdk Þ : > > > h 0 h2 > > k2 ¼ f d k þ dk þ k 1 > > 2 8 > > > > 2 > h > 0 > ; k3 ¼ f dk þ hdk þ k2 2 hd0k

ð7:113Þ

The classical model for transient stability analysis applies to ‘‘first swing’’ (about 1.5 s after the disturbance). This model is free from the stiffness problem and therefore permits the use of larger step size (0.1–0.2 s).

7.5

Transient Stability Analysis with FACTS Devices

To study in detail the transient stability of a large scale interconnected power system experiencing various large disturbances and to analyze the effects of control devices on system stability, often for the purpose of seeking mechanisms for improving stability, a detailed component model for transient stability analysis is required. As the technology of HVDC develops, HVDC systems are widely used in long‐ distance transmission, under‐sea cable transmission and system interconnection. The technology of flexible AC transmission (FACTS), matured only in recent years, is also receiving much acceptance from the industry. FACTS devices not only help to improve system steady‐state performance, they also improve the dynamic performance of power systems to a certain degree, as a result system transfer capabilities are enlarged considerably. The dynamic performance of a power system is also affected by generator prime movers and speed‐governing systems, excitation systems, PSSs, and other control devices. A power system with increasing scale, and increasing installations of dynamic devices, exhibits complex behavior after it experiences a disturbance. The mechanical–electrical interaction of such a system lasts longer, and the duration of oscillation of the system before loss of stability occurs can be as long as several seconds to a dozen seconds. This section introduces the basic transient stability analysis method for large‐ scale interconnected power systems with many dynamic devices which are modeled in detail. It should be noted that the material presented does not address the detailed implementation of a commercial code, rather it concerns the basic principles.

464

7 Power System Transient Stability Analysis

The mathematical models for the dynamic devices are as follows: synchronous machines which are modeled by a sixth‐order model with varying Eq0 , E00q, Ed0 , E00d, and rotor variables; hydroprime mover and their speed‐governing system; excitation systems with thyristor‐based DC excitors, PSSs with generator speed deviation as input; two‐terminal HVDC; SVC and TCSC of the FACTS family; constant impedance loads or loads with second‐order voltage characteristics. If a different model other than those described above is used for a component, the same principle applies. The above large scale dynamic system is a typical stiff system because of the existence of dynamic devices with drastically different time constants. To solve such systems with an explicit numerical method, a very small step size has to be assumed because the stability region is relatively small. The implicit trapezoidal rule is a second‐order algorithm with the left half plane being the stability region, therefore it allows for the use of a larger step size. In early commercial codes explicit methods such as forth‐order Runge–Kutta method were quite popular. Because of their better numerical properties, adaptability to stiffness, and the introduction of fast control schemes with small time constants, the second‐order trapezoidal rule has become almost an industry standard since the 1970s. Many commercial grade codes, for example, the transient stability analysis package developed by Bonneville power administration (BPA), the power system analysis software package, are based on this method. In a typical transient stability analysis, the trapezoidal rule with constant step size, between 0.01 and 0.02 s (or even longer), is assumed. The difference and algebraic equations are solved by a simultaneous method or a sequential method. In the large‐scale transient stability analysis procedure to be presented below, the implicit trapezoidal rule is used to solve the differential equations, while a Newton method is used to solve the simultaneous difference‐algebraic equations of the detailed system model.

7.5.1 7.5.1.1.

Initial Values and Difference Equations of Generators Generators

The mathematical model of a synchronous machine comprises rotor motion equations, rotor electromagnetic equations, etc., together with stator voltage equations and the expressions for electromagnetic powers. Based on (6.1)–(6.4), these equations can be rewritten as follows: Rotor motion equations: 9 dd > > ¼ os ðo 1Þ = dt : do 1 > ; ¼ ðPm Pe DoÞ > dt TJ

ð7:114Þ

7.5 Transient Stability Analysis with FACTS Devices

465

Rotor electromagnetic equations: 9 dE0q 1 0 00 > ¼ 0 ½Efq kd Eq þ ðkd 1ÞEq > > > > dt Td0 > > > > 00 > dEq 1 0 > 00 0 00 > ¼ 00 ½Eq Eq ðXd Xd ÞId > = dt Td0 ; > dE0d 1 > > ¼ 0 ½kq E0d þ ðkq 1ÞE00d > > dt Tq0 > > > > > 00 > dEd 1 0 00 0 00 > ¼ 00 ½Ed Ed þ ðXq Xq ÞIq > ; dt Tq0

ð7:115Þ

Xq Xq00 Xd Xd00 and k ¼ q Xd0 Xd00 Xq0 Xq00 Stator voltage equations:

where kd ¼

Vd ¼ E00d Ra Id þ Xq00 Iq Vq ¼ E00q Xd00 Id Ra Iq

) :

ð7:116Þ

The electrical power is equal to the output power plus stator copper loss: 2 Pe ¼ Pout þ I_ Ra ¼ Vx Ix þ Vy Iy þ ðIx2 þ Iy2 ÞRa :

ð7:117Þ

Given a load flow solution, some of the initial values of generators can be computed based on (7.74)–(7.78). Note that the current flows in damper windings under steady‐state operation are equal to zero, based on (6.60), (6.64), and (6.65), the initial values of generator no‐load synchronous voltages, transient voltages, and sub‐transient voltages can be easily obtained as Efqð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd Idð0Þ ; E0qð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd0 Idð0Þ

) ;

E0dð0Þ ¼ Vdð0Þ þ Ra Idð0Þ Xq0 Iqð0Þ E00qð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd00 Idð0Þ

ð7:118Þ ð7:119Þ

)

E00dð0Þ ¼ Vdð0Þ þ Ra Idð0Þ Xq00 Iqð0Þ

:

ð7:120Þ

Besides, the electrical power Pe(0) of generators under steady‐state operation can be computed directly from (7.117): 2 2 Peð0Þ ¼ Pð0Þ þ ðIxð0Þ þ Iyð0Þ ÞRa :

ð7:121Þ

466

7 Power System Transient Stability Analysis

Set

do dt

¼ 0 in (7.114), the prime mover outputs Pm(0) are equal to Pmð0Þ ¼ Peð0Þ þ D:

ð7:122Þ

To solve the difference equations we first apply the trapezoidal rule for the rotor motion (7.114), dðtþDtÞ ¼ dðtÞ þ

os Dt ðoðtþDtÞ þ oðtÞ 2Þ; 2

oðtþDtÞ ¼ oðtÞ þ

ð7:123Þ

Dt ðPmðtþDtÞ PeðtþDtÞ DoðtþDtÞ þ PmðtÞ PeðtÞ DoðtÞ Þ: 2TJ ð7:124Þ

From (7.124) one obtains the expression for o(tþDt), substituting this into (7.123), it follows that: dðtþDtÞ ¼ aJ ðPmðtþDtÞ PeðtþDtÞ Þ þ d0 ;

ð7:125Þ

where aJ ¼

os ðDtÞ2 ; 4TJ þ 2DDt

ð7:126Þ

4TJ d0 ¼ dðtÞ þ aJ PmðtÞ PeðtÞ þ oðtÞ os Dt: Dt

ð7:127Þ

In (7.126), aJ is a function of step size Dt and some other constants. If a fixed step size is assumed, it becomes a constant. As for d0 in (7.127), it is a constant only in difference equation (7.125), it takes different values in each computational step. After d(tþDt) is found, o(tþDt) is calculated based on (7.123): oðtþDtÞ ¼

2 ðdðtþDtÞ dðtÞ Þ oðtÞ þ 2: os Dt

ð7:128Þ

Now applying the trapezoidal rule to the electromagnetic equation (7.115), it follows: 9 Dt 0 00 > ½E k E þ ðk 1ÞE d qðtþDtÞ d fqðtþDtÞ > qðtþDtÞ > 0 > 2Td0 > > > > 0 00 = þ EfqðtÞ kd E þ ðkd 1ÞE

E0qðtþDtÞ ¼ E0qðtÞ þ

qðtÞ

E00qðtþDtÞ ¼ E00qðtÞ þ

qðtÞ

Dt 0 00 0 00 00 ½EqðtþDtÞ EqðtþDtÞ ðXd Xd ÞIdðtþDtÞ 2Td0

þ E0qðtÞ E00qðtÞ ðXd0 Xd00 ÞIdðtÞ

> > > > > > > > ;

;

ð7:129Þ

7.5 Transient Stability Analysis with FACTS Devices

E0dðtþDtÞ

E00dðtþDtÞ

9 > > ¼ > > > > > > > 0 00 = kq EdðtÞ þ ðkq 1ÞEdðtÞ : > Dt > ¼ E00dðtÞ þ 00 ½E0dðtþDtÞ E00dðtþDtÞ þ ðXq0 Xq00 ÞIqðtþDtÞ > > > 2Tq0 > > > > ; 0 00 0 00 þ E E þ ðX X ÞI E0dðtÞ

467

Dt þ 0 ½kq E0dðtþDtÞ þ ðkq 1ÞE00dðtþDtÞ 2Tq0

dðtÞ

dðtÞ

q

q

ð7:130Þ

qðtÞ

Eliminating variables E0 q(tþDt) and E0 d(tþDt) in (7.129) and (7.130), we have E00qðtþDtÞ ¼ a00d ðXd0 Xd00 ÞIdðtþDtÞ þ a00d ad1 EfqðtþDtÞ þ E00q0 ;

ð7:131Þ

E00dðtþDtÞ ¼ a00q ðXq0 Xq00 ÞIqðtþDtÞ þ E00d0 ;

ð7:132Þ

where n 9 E00q0 ¼ a00d ad1 EfqðtÞ ðXd0 Xd00 ÞIdðtÞ þ 2ð1 kd ad1 ÞE0qðtÞ > > > > >

> > 1 > 00 > > þ ad1 ðkd 1Þ þ 2 EqðtÞ = ad2 n ; > > E00d0 ¼ a00q ðXq0 Xq00 ÞIqðtÞ þ 2ð1 kq aq1 ÞE0dðtÞ > > > >

> > 1 > 00 > þ aq1 ðkq 1Þ þ 2 EdðtÞ ; aq2

ð7:133Þ

9 > > > > > > = Dt Dt : ad2 ¼ 00 ; ad2 ¼ 00 > > 2Td0 þ Dt 2Td0 þ Dt > > > > a00d ¼ ½ad1 ð1 kd Þ þ 1=ad2 1 ; a00q ¼ ½aq1 ð1 kq Þ þ 1=aq2 1 ;

ð7:134Þ

ad1 ¼

Dt ; 0 2Td0 þ kd Dt

aq1 ¼

Dt 0 2Tq0 þ kq Dt

The coefficients ad1, ad2, a00d, aq1, aq2, and a00q in (7.134) are all constants if a fixed step size Dt is assumed, while in (7.133), E00q0 and E00d0 are known quantities at step t, although they take different values in each step. After E00 q(tþDt) and E00 d(tþDt) are calculated, now based on (7.129) and (7.130), 0 E q(tþDt) and E0 d(tþDt) can be obtained by

E0dðtþDtÞ

0 2Tdo kd Dt 0 EqðtÞ þ EfqðtþDtÞ þ EfqðtÞ þ ðkd 1ÞðE00qðtþDtÞ þ E00qðtÞ Þ Dt 0 2Tqo kq Dt 0 00 00 ¼ aq1 EdðtÞ þ ðkq 1ÞðEdðtþDtÞ þ EdðtÞ Þ Dt

E0qðtþDtÞ ¼ ad1

9 > > > = : > > > ;

ð7:135Þ

468

7 Power System Transient Stability Analysis

7.5.1.2

Exitation System and PSS

Taking the DC exitor with thyristor‐based regulator, illustrated in Fig. 6.16, as an example, let us derive the differential‐algebraic equations based on the transfer function diagram. We will neglect the effects of RC, and equivalent time constants TB and TC of the analog regulator. Under the ‘‘one per unit exitation voltage/one per stator voltage’’ system, by (6.51) it follows that Vf ¼ Efq. The measurement and filter system dVM 1 ¼ ðVC VM Þ; dt TR

VC ¼ V_ þ jXC I_

ð7:136Þ

The transient droop feedback: dðKF Efq TF VF Þ ¼ VF dt

ð7:137Þ

The amplifier: 9 1 > > f ¼ ½KA ðVREF þ VS VM VF Þ VR > > TA > > > > dVR > = if VR ¼ VRMAX and f > 0; ¼ 0; VR ¼ VRMAX > dt > dVR > > if VRMIN < VR < VRMAX ; ¼f > > dt > > > > dVR > if VR ¼ VRMIN and f < 0; ¼ 0; VR ¼ VRMIN ; dt

ð7:138Þ

The exiter dEfq 1 ¼ ½VR ðKE þ SE ÞEfq ; dt TE

ð7:139Þ

where the saturation coefficient SE is modeled as an exponential function according to (6.101). Under one per unit excitation voltage/one per unit stator voltage system, (6.101) is simplified to SE ¼ CE ENfqE 1 :

ð7:140Þ

The saturation function can also be piece‐wise linearized as follows: SE Efq ¼ K1 Efq K2 :

ð7:141Þ

7.5 Transient Stability Analysis with FACTS Devices

469

From Fig. 6.14 we have the PSS equations:

9 dV1 1 > > ¼ ðKS VIS V1 Þ > > > dt T6 > > > > dðV1 V2 Þ 1 > > ¼ V2 = dt T5 : > dðT1 V2 T2 V3 Þ > > ¼ V3 V2 > > > dt > > > > dðT3 V3 T4 V4 Þ ; ¼ V4 V3 > dt

ð7:142Þ

The limits of PSS output are If V4 VSmax ;

9 > = VS ¼ V4 : > ;

VS ¼ VSmax

If VSmin < V4 < VSmax ; If V4 VSmin ; VS ¼ VSmin

ð7:143Þ

The initial values of excitation system variables can be found by setting, in the transfer function diagram, s ¼ 0, or alternatively setting the left‐hand side of the differential equations of the excitation system to zero. The effects of limiters can in general be ignored since the variables with limiters under normal operation do not in general exceed their corresponding limits. In the following, we describe how to compute the initial values of the excitation system mentioned above, the other excitation systems can be dealt with likewise. Setting the left‐hand side of (7.139) to zero, one obtains the initial value for the amplifier VRð0Þ ¼ ðSEð0Þ þ KE ÞEfqð0Þ ;

ð7:144Þ

where the saturation coefficient is calculated based on (7.140), that is, E 1 SEð0Þ ¼ CE ENfqð0Þ :

Setting the left‐hand side of (7.136), (7.137), and (7.138) to zero, it follows: 9 VFð0Þ ¼ 0; VMð0Þ ¼ V_ð0Þ þ jXC I_ð0Þ = : ð7:145Þ VRð0Þ ; VREF ¼ VMð0Þ þ KA Setting the left‐hand side of (7.142) to zero, and taking into account the relationship expressed in (7.142), we have the initial value of PSS: ) VSð0Þ ¼ V4ð0Þ ¼ V3ð0Þ ¼ V2ð0Þ ¼ 0 ; ð7:146Þ V1ð0Þ ¼ KS VISð0Þ ¼ 0 where VIS is equal to zero since it often takes the form of speed, or change of active power.

470

7 Power System Transient Stability Analysis

Applying the trapezoidal rule to (7.136), we have the difference equations of measurement and filter systems: VMðtþDtÞ ¼ aR VCðtþDtÞ þ VM0 ;

ð7:147Þ

in which aR ¼

Dt ; 2TR þ Dt

ð7:148Þ

2TR Dt VMðtÞ ; 2TR þ Dt ) ¼ V_ ðtþDtÞ þ jXC I_ðtþDtÞ : ¼ V_ ðtÞ þ jXC I_ðtÞ

VM0 ¼ aR VCðtÞ þ VCðtþDtÞ VCðtÞ

ð7:149Þ ð7:150Þ

Applying the trapezoidal rule to (7.137), we have VFðtþDtÞ ¼ aF EfqðtþDtÞ þ VF0 ;

ð7:151Þ

where aF ¼ VF0 ¼

2KF ; 2TF þ Dt

ð7:152Þ

2TF Dt VFðtÞ aF EfqðtÞ : 2TF þ Dt

ð7:153Þ

When limiters are not taken into consideration, applying the trapezoidal rule to (7.138), we have the difference equation: VRðtþDtÞ ¼ aA ðVSðtþDtÞ VMðtþDtÞ VFðtþDtÞ Þ þ VR0 ;

ð7:154Þ

where aA ¼

KA Dt ; 2TA þ Dt

VR0 ¼ aA ð2VREF þ VSðtÞ VMðtÞ VFðtÞ Þ þ

ð7:155Þ 2TA Dt VRðtÞ : 2TA þ Dt

ð7:156Þ

Substituting (6.141) into (6.139), and applying trapezoidal rule, we have the difference equations of the excitor: EfqðtþDtÞ ¼ aE VRðtþDtÞ þ VE0 ;

ð7:157Þ

where aE ¼

Dt ; 2TE þ ðKE þ K1 ÞDt

ð7:158Þ

7.5 Transient Stability Analysis with FACTS Devices

471

VE0 ¼ aE ½VRðtÞ 2ðKE þ K1 ÞEfqðtÞ þ 2K2 þ EfqðtÞ : Applying the trapezoidal rule to (7.142), it follows: 9 V1ðtþDtÞ ¼ a1 VISðtþDtÞ þ V10 > > > > V2ðtþDtÞ ¼ a2 V1ðtþDtÞ þ V20 = ; V3ðtþDtÞ ¼ a3 V2ðtþDtÞ þ V30 > > > > ; V4ðtþDtÞ ¼ a4 V3ðtþDtÞ þ V40

ð7:159Þ

ð7:160Þ

in the above formula a1 ¼

KS Dt ; 2T6 þ Dt

a2 ¼

a4 ¼

2T3 þ Dt ; ð7:161Þ 2T4 þ Dt

9 > > > > > > > > > 2T5 Dt > > ¼ V2ðtÞ a2 V1ðtÞ = 2T5 þ Dt : > 2T2 Dt 2T1 Dt > ¼ V3ðtÞ V2ðtÞ > > > 2T2 þ Dt 2T2 þ Dt > > > > > 2T4 Dt 2T3 Dt ; ¼ V4ðtÞ V3ðtÞ > 2T4 þ Dt 2T4 þ Dt

ð7:162Þ

2T5 ; 2T5 þ Dt

V10 ¼ a1 VISðtÞ þ V20 V30 V40

a3 ¼

2T1 þ Dt ; 2T2 þ Dt

2T6 Dt V1ðtÞ 2T6 þ Dt

Eliminating the intermediate variables V1(tþDt), V2(tþDt), and V3(tþDt) in (7.160), it follows: V4ðtþDtÞ ¼ a4 a3 a2 a1 VISðtþDtÞ þ V40 þ a4 ½V30 þ a3 ðV20 þ a2 V10 Þ:

ð7:163Þ

If the input of the PSS is set to VIS = o os, apparently VIS(t) = o(t) os. Substituting VIS(tþDt)¼o(t¼Dt) os into (7.163), and making use of (7.128) to eliminate variable o(tþDt), we have V4ðtþDtÞ ¼ aS dðtþDtÞ þ VS0 ;

ð7:164Þ

where aS ¼

2a4 a3 a2 a1 ; os Dt

ð7:165Þ

VS0 ¼ V40 þ a4 ½V30 þ a3 ðV20 þ a2 V10 Þ aS dðtÞ þ a4 a3 a2 a1 ð2 os oðtÞ Þ: ð7:166Þ If the limits of PSS outputs are not considered, obviously we get VSðtþDtÞ ¼ V4ðtþDtÞ :

ð7:167Þ

If PSS takes other forms of input signals, following the same derivations, we should be able to find the corresponding expressions.

472

7 Power System Transient Stability Analysis

Eliminating the intermediate variables V4(tþDt), VS(tþDt), VM(tþDt), VF(tþDt), and VR(tþDt) in (7.164), (7.167), (7.147), (7.151), (7.154), and (7.157), the difference equations of the excitation system without taking into account the affects of limiters are obtained as EfqðtþDtÞ ¼ b1 dðtþDtÞ b2 V_ðtþDtÞ þ jXC I_ðtþDtÞ þ Efq0 ; ð7:168Þ where b1 ¼ Efq0 ¼ 7.5.1.3

aE aA aS ; 1 þ aE aA aF

b2 ¼

aE aA aR ; 1 þ aE aA aF

VE0 þ aE ½VR0 þ aA ðVS0 VM0 VF0 Þ : 1 þ aE aA aF

ð7:169Þ ð7:170Þ

The Prime Movers and Their Speed-Governing Systems

Taking the hydrogenerator and its speed-governing system illustrated in Fig. 6.24 as an example, based on the transfer function we have the differential-algebraic equations: The acentric flyball ¼ Kd ðoREF oÞ

ð7:171Þ

9 eKd eKd > > > 2 2 > = eKd eKd : If x

; s¼x > 2 2 > > > eKd eKd > ; If x ; s¼xþ 2 2

ð7:172Þ

The valve The dead zones are If

The limits of value position are If sMIN < s < sMAX ; If s sMAX ; If s sMAX ;

9 s ¼ s> =

s ¼ sMAX s ¼ sMIN

: > ;

ð7:173Þ

The servo system dm s ¼ dt TS

ð7:174Þ

7.5 Transient Stability Analysis with FACTS Devices

473

The limits of the valve 9 If mMIN < m < mMAX ; m ¼ m > = If m mMAX ; m ¼ mMAX > ; If m mMAX ; m ¼ mMIN

ð7:175Þ

d½x ðKb þ Ki Þm 1 ¼ ðKi m xÞ dt Ti

ð7:176Þ

dðPm þ 2KmH mÞ 2 ¼ ðKmH m Pm Þ; dt To

ð7:177Þ

The feedback system

The hydrogenerator

where parameter KmH is defined as follows: KmH ¼

PH ðMWÞ : SB ðMVAÞ

ð7:178Þ

In general, the parameters of a prime mover and its speed-governing system are provided in the per unit system with the nominal capability of the generator being the base. With the introduction of parameters KmH, Pm, and Pe all expressed in per unit system with system base SB. Similar to the calculation of excitation system initial values, the initial values of prime mover and speed-governing systems can be found by setting, in the transfer functions, s = 0, or alternatively by setting the left-hand side of the differential equations to zero. Again the dead zones of measurement systems and various limiters need not be considered in general. Setting the left-hand side of (7.177), (7.176), and (7.174) to zero, and making use of the linear relationships in (7.171), (7.172), (7.173), and (7.175), together with (7.77), we have the initial values of each state variable: 9 Pmð0Þ > mð0Þ ¼ mð0Þ ¼ ; ð0Þ ¼ xð0Þ ¼ Ki mð0Þ ; sð0Þ ¼ sð0Þ ¼ 0; > = KmH ð7:179Þ xð0Þ xð0Þ > > ; oREF ¼ oð0Þ þ ¼1þ : Kd Kd Based on (7.171), the equation corresponding to instant t + Dt for the acentric flyball is as follows: ðtþDtÞ ¼ Kd ðoREF oðtþDtÞ Þ:

ð7:180Þ

474

7 Power System Transient Stability Analysis

Neglecting the measurement dead zone, based on (7.172), it follows: sðtþDtÞ ¼ ðtþDtÞ xðtþDtÞ :

ð7:181Þ

Also neglecting the limits on valve position, and based on (7.173), obviously we have: sðtþDtÞ ¼ sðtþDtÞ :

ð7:182Þ

Applying the trapezoidal rule to (7.174), we obtain the following difference equation: mðtþDtÞ ¼ aS sðtþDtÞ þ m0 ;

ð7:183Þ

where aS ¼

Dt ; 2TS

m0 ¼ aS sðtÞ þ mðtÞ :

ð7:184Þ ð7:185Þ

Neglect again the limit on valve position, based on (7.175), we have mðtþDtÞ ¼ mðtþDtÞ :

ð7:186Þ

Applying the trapezoidal rule to (7.176), we have the difference equation of the feedback block: xðtþDtÞ ¼ ai mðtþDtÞ þ x0 ;

ð7:187Þ

where ai ¼ K i þ x0 ¼

2Ti Kb ; 2Ti þ Dt

2T i Dt 2T i Kb ½xðtÞ Ki mðtÞ m : 2T i þ Dt 2T i þ Dt ðtÞ

ð7:188Þ ð7:189Þ

Applying the trapezoidal rule to (7.177), the difference equations of hydrogenerators are obtained, as follows: PmðtþDtÞ ¼ aH mðtþDtÞ þ P0 ;

ð7:190Þ

where aH ¼

KmH ð2To DtÞ ; To þ Dt

ð7:191Þ

7.5 Transient Stability Analysis with FACTS Devices

P0 ¼

To Dt KmH ð2To þ DtÞ PmðtÞ þ mðtÞ : To þ Dt To þ Dt

475

ð7:192Þ

Eliminating the intermediate variables Z(tþDt), sðtþDtÞ , s(tþDt), mðtþDtÞ , m(tþDt), and x(tþDt) in (7.180), (7.182), (7.183), (7.186), (7.187), and (7.190), and eliminating variable o(tþDt) based on (7.128), we find the difference equations for step tþDt of hydrogenerators and their speed-governing systems, without consideration of limiters: PmðtþDtÞ ¼ b3 dðtþDtÞ þ Pm0 ;

ð7:193Þ

in which b3 ¼ Pm0 ¼ P0 b3 dðtÞ þ

2aH aS Kd ; ð1 þ aS ai Þos Dt

aH ½aS Kd ð2 oREF oðtÞ Þ þ aS x0 m0 : 1 þ aS ai

ð7:194Þ ð7:195Þ

Finally, substituting (7.117) of Pe(t + Dt) and (7.193) of Pm(t + Dt) into (7.125), and substituting difference equation (7.168) of Efq(t + Dt) into (7.131), together with (7.132) we obtain the difference equations of generators for step t + Dt. Let us transform the state currents under d–q coordinates into those under x–y coordinates, for notational simplicity, neglecting the subscripts (t + Dt), it follows: 9 ð1 aJ b3 Þd þ aJ ½Vx Ix þ Vy Iy þ Ra ðIx2 þ Iy2 Þ aJ Pm0 d0 ¼ 0 > > > > 00 00 0 00 00 > = Eq þ ad ðXd Xd ÞðIx sin d Iy cos dÞ ad ad1 b1 d qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ : > þ a00d ad1 b2 ðVx XC Iy Þ2 þ ðVy þ XC Ix Þ2 a00d ad1 Efq0 E00q0 ¼ 0 > > > > ; E00d a00q ðXq0 Xq00 ÞðIx cos d þ Iy sin dÞ E00d0 ¼ 0 ð7:196Þ The set of simultaneous (7.196) consists of three equations, the first reflects the mechanical motion of the generators, while the other two reflect the electromagnetic interactions. Based on (7.39), generator currents Ix and Iy are functions of Vx, Vy, d, E00q, and E00d (refer to (7.258) for details), and therefore can be eliminated. The above set of simultaneous equations thus has three state variables d, E00q , and E00d , plus two operating parameters Vx and Vy.

7.5.2

Initial Values and Difference Equations of FACTS and HVDC

7.5.2.1

SVC

Here we will focus on an SVC model comprising a fixed capacitor (FC) and a thyristor-controlled reactor (TCR). For ease of exposition, we will take a proportional regulator-based SVC as an example; its transfer function is illustrated in (7.16).

476

7 Power System Transient Stability Analysis VRef V

−

Σ

+

KS

BS1

1+sTS2

1+sTS

BC

BS2

1+sTS1

BSVC BC − BL

Fig. 7.16 A simple model of SVC

An SVC is generally connected to a high-voltage system via a transformer. The equivalent admittance of TCR is controlled by the firing angle a of a thyristor, thus the equivalent admittance BSVC of the SVC is manipulated. This mechanism facilitates the control of voltage V given the input VREF. The mathematical model of the SVC is obtained easily from Fig. 7.16 as 9 dBS1 1 > ¼ ½KS ðVREF VÞ BS1 > = dt TS : > dðTS2 BS2 TS1 BS1 Þ > ; ¼ BS1 BS2 dt

ð7:197Þ

The limit on SVC output is 9 BSVC ¼ BS2 > > = ¼ BC ; > > ; BSVC ¼ BC BL

If BC BL < BS2 < BC ; If BS2 BC ;

BSVC

If BS2 BC BL ;

ð7:198Þ

where BC = oC is the susceptance of the fixed capacitor, BL = 1/oL is the susceptance of the reactor, the output BSVC is the equivalent susceptance of the SVC. The upper limit of the SVC corresponds to the point at which the thyristor is completely shut off, while the lower limit corresponds to the point at which the thyristor is like a lossless conductor. The position between the limits corresponds to a point at which the thyristor is partially closed. Although an SVC is connected at the low-voltage side of a transformer, it can still be viewed as a reactive power source at the high-voltage side, intended to control the voltage at the high-voltage side bus of the transformer. Therefore, the high-voltage bus can be effectively set as a PV node in load flow studies (P = 0, V = VSP). From the result of a load flow study, one obtains V_ ð0Þ ¼ V SP ﬀyð0Þ and the power injection from the SVC S(0) = jQ(0). Let the reactance of the transformer be XT, the power injected into the network from the SVC is given by Qð0Þ ¼

2 Vð0Þ

1 BSVCð0Þ

XT

:

ð7:199Þ

7.5 Transient Stability Analysis with FACTS Devices

477

Setting both sides of (7.197) to zero, and noticing the relationship in (7.198) and (7.199), we find the initial values of the SVC as BSVCð0Þ ¼ BS2ð0Þ ¼ BS1ð0Þ VREF

BSVCð0Þ ¼ V SP þ KS

9 > > ¼ 2 > > Vð0Þ = XT þ Qð0Þ : > > > > ; 1

ð7:200Þ

Applying the trapezoidal rule to the first (7.197), it follows: BS1ðtþDtÞ ¼ n1 VðtþDtÞ þ BS10 ;

ð7:201Þ

where n1 ¼

KS Dt ; 2TS þ Dt

BS10 ¼ n1 ð2VREF VðtÞ Þ þ

2TS Dt BS1ðtÞ : 2TS þ Dt

ð7:202Þ ð7:203Þ

And applying the trapezoidal rule to the second equation in (7.197), and eliminating BS1(t + Dt) from (7.201), it follows: BS2ðtþDtÞ ¼ BSVC0 nS

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxðtþDtÞ þ VyðtþDtÞ ;

ð7:204Þ

where nS ¼ n1 BSVC0 ¼

2TS1 þ Dt ; 2TS2 þ Dt

2TS1 þ Dt 2TS2 Dt 2TS1 Dt BS10 þ BS2ðtÞ BS1ðtÞ : 2TS2 þ Dt 2TS2 þ Dt 2TS2 þ Dt

ð7:205Þ ð7:206Þ

If the limit of the SVC is ignored, then BS(tþDt) ¼ BS2(tþDt), thus: BSVCðtþDtÞ ¼ BSVC0 nS 7.5.2.2

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxðtþDtÞ þ VyðtþDtÞ :

ð7:207Þ

TCSC

A thyristor-controlled series compensator (TCSC) is connected into a transmission line in series, it changes its equivalent admittance thus achieving the goal of controlling the equivalent admittance of the transmission line. Here we will only

478

7 Power System Transient Stability Analysis

give the mathematical model of a TCSC composed of FC and TCR connected in parallel (similar to an SVC): 9 dBT1 1 > ¼ ½KT ðPREF PT Þ BT1 > = dt TT ; > dðTT2 BT2 TT1 BT1 Þ > ; ¼ BT1 BT2 dt

ð7:208Þ

where the input signal PT is the power flowing through the line in which the TCSC is connected, the output BTCSC is the equivalent susceptance of the TCSC. The limits of the TCSC are MAX If BMIN TCSC < BT2 < BTCSC ;

If BT2 BMAX TCSC ;

BTCSC

If BT2 BMIN TCSC ;

BTCSC

9 BTCSC ¼ BT2 > > = MAX ; ¼ BTCSC > > ; ¼ BMIN TCSC

ð7:209Þ

MIN where the specific values of BMAX TCSC and BTCSC depend on the sizes of L and C. They can be computed based on (5.153)–(5.155). As usual a load flow study provides BTCSC(0) and PT(0) = PSP, similar to computing the initial values of an SVC, we have

9 BTCSCð0Þ ¼ BT2ð0Þ ¼ BT1ð0Þ = : BTCSCð0Þ ; PREF ¼ PTð0Þ þ KT

ð7:210Þ

If the measured value of PT of TCSC flows from bus i to bus j, the expression of PT is easily obtained as PT ¼ BTCSC ðVxi Vyj Vyi Vxj Þ:

ð7:211Þ

Apply the trapezoidal rule to the first equation in (7.208), one obtains BT1ðtþDtÞ ¼ z1 PTðtþDtÞ þ BT10 ;

ð7:212Þ

in which z1 ¼

KT Dt ; 2TT þ Dt

BT10 ¼ z1 ð2PREF PTðtÞ Þ þ

2TT Dt BT1ðtÞ : 2TT þ Dt

ð7:213Þ ð7:214Þ

7.5 Transient Stability Analysis with FACTS Devices

479

Now applying the trapezoidal rule to the second equation of (7.208), eliminating BT1(tþDt) and PT(tþDt) based on (7.212) and (7.211), we have ½1 þ zT ðVxiðtþDtÞ VyjðtþDtÞ VyiðtþDtÞ VxjðtþDtÞ ÞBT2ðtþDtÞ BTCSC0 ¼ 0;

ð7:215Þ

in which zT ¼ z1 BTCSC0 ¼

2TT1 þ Dt ; 2TT2 þ Dt

2TT1 þ Dt 2TT2 Dt 2TT1 Dt BT10 þ BT2ðtÞ BT1ðtÞ : 2TT2 þ Dt 2TT2 þ Dt 2TT2 þ Dt

ð7:216Þ ð7:217Þ

If the output limits of the TCSC are neglected, obviously we have BTCSC(t + Dt) = BT2(t + Dt), then ½1 þ zT ðVxiðtþDtÞ VyjðtþDtÞ VyiðtþDtÞ VxjðtþDtÞ ÞBTCSCðtþDtÞ BTCSC0 ¼ 0: 7.5.2.3

ð7:218Þ

HVDC Systems

In stability studies, the network equations of the AC system appear in terms of positive sequence quantities, this places a fundament limitation on the model of an HVDC system. In particular, commutation failure in the HVDC system cannot be predicted. A commutation failure may be the result of a severe three-line-to-ground fault occurring close to the rectifier, an unsymmetrical fault on the AC side of the rectifier, or saturation of HVDC transformer operating during a transient period. Earlier HVDC models included the dynamic characteristics of transmission lines and converter dynamics. In recent years, there is a trend toward adopting simpler models. Two models for an HVDC system are popular, these are, a simplified model and a steady-state model.

1. The simplified model An HVDC system some distance away from the study area has little impact on the results of a stability study, and thus can be modeled using a simple model: the system is viewed as a pair of active and reactive power sources connected at the converter AC substation. A more realistic model is termed the steady-state model. Based on (5.2), the DC line is modeled by the algebraic equation of a resistor: VdR ¼ VdI þ Rdc Id ; where Rdc denotes the resistance of the line.

ð7:219Þ

480

7 Power System Transient Stability Analysis

Noticing that IdR ¼ IdI ¼ Id, from (7.52) and (7.53), eliminating VdR and VdI in (7.219), it follows: RId ¼ kR VR cos a kI VI cos b;

ð7:220Þ

R ¼ Rdc þ XcR þ XcI :

ð7:221Þ

where

The pole control action is assumed to be instantaneous; many of the control functions are represented in terms of their net effects, rather than actual characteristics of the hardware. This model appears in the form of an algebraic equation, the interaction between AC and DC system is similar to that in a load flow model. 2. Quasi-steady-state model If the short circuit currents in any of the converters are relatively low, then the dynamics of DC system elements has a non-negligible impact on the AC system. As a result, a more detailed DC model is necessary for conducting a transient stability analysis. In a quasi-steady-state model, the converter characteristics are still modeled by the equation governing the relationship between average DC values and the nominal values of fundamental frequency components. In this setting, the DC transmission line can adopt different models given different requirements on precision. The simplest DC line model is just that of a steady-state model, as in (7.220). A more detailed model is based on an R–L circuit: L

dId þ RId ¼ kR VR cos a kI VI cos b; dt

ð7:222Þ

where R is defined in (7.221), besides, L ¼ Ldc þ LR þ LI ;

ð7:223Þ

where Ldc, LR, and LI are the reactance of DC line, and the smoothing reactors. For the control system, taking the control mode of constant current and constant voltage as an example, from the transfer function given in Fig. 5.18, we have the differential equations: 9 > > = : > dðKc1 x1 aÞ Kc2 > ; ¼ ðIdREF x1 Þ dt Tc2 dx1 1 ¼ ðId x1 Þ dt Tc3

ð7:224Þ

7.5 Transient Stability Analysis with FACTS Devices

481

The limits on delayed ignition angle include 9 < aMAX ; a ¼ a > If aMIN < a = aMAX ; a ¼ aMAX If a ; > ; aMIN ; a ¼ aMIN If a

ð7:225Þ

9 dx4 1 > > ¼ ðVdI x4 Þ = dt Tv3 : > dðKv1 x4 bÞ Kv2 ; ¼ ðVdREF x4 Þ > dt Tv2

ð7:226Þ

The limits on ignition advance angle include MAX ; = b : If b ; b ¼ b MAX MAX > ; If b bMIN ; b ¼ bMIN

ð7:227Þ

When the rectifier is under constant current control, and the inverter is under SP constant voltage control, we have Idð0Þ ¼ IdSP and VdIð0Þ ¼ VdI . From a load flow study we have VR(0) and VI(0). Based on (7.224)–(7.227), noticing the relationships in (7.219) and/or (7.222), and (7.52) and (7.53), it follows: 9 IdREF ¼ x1ð0Þ ¼ Idð0Þ > > > > > V þ ðR þ X ÞI dc cR dIð0Þ dð0Þ > 1 > að0Þ ¼ að0Þ ¼ cos > = kR VRð0Þ : > VdREF ¼ x4ð0Þ ¼ VdIð0Þ > > > > > > 1 VdIð0Þ XcI Idð0Þ > ; bð0Þ ¼ bð0Þ ¼ cos kI VIð0Þ

ð7:228Þ

Applying trapezoidal rule to (7.224), we find x1ðtþDtÞ ¼ g1 IdðtþDtÞ þ x10 ;

ð7:229Þ

where g1 ¼

Dt ; 2Tc3 þ Dt

x10 ¼ g1 IdðtÞ þ

2Tc3 Dt x1ðtÞ : 2Tc3 þ Dt

ð7:230Þ ð7:231Þ

482

7 Power System Transient Stability Analysis

Using the second formula in (7.224), and eliminating x1(tþDt) by making use of (7.229), we have aðtþDtÞ ¼ g2 IdðtþDtÞ þ a0 ;

ð7:232Þ

where g2 ¼ g1

Kc2 Dt Kc1 þ ; 2Tc2

Kc2 Dt Kc2 Dt Kc2 Dt Kc1 þ x10 þ aðtÞ IdREF Kc1 x1ðtÞ : 2Tc2 Tc2 2Tc2

a0 ¼

ð7:233Þ ð7:234Þ

Neglecting the limits on ignition angle a, it is obvious that: aðtþDtÞ ¼ aðtþDtÞ :

ð7:235Þ

Applying the trapezoidal rule to the first formula of (7.226), it follows: x4ðtþDtÞ ¼ g3 VdIðtþDtÞ þ x40 ;

ð7:236Þ

where g3 ¼

Dt ; 2Tv3 þ Dt

x40 ¼ g3 VdIðtÞ þ

2Tv3 Dt x4ðtÞ : 2Tv3 þ Dt

ð7:237Þ ð7:238Þ

Applying the trapezoidal to the second formula in (7.226), making use of (7.236) to eliminate x4(tþDt), and noticing the relationship in (7.53) allows VdI(tþDt) to be eliminated, after simple manipulations, we have b ðtþDtÞ ¼ g4 VIðtþDtÞ cos bðtþDtÞ þ g5 IdðtþDtÞ þ b0 ;

ð7:239Þ

in which g3 Kv2 Dt g4 ¼ Kv1 þ ; nI 2Tv2 b0 ¼

g5 ¼ g4 nI RcI ;

Kv2 Dt Kv2 Dt VdREF Kv1 Kv2 Dt x4ðtÞ : Kv1 þ x40 þ b ðtÞ 2Tv2 Tv2 2Tv2

ð7:240Þ ð7:241Þ

7.5 Transient Stability Analysis with FACTS Devices

483

Neglecting the limits on ignition angle b, it follows: bðtþDtÞ ¼ b ðtþDtÞ :

ð7:242Þ

Under a quasi-steady-state model, different difference equations can be developed, with or without consideration for the transient duration of the DC transmission line. If the transient duration of the DC line is not considered, the DC line is modeled based on (7.220), where Id can be expressed as a function of a, b, VxR, VyR, VxI, and VyI: Id ¼

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ kR qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 þ V 2 cos a kI VxR VxI2 þ VyI2 cos b: yR R R

ð7:243Þ

ðtþDtÞ and Id(tþDt) in (7.232), (7.235), and (7.243), we then Let us eliminate a obtain the difference equation of the rectifier under constant current control, when the limits on a are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 þ r2 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ a0 ¼ 0;

aðtþDtÞ r1

ð7:244Þ

where r1 ¼

kR g; R 2

r2 ¼

kI g : R 2

ð7:245Þ

Similarly, eliminating variables b ðtþDtÞ and Id(tþDt) in (7.239), (7.242), and (7.243), we get the difference equation of the inverter under constant voltage control, when the limits on b are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 r4 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ b0 ¼ 0;

bðtþDtÞ r3

ð7:246Þ

where r3 ¼

kR g ; R 5

r 4 ¼ g4

kI g: R 5

ð7:247Þ

If the transient response of the DC line is considered, the DC line is modeled by (7.222), applying the trapezoidal rule to this equation, it follows: IdðtþDtÞ ¼ g6 VRðtþDtÞ cos aðtþDtÞ g7 VIðtþDtÞ cos bðtþDtÞ þ Id0 ;

ð7:248Þ

484

7 Power System Transient Stability Analysis

where g6 ¼

kR Dt ; 2L þ RDt

g7 ¼ g6

Id0 ¼ g6 VRðtÞ cos aðtÞ g7 VIðtÞ cos bðtÞ þ

kI ; kR

2L RDt IdðtÞ : 2L þ RDt

ð7:249Þ ð7:250Þ

Now eliminating aðtþDtÞ and Id(tþDt) in (7.232), (7.235), and (7.248), we find the difference equation of the rectifier, under constant current control, when the limits on a are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 þ r6 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ u0 ¼ 0;

aðtþDtÞ r5

ð7:251Þ

where r5 ¼ g 2 g 6 ;

r6 ¼ g 2 g 7 ;

u0 ¼ a0 þ g2 Id0 :

ð7:252Þ ð7:253Þ

By the same token, eliminating b ðtþDtÞ and Id(tþDt) in (7.239), (7.242), and (7.248), we find the difference equation of the inverter, under constant current control, when the limits on b are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 þ r8 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ n0 ¼ 0;

aðtþDtÞ r7

ð7:254Þ

where r7 ¼ g 5 g 6 ;

r8 ¼ g 4 g 5 g 7 ;

v0 ¼ b0 þ g5 Id0 :

7.5.3

ð7:255Þ ð7:256Þ

Forming Network Equations

The network equations expressed in the domain of real numbers are provided in (7.36). In transient stability studies, the nodes in the network are divided into three classes: nodes connected in parallel with dynamic devices (including generator nodes, SVC nodes, and load nodes); nodes connected in series with dynamic

7.5 Transient Stability Analysis with FACTS Devices

485

devices (including the AC buses of an HVDC system, TCSC nodes, etc.); and the faulted nodes or nodes not connected with any device. Substitute the network current expressions of each dynamic device, illustrated in Sect. 7.3.1, into the network equations, and properly dealing with a fault or switch, as described in Sect. 7.3.2, we obtain the network equations ready for subsequent simulation.

7.5.3.1

Nodes Connected in Parallel with Dynamic Devices

If a dynamic device is connected at node i, then the network equation for this node is DIxi ¼ Ixi

X k2i

DIyi ¼ Iyi

X k2i

9 ðGik Vxk Bik Vyk Þ ¼ 0 > > = ðGik Vyk þ Bik Vxk Þ ¼ 0 > > ;

:

ð7:257Þ

The expressions for currents Ixi and Iyi at node i depend on what device is connected. (1) Connected with a generator: Note that a generators is represented using a varying Eq0 , E00q, Ed0 , and E00d model, therefore assigning the corresponding values to the elements in (7.40), based on Table 7.1, it turns out that the expression for the generator current (7.39) can be rewritten as 9 n 1 00 00 00 00 > > ðR cos d þ X sin d ÞE þ ðR sin d X cos d ÞE ai i i ai i i > qi qi di di 00 00 > R2ai þ Xdi Xqi > > o > > > 00 00 00 2 00 2 > = ½Rai ðXdi Xqi Þ sin di cos di Vxi ðXdi cos di þ Xqi sin di ÞVyi : n 1 00 00 00 00 > > > Iyi ¼ 2 ðR sin d X cos d ÞE ðR cos d þ X sin d ÞE ai i i qi ai i i di > qi di 00 X00 > Rai þ Xdi > qi > o > > > 00 00 00 00 ; þ ðXdi sin2 di þ Xqi cos2 di ÞVxi ½Rai þ ðXdi Xqi Þ sin di cos di Vyi

Ixi ¼

ð7:258Þ (2) Connected with a load: As described in Sect. 7.3.1, if the load is a constant impedance, it can be incorporated in the network. If the load is modeled as a second-order polynomial function, it is modeled as a current injection in (7.48), note that the constant impedance part of the load can also be incorporated into the network. The last two terms in (7.48) are treated as current injections. If the load is modeled as an exponential function of voltage, it can be viewed as a current injection (7.49).

486

7 Power System Transient Stability Analysis

(3) Connected with an SVC: The expression of the current injection of an SVC is given in (7.50). 7.5.3.2

Nodes Connected with Series Devices

If a dynamic device is connected in series between node i and node j, the network equations between node i and j are X

9 ðGik Vxk Bik Vyk Þ ¼ 0 > > > > k2i > > X > > DIyi ¼ Iyi ðGik Vyk þ Bik Vxk Þ ¼ 0 > > > = k2i X : DIxj ¼ Ixj ðGjk Vxk Bjk Vyk Þ ¼ 0 > > > > > k2j > > X > > DIyj ¼ Iyj ðGjk Vyk þ Bjk Vxk Þ ¼ 0 > > ; DIxi ¼ Ixi

ð7:259Þ

k2j

The currents Ixi, Iyi, Ixj, and Iyj take different forms, depending upon what device is connected between node i and j. 1. Series TCSC The expressions for currents Ixi, Iyi, Ixj, and Iyj are given by (7.51). 2. Series HVDC system If the AC–DC system is solved by a simultaneous approach, the expressions for DC currents Ixi, Iyi, Ixj, and Iyj injected into the AC nodes are given in (7.56). In the formula of (7.54) and (7.55), the DC current Id can be replaced by (7.243) or (7.248), thus the current injections are functions of a, b, VxR, VyR, VxI, and VyI only. 7.5.3.3

A Connection Node or Faulted Node

A connection node has zero current injection. As discussed before, any type of fault can be modeled by adjusting the admittance matrix of the positive sequence network based on the concept of the synthesized impedance matrix. Therefore, in the faulted node of the extended positive sequence network, there is no current injection. The faulted node is therefore a connection node. The network equation of a connection node or faulted node is X

9 ðGfk Vxk Bfk Vyk Þ ¼ 0 > > = k2f X : DIyf ¼ 0 ðGfk Vyk þ Bfk Vxk Þ ¼ 0 > > ; DIxf ¼ 0

k2f

ð7:260Þ

7.5 Transient Stability Analysis with FACTS Devices

7.5.4

487

Simultaneous Solution of Difference and Network Equations

All the equations of the system for step t þ Dt have been given, they include the network equations and the difference equations of each dynamic device. In this system of equations, the unknowns include the operating variables of the power system under study Vx and Vy; all the state variables of dynamic devices, for example, d, E00q, and E00d of each generator, the BSVC of each SVC, the BTCSC of each TCSC, and the a and b of the HVDC system. Assume that there are n nodes, nG generators, nS SVCs, nT TCSCs, nD HVDC systems, then the number of unknowns is equal to 2n þ 3nG þ nS þ nT þ 2nD, which is just equal to the number of equations. The system of equations is well defined. Network equations (7.257), (7.259), (7.260), generator difference (7.196), SVC (7.207), TCSC (7.218), together with HVDC difference equations (7.244), (7.246) or (7.251), (7.254) comprise a set of nonlinear equations. The current injections and the difference equations are time varying, while the network equations assume the same structure, except for the steps at which a disturbance (either a fault or a switch operation) occurs. To compute the system states immediately after a disturbance, only the network equations need to be resolved. The state variables d, E00q, E00d, BSVC, BTCSC, a, and b of the dynamic devices should take the values obtained before the disturbance. The set of nonlinear equations comprising the difference and network equations is solved in a recursive manner to provide the states of the study system at each integration step. The above set of nonlinear equations is typically solved using a Newton method. Since the Newton method is already fairly familiar, the computational procedure of the method will only be briefly described below: 1. Set, for step t + Dt, the initial values of generator state variables d, E00q, E00d, the initial values of SVC state variables BSVC, the initial values of TCSC state variables BTCSC, the initial values of HVDC state variables a, b, and the initial values of network voltage Vx and Vy. These initial values either can be set to the values at step t, or may be extrapolated from the values of the previous steps. 2. For the set of nonlinear equations comprising generator difference equations, SVC difference equations, TCSC difference equations, HVDC difference equations, and the network equations, compute the Jacobian matrix and mismatches given the initial values obtained in step (1), then solve the linear equations for the updates to the variables. 3. Check if the iteration has converged. If yes, stop; otherwise, return to step (2). The iteration continues until convergence is reached. 4. After the quantities of the state variables for t + Dt are obtained, proceed to compute the values of the other dynamic variables according to the difference/ algebraic equations derived in Sects. 7.5.1 and 7.5.2. These values will be useful for the computation of the next step. It should be noted that, the effects of limiters should be considered in this step.

488

7 Power System Transient Stability Analysis

Thinking and Problem Solving 1. What is meant by the transient stability of electrical power systems? What methods can be adopted to analyze it? How can we judge the transient stability of electrical power systems? 2. What are the consequences of loss of transient stability in a power system? 3. What suppositions are made within the transient stability analysis model of electrical power systems and what is the underlying theory? 4. What main measures exist to improve the transient stability of electrical power systems? What is the principle of each measure? 5. Give the method used to modify an admittance matrix when short-circuit failures at different locations and of different fault types occur on one transmission line in an electrical network, and list essential calculation formulas. 6. What aspects should be considered in choosing appropriate integration methods when numerical integration is used to analyze the transient stability of electrical power systems? 7. What kinds of initial value problems of differential equations belong to the class of ‘‘stiff’’ problems? What requirements are there for numerical integration methods to solve stiff problems? 8. What are the advantages and disadvantages of the alternating solution method and the simultaneous solution method in solving initial value problems of differential-algebraic equations? 9. How can we deal with limiters when a numerical integration method is used to find the time-domain solution of each state variable in an electrical power system? 10. Although there are many numerical integration methods, the implicit trapezoidal integration method obtains wide application in transient stability analysis of electrical power systems, why? 11. During dynamic security evaluation, it is required to carry out rapid transient stability analysis of electrical power systems under each contingency. What aspects can be considered to improve the speed of the transient stability analysis? 12. When analyzing the transient stability of electrical power systems, each generator can be represented by one of the following models: E0 = C; Eq0 = C; Eq0 vary; Eq0, Ed0 vary; Eq0, E00q, E00d vary; Eq0, E00q, Ed0, E00d vary. Explain the applicability of each model. 13. It can be seen from the numerical solution of the transient stability analysis of a real electrical power system that the current on an inductance and the voltage across the two terminals of a capacitance will change significantly at the second that failure occurs, which seems to not satisfy the law of electromagnetic induction. Why? 14. During transient stability calculation using the improved Euler’s method, when considering the transient process of the excitation winding and the influence of excitation system, select a type of excitation system, and list the relevant formulas of the transient process calculation for one step.

Chapter 8

Small-Signal Stability Analysis of Power Systems

8.1

Introduction

Small-signal stability analysis is about power system stability when subject to small disturbances. If power system oscillations caused by small disturbances can be suppressed, such that the deviations of system state variables remain small for a long time, the power system is stable. On the contrary, if the magnitude of oscillations continues to increase or sustain indefinitely, the power system is unstable. Power system small-signal stability is affected by many factors, including initial operation conditions, strength of electrical connections among components in the power system, characteristics of various control devices, etc. Since it is inevitable that power system operation is subject to small disturbances, any power system that is unstable in terms of small-signal stability cannot operate in practice. In other words, a power system that is able to operate normally must first be stable in terms of small-signal stability. Hence, one of the principal tasks in power system analysis is to carry out small-signal stability analysis to assess the power system under the specified operating conditions. The dynamic response of a power system subject to small disturbances can be studied by using the method introduced in Chap. 7 to determine system stability. However, when we use the method for power system small-signal stability analysis, in addition to slow computational speed, the weakness is that after a conclusion of instability is drawn, we cannot carry out any deeper investigation into the phenomenon and cause of system instability. The Lyapunov linearized method has provided a very useful tool for power system small-signal stability analysis. Based on the fruitful results of eigensolution analysis of linear systems, the Lyapunov linearized method has been widely used in power system small-signal stability analysis. In the following, we shall first introduce the basic mathematics of power system small-signal stability analysis. The Lyapunov linearized method is closely related to the local stability of nonlinear systems. Intuitively speaking, movement of a nonlinear system over a small range should have similar properties to its linearized approximation.

X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

489

490

8 Small-Signal Stability Analysis of Power Systems

Assume the nonlinear system described by dx ¼ f ðxÞ: dt Its Taylor expansion at the origin is dDx ¼ ADx þ hðDxÞ; dt

ð8:1Þ

where @fðxe þ DxÞ @fðxÞ A¼ ¼ : @Dx @x x¼xe Dx¼0 If in the neighborhood of Dx = 0, h(Dx) is a high-order function of Dx, we can use the stability of the following linear system dDx ¼ ADx: dt

ð8:2Þ

To study the stability of the nonlinear system at point xe (1) If the linearized system is asymptotically stable, i.e., all eigenvalues of A have negative real parts, the actual nonlinear system is asymptotically stable at the equilibrium point. (2) If the linearized system is unstable, i.e., at least one of eigenvalues of A has a positive real part, the actual nonlinear system is unstable at the equilibrium point. (3) If the linearized system is critically stable, i.e., real parts of all eigenvalues of A are nonpositive but the real part of at least one of them is zero, no conclusion can be drawn about the stability of the nonlinear system from its linearized approximation. The basic principle of the Lyapunov linearized method is to draw conclusions about the local stability of the nonlinear system around the equilibrium point from the stability of its linear approximation. When carrying out small-signal stability analysis of a power system, we always assume that the system at normal operation at equilibrium point x ¼ xe or Dx ¼ 0 is disturbed instantly at the moment t ¼ t0 when system state moves from 0 to Dx(t0). Dx(t0) is the initial state of system free movement after disappearance of the disturbance. Because the disturbance is sufficiently small, Dx(t0) is within a sufficiently small neighborhood of Dx ¼ 0. Thus in the neighborhood of Dx ¼ 0, h(Dx) is a high-order indefinitely small variable. Hence according to the Lyapunov linearized method, we can study the stability of the linearized system to investigate that of the actual nonlinear power system. Linearizing the differential-algebraic dynamic description of a power system of (8.1) and (8.2) at steady-state operating point (x(0), y(0)), we can obtain

8.1 Introduction

491

~ ~ dDx=dt A B Dx ¼ ~ ~ ; 0 C D Dy

ð8:3Þ

where 2

@f1 6 @x1 6 ~¼6 . A 6 .. 6 4 @f n @x1 2 @g1 6 @x1 6 ~ 6 C ¼ 6 .. 6 . 4 @g m @x1

3 @f1 @xn 7 7 .. 7 ; . 7 7 5 @fn @xn x¼xð0Þ y¼yð0Þ 3 @g1 @xn 7 7 .. 7 ; . 7 7 5 @gm @xn x¼xð0Þ y¼yð0Þ

2 @f

1

6 @y1 6 6 ~ B ¼ 6 .. 6 . 4 @f n

@y1 2 @g

1

6 @y1 6 ~ 6 . D¼6 . 6 . 4 @g m @y1

@f1 3 @ym 7 7 .. 7 ; . 7 7 5 @fn @ym x¼xð0Þ y¼yð0Þ : @g 3 1

@ym 7 7 .. 7 . 7 7 @gm 5 @ym x¼xð0Þ y¼yð0Þ

R denotes the set of real numbers, Rn is the n-dimensional space of real vectors, Rmn is the set of m-row n-column real matrices. We define Rn to be Rn1 , i.e., elements in Rn are column vectors. On the other hand, elements in R1n are row ~ ~ ~ vectors. Obviously in the above equation, A 2 Rnn , B 2 Rnm , C 2 Rmn , ~ mm D2R . Deleting Dy in (8.3), we have dDx ¼ ADx; dt

ð8:4Þ

~ ~ ~ 1 ~ A ¼ A BD C:

ð8:5Þ

where

Matrix A 2 Rnn is often referred to as the state matrix or coefficient matrix. Therefore, small-signal stability studies local characteristics of the power system, i.e., asymptotic stability of an equilibrium point before the system is disturbed. Obviously, the theoretical basis to study power system small-signal stability by using the Lyapunov linearized method is that the disturbance must be sufficiently small. When the power system is subject to any such disturbance, state variables of the transient system model vary over a very small range. Hence asymptotic stability of the linearized system can guarantee a certain type of asymptotic stability of the actual nonlinear system.

492

8 Small-Signal Stability Analysis of Power Systems

We know that when the power system is subject to a sufficiently small disturbance at steady-state operation, there can be two consequences. One is that the disturbance approaches zero with time (i.e., disturbed movement approaches the undisturbed movement and all eigenvalues of corresponding matrix A have negative real parts) in this case the system is asymptotically stable at steady-state operation. The disturbed system will eventually return to the steady-state operation before the occurrence of disturbance. Another possible consequence is that disturbance Dx increases indefinitely with time, no matter how small the disturbance is (i.e., the real part of at least one of the eigenvalues of A is positive). Obviously the system is then unstable at this steady-state operating point. For the operation of a real power system, study of the critically stable situation is not so important, except that we can see it as the limiting case of small-signal stability. Here we need to point out that in our previous discussion about system stability, we assumed that the disturbance was instantaneous. That is, the system state moves instantly from Dx ¼ 0 to Dx(t0), and the disturbance disappears when the movement happens. However, the same theory is applicable to the study of stability when the system is subject to a permanent disturbance, because we can consider this as a case of stability subject to an instantaneous disturbance but operating at a new equilibrium point. Furthermore, for certain operating conditions in which the system is unstable in terms of small-signal stability or lacks damping, we can determine relationships between some controller parameters and the system eigenvalues (representing system stability) by using eigensolution analysis. In doing so, we can find certain ways to improve the power system small-signal stability. Hence small-signal stability analysis is a very important aspect of power system analysis and control. Therefore, power system analysis for operation at a steady state and subject to small disturbances includes: (1) Computation of steady-state values of various variables of the power system at a given steady-state operating condition, (2) Linearization of the differential-algebraic description of power system nonlinear dynamics to obtain the linearized differential-algebraic equations, (3) Formation of system state matrix A from the system linearized differentialalgebraic equations to determine system stability by calculation of the eigenvalues of A. In our above discussion, only the electromechanical oscillations between generators are considered in small-signal stability analysis. That is, we consider generators to be lumped rigid masses. However, the mechanical structure of real large-scale steam-turbine generation units is very complicated. It consists of several major lumped masses, such as turbine rotor, generator rotor, exciter rotor, etc. These lumped masses are connected by a rigid shaft of limited length. When generation units are disturbed, rotational speeds of the lumped masses are different during the system transient, due to elasticity between the lumped mass. This leads to torsional oscillations between each lumped mass. Because the inertia of each lumped mass is smaller than the total inertia of generation units, and taking into account the

8.2 Linearized Equations of Power System Dynamic Components

493

relevant stiffness, the frequency of torsional oscillations between each lumped mass is higher than that of electromechanical oscillations between generation units. Frequency of torsional oscillation is between about 10 and 50 Hz. This oscillation is often referred to as sub-synchronous oscillation (SSO). When SSO occurs, there is an oscillating torsional torque between lumped masses connected by the common shaft of a generation unit. Fatigue accumulation due to repeated episodes of torsional oscillation on the shaft will reduce shaft operating life. If the torsional torque exceeds a certain limit, shaft cracking, even breaking, can happen. Occurrence of SSO is mainly related to excitation control, governing control, HVDC control, and interaction between transmission line and series compensation of the line. When carrying out torsional oscillation analysis, we need to first establish a mathematical model of the shaft system of the steam turbine and generator. In addition, because frequency of torsional oscillation is high, a quasi-steady-state model of various components cannot be used. Instead, the electromagnetic transients in the power system need to be considered. Detailed analysis on torsional oscillation is outside the scope of this book. In this chapter, we first derive linearized models of various dynamic components in power systems, to establish the linearized equations of the whole system, in order to demonstrate the basic steps for computation of small-signal stability in power systems. Then, we will discuss the eigensolution problem in power system smallsignal stability analysis and the analytical methods required to study power system oscillations.

8.2

Linearized Equations of Power System Dynamic Components

In power system small-signal stability analysis, we need to linearize various dynamic components in the power system. In linearization, limiters in control devices often need not be considered. This is because in normal steady-state operation, the values of state variables associated with control devices are within the range determined by the limiters. If disturbances are sufficiently small, variations of state variables will not go beyond these limitations. As far as dead zones associated with certain control devices are concerned, we generally consider the dead zone to be small and hence ignored. If the dead zone is large, we can simply consider that in this case the control device does not function.

8.2.1

Linearized Equation of Synchronous Generator

8.2.1.1

Linearized Equation of Each Part of a Synchronous Generator

(1) Synchronous generator: For a synchronous generator described by (7.114)– (7.116) at a given steady-state operating condition, steady-state values of

494

8 Small-Signal Stability Analysis of Power Systems

various variables are d(0), o(0), E0 q(0), E00 q(0), E0 d(0), E00 d(0), Id(0), Iq(0), Vd(0), Vq(0), Pm(0), Pe(0), Efq(0) which can be calculated from (7.74–7.78) and (7.118–7.122). Linearizing each equation at these steady-state values, we obtain the linearized equation of a synchronous generator 9 dDd > > ¼ os Do > > > dt > > n h i > > dDo 1 00 00 00 00 00 > ¼ DDo Iqð0Þ DEq Idð0Þ DEd þ DPm Edð0Þ Xd Xq Iqð0Þ > > > dt TJ > > h i o > > > 00 00 00 > DId Eqð0Þ Xd Xq Idð0Þ DIq > > > > > 0 h i > dDEq > 1 = 0 00 ¼ 0 kd DEq þ ðkd 1ÞDEq þ DEfq dt Td0 ; > > 00 h i > dDEq > 1 > > ¼ 00 DE0q DE00q Xd0 Xd00 DId > > dt Td0 > > > > 0 >

dDEd 1 > 0 00 > > ¼ 0 kq DEd þ ðkq 1ÞDEd > > dt Tq0 > > > > h i 00 > dDEd 1 > 0 00 0 00 > > ¼ 00 DEd DEd þ Xq Xq DIq ; dt T q0

ð8:6Þ DVd ¼ DE00d Ra DId þ Xq00 DIq DVq ¼ DE00q Xd00 DId Ra DIq

) :

ð8:7Þ

(2) Excitation system: Taking an excitation system consisting of a DC exciter with thyristor-controlled regulator as an example, we can derive the linearized equation of (7.136–7.140) as follows. For measurement unit with VC ¼ V_ þ jXC I_, from coordinate transformation of (6.63), d, q components of voltage and current at generator terminals can be represented as V_ ¼ ðVd þ jVq Þejðdp=2Þ ;

I_ ¼ ðId þ jIq Þejðdp=2Þ :

ð8:8Þ

Obviously we have VC ¼ ½ðVd þ jVq Þ þ jXC ðId þ jIq Þejðdp=2Þ ¼ ðVd þ jVq Þ þ jXC ðId þ jIq Þ : qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ ðVd XC Iq Þ2 þ ðVq þ XC Id Þ2

ð8:9Þ

Linearizing the above equations at steady-state values, we can obtain DVC ¼ Kcd ðDVd XC DIq Þ þ Kcq ðDVq þ XC DId Þ;

ð8:10Þ

8.2 Linearized Equations of Power System Dynamic Components

495

where

9 Kcd ¼ ðVdð0Þ XC Iqð0Þ Þ=VCð0Þ ; Kcq ¼ ðVqð0Þ þ XC Idð0Þ Þ=VCð0Þ = qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ : ; VCð0Þ ¼ ðVdð0Þ XC Iqð0Þ Þ2 þ ðVqð0Þ þ XC Idð0Þ Þ2

ð8:11Þ

Linearizing (7.136), substituting (8.10) into it to cancel DVC, we can obtain the linearized equation of measuring and filtering unit to be dDVM 1 ¼ ðDVM þ Kcq XC DId Kcd XC DIq þ Kcd DVd þ Kcq DVq Þ: dt TR

ð8:12Þ

Representing the saturation characteristic of the exciter by (7.140) and linearizing (7.139) at steady-state operating point, we can have the linearized equation of the exciter to be i dDEfq 1 h E 1 ¼ KE þ nE cE Enfqð0Þ DEfq þ DVR : ð8:13Þ dt TE Finally, linearizing (7.137, 7.138) and rearranging them together with (8.12) and (8.13), we can obtain the linearized equation of the whole DC excitation system to be 9 E 1 KE þ nE cE Enfqð0Þ > dDEfq 1 > > ¼ DEfq þ DVR > > > dt TE TE > > > > dDVR 1 KA KA KA > > ¼ DVR DVF DVM þ DVS > = dt TA TA TA TA : ð8:14Þ E 1 > KF KE þ nE cE Enfqð0Þ > > dDVF KF 1 > > ¼ DEfq þ DVR DVF > > dt TE TF TE TF TF > > > > > dDVM 1 Kcq XC Kcd XC Kcd Kcq ; ¼ DVM þ DId DIq þ DVd þ DVq > dt TR TR TR TR TR (3) Power system stabilizer: For a Power system stabilizer (PSS) of Fig. 6.14, from (7.142) and (7.143) we can establish the following linearized equations when input to PSS is the deviation of rotor speed, VIS = o os 9 dDV1 KS 1 > > ¼ Do DV1 > > > dt T6 T6 > > > > dðDV1 DV2 Þ 1 > > ¼ DV2 = dt T5 : > dðT1 DV2 T2 DV3 Þ > > ¼ DV3 DV2 > > > dt > > > > dðT3 DV3 T4 DVS Þ ; ¼ DVS DV3 > dt

ð8:15Þ

After rearrangement, we can obtain linearized state equations of the PSS as follows

496

8 Small-Signal Stability Analysis of Power Systems

dDV1 dt dDV2 dt dDV3 dt dDVS dt

9 KS 1 > > Do DV1 > > T6 T6 > > > > > KS 1 1 > > ¼ Do DV1 DV2 = T6 T6 T5 : K S T1 T1 T1 T5 1 > > > ¼ Do DV1 DV2 DV3 > > T2 T6 T 2 T6 T2 T5 T2 > > > > > K S T1 T 3 T1 T3 T3 ðT1 T5 Þ T3 T2 1 ; ¼ Do DV1 DV2 DV3 DVS > T2 T4 T6 T2 T4 T6 T2 T4 T5 T2 T4 T4 ð8:16Þ ¼

(4) Prime mover and governing system: For the hydraulic turbine and its governing system, of Fig. 6.24, we can obtain its linearized equation from (7.171)–(7.177) to be 9 dDm Kd 1 > > ¼ Do Dx > > > dt TS TS >

> = dDx Kd ðKi þ Kb Þ Ki 1 Ki þ Kb ¼ Do þ Dm þ Dx : ð8:17Þ > dt TS Ti Ti TS > > > > dDPm 2KmH Kd 2KmH 2KmH 2 > ; ¼ Do þ Dm þ Dx DPm > dt TS To TS To 8.2.1.2

Matrix Description of Linearized Equation of Synchronous Generator Unit and Coordinate Transformation

(1) Matrix description of generation unit: For a generation unit described by (8.6), (8.7), (8.9), (8.15) and (8.17), its state variables can be arranged to form the following vector: Dxg ¼ ½Dd; Do; DE0q ; DE00q ; DE0d ; DE00d ; DEfq ; DVR ; DVF ; DVM ; DV1 ; DV2 ; DV3 ; DVS ; Dm; Dx; DPm T

:

ð8:18Þ

We define DVdqg ¼ ½DVd ; DVq T ;

DIdqg ¼ ½DId ; DIq T :

ð8:19Þ

Linearized differential equations of each generation unit can be written in the following matrix form dDxg Ig DIdqg þ B Vg DVdqg : ¼ Ag Dxg þ B dt

ð8:20Þ

Linearized equations of armature voltage equations can be arranged as g Dxg þ Z g DIdqg : DVdqg ¼ P

ð8:21Þ

g; B Ig ; In the two equations above, elements in the coefficient matrices A Vg ; P g; Z g , can be obtained easily by comparing (8.20) and (8.6), (8.9), B (8.15), (8.17) and comparing (8.21) and (8.7) as follows:

⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ Ag = ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢⎣

−

1 Td′′0

1 Td′′0

1 Tq′′0

1 Tq′′0

1 TA

−

1 T5

TS

2 K mH Tω

1 T4

2 K mH Kδ TS

−

Ki Ti

Kδ TS

1 T2 T2 − T3 T2T4

−

KA TA

Kδ ( K i + K β )

−

−

1 T6

T3 (T5 − T1 ) T2T4T5

1 T6

T1T3 T2T4T6

1 TR

KA TA

K ST1T3 T2T4T6

−

−

T5 − T1 T2T5

1 TF

KA TA

T1 T2T6

−

−

−

KF TETF

−

1 TE

K S T1 T2T6

K F K E′ TETF

K E′ TE

−

−

−

1 Td′0

KS T6

−

kq − 1 Tq′0

kq

TJ

Tq′0

−

−

I d (0)

−

−

kd − 1 Td′0

TJ

k − d Td′0

−

I q (0)

KS T6

D − TJ

ωs ⎤ ⎥ 1 ⎥ TJ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥, ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ 1 ⎥ − ⎥ TS ⎥ ⎥ ⎛ 1 K + Kβ ⎞ −⎜ + i ⎥ ⎟ T T ⎥ S ⎝ i ⎠ 2 K mH 2⎥ − ⎥ TS Tω ⎥ ⎦

8.2 Linearized Equations of Power System Dynamic Components 497

498

8 Small-Signal Stability Analysis of Power Systems

⎡ ⎢ ( X ′′ − X ′′) I − E ′′ q q (0) d (0) ⎢ d ⎢ TJ ⎢ ⎢ ⎢ X d′′ − X d′ ⎢ Td′′0 ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ B Ig = ⎢ ⎢ ⎢ K cq X C ⎢ ⎢ TR ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣⎢ ⎡ Pg = ⎢ ⎣

⎤ ( X d′′ − X q′′) I d (0) − Eq′′(0) ⎥⎥ ⎡ ⎥ TJ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ X q′ − X q′′ ⎥ ⎢ ⎥ ⎢ Tq′′0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ , B = ⎥ Vg ⎢ K cd ⎥ ⎢ ⎥ ⎢ TR ⎥ K cd X C ⎢ − ⎥ ⎢ TR ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢⎣ ⎥ ⎥ ⎦⎥ 1

1

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥, K cq ⎥ ⎥ TR ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦⎥

⎤ ⎥, ⎦

X q′′ ⎤ ⎡ −R Zg = ⎢ a ⎥ ⎣ − X d′′ − Ra ⎦

If different models of synchronous generator, excitation system and governing system are adopted, using the same procedure we can always first derive the linearized equation of each unit and then represent them in the form of (8.20) and (8.21). Of course, in (8.18), the sequential arrangement of state variables can be different, which would lead to different coefficient matrices.

8.2 Linearized Equations of Power System Dynamic Components

499

(2) Coordinate transformation: In (8.20) and (8.21), DVdqg and DIdqg are the deviation of d, q voltage and current components of each generator, respectively. Hence we must convert them into a representation in a unified x–y coordinate rotating at the same speed, so that they can then be connected to a common power network. Coordinate transformation for the terminal voltages of the generator is given by (6.62)

Vd Vq

sin d cos d ¼ cos d sin d

Vx : Vy

ð8:22Þ

Steady-state values Vd(0), Vq(0), Vx(0), Vy(0), and d(0) should also satisfy (8.22). That is Vdð0Þ sin dð0Þ cos dð0Þ Vxð0Þ ¼ : ð8:23Þ Vqð0Þ cos dð0Þ sin dð0Þ Vyð0Þ Linearizing (8.22) at steady-state values we have sin dð0Þ cos dð0Þ DVx cos dð0Þ DVd ¼ þ cos dð0Þ sin dð0Þ sin dð0Þ DVy DVq V xð0Þ Dd: Vyð0Þ

sin dð0Þ cos dð0Þ

ð8:24Þ

From (8.23) we can rewrite (8.24) as

DVd DVq

sin dð0Þ ¼ cos dð0Þ

cos dð0Þ sin dð0Þ

Vqð0Þ DVx þ Dd Vdð0Þ DVy

ð8:25Þ

that can be written simply as DVdqg ¼ Tgð0Þ DVg þ RVg Dxg ;

ð8:26Þ

where DVg ¼ Tgð0Þ

Vqð0Þ DVx ; RVg ¼ Vdð0Þ DVy sin dð0Þ cos dð0Þ ¼ : cos dð0Þ sin dð0Þ

0 0

0 ; 0

Note that Tg(0) is an orthogonal matrix, satisfying T T1 gð0Þ ¼ Tgð0Þ :

ð8:27Þ

500

8 Small-Signal Stability Analysis of Power Systems

Similarly, for generator current we can obtain DIdqg ¼ Tgð0Þ DIg þ RIg Dxg ; where

DIg ¼

DIx ; DIy

RIg ¼

Iqð0Þ Idð0Þ

ð8:28Þ 0 : 0

0 0

Substituting (8.26) and (8.28) into (8.21) to cancel DVdqg and DIdqg, we have DIg ¼ Cg Dxg þ Dg DVg ; where

ð8:29Þ

9 1 ðRVg P g Þ RIg = Cg ¼ TTgð0Þ ½Z g ;

1

Tgð0Þ Dg ¼ TTgð0Þ Z g

:

ð8:30Þ

Substituting (8.26) and (8.28) into (8.20) to cancel DVdqg and DIdqg and using (8.29), (8.30) to cancel DIg, we can obtain

where

dDxg ¼ Ag Dxg þ Bg DVg ; dt

ð8:31Þ

9 g þ B Ig Z 1 ðRVg P gÞ þ B Vg RVg = Ag ¼ A g : ; Ig Z 1 þ B Vg ÞTgð0Þ Bg ¼ ðB g

ð8:32Þ

Equations (8.31) and (8.29) consist of linearized equations of every generator, in the form of the state equation and output equation for a general time-invariant linear system.

8.2.2

Linearized Equation of Load

In small-signal stability analysis, a static load model is usually adopted. If a certain amount of induction motor load needs to be considered, we can use procedures similar to those used to derive the linearized equations of a synchronous generator, to establish the linearized equations of an induction motor. No matter which form is adopted to model the static voltage characteristics of load, deviation of injected current into the load has the following relationship to nodal voltage: DIl ¼ Yl DVl ;

ð8:33Þ

where

DIx DIl ¼ ; DIy

Gxx Yl ¼ Byx

Bxy ; Gyy

DVx DVl ¼ : DVy

ð8:34Þ

8.2 Linearized Equations of Power System Dynamic Components

501

The coefficients can be calculated from the following relationship between injected current and nodal voltage at the load node @Ix @Ix Gxx ¼ ; Bxy ¼ ; @Vx Vx ¼Vxð0Þ @Vy Vx ¼Vxð0Þ Vy ¼Vyð0Þ Vy ¼Vyð0Þ @Iy @Iy Byx ¼ ; Gyy ¼ : @Vx Vx ¼Vxð0Þ @Vy Vx ¼Vxð0Þ Vy ¼Vyð0Þ

ð8:35Þ

Vy ¼Vyð0Þ

If the static voltage characteristic of the load is modeled by a quadratic polynomial, we can use the relationship of (8.48) between injected current and node voltage and (8.35) to calculate relevant coefficients in (8.34) directly 9 > > > Gxx ¼ > 4 > > Vð0Þ > > > > 2 > Qð0Þ Vyð0Þ ðbQ þ 2cQ Þ þ Pð0Þ Vxð0Þ Vyð0Þ ðbP þ 2cP Þ Qð0Þ > > > Bxy ¼ 2 > 4 = V V > 2 Pð0Þ Vxð0Þ ðbP þ 2cP Þ þ Qð0Þ Vxð0Þ Vyð0Þ ðbQ þ 2cQ Þ

Pð0Þ 2 Vð0Þ

ð0Þ

ð0Þ

> Qð0Þ > > Byx ¼ 2 > > 4 Vð0Þ Vð0Þ > > > > > > 2 > Pð0Þ Vyð0Þ ðbP þ 2cP Þ Qð0Þ Vxð0Þ Vyð0Þ ðbQ þ 2cQ Þ Pð0Þ > > Gyy ¼ 2 > > 4 V V ; 2 Qð0Þ Vxð0Þ ðbQ þ 2cQ Þ Pð0Þ Vxð0Þ Vyð0Þ ðbP þ 2cP Þ

ð0Þ

:

ð8:36Þ

ð0Þ

When an exponential function is used to model static voltage characteristics of the load, the relationship between load injected current and node voltage, of (7.49), can be used jointly with (8.35) to derive relevant coefficients in (8.34) directly as Pð0Þ Gxx ¼ 2 Vð0Þ Qð0Þ Bxy ¼ 2 Vð0Þ Qð0Þ Byx ¼ 2 Vð0Þ Pð0Þ Gyy ¼ 2 Vð0Þ

ð2 mÞ ð2 nÞ ð2 nÞ ð2 mÞ

2 Vxð0Þ 2 Vð0Þ 2 Vyð0Þ 2 Vð0Þ 2 Vxð0Þ 2 Vð0Þ 2 Vyð0Þ 2 Vð0Þ

! 1 ! 1

Qð0Þ þ 2 Vð0Þ þ

! 1 ! 1

Pð0Þ 2 Vð0Þ

Pð0Þ 2 Vð0Þ

Qð0Þ 2 Vð0Þ

!9 > > > > > > > > !> > > Vxð0Þ Vyð0Þ > > > > ð2 mÞ > 2 = Vð0Þ ! : > Vxð0Þ Vyð0Þ > > > ð2 mÞ > 2 > Vð0Þ > > > !> > > Vxð0Þ Vyð0Þ > > > > ð2 nÞ ; 2 V Vxð0Þ Vyð0Þ ð2 nÞ 2 Vð0Þ

ð8:37Þ

ð0Þ

Especially, when there is not enough information about the static voltage characteristics of the load, a normally acceptable load model is to represent load active power by a constant current (i.e., taking m ¼ 1) and load reactive power by a constant impedance (i.e., taking n ¼ 2).

502

8 Small-Signal Stability Analysis of Power Systems

8.2.3

Linearized Equation of FACTS Components

1. SVC From (7.197) and (7.198) we can obtain the following linearized equation directly 9 dDBS1 KS 1 > > ¼ DV DBS1 = dt TS TS : > dðTS2 DBSVC TS1 DBS1 Þ > ; ¼ DBS1 DBSVC dt

ð8:38Þ

Because V 2 ¼ Vx2 þ Vy2 , after linearization we have DV ¼

Vxð0Þ Vyð0Þ DVx þ DVy : Vð0Þ Vð0Þ

ð8:39Þ

Substituting the above equation into (8.38) and after rearrangement we obtain dDxs ¼ As Dxs þ Bs DVs ; dt

ð8:40Þ

where Dxs ¼ " As ¼

DBS1 DBSVC

;

DVx

DVs ¼ DVy #

T1S

0

TS TS1 TS TS2

T1S1

;

KS Bs ¼ TS Vð0Þ

"

Vxð0Þ TS1 TS2

Vxð0Þ

Vyð0Þ TS1 TS2

Vyð0Þ

#

9 > > > > = > > > > ;

:

ð8:41Þ

In addition, from (7.50) we can obtain the relationship of deviation between SVC injected current and nodal voltage to be DIs ¼ Cs Dxs þ Ds DVs ; where

DIx DIs ¼ ; DIy

DVx DVs ¼ DVy " # 0 Vyð0Þ 1 Cs ¼ ; ð1 XT BSVCð0Þ Þ2 0 Vxð0Þ

BSVCð0Þ 0 Ds ¼ 1 XT BSVCð0Þ 1

ð8:42Þ 9 > > > > = : 1 > > > > 0 ;

ð8:43Þ Hence (8.40) and (8.42) form the linearized equation of the SVC. 2. TCSC From (7.208) and (7.209) we can obtain the following linearized equation directly

8.2 Linearized Equations of Power System Dynamic Components

503

9 dDBT1 KT 1 > > ¼ DPT DBT1 = dt TT TT : > dðTT2 DBTCSC TT1 DBT1 Þ ; ¼ DBT1 DBTCSC > dt

ð8:44Þ

From (7.211) we have DPT ¼ ðVxið0Þ Vyjð0Þ Vyið0Þ Vxjð0Þ ÞDBTCSC þ BTCSCð0Þ Vyjð0Þ DVxi BTCSCð0Þ Vxjð0Þ DVyi BTCSCð0Þ Vyið0Þ DVxj þ BTCSCð0Þ Vxið0Þ DVyj

:

ð8:45Þ

Substituting the above equation into (8.44) and after rearrangement we have dDxt ¼ At Dxt þ Bt DVt ; dt where

Dxt ¼

DBT1

ð8:46Þ

9 > > > > > > > > > > > > > > =

T

; DVt ¼ DVxi DVyi DVxj DVyj DBTCSC 3 1 KT ðV V V V Þ yið0Þ xjð0Þ xið0Þ yjð0Þ 6 7 TT TT 7 At ¼ 6 4 1 : TT1 1 TT1 KT 1 5 > ðVyið0Þ Vxjð0Þ Vxið0Þ Vyjð0Þ Þ > > TT2 TT2 TT TT2 TT TT2 > 2 3> > > Vyjð0Þ Vxjð0Þ Vyið0Þ Vxið0Þ > > > KT BTCSCð0Þ > 4 5 > Bt ¼ TT1 TT1 TT1 TT1 > TT ; Vyjð0Þ Vxjð0Þ Vyið0Þ Vxið0Þ > TT2 TT2 TT2 TT2 ð8:47Þ 2

In addition, from (7.51) we can directly obtain the relationship of deviation in TCSC injected current and nodal voltage to be DIt ¼ Ct Dxt þ Dt DVt ; where DIt ¼ DIxi 2 0 60 6 Ct ¼ 6 40 0

DIyi

DIxj

Vyið0Þ Vyjð0Þ

DIyj 3

Vxjð0Þ Vxið0Þ 7 7 7; Vyjð0Þ Vyið0Þ 5 Vxið0Þ Vxjð0Þ

ð8:48Þ

T 2

0 6 1 6 Dt ¼ BTCSCð0Þ 6 4 0 1

1

0

0 1 1 0 0 1

9 > > > > > 3> = 1 > 7 0 7 >: ð8:49Þ 7> > > 1 5> > > ; 0

Thus (8.46) and (8.48) form the linearized equations of a TCSC.

8.2.4

Linearized Equation of HVDC Transmission System

When transient behavior of an HVDC transmission line is considered, the control equations of HVDC transmission line, rectifier, and inverter are given by (7.222),

504

8 Small-Signal Stability Analysis of Power Systems

and (7.224)–(7.227). Canceling VdI in (7.226) by using the first equation in (7.53) and ignoring the limitation on a and b, we can obtain the following linearized equation around steady state 9 kI VIð0Þ sin bð0Þ kR VRð0Þ sin að0Þ dDId R > > ¼ DId Da þ Db > > dt L L L > > > > kI cos bð0Þ kR cos að0Þ > > > þ DVR DVI > > L L > > > > dDx1 1 > > > ¼ ðDId Dx1 Þ = dt Tc3 : dðKc1 Dx1 DaÞ Kc2 > > > ¼ Dx1 > > dt Tc2 > > > > kI VIð0Þ sin bð0Þ kI cos bð0Þ > dDx4 XcI 1 > ¼ DId Dx4 Db þ DVI > > > dt Tv3 Tv3 Tv3 Tv3 > > > > > dðKv1 Dx4 DbÞ Kv2 > ; ¼ Dx4 dt Tv2 ð8:50Þ

Relationships between the magnitude of AC bus voltages of rectifier and inverter 2 2 and their x, y components are VR2 ¼ VxR þ VyR , VI2 ¼ VxI2 þ VyI2 . Linearizing the above equations, we have 9 VxRð0Þ VyRð0Þ > DVR ¼ DVxR þ DVyR > > = VRð0Þ VRð0Þ : VxIð0Þ VyIð0Þ > > DVI ¼ DVxI þ DVyI > ; VIð0Þ VIð0Þ

ð8:51Þ

Substituting (8.51) into (8.50) to cancel DVR and DVI and after rearrangement we obtain dDxd ¼ Ad Dxd þ Bd DVd ; dt

ð8:52Þ

where Dxd ¼ ½DId Dx1 Dx4 Da DbT

T DVd ¼ DVxR DVyR DVxI DVyI

) ;

ð8:53Þ

where coefficient matrices Ad and Bd can be easily obtained by comparing (8.52) and the original equation.

8.2 Linearized Equations of Power System Dynamic Components

505

Algebraic equations of a two-terminal HVDC transmission system can be derived from relationships of power and current on the AC and DC sides of the converter. For the rectifier, the power relationship is VxR IxR þ VyR IyR ¼ XcR Id2 kR Id VR cos a:

ð8:54Þ

Linearizing the above equation we have VxRð0Þ DIxR þ VyRð0Þ DIyR ¼ IxRð0Þ DVxR IyRð0Þ DVyR þ 2XcR Idð0Þ DId kR VRð0Þ cos að0Þ DId kR Idð0Þ cos að0Þ DVR : ð8:55Þ þ kR Idð0Þ VRð0Þ sin að0Þ Da In addition, from the third equation in (8.52) we have 2 2 IR2 ¼ IxR þ IyR ¼ kR2 Id2 :

ð8:56Þ

The linearized form of the above equation is IxRð0Þ DIxR þ IyRð0Þ DIyR ¼ kR2 Idð0Þ DId :

ð8:57Þ

Substituting (8.51) into (8.55) to cancel DVR and noting the reactive power injection into the AC system from the rectifier, QR(0) = VyR(0) IxR(0) VxR(0) IyR(0), is always nonzero, we can derive the deviation of node injected current from (8.55) and (8.57) and have the following matrix form DIR ¼ CR Dxd þ DR DVR ; " DIR ¼

DIxR DIyR

#

" ;

DVR ¼

DVxR DVyR "

#

C11 1 CR ¼ VxRð0Þ IyRð0Þ VyRð0Þ IxRð0Þ C 21

0 0

C14

0

0 0

C24

0

#

C11 ¼ 2XcR Idð0Þ IyRð0Þ kR VRð0Þ IyRð0Þ cos að0Þ kR2 Idð0Þ VyRð0Þ C14 ¼ kR VRð0Þ IyRð0Þ Idð0Þ sin að0Þ C21 ¼ 2XcR Idð0Þ IxRð0Þ þ kR VRð0Þ IxRð0Þ cos að0Þ þ kR2 Idð0Þ VxRð0Þ

ð8:58Þ 9 > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > =

> > > > > C24 ¼ kR VRð0Þ IxRð0Þ Idð0Þ sin að0Þ > > > > > " # > > D D > 11 12 1 > > > DR ¼ > > VxRð0Þ IyRð0Þ VyRð0Þ IxRð0Þ D > D22 > 21 > > > > > kR VxRð0Þ Idð0Þ cos að0Þ kR VyRð0Þ Idð0Þ cos að0Þ > > > > D11 ¼ IyRð0Þ IxRð0Þ þ ; D12 ¼ IyRð0Þ IyRð0Þ þ > > VRð0Þ VRð0Þ > > > > > > kR VxRð0Þ Idð0Þ cos að0Þ kR VyRð0Þ Idð0Þ cos að0Þ > > > D21 ¼ IxRð0Þ IxRð0Þ þ ; D22 ¼ IxRð0Þ IyRð0Þ þ > ; VRð0Þ VRð0Þ

:

ð8:59Þ

506

8 Small-Signal Stability Analysis of Power Systems

Power relationship of the inverter is VxI IxI þ VyI IyI ¼ XcI Id2 þ kI Id VI cos b:

ð8:60Þ

After linearization we have VxIð0Þ DIxI þ VyIð0Þ DIyI ¼ IxIð0Þ DVxI IyIð0Þ DVyI þ 2XcI Idð0Þ DId þ kI VIð0Þ cos bð0Þ DId þ kI Idð0Þ cos bð0Þ DVI

ð8:61Þ

kI Idð0Þ VIð0Þ sin bð0Þ Db: Similarly from the third equation in (8.53) and after linearization we have IxIð0Þ DIxI þ IyIð0Þ DIyI ¼ kI2 Idð0Þ DId :

ð8:62Þ

Again, substituting (8.51) into (8.61) to cancel DVI, from (8.61) and (8.62) we obtain the following matrix form for current and voltage deviation DII ¼ CI Dxd þ DI DVI ;

ð8:63Þ

where

DIxI DII ¼ ; DIyI

DVxI DVI ¼ : DVyI

ð8:64Þ

Equation (8.58) and (8.63) form the algebraic equation of the DC system DId ¼ Cd Dxd þ Dd DVd ;

ð8:65Þ

where

DIR DId ¼ ; DII

DVR DVd ¼ ; DVI

CR Cd ¼ ; CI

DR Dd ¼ 0

0 : ð8:66Þ DI

When different mathematical models of the DC system are used, we can follow similar procedures to derive linearized equations like (8.52) and (8.65).

8.3 8.3.1

Steps in Small-Signal Stability Analysis Network Equation

For convenience of expression, we write the network equation of (8.36) in the form of block matrices. Noting that the network equation is itself linear, we can write the linear equation for the relationship between deviation of node injection current and node voltage in x–y coordinates directly, to be

8.3 Steps in Small-Signal Stability Analysis

2

3 2 DI1 Y11 6 .. 7 6 .. 6 . 7 6 . 6 7 6 6 DIi 7 ¼ 6 Yi1 6 . 7 6 . 4 . 5 4 . . . DIn Yn1

507

Y1i .. . Yii .. . Yni

32 3 Y1n DV1 .. 76 .. 7 . 76 . 7 76 7 Yin 76 DVi 7; 7 6 .. 54 .. 7 5 . . DVn Ynn

ð8:67Þ

where DIi ¼

DIxi ; DIyi

DVi ¼

DVxi ; DVyi

Yij ¼

Gij Bij

Bij ; Gij

ð8:68Þ

i; j ¼ 1; 2; . . . ; n: For load nodes, we can substitute the relationship between deviation of injected current and node voltage into the above equation to cancel the current deviation at the load node. Assuming load is connected at node i, then the network equation after canceling this node is just a simple correction to the original network equation of (8.67): current deviation at node i becomes zero, the ith diagonal block in the network admittance matrix changes to Yii Yli and nothing more. Without loss of generality, we assume the sequence of nodes in the network is: firstly each generator node, then each SVC node followed by two-terminal nodes of TCSC, then AC bus nodes of each HVDC transmission line (the node on the rectifier side first and inverter side second), and finally the remaining nodes. Canceling current deviation of all load nodes, we have the following block-matrix form of the network equation 2

3 2 DIG YGG 6 DIS 7 6 YSG 6 7 6 6 DIT 7 ¼ 6 YTG 6 7 6 4 DID 5 4 YDG 0 YLG

YGS YSS YTS YDS YLS

YGT YST YTT YDT YLT

YGD YSD YTD YDD YLD

32 3 YGL DVG 6 7 YSL 7 76 DVS 7 7 6 YTL 76 DVT 7 7; YDL 54 DVD 5 YLL DVL

ð8:69Þ

where DIG and DVG are vectors consisting of deviation of injected current and node voltage of all generators, respectively; DIS and DVS vectors of deviation of node injection current and node voltage of all SVC nodes; DIT and DVT all TCSC nodes; DID and DVD those at AC busbars of all converters; DVL associated with voltage of remaining nodes. All those vectors can be written as T T 9 DIG ¼ ½ DIg1 DIg2 ; DVG ¼ ½ DVg1 DVg2 > > > > DIS ¼ ½ DIs1 DIs2 T ; DVS ¼ ½ DVs1 DVs2 T > > = T T : ð8:70Þ DIT ¼ ½ DIt1 DIt2 ; DVT ¼ ½ DVt1 DVt2 > T T> > DID ¼ ½ DId1 DId2 ; DVD ¼ ½ DVd1 DVd2 > > > ; DVL ¼ ½ DV1 DV2 T

508

8.3.2

8 Small-Signal Stability Analysis of Power Systems

Linearized Differential Equations of Whole Power System

Equations of all generation units are formed from (8.31) and (8.29) to be dDxG ¼ AG DxG þ BG DVG ; dt

ð8:71Þ

DIG ¼ CG DxG þ DG DVG ;

ð8:72Þ

where )

AG ¼ diagfAg1

Ag2

g;

BG ¼ diagfBg1

Bg2

g

CG ¼ diagfCg1

Cg2

g;

DG ¼ diagfDg1

Dg2

g

:

ð8:73Þ

Equations (8.40) and (8.42) of each SVC can form equations of all SVCs to be dDxS ¼ AS DxS þ BS DVH ; dt

ð8:74Þ

DIS ¼ CS DxS þ DS DVS ;

ð8:75Þ

where AS ¼ diagfAs1

As2

g;

BS ¼ diagfBs1

Bs2

g

CS ¼ diagfCs1

Cs2

g;

DS ¼ diagfDs1

Ds2

g

) :

ð8:76Þ

TCSC equations are formed from (8.46) and (8.48) ðA s1 IÞvA ¼ ðlA s1 ÞvA ðA s2 IÞvA ¼ ðlA s2 ÞvA

) ;

ð8:77Þ

DIT ¼ CT DxT þ DT DVT ;

ð8:78Þ

where AT ¼ diagfAt1

At2

g;

BT ¼ diagfBt1

Bt2

g

CT ¼ diagfCt1

Ct2

g;

DT ¼ diagfDt1

Dt2

g

) :

ð8:79Þ

All two-terminal HVDC transmission lines have the following equations dDxD ¼ AD DxD þ BD DVD ; dt

ð8:80Þ

DID ¼ CD DxD þ DD DVD ;

ð8:81Þ

8.3 Steps in Small-Signal Stability Analysis

509

where AD ¼ diagfAd1 CD ¼ diagfCd1

Ad2 Cd2

g; g;

BD ¼ diagfBd1 Bd2 DD ¼ diagfDd1 Dd2

) g : g

ð8:82Þ

Substituting (8.72), (8.75), (8.78), and (8.81) into (8.69) to cancel DIG, DIS, DIT, and DID, together with (8.71), (8.74), (8.77), and (8.80), we can obtain the matrix formulations, as required by (8.3): 9 9 = Dx ¼ ½DxG DxS DxT DxD T > > > > > > T; > Dy ¼ ½DVG DVS DVT DVD DVL > > > > 2 3 > > AG 0 0 0 > > > > 6 0 A 7 > 0 0 7 S > ~ 6 > > A¼6 7 > > 4 0 5 0 AT 0 > > > > > 0 0 0 AD > > 2 3> > > CG 0 0 0 > 2 3 > BG 0 0 0 0 6 7= 0 C 0 0 7 6 S 6 7 >: ð8:83Þ 0 07 7 ~ 6 ~ 6 0 BS 0 6 7> B¼6 ; C ¼ 0 0 C 0 7 T > 6 7> 4 0 0 BT 0 0 5 > 6 7> 0 0 CD 5 > > 4 0 > > 0 0 0 BD 0 > > > 0 0 0 0 > > 2 3 > > > YGG DG YGS YGT YGD YGL > > > 6 7 > > Y Y D Y Y Y 6 SG SS S ST SD SL 7 > > 6 7 > ~ 6 > 7 > D ¼ 6 YTG YTS YTT DT YTD YTL 7 > > > 6 7 > > YDS YDT YDD DD YDL 5 4 YDG > > ; YLG YLS YLT YLD YLL ~ ~ ~ ~ Obviously, A, B, and C are sparse block matrices, as is D which is also an ~ ~ ~ ~ admittance matrix. Using matrices A, B, C and D, and from (8.5) we can obtain the system state matrix A. By now, we have obtained the linearized equations of a power system at a steady-state operating point. Finally, we would like to point out: 1. If this linearized system is asymptotically stable, i.e., the real part of all eigenvalues of matrix A are negative, the actual nonlinear system is asymptotically stable at this equilibrium point. 2. The method used to form matrix A is different in various commercial software packages. In the above, we only give one way to form it to introduce the principles and techniques used in forming matrix A [189, 190]. There are various ~ ~ ~ ~ alternative formats of matrices A, B, C, and D in (8.83) that are related with the sequence order of state variables, format of network equations, algebraic equations of various dynamic components and ways to treat the network equations.

510

8 Small-Signal Stability Analysis of Power Systems

Different methods determine the complexity of, and flexibility in developing, the program, but do not change the resulting eigensolution. 3. In the formation of the above linearized equations, we have considered generation units, SVC, TCSC, two-terminal HVDC transmission lines. We can also treat other dynamic components in power systems in similar ways. For example, for dynamic components (such as induction motor loads) we can derive their linearized equation in the same way as treating generators; for multiterminal HVDC transmission lines, we can obtain the linearized equation as we have done in treating two-terminal HVDC transmission lines. We can then arrange the linearized equations into the equation of the whole power system. 4. Matrix A, as formed, must have a zero eigenvalue. A zero eigenvalue exists because the absolute angle of the generator rotors is not unique. In other words, there is a redundant rotor angle in a power system model. In fact, power distribution among generators is determined by the relative rotor angle of generators. If the absolute rotor angle of all generators is added to by a fixed value, the power distribution does not change at all. Hence this does not affect system stability. To eliminate the zero eigenvalue, we only need to choose the rotor angle of any particular generator as a reference and then use the relative rotor angle of other generators as the new state variable. In doing so, the dimension of state matrix A and the corresponding state variable vector is reduced by one. 5. In the case that all generator torques are not directly related to rotor speed, i.e., when there is no damping term in the swing equation and the governing effect is ignored; matrix A will have another zero eigenvalue. Similarly, to remove this zero eigenvalue, we only need to choose the rotor speed of any generator as a reference and use the relative rotor speed of the other generators as new state variables. Again, in doing so, the order of matrix A and the corresponding state variable vector is reduced by one. Knowing the origin of the zero eigenvalues, we do not have to apply the treatment of (4) and (5) above, but simply eliminate the corresponding zero eigenvalues in our computational results. However, due to errors in load flow calculation and in the computation of eigenvalues, we should note that the theoretically zero eigenvalues will be computed as eigenvalues with very small magnitude.

8.3.3

Program Package for Small-Signal Stability Analysis

From what we have discussed previously, we can develop a program package for small-signal stability analysis of an AC/DC power system with FACTS devices such as SVC and TCSC installed. Basic steps in developing the stability analysis program are: 1. Load flow calculation at a given steady-state operating condition of the power system. This includes finding the voltage, current, and power at each node in the system. 2. Formation of the admittance matrix in (8.67). 3. Treatment of load. Load power and voltage at steady-state operation are known to be P(0), Q(0), Vx(0), and Vy(0). From parameters of the static voltage characteristics of load, we calculate matrix elements Gxx, Bxy, Byx, and Gyy in (8.34) from

8.3 Steps in Small-Signal Stability Analysis

511

(8.36) or (8.37). These will be used to adjust diagonal blocks related to loads in the admittance matrix. 4. Establishment of linearized equations of dynamic components in the system. Firstly we calculate initial values of all variables of generators from (7.74)– (7.78) and (7.118)–(7.122). Then we can form matrices Ag , BIg , BVg , Pg , and Zg in (8.20) and (8.21) as well as matrices Tg(0), RVg, and RIg in (8.26) and (8.28). Finally we calculate matrices from (8.30) and (8.32) to establish the linearized equation of generators. In a similar way we can obtain linearized equations of all dynamic components in the power system. 5. Formation of system state matrix A from (8.5). This is obtained by forming ~ ~ ~ ~ matrices A, B, C, and D from (8.71)–(8.83). 6. Calculation of all eigenvalues of the state matrix A by using the QR method [187, 188]. The result of this calculation is used to determine system smallsignal stability. The QR method to calculate all eigenvalues of matrix A will be introduced in Sect. 8.4. [Example 8.1] Single-line diagram of the 9-node power system, line data, generator parameters, and load flow at a steady-state operating condition is given in Fig. 7.7, Tables 7.5–7.7, respectively. System frequency is 60 Hz. All loads in the system are modeled by constant impedance. Generator 1 uses the classical model, generators 2 and 3 the double-axis model with self-excited potential-source excitation system. Parameters of the exciter are: XC ¼ 0;

KA ¼ 200;

TR ¼ 0:03 s;

TA ¼ 0:02 s;

TB ¼ 10:0 s;

TC ¼ 1:0 s:

In addition, the damping coefficient of each generator Di is 1.0. [Solution] In the following, we will demonstrate the process of small-signal stability analysis for the example power system. For simplicity of expression, a blank in the matrix will represent either zero or a zero matrix: (1) From load flow calculation, (7.74)–(7.78) and (7.118)–(7.122) we calculate initial values of all variables of generators shown in Table 8.1. Equivalent admittance of loads has been calculated in example 8.1 which is included in the power network model. (2) Establishment of linearized equations of generators using the method introduced in Sect. 8.2.1. Generator 1 We can calculate coefficient matrices in (8.20), (8.21), (8.26), and (8.28) as follows: Table 8.1 Initial values of generator variables

1 2 3

d(0)

Vq(0)

Vd(0)

Iq(0)

Id(0)

Efq(0)

E0 q(0)

E0 d(0)

2.27165 61.09844 54.13662

1.03918 0.63361 0.66607

0.04122 0.80571 0.77909

0.67801 0.93199 0.61941

0.28716 1.29015 0.56147

1.78932 1.40299

1.05664 0.78817 0.76786

0.62220 0.62424

512

8 Small-Signal Stability Analysis of Power Systems

Dd1 0:0 376:99112 0:0 0:0 0:0 0:0 ; BI1 ¼ ; BV1 ¼ ; Do1 0:0 0:02115 0:0 0:02235 0:0 0:0 0:0 0:0 0:0 0:0608 P1 ¼ ; Z1 ¼ ; 0:0 0:0 0:0608 0:0 0:03964 0:99921 1:03918 0:0 T1ð0Þ ¼ ; RV1 ¼ ; 0:99921 0:03964 0:04122 0:0 0:67801 0:0 RI1 ¼ : 0:28716 0:0 A1 ¼

From (8.32), (8.30), and the matrices above, we can obtain matrices in the linearized equation of the generator of (8.29) and (8.31) to be A1 ¼

0:0

376:99112

;

0:0

0:0

B1 ¼ ; 0:38198 0:02115 0:01457 0:36729 17:36532 0:0 0:0 16:44737 C1 ¼ ; D1 ¼ : 0:68886 0:0 16:44737 0:0 Generator 2 In a similar way, we can obtain coefficient matrices of the linearized equation of generator 2 to be ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ΔE fq 2 ⎢ ⎢ ΔVR 2 ⎢ ΔVM 2 ⎢⎣

Δδ 2 Δω2 ΔEq′ 2 A 2 = ΔEd′ 2

376.99112 −0.07813

⎤ ⎥ ⎥ ⎥ −0.16667 0.16667 ⎥ −1.86916 ⎥, −0.10000 −4.90000 −1000.0 ⎥ ⎥ −50.0000 −10000.0 ⎥ −33.33333⎥⎦ −0.07281 −0.10079

⎡ ⎤ ⎡ ⎤ ⎢ −0.05422 −0.06935⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ −0.12933 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 1.24785 ⎥ , BV 2 = ⎢ BI 2 = ⎢ ⎥, ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 26.20186 20.60518⎥ ⎣ ⎦ ⎣ ⎦

8.3 Steps in Small-Signal Stability Analysis

⎡ P2 = ⎢ ⎣

1.0 1.0

513

⎤ ⎡ 0.0000 0.1969⎤ ⎥ , Z 2 = ⎢ −0.1198 0.0000 ⎥ , ⎦ ⎣ ⎦

⎡0.87545 −0.48331⎤ T2(0) = ⎢ ⎥, ⎣ 0.48331 0.87545 ⎦

⎡ 0.63361 RV 2 = ⎢ ⎣ −0.80571

⎤ ⎡ 0.93199 ⎥ , RI 2 = ⎢ −1.29015 ⎦ ⎣

⎤ ⎥, ⎦

376.99112 ⎡ ⎤ ⎢ −0.58783 −0.07813 −0.52542 0.25140 ⎥ ⎢ ⎥ ⎢ −0.86982 ⎥ −1.24624 0.16667 ⎢ ⎥ −8.20664 A 2 = ⎢ 4.01549 ⎥, ⎢ −0.10000 −4.90000 −1000.0 ⎥ ⎢ ⎥ −50.0000 −10000.0 ⎥ ⎢ ⎢ −33.33333⎥⎦ ⎣

⎡ ⎢ −0.08958 ⎢ ⎢ 0.52177 ⎢ B 2 = ⎢ 5.54816 ⎢ ⎢ ⎢ ⎢ 32.89707 ⎣

⎡ 7.25066 C2 = ⎢ ⎣1.14659

⎤ 0.56646 ⎥⎥ 0.94512 ⎥ ⎥ −3.06294 ⎥ , ⎥ ⎥ ⎥ 5.37531 ⎥⎦

7.30761 −2.45458 −4.03428 −4.44617

⎡ −1.38295 −7.58377 ⎤ D2 = ⎢ ⎥. ⎣ 5.84220 1.38295 ⎦

⎤ ⎥, ⎦

514

8 Small-Signal Stability Analysis of Power Systems

Generator 3 Similarly we have Δδ 3 ⎡ ⎢ Δω3 ⎢ ΔEq′ 3 ⎢ ⎢ A 3 = ΔEd′ 3 ⎢ ΔE fq 3 ⎢ ⎢ ΔVR 3 ⎢ ΔVM 3 ⎢⎣

376.99112 −0.16611 −0.10289 −0.09327 −0.16978

⎤ ⎥ ⎥ ⎥ 0.16978 ⎥ −1.66667 ⎥, −0.10000 −4.90000 −1000.0 ⎥ ⎥ −50.0000 −10000.0 ⎥ −33.33333⎥⎦

⎡ ⎤ ⎡ ⎤ ⎢ −0.11076 −0.13396 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ −0.19205 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 1.67967 ⎥ , BV 3 = ⎢ BI 3 = ⎢ ⎥, ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 25.33623 21.66071⎥ ⎣ ⎦ ⎣ ⎦ ⎡ P3 = ⎢ ⎣

1.0 1.0

⎤ ⎡ 0.0000 0.2500 ⎤ ⎥ , Z3 = ⎢ −0.1813 0.0000 ⎥ , ⎦ ⎣ ⎦

⎡0.81042 −0.58586 ⎤ T3(0) = ⎢ ⎥, ⎣0.58586 0.81042 ⎦ ⎡ 0.66607 RV 3 = ⎢ ⎣ −0.77907

⎤ ⎡ 0.61941 ⎥ , RI 3 = ⎢ −0.56147 ⎦ ⎣

376.99112 ⎡ ⎢ −0.83288 −0.16611 −0.71383 0.44257 ⎢ ⎢ −0.82530 −1.22910 0.16978 ⎢ −8.38533 A 3 = ⎢ 4.47508 ⎢ −0.10000 −4.90000 ⎢ −50.0000 ⎢ ⎢ ⎣

⎤ ⎥, ⎦ ⎤ ⎥ ⎥ ⎥ ⎥ ⎥, −1000.0 ⎥ ⎥ −10000.0 ⎥ − 33.33333⎥⎦

8.3 Steps in Small-Signal Stability Analysis

515

⎡ ⎤ ⎢ −0.07633 0.80903 ⎥ ⎢ ⎥ ⎢ 0.62061 0.85849 ⎥ ⎢ ⎥ B3 = ⎢ 5.44492 −3.93616 ⎥ , ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 33.22292 2.71085 ⎥ ⎣ ⎦ ⎡ 4.87039 C3 = ⎢ ⎣ 0.45951

4.47003

−2.34342

⎤ ⎥, ⎦

−3.23141 −3.24166

⎡ −0.71964 −4.99548⎤ D3 = ⎢ ⎥. ⎣ 4.52023 0.71964 ⎦

(3) Linearized equations of the system Obviously, matrices in (8.3) (see 8.83) are " A~ ¼ AG ;

B~ ¼ ½ BG

0 ;

C~ ¼

CG

# ;

" D~ ¼

0 2

A1

0

6 6 AG ¼ 6 0 4

A2

0

0

2

D1 6 6 DG ¼ 6 0 4 0

0 D2 0

0

3

7 7 0 7; 5

A3 0

2

B1

0

0

3

6 6 BG ¼ 6 0 4

B2

7 7 0 7; 5

0

0

B3

YGG DG

YGL

YLG 2

YLL C1

0

6 6 CG ¼ 6 0 4

C2

0

0

# ; 0

3

7 7 0 7; 5

C3

3

7 7 0 7; 5

D3

33.80848 ⎡ ⎤ ⎢ −33.80848 ⎥ ⎢ ⎥ ⎢ ⎥ 1.38295 23.58377 YGG − DG = ⎢ ⎥, − 21.84220 − 1.38295 ⎢ ⎥ ⎢ 0.71964 22.06033 ⎥ ⎢ ⎥ −21.58508 −0.71964 ⎥⎦ ⎣⎢

39.30889 −1.36519 −11.60410 −1.94219 −10.51068 ⎡ 3.30738 ⎢ −39.30889 3.30738 11.60410 −1.36519 10.51068 −1.94219 ⎢ ⎢ −1.36519 −11.60410 3.81379 17.84263 −1.18760 −5.97513 ⎢ 17.84263 3.81379 11.60410 − 1.36519 − 5.97513 −1.18760 ⎢ ⎢ −1.94219 −10.51068 4.10185 16.13348 −1.28201 ⎢ 10.51068 − 1.94219 − 16.13348 4.10185 5.58824 ⎢ YLL = ⎢ −1.18760 −5.97513 2.80473 35.44561 1.61712 −13.69798 ⎢ 5.97513 −1.18760 −35.44561 2.80473 13.69798 −1.61712 ⎢ ⎢ −1.61712 −13.69798 3.74119 23.64239 −1.15509 ⎢ ⎢ 13.69798 −1.61712 −23.64239 3.74119 9.78427 ⎢ 1.28201 5.58824 1.15509 9.78427 2.43710 − − − − ⎢ ⎢ 5.58824 1.28201 9.78427 1.15509 32.15386 − − − ⎣

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ −5.58824 ⎥ ⎥ −1.28201⎥ ⎥, ⎥ ⎥ −9.78427 ⎥ ⎥ −1.15509 ⎥ ⎥ 32.15386 ⎥ 2.43710 ⎦⎥

516 8 Small-Signal Stability Analysis of Power Systems

376.99112 ⎡ ⎢−0.06244 −0.02115 0.03518 0.02485 ⎢ ⎢ 376.99112 ⎢ −0.19905 −0.07813 −0.24571 ⎢ 0.11924 ⎢ 0.20108 −0.32411 −0.52072 ⎢ 1.13669 0.44018 ⎢−0.62862 ΔE fq 2 ⎢ ⎢ ΔVR 2 ⎢ A= ⎢ −1.48647 15.66751 ΔVM 2 ⎢ 1.23883 Δδ 3 ⎢ ⎢ 0.17777 0.10937 Δω3 ⎢ 0.20654 0.16072 0.21160 ΔEq′ 3 ⎢ 0.22459 ⎢ 0.15877 −1.06680 ΔEd′ 3 ⎢ −0.96568 ΔE fq 3 ⎢ ⎢ ΔVR 3 ⎢ 4.92556 −0.73652 ΔVM 3 ⎢⎣ 0.95082

Δδ1 Δω1 Δδ 2 Δω2 ΔEq′ 2 ΔEd′ 2

4.68110

−0.11476 −0.01561 1.32599

13.92522

−0.10000 −4.90000 −50.0

−1000.0 −10000.0 −33.33333

0.03553 0.13428 0.37374

0.01775

−0.06960 −0.04154 0.93904

−0.01913

−0.21429

13.98229

11.82683

4.10804 3.08961 376.99112 −0.38432 −0.16611 −0.40080 0.14517 −0.38531 −0.62589 −0.04685 2.03248 0.43705 −4.99508

0.24764

0.07982 0.12303 −0.50807

0.03212 −0.04750 −4.61925 0.16667

0.02726

−0.02020

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥, ⎥ ⎥ ⎥ ⎥ ⎥ 0.16978 ⎥ ⎥ −0.10000 −4.90000 −1000.0 ⎥ ⎥ −50.0 −10000.0 ⎥ −33.3333 ⎥⎦

8.3 Steps in Small-Signal Stability Analysis 517

518

8 Small-Signal Stability Analysis of Power Systems

YLG

−17.36111 ⎡ ⎤ ⎢17.36111 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ =⎢ ⎥, −16.00000 ⎢ ⎥ 16.00000 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ − 17.06485 ⎢ ⎥ ⎢ ⎥ 17.06485 ⎣ ⎦

−17.36111 ⎡ ⎢17.36111 ⎢ ⎢ YGL = ⎢ ⎢ ⎢ ⎢ ⎣⎢

⎤ ⎥ ⎥ ⎥ ⎥. ⎥ −17.06485⎥ ⎥ 17.06485 ⎦⎥

−16.00000 16.00000

From (8.5) we can obtain the state matrix A ¼ AG ½ BG

0

YGG DG YLG

YGL YLL

1

CG 0

1 ¼ AG þ BG ½YGG DG YGL Y1 LL YLG CG :

(4) Eigenvalues and eigenvectors of state matrix A All eigenvalues of A are obtained by using the QR method as l1 ¼ 53:05299;

l2 ¼ 51:80217;

l5;6 ¼ 0:75497 j12:86370; l9 ¼ 5:58205;

l4 ¼ 28:21401;

l7;8 ¼ 0:15154 j8:67125;

l10 ¼ 3:72276;

l11;12 ¼ 1:13701 j0:91540; l15 ¼ 0:04571;

l3 ¼ 30:41762;

l13;14 ¼ 0:48432 j0:657417;

l16 ¼ 0:00000:

Obviously, except the zero eigenvalue we expect, the real part of the remaining eigenvalues is negative. Hence the power system is stable in terms of small-signal stability.

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

8.4

519

Eigenvalue Problem in Small-Signal Stability Analysis

Nonlinear system stability, when the system is subject to small disturbances, can be analyzed from the stability of its linearized system as determined by the eigenvalues of state matrix A. Hence, in the following, we shall introduce the method of eigensolution analysis for a state matrix A. From the discussion above we can see that state matrix A is a real asymmetric matrix. Hence, in the following, all our discussion will be under the condition that A 2 Rnn . We denote the set of complex numbers by C, n-dimensional complex vector space (column vector) by Cn , and set of all m-row n-column complex matrices by Cmn . Operations of scalar multiplication, addition and multiplication of complex matrices are similar to those for real matrices. However, transposition of a complex matrix is taken as conjugate transposition (denoted by superscript H), i.e., C ¼ AH ) cij ¼ a^ji . Dot product of n-dimensional vector x and y is n P s ¼ xH y ¼ x^i yi . In addition, unit vector (normalized vector) under norm p is a i¼1

vector x satisfying kxkp = 1. For example, unit vectors x under 1-norm, 2-norm, and infinite norm, respectively, are 9 jxj1 ¼ jx1 j þ þ jxn j ¼ 1 > > > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ pﬃﬃﬃﬃﬃﬃﬃﬃ = 2 2 H jxj2 ¼ jx1 j þ þ jxn j ¼ x x ¼ 1 : ð8:84Þ > > > ; jxj ¼ max jx j ¼ 1 1

1in

i

The process to convert a vector to a unit vector is called normalization.

8.4.1

Characteristics of State Matrix Given by Its Eigensolution

8.4.1.1

Eigenvalue

For a scalar l 2 C and vector v 2 Cn , if equation Av ¼ lv

ð8:85Þ

has a nonsingular solution (i.e., v 6¼ 0), l is an eigenvalue of matrix A. To calculate eigenvalues, (8.85) can be written as ðA lIÞv ¼ 0:

ð8:86Þ

A sufficient and necessary condition for existence of a nonsingular solution of the equation is detðA lIÞ ¼ 0:

ð8:87Þ

520

8 Small-Signal Stability Analysis of Power Systems

Expansion of the determinant in the above equation gives the following polynomial equation a0 þ a1 l þ þ an1 ln1 þ ð1Þn ln ¼ 0:

ð8:88Þ

It is called the characteristic equation of matrix A. The polynomial on the left side of the above equation is called the characteristic polynomial. Because the coefficient of ln is nonzero, there are a total of n roots. The set of all roots is called the spectrum and is denoted by l(A). If lðAÞ ¼ fl1 ; ; ln g, we have detðAÞ ¼ l1 l2 ln : In addition, if we define the trace of A to be trðAÞ ¼

n X

aii :

i¼1

Then trðAÞ ¼ l1 þ l2 þ þ ln , can be proved. Eigenvalues of a real asymmetric matrix can be real or complex numbers. Complex eigenvalues always appear in the form of conjugate pairs. Moreover, similar matrices have the same eigenvalues and transposition of a matrix does not change its eigenvalues.

8.4.1.2

Eigenvectors

For any eigenvalue li, any nonzero vector vi 2 Cn satisfying equation Avi ¼ li vi

i ¼ 1; 2; . . . ; n

ð8:89Þ

is called a right eigenvector of matrix A corresponding to eigenvalue li. Since it is a homogenous equation, kvi (k is a scalar) is also the solution of the equation to be a right eigenvector of matrix A corresponding to eigenvalue li. In the following (unless explicitly stated otherwise) ‘‘eigenvector’’ refers to ‘‘right eigenvector.’’ An eigenvector defines a one-dimensional subspace that remains invariable under the operation of left multiplication by matrix A. Similarly, any nonzero vector ui 2 Cn satisfying equation AT ui ¼ li ui

i ¼ 1; 2; . . . ; n

ð8:90Þ

is called a right eigenvector of matrix AT corresponding to eigenvalue li. Taking transposition on both sides of equation, we have uTi A ¼ li uTi ;

i ¼ 1; 2; . . . ; n:

ð8:91Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

521

We call row vector uTi the left eigenvector of matrix A corresponding to eigenvalue li. To express the eigensolution of matrix A clearly, we form a diagonal matrix L consisting of all eigenvalues of matrix A, a matrix XR of all right eigenvectors arranged in columns, a matrix XL of all left eigenvectors in rows. That is 9 l2 ln g > = vn : > ; T un

L ¼ diagfl1 XR ¼ ½v1 v2 X L ¼ ½ u1

u2

ð8:92Þ

These three n-dimensional square matrices are called modal matrices. Using (8.92), (8.89), and (8.91) can be expressed in the following matrix form: ) AXR ¼ XR L : XL A ¼ LXL

ð8:93Þ

Premultiplying the first equation above by XL, and postmultiplying the second by XR, we have ðXL XR ÞL ¼ LðXL XR Þ

ð8:94Þ

or lj uTi vj ¼ li uTi vj ;

i; j ¼ 1; 2; . . . ; n:

Obviously, left and right eigenvectors corresponding to different eigenvalues are orthogonal; for the same eigenvalue their product is a nonzero number that can be converted to 1 after normalization of left and right eigenvectors. That is uTi vj ¼

0 1

i 6¼ j : i¼j

ð8:95Þ

Please note that uTi vj is not the normal inner product of two vectors. The matrix form of above equation is XL XR ¼ I;

X1 L ¼ XR :

ð8:96Þ

From (8.93) and (8.96) we have X1 R AXR ¼ L:

ð8:97Þ

522

8 Small-Signal Stability Analysis of Power Systems

8.4.1.3

Free Movement of Dynamic System

From the state equation, (8.4), we can see that the rate of change of every state variable is a linear combination of all state variables. Hence due to the coupling among state variables, it is difficult to clearly see the system movement. To cancel the coupling among state variables, we introduce a new state variable vector z. Its relationship with the original state variable vector Dx is defined to be Dx ¼ XR z:

ð8:98Þ

Substituting the above equation into (8.4) and using (8.14), the state equation can be written as dz ¼ Lz: dt

ð8:99Þ

The difference from the original state equation is that L is a diagonal matrix, while A usually is not. Equation (8.99) can be expressed as n decoupled first-order differential equations dzi ¼ li z i ; dt

i ¼ 1; 2; . . . ; n:

ð8:100Þ

Its solution in the time domain is zi ðtÞ ¼ zi ð0Þeli t ;

ð8:101Þ

where initial values of zi, zi(0) can be expressed from (8.98) by uTi and Dx(0) zi ð0Þ ¼ uTi Dxð0Þ:

ð8:102Þ

Substituting (8.101) and (8.102) into the transformation of (8.98), we have the solution of the original state vector in the time domain to be Dx ¼

n X

vi zi ð0Þeli t ;

i¼1

where solution of the ith state variable in the time domain is Dxi ðtÞ ¼ vi1 z1 ð0Þel1 t þ vi2 z2 ð0Þel2 t þ þ vin zn ð0Þeln t ; ¼ 1; 2; . . . ; n;

ð8:103Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

523

where vik is the ith element of vector vk. The above equation is the time response of system free movement expressed by eigenvalues, left and right eigenvectors. Eigenvalue li represents the ith mode of the system, with corresponding time characteristic eli t . Hence, time response of system free movement is the linear combination of n system modes. Therefore, system stability is determined by the eigenvalues: (1) A real eigenvalue represents a nonoscillatory mode. A negative real eigenvalue is a decaying mode and the bigger its absolute value, the faster it decays. A positive real eigenvalue indicates nonperiodic instability. Eigenvectors, and z (0), corresponding to real eigenvalues are real valued. (2) Complex eigenvalues always appear in conjugate pairs, i.e., l ¼ s jo:

ð8:104Þ

Each pair of complex eigenvalues represents an oscillation mode. Eigenvectors, and z(0), corresponding to complex eigenvalues are complex valued. Hence ða þ jbÞeðsjoÞt þ ða jbÞeðsþjoÞt ¼ est ð2a cos ot þ 2b sin otÞ should exhibit as est sin(ot + y). Obviously, the real part of the eigenvalue describes system oscillation damping and the imaginary part gives the frequency of oscillation. A negative real part is a decaying oscillation mode and positive an increasing oscillation mode. Oscillation frequency (Hz) is f ¼

o : 2p

ð8:105Þ

Damping ratio is defined to be s z ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ : 2 s þ o2

ð8:106Þ

This determines the decay rate property of the oscillation magnitude.

8.4.2

Modal Analysis of Linear Systems

8.4.2.1

Mode and Eigenvector

From the discussion above we know that the relationship among system time response, vectors Dx and z are 9 n X > DxðtÞ ¼ XR zðtÞ ¼ ½v1 v2 vn zðtÞ ¼ vi zi ðtÞ = : ð8:107Þ i¼1 > ; T zðtÞ ¼ XL DxðtÞ ¼ ½u1 u2 un DxðtÞ

524

8 Small-Signal Stability Analysis of Power Systems

Variables Dx1, Dx2, . . ., Dxn are the original state variables depicting system dynamics. Variables z1 ; z2 ; ; zn are state variables after transformation, each of which represents a mode of the system. From the first equation of (8.107) we can see that right eigenvectors decide the form of exhibition of each mode, i.e., when a specific mode is excited, the relative activity of each state variable is described by the right eigenvector. For example, when the ith mode is excited, the kth element vki of right eigenvector vi gives the level of influence of this mode on state variable xk. Magnitude of each element in vi represents the level of activity of each of the n state variables resulting from the ith mode; while the angle of each element represents the effect of the mode on the phase shift of each state variable. From the second equation of (8.107) we can see that the left eigenvector uTi represents the way the original state variables combine to effect the ith mode. Therefore, the kth element in the right eigenvector vi measures the level of activity of state variable xk in the ith mode; while the kth element of the left eigenvector uTi weights the contribution of the exhibited activity to the ith mode.

8.4.2.2

Eigenvalue Sensitivity

Firstly we consider the sensitivity of an eigenvalue to each element akj in matrix A (the k-row, j-column element in A). Taking partial derivatives to akj on both sides of (8.89), we have @A @vi @li @vi vi þ A ¼ vi þ li : @akj @akj @akj @akj

ð8:108Þ

Premultiplying both sides of the above equation by uTi and from (8.91) and (8.95) we can obtain @li @A ¼ uTi vi : @akj @akj

ð8:109Þ

Obviously, in @A=@akj the kth-row, jth-column element is 1 and remaining elements are zero. Hence @li ¼ uki vji @akj

ð8:110Þ

where, nji is the jth element in vi and uki is the kth element in ui. Assuming a is a scalar, A(a) is an n-order square matrix with elements being akj(a) and for all k and j, akj (a) is a differentiable function, we have dAðaÞ ¼ da

dakj ðaÞ : da

ð8:111Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

525

Therefore, similarly we can find the eigenvalue sensitivity to scalar a to be @li @A ¼ uTi vi : @a @a 8.4.2.3

ð8:112Þ

Participation Factor

To determine the relationship between state variables and system modes, we establish a so-called participation matrix P by combining right and left eigenvectors to measure the level of coupling between state variables and system modes. l1 u11 n11 Dx1 6 .. 6 . 6 6 P ¼ Dxk 6 uk1 nk1 6 6 .. 4 . Dxn un1 nn1 2

.. .

li u1i n1i .. .

.. .

.. .

uki nki .. . uni nni

.. .

ln 3 u1n n1n .. 7 . 7 7 7 ukn nkn 7: 7 .. 7 . 5

ð8:113Þ

unn nnn

Element pki = uki nki in matrix P is called a participation factor [193] that measures the level of participation of the ith mode and the kth state variable Dxk with each other. The ith row of matrix P, pi, is the participation vector of the ith mode. Since nki measures the level of activity of Dxk in the ith mode and uki weights the contribution of the activity to the mode, their product pki can measure the pure participation. The product of corresponding elements in left and right eigenvectors is a dimensionless result, independent of the dimensions selected for the eigenvectors. Assuming Dx(0) = ek, i.e., Dxk (0) = 1 and Dxj 6¼ k (0) = 0, from (8.102) we have zi (0) = uki. From (8.103) we can obtain Dxk ðtÞ ¼

n X

vki uki eli t ¼

i¼1

n X

pki eli t :

ð8:114Þ

i¼1

This equation shows that the ith mode excited by initial value Dxk (0) = 1 participates in response Dxk (t) with a participation coefficient pki. That is why it is called a participation factor. For all modes or all state variables, it is easy to prove that n X i¼1

pki ¼

n X

pki ¼ 1:

ð8:115Þ

k¼1

To set t = 0 in (8.114), we can easily obtain the summation of the kth-row elements of P to be 1. Summation of the ith-column elements of matrix P is equal to uTi vi

526

8 Small-Signal Stability Analysis of Power Systems

which is 1 according to (8.95). In addition, from (8.110) we can see that participation factor pki in fact is the sensitivity of eigenvalue li to diagonal element akk of matrix A, i.e., pki ¼

@li : @akk

8.4.3

Computation of Eigenvalues

8.4.3.1

QR Method

ð8:116Þ

Among numerical methods to compute all the eigenvalues of a general matrix, the QR method is usually the first choice. It was proposed by J. G. F. Francis in 1962, and has advantages such as strong robustness and fast speed of convergence. It has been found to be the most effective method of eigensolution so far. For a given A 2 Rnn and orthogonal matrix Q0 2 Rnn , we have the following iteration: A0 ¼ QT0 AQ0 ; k ¼ 1; 2; . . . ; Ak1 ¼ Qk Rk

ðQR decompositionÞ;

ð8:117Þ

A k ¼ R k Qk ; where each Qk 2 Rnn is an orthogonal matrix and Rk 2 Rnn upper triangular matrix. By an inductive approach we have Ak ¼ ðQ0 Q1 Qk ÞT AðQ0 Q1 Qk Þ:

ð8:118Þ

Hence each Ak is similar to A. Because matrix A has complex eigenvalues, Ak will not converge to a strict ‘‘eigenvalue exposed’’ upper triangular matrix, but satisfy a computational real Schur decomposition. An upper triangular matrix with diagonal elements being 1 1 blocks or 2 2 blocks is called an upper quasi-triangular matrix. Real Schur decomposition is a real operation to reduce a matrix to an upper quasi-triangular matrix. If A 2 Rnn , there exists an orthogonal matrix Q 2 Rnn to lead to 2

R11 6 0 6 QT AQ ¼ 6 .. 4 . 0

R12 R22 .. . 0

.. .

3 R1m R2m 7 7 .. 7; . 5 Rmm

ð8:119Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

527

where Rii is either a 1 1 matrix or 2 2 matrix. If it is 1 1, the element is an eigenvalue of matrix A; if 2 2, the eigenvalues of Rii are a pair of conjugate complex eigenvalue of A. To effectively complete a Schur decomposition, we select the initial orthogonal matrix of a similarity transformation as in (8.117) Q0 to make A0 become an upper Hessenberg matrix. Doing so, the computational complexity of one iteration is reduced from O(n3) to O(n2). In an upper Hessenberg matrix, except for the sub-diagonal elements, those below the diagonal are zero. For example, in a 6 6 upper Hessenberg matrix, nonzero elements are distributed as shown below: 2

6 6 6 6 6 6 4

3 7 7 7 7: 7 7 5

This form of matrix can be obtained by performing a series of Householder transformations. Since Householder transformation is a symmetrical orthogonal similarity transformation, the upper Hessenberg matrix obtained has the same eigenvalues as the original matrix. Finally, if values of elements of A have large differences, implementation of the iterative method could result in large computational errors in eigenvalues. The level of sensitivity of eigenvalue computation to round off can be reduced by a balancing operation. Since usually errors of eigensolution from numerical computation are proportional to a Euclidean norm, the idea of the balancing operation is to make the norm of corresponding rows and columns as close as possible through similarity transformation. Thus, the total norm of the matrix is reduced without changing the eigenvalues of the matrix. Implementation of the balancing operation is to determine the diagonal matrix D through O(n2) computation such that ~ A ¼ D1 AD ¼ ½c1 ; c2 ; . . . ; cn ¼ ½r1 ; r2 ; . . . ; rn T

ð8:120Þ

with k rik1 k cik1, i = 1,2, . . ., n. Diagonal matrix D is selected to have the form D ¼ diagfbi1 ; bi2 ; . . . ; bin g, where b is the floating-point base. Thus round off in ~ computing A is avoided. After A goes through the balancing operation, computation of eigenvalues will become more accurate. 8.4.3.2

The Power Method

In practical applications, often we do not need to compute all eigenvalues of matrix A, but only that with largest modulus (often called the dominant eigenvalue). The

528

8 Small-Signal Stability Analysis of Power Systems

power method is a very effective iterative method to calculate the dominant eigenvalue. Assuming that A 2 Cnn can be diagonalized and X1 A X = diag (l1, l2, . . ., ln), where X = [x1, x2, . . ., xn], jl1 j > jl2 j jln j. For a given initial unit vector under the 2-norm vð0Þ 2 Cn , the power method generates the following series of vectors v(k) 9 > zðkÞ ¼ Avðk1Þ > = ðkÞ ðkÞ ðkÞ ; v ¼z = z 2> > ðkÞ ðkÞ H ðkÞ ; l ¼ ½v Av

k ¼ 1; 2; . . . :

ð8:121Þ

Obviously, the series of vectors in the above iteration v(k) are unit vectors under the 2-norm. Because k ! l2 distðspanfvðkÞ g; spanfx1 gÞ ¼ O l1 and

l1 lðkÞ ¼ O

k ! l2 : l 1

Obviously, only l2/l1 < 1, when k ! 1, we have lðkÞ ! l1 ;

vðkÞ ! x1 :

ð8:122Þ

The power method is of linear convergence and its applicability depends on the ratio |l2| / |l|1, which reflects the rate of convergence. After the dominant eigenvalue of A is obtained by using the power method, we can compute the remaining eigenvalues through a deflation technique. There are many deflation methods but only a few of them are numerically stable. In the following, we shall introduce a deflation method based on similarity transformation. Assuming l1 and v1 are known, we can find a Householder matrix H1 to satisfy H1 v1 = k1 e1 and k1 6¼ 0. From A1 v1 = l1 v1, we have H1 A1 ðH1 1 H1 Þv1 ¼ l1 H1 v1 . 1 Obviously, H1 A1 H1 e ¼ l e , that is, the first column of H 1 1 1 A1 H1 is l1 e1. 1 1 Denoting l1 bT1 1 A2 ¼ H1 A1 H1 ¼ ; ð8:123Þ 0 B2 where B2 is an (n 1)th-order square matrix that obviously has eigenvalues to be l2, . . ., ln. Under the condition that |l2| > |l3|, we can use power method to compute the dominant eigenvalue of B2, l2, and corresponding eigenvector, y2, where

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

529

B2 y2 = l2 y2. Assuming A2 z2 = l2 z2 and to calculate z2, assuming a a constant to be found and y an (n 1)th dimensional vector, we have ( l1 a þ bT1 y ¼ l2 a a z2 ¼ ; : y B 2 y ¼ l2 y bT y

1 Because l1 6¼ l2, we can choose y = y2, a ¼ l2 l , thus we can find z2. v2 ¼ H1 1 z2 1 is the eigenvector of A corresponding to l2. With application of the above method and Householder matrix, we have 9 k1 ¼ sgnðeT1 v1 Þjjv1 jj2 > > > T 1 > = b ¼ jjv1 jj2 jjv1 jj2 þe1 v1 : ð8:124Þ > u ¼ v1 k1 e1 > > > T T ; A2 ¼ H1 A1 H1 1 ¼ ðI buu ÞA1 ðI buu Þ

After l2 and v2 are computed, we can continue to deflate B2 to calculate the rest of the eigenvalues and eigenvectors. In theory, if eigenvalues of A are arranged according to their modulus and those with higher values can be separated, we can use the above method to compute those eigenvalues. A drawback of the deflation method is that it changes elements of the original matrix, so that any sparsity in the matrix cannot be maintained during deflation. Finally, we would like to point out that it is not so straightforward to compute the dominant eigenvalue and corresponding eigenvector by using the power method as introduced above, because we only discussed the case of a single dominant eigenvalue. In fact, l1 could be one of a set of multiple real eigenvalues or l1 and l2 could have the same modulus but are real eigenvalues with opposite sign, or l1 and l2 are a pair of conjugate complex eigenvalues. For those different cases, the power method will be slightly different. Details can be found in [187].

8.4.3.3

The Inverse Power Method

Eigenvalues of inverse matrix A1 of a nonsingular matrix A are reciprocal values of the eigenvalues of A. Hence the reciprocal of the dominant eigenvalue of A1 is the eigenvalue of A with smallest modulus. Applying the power method on A1 is called the inverse power method (or inverse iterative method) to compute the eigenvalue of the nonsingular matrix A with smallest modulus and corresponding eigenvector. For a given initial unit vector under the 2-norm, vð0Þ 2 Cn , the inverse power method generates the following iterative series 9 > AzðkÞ ¼ vðk1Þ > = ðkÞ ðkÞ ðkÞ ; k ¼ 1; 2; . . . : ð8:125Þ v ¼z = z 2> > ; lðkÞ ¼ ½vðkÞ H AvðkÞ

530

8 Small-Signal Stability Analysis of Power Systems

When k ! 1, lðkÞ !

1 ðkÞ v ! xn ln

ð8:126Þ

Another more useful form of inverse power method is to apply the power method to matrix (A tI)1, where t is a real or complex constant. For a given initial unit vector under the 2-norm, vð0Þ 2 Cn , the iterative process is as the following: 9 ðA tIÞzðkÞ ¼ vðk1Þ > > = ðkÞ ðkÞ ðkÞ ; k ¼ 1; 2; . . . : ð8:127Þ v ¼z = z 2 > > ; ðkÞ ðkÞ H ðkÞ l ¼ ½v Av When k ! 1,

9 1 1 > t þ ðkÞ ! lp = lp t l ; > ; !x

lðkÞ ! v

ðkÞ

ð8:128Þ

p

where lp is that closest to t among all eigenvalues of A and xp is the corresponding eigenvector. We need to explain (8.128) further as follows. Because eigenvalues of nonsingular matrix A tI are lj t (j = 1, 2, . . ., n), those corresponding to matrix (A t I)1 are lj 1t ðj ¼ 1; 2; . . . ; nÞ. Applying the power method to matrix (A tI)1, we obtain eigenvalue

1 lp t

with largest

modulus that means lp t with smallest modulus. Hence lp is the closest to t. Hence if we need to compute the eigenvalue of matrix A with a value closest to number t and corresponding eigenvector, we can use the inverse power method given by (8.127). Another application of the inverse power method is that with a known approximation t of an eigenvalue of matrix A, we can use the inverse power method to compute the corresponding eigenvector and improve the accuracy of computation of the eigenvalue. Using (8.127), we can apply triangular decomposition on matrix A tI, A tI ¼ LU; where L is a unit lower triangular matrix and U upper triangular. Then equation of solution becomes LUzðkÞ ¼ vðk1Þ :

8.4.4

Eigensolution of Sparse Matrix

In small-signal stability analysis, the dynamics of a power system are described by ~ ~ ~ differential-algebraic equations of (8.3). From (8.83) we can see that A, B, C, and ~ D are all sparse matrices. When we obtain matrix A from (8.5) to compute its

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

531

eigenvalues, we can find that matrix A has lost its sparsity almost completely. Since the implementation of the QR method cannot take advantage of matrix sparsity, it is not important whether A is sparse or not when we compute its eigenvalues. However, when other iterative methods, such as power method, inverse power method, and sub-space method (to be introduced later), are used to compute part of the eigenvalues of matrix A; if we can take full advantage of the sparsity of the original matrices to compute those eigenvalues directly from (8.3), computational efficiency will be greatly enhanced. For an eigenvalue of A, l, a nonzero vector v 2 Cn satisfying the following equation " # ~ ~ v v A B ¼ l ð8:129Þ ~ ~ 0 C D w is the right eigenvector of A, corresponding to this eigenvalue. The matrix on the left-hand side of above equation is called the augmented state matrix. It is not difficult to prove the above conclusion. In fact, from (8.129) we have ~ ~ w ¼ D1 Cv. Canceling w and from (8.5), we can obtain ~ ~ ~ 1 ~ ð A B D CÞv ¼ Av ¼ lv:

ð8:130Þ

Hence we can compute eigenvalues and eigenvectors of matrix A from the eigensolution of the augmented state matrix of (8.129) without destroying system sparsity. In the following, we shall introduce the sparse realization of the power method and the inverse power method. In addition, the sparse expression of eigenvalue sensitivity to scalar a will be presented. 8.4.4.1

Sparse Realization of Power Method of (8.121)

Since the relationship between z(k) and v(k1) given by the equation "

zðkÞ 0

#

" ¼

~ A ~ C

~ B ~ D

#"

vðk1Þ

#

wðk1Þ

is equivalent to z(k) = Av(k 1), computation of the first equation in (8.121) can be replaced by the following equations: 9 ~ ðk1Þ = ~ ðk1Þ Dw ¼ Cv : ~ ðk1Þ ~ðk1Þ ; zðkÞ ¼ Av þB

ð8:131Þ

Before the iteration of (8.121) is implemented, we only apply sparse triangular ~ ~ decomposition on D once, i.e., D. Hence computation of (8.131) in each iteration is

532

8 Small-Signal Stability Analysis of Power Systems

only the multiplication of some sparse matrices and vectors and solution of two triangular equations.

8.4.4.2

Sparse Realization of Inverse Power Method of (8.127)

Since the relationship between z(k) and v(k 1) given by the equation "

~ A tI ~ C

~ B ~ D

#"

zðkÞ

#

wðkÞ

" ¼

vðk1Þ

#

0

ð8:132Þ

is equivalent to (A tI)z(k) = v(k 1), solution of the first equation in (8.127) can be replaced by that of (8.132) to obtain vector z(k). ~ ~ ~ ~ ~ For a given number t, we first calculate D ¼ D C ðA tIÞ1 B and apply ~ ~ sparse triangular decomposition D ¼ LU. Noting ðDÞ is a diagonal block matrix (a diagonal block is from a dynamic component in power system), we can obtain ~ ðA tIÞ1 by calculating the inverse of diagonal block matrices directly. In ~ ~ addition, D and D have the same sparse structure (2 2 block sparse matrix). Hence solution of (8.132) can be summarized in the following steps: (1) Calculate w(k) from solution of the equation ~ ~ ~ D wðkÞ ¼ CðA tIÞ1 vðk1Þ ~ ~ (2) Calculate zðkÞ ¼ ðA tIÞ1 ðvðk1Þ BwðkÞ Þ

8.4.4.3

Eigenvalue Sensitivity to Scalar a

Similar to (8.129), for a left eigenvector we have

uT

yT

~

A ~ C

~ B ¼ l T u ~ D

0 :

ð8:133Þ

Hence using a similar derivation, we can obtain

@li T ¼ ui @a

2 ~ ~3 @A @B 7

6 6 @a @a 7 vi yTi 6 ~ 7 ~ 4 @ C @ D 5 wi @a @a

ð8:134Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

8.4.5

533

Application of Eigenvalue Sensitivity Analysis

In analysis of power system operation and design of power system controllers, we often need to investigate the influence of certain parameters, such as the gain and time constant of a controller, on power system stability. This will help in the selection or setting of those parameters to stabilize the power system or to improve system stability. Since system state matrix A is a function of a system parameter a, i.e., A(a), any eigenvalue of matrix A, li, is also a function of parameter a, i.e., li (a), i = 1, 2, . . ., n. When parameter a varies, li (a) will accordingly change. Variation of li (a) represents the influence of variation of parameter a on power system stability. Assuming that parameter a changes from a(0) to a(0) + D a, the corresponding change of system eigenvalue is from li (a(0)) to li (a(0) = D a). Taylor expansion of li (a(0) + D a) at a(0) is li ðað0Þ þ DaÞ ¼ li ðað0Þ Þ þ

@li ðaÞ @ 2 li ðaÞ Da þ ðDaÞ2 þ : @a a¼að0Þ @a2 a¼að0Þ

When D a is very small, change of li can be approximately expressed as Dli ðað0Þ ; DaÞ ¼ li ðað0Þ þ DaÞ li ðað0Þ Þ ¼

@li ðaÞ Da; @a a¼að0Þ

ð8:135Þ

where partial derivative ∂ li/∂ a is the first-order sensitivity of eigenvalue li to parameter a, referred to simply as eigenvalue sensitivity. Hence if we can calculate ∂ li/∂a, D a can be approximately determined from the required change of eigenvalue D li. Calculation of the first-order sensitivity of eigenvalue li to parameter a can be summarized as follows: (1) Set a = a(0) to form state matrix A(a(0)) (2) Calculate eigenvalue of A(a(0)), li, and corresponding left and right eigenvector H uH i and vi such that ui vi ¼ 1 (3) Calculate @AðaÞ @a a¼a ð0Þ (4) @li ðaÞ ¼ uH @AðaÞ vi i @a @a a¼að0Þ

In the following, we shall give an example taking the gain KS, time constant T1, T2, T3, and T4 of lead-lag network of PSS as parameter a to demonstrate the calculation of @AðaÞ @a . In the equations of generation unit g of (8.20) and (8.21), except Ag ; BIg ; BVg ; Pg , and Zg are independent of a. In addition, RIg, RVg, and Tg(0) are also independent of a. Hence from (8.30) and (8.32) we have

534

8 Small-Signal Stability Analysis of Power Systems

@Bg @Cg @Dg ¼ ¼ ¼ 0; @a @a @a @A

where matrix @ag can be calculate from matrix Ag in (8.20). Obviously in the equation of the whole system, from (8.83) we can obtain 2 @A G @a ~ @A 6 0 ¼6 @a 4 0 0

0 0 0 0

0 0 0 0

3 0 07 7; 05 0

~ @B ¼ 0; @a

~ @C ¼ 0; @a

~ @D ¼ 0; @a

where ∂AG/∂a can be calculated from (8.73). Moreover, from (8.5) and the equation above we have ~ @A @ A ¼ : @a @a The partial derivative of matrix A to other parameters can be calculated similarly. In the analysis of eigenvalue sensitivity, in addition to the eigenvalue sensitivity to parameters introduced above, eigenvalue sensitivity to power system operating conditions has been proposed. To enhance computational accuracy, second-order eigenvalue sensitivity has also been suggested, with some effective computational methods proposed. Details can be found in references [215–218].

8.5

Oscillation Analysis of Power Systems

A power system cannot operate without proper control. System operators can satisfy the predicted load demand through automatic generation control, and also through switching on, or off, various other controllable devices. Certain automatic control devices, such as the governor and AVR of a generator, HVDC control and FACTS control, etc., carry out the task of fast automatic regulation to maintain system frequency and voltage within required limits, when the power system is subject to disturbances. Since the middle of the twentieth century, the power industry has found that interconnection of regional power systems can lead to more reliable and economical operation of power systems. This has resulted in the increasing scale of modern power systems. In the 1960s, the interconnection of two Northern American power systems suffered from increasing oscillations. Power system oscillations have subsequently been reported in many countries. Investigation into power system oscillations has revealed that when regional power networks are connected through long-distance transmission lines, the resulting weak coupling of large power generation centers implies weak damping of interarea power oscillations. Another cause of reduced, or even negative, damping of power system oscillation is the application

8.5 Oscillation Analysis of Power Systems

535

of high-gain, fast-acting excitation systems. Electrical engineers have found that through the introduction of a supplementary control signal from PSS, system damping can be increased. Experience of Northern American power system interconnection has shown that application of PSS is very effective in damping power system oscillations. Increasing oscillations prevent power networks from exploiting interconnection. In some interconnected power systems, power exchange between interconnected networks has to be kept below a certain limit to avoid the occurrence of oscillations. This greatly reduces the value of interconnecting regional power networks. In some interconnected power systems, low-gain AVR have to be adopted to avoid the oscillation problem. Hence before the scheme of asynchronous interconnection via HVDC was proposed, further interconnection was abandoned in some power systems. Since the 1940s, it has been known that excitation control can enhance the stability limit of a synchronous generator. Since in some cases, excitation control can successfully improve power system dynamic performance, in addition to the control being fast and efficient, electrical engineers have held high expectations of the function of excitation control. However, effectiveness of excitation control is not unlimited. Fast-acting excitation systems can improve synchronous torque to enhance system first swing stability. However, fact-acting excitation is often a negative feedback system with high gain that has little influence on oscillation damping after the first swing. Sometimes it could provide negative damping. When a power system exhibits negative oscillation damping, fast-acting excitation control (usually with high gain) often increases the negative damping to the detriment of system operating conditions. In an m machine interconnected power system, there are a total of m 1 electromechanical oscillation modes. From field records of real power system oscillation [229] and extensive experience from power system simulation, these oscillation modes can be classified, according to the area of coverage, into two types [189], local modes and interarea modes: (1) Local modes only involve power swings of generation units in a power plant to the rest of the power system. Oscillation frequency usually is between 1 and 2 Hz. (2) Interarea modes are power swings of a group of generators in an area to another group of generators in another area. This interarea oscillation often occurs between two or more generators in a weakly connected power system. Because the moment of inertia of the equivalent generator in each area is very large, the oscillation frequency of an interarea oscillation is lower than that of local-mode oscillation, being in the range of 0.1–0.7 Hz. When the oscillation is exhibited between two groups of generators, oscillation frequency is between about 0.1 and 0.3 Hz. When it is an oscillation among multiple groups of generators, oscillation frequency is about 0.4–0.7 Hz. Since the frequency of those two types of oscillation is low, they are often called power system low-frequency oscillations. In addition to electromechanical oscillation modes, control modes and torsional oscillation modes may exist in a power

536

8 Small-Signal Stability Analysis of Power Systems

system. Torsional modes have been previously introduced. Control modes are related to various control devices installed in the power system. Since regulation of control devices is fast and controllers have small time constant, frequency of control modes is usually high. Here we are only concerned about electromechanical oscillation modes. Analysis regarding control modes and torsional modes is out of the scope of this book. Small disturbances can lead to power system low-frequency oscillations. If the oscillations of all modes are decaying, the power system is stable in terms of smallsignal stability. However, in real power system operation, usually only where the damping ratio of electromechanical oscillation modes is greater than 0.05, is the power system operation acceptable. Of course this value is not fixed. With variations of system operating conditions and small changes of oscillation modes, lower damping ratios (such as 0.03) could also be acceptable. It is apparent that small-signal instability of real power systems is mainly due to system oscillations caused by lack of damping. In 1969, Demello and Concordia [218] obtained conditions of power system small-signal stability with regard to the operation of a thyristor-controlled excitation system for a single-machine infinitebus power system. These are certain requirements on the setting of AVR gain and the introduction of an auxiliary control signal of generator rotor deviation. Their work clearly revealed the cause of power system oscillation in the single-machine infinite-bus model and laid down a solid theoretical foundation for the design of PSSs. Based on their idea and principles proposed, researchers have attempted extensions into multimachine power systems for the analysis of local-mode oscillations, and further to interarea oscillations in interconnected power systems. However, we have to point out that some of the simple extensions are often found to be inappropriate. A large-scale multimachine power system is a typical nonlinear dynamic system. Increasing oscillations caused by disturbances are dependant on many factors. Network topology and parameters, characteristics of dynamic components, system operating conditions, control strategies and parameters of various controllers all play an important role in system oscillations. It is a challenging task to clearly analyze the cause of power system electromechanical oscillations and to propose effective measure to overcome the problem. With the increasing demand of economics in a modern society, especially with the trend toward electricity markets, more and more load is required to be carried over existing power networks. However, economics and security of the power system are two conflicting requirements. When a power system operates under a light load condition before it is disturbed, damping windings of generators can provide adequate torque proportional to rotor speed. This damping can usually absorb the energy involved in system oscillations and thus the magnitude of oscillations decays continuously. The power system is stable in terms of small-signal stability. If the power system operates at heavy load conditions before it is disturbed, damping windings of generators cannot completely dissipate the energy involved in the system oscillations, so that the oscillations can grow continuously. The power system is unstable in terms of smallsignal stability. Moreover, to increase the capability of power transmission and to

8.5 Oscillation Analysis of Power Systems

537

improve system transient stability or other system performance aspects, large numbers of various types of controller are installed in the power system. Some of these may clash with the damping of system oscillations, due to improper control strategies or parameters, or mismatch among controller functions. This may again lead to unstable system oscillations. The purpose of oscillation analysis of power systems is to study key factors affecting oscillation modes, so that useful measures can be worked out to suppress oscillations effectively. [Example 8.2] In Example 8.1, all the eigenvalues of the system state matrix have been calculated. In the following, we shall study system oscillation modes. [Solution] Table 8.2 gives oscillation frequency and damping ratio of several oscillation modes; corresponding left and right eigenvectors and participation vectors are given in Tables 8.3 and 8.4. In the following we will carry out modal analysis from the results in Tables 8.3 and 8.4, where all vectors have been normalized to unit vectors under the infinite norm. Firstly, we identify electromechanical oscillation modes from the participation vector of specified modes: if the component with largest modulus in a participation vector is related to generator speed, we identify that the mode is an electromechanical oscillation mode. Then we can observe the exhibition of modes from right eigenvectors: for those components in right eigenvectors related to generator speed, a group of components with similar modulus and directional phase identifies a group of coherent generators. Incoherent generators are associated with those components with opposite phase. The right eigenvector of a local mode is dominated by variables related to one or a group of closely located generators. Components of the right eigenvector of an interarea mode evenly distribute in all regions in a power system. For oscillation mode l5,6, the element with largest modulus in its participation vector is related to Do3. Hence it is an electromechanical oscillation mode. Besides, in its right eigenvector, components associated with Do1, Do2 have small modulus (being 0.00018 and 0.00121, respectively) and similar phase (being 170.62 and 166.98 , respectively); the component associated with Do3 has large modulus (0.00411) and opposite directional phase to those above (being 10.52 ). Hence this mode will exhibit as an electromechanical oscillation between generator 1, 2 and generator 3. It is a local oscillation mode with oscillation frequency being 2.04732 Hz.

Table 8.2 Oscillation frequency and damping ratio of several oscillation modes

f x

l5,6

l7,8

l11,12

l13,14

2.04732 0.05859

1.38007 0.01747

0.14569 0.77893

0.10463 0.59313

Dd1 Do1 Dd2 Do2 DE0 q2 DE0 d2 DEfq2 DVR2 DVM2 Dd3 Do3 DE0 q3 DE0 d3 DEfq3 DVR3 DVM3

0.01137 0.33268 0.02281 0.66769 0.02124 0.00888 0.00028 0.00003 0.00198 0.03416 1.00000 0.03534 0.01302 0.00047 0.00005 0.00336

Modulus

u6

90.60 176.13 85.77 178.79 85.82 110.43 178.73 13.37 53.99 92.62 0.00 94.07 78.12 1.15 166.51 126.13

Phase

0.00532 0.00018 0.03527 0.00121 0.00156 0.00688 0.03610 0.36141 0.00184 0.12030 0.00411 0.00411 0.02094 0.09989 1.0000 0.00509

Modulus

v6

l6

96.02 170.62 99.66 166.98 162.26 36.04 72.77 76.77 88.59 82.84 10.52 12.05 152.38 4.01 0.00 165.36

Phase 0.01470 0.01470 0.19568 0.19574 0.00808 0.01485 0.00241 0.00233 0.00089 0.99933 1.0000 0.03536 0.06632 0.01132 0.01091 0.00416

Modulus

p6 5.10 5.00 3.36 3.71 101.41 135.95 116.48 52.88 45.12 0.74 0.00 71.50 118.98 13.38 155.99 57.99

Phase

Table 8.3 Left and right eigenvectors and participation vectors of oscillation modes

0.02300 1.00000 0.01871 0.81346 0.02567 0.00413 0.00049 0.00005 0.00248 0.00430 0.18692 0.00870 0.00257 0.00017 0.00002 0.00086

Modulus

u8

90.86 0.00 89.81 179.71 77.72 42.44 168.06 2.07 71.83 86.22 176.13 68.23 12.05 158.57 11.56 81.33

Phase 0.04052 0.00093 0.11332 0.00261 0.00265 0.01286 0.10048 1.00000 0.00506 0.06144 0.00141 0.00054 0.00513 0.07102 0.70681 0.00358

Modulus

v8

l8

155.40 113.60 26.92 64.08 48.13 91.43 5.93 0.00 170.13 23.80 67.20 19.43 94.98 6.82 0.89 171.02

Phase

0.43955 0.43957 0.99988 1.0000 0.03212 0.02506 0.02337 0.02253 0.00591 0.12456 0.12458 0.00222 0.00621 0.00570 0.00550 0.00144

Modulus

p8

2.47 2.61 0.52 0.00 86.62 67.22 57.78 118.28 17.92 6.19 7.29 67.42 9.19 49.18 126.88 26.52

Phase

538 8 Small-Signal Stability Analysis of Power Systems

Dd1 Do1 Dd2 Do2 DE0 q2 DE0 d2 DEfq2 DVR2 DVM2 Dd3 Do3 DE0 q3 DE0 d3 DEfq3 DVR3 DVM3

0.00383 1.00000 0.00240 0.64734 0.36458 0.15380 0.04393 0.00440 0.02583 0.00149 0.41946 0.16500 0.10380 0.02025 0.00203 0.01191

Modulus

u12

140.64 0.00 31.48 170.64 34.69 156.74 103.88 75.05 171.93 52.19 171.13 42.95 156.78 95.61 83.32 179.80

Phase

0.05071 0.00020 0.05334 0.00021 0.00812 0.00066 0.06692 1.00000 0.00489 0.05286 0.00021 0.00685 0.00099 0.05590 0.83532 0.00408

Modulus

v12

l12

85.09 133.74 73.44 145.39 178.71 13.57 40.05 0.00 178.93 75.02 143.82 177.95 11.69 39.27 0.78 178.14

Phase 0.04408 0.04458 0.02911 0.03036 0.67247 0.02317 0.66740 1.00000 0.02866 0.01782 0.01949 0.25677 0.02343 0.25702 0.38511 0.01104

Modulus

p12

150.68 151.21 33.09 31.08 140.93 68.12 141.02 0.00 82.04 52.22 47.74 149.95 70.04 150.07 9.05 72.99

Phase 0.00049 0.23071 0.00194 0.94544 0.43208 0.08490 0.09457 0.00936 0.04856 0.00194 1.00000 0.50272 0.14005 0.11208 0.01109 0.05756

Modulus

Table 8.4 Left and right eigenvectors and participation vectors of oscillation modes

u14

146.95 87.88 49.58 171.29 38.37 153.90 81.94 97.30 148.04 115.83 0.00 141.75 30.18 97.94 82.82 32.08

Phase 0.04062 0.00009 0.04627 0.00010 0.01202 0.00528 0.05319 0.48473 0.00240 0.04028 0.00009 0.02494 0.01276 0.10972 1.00000 0.00495

Modulus

v14

l14

106.90 126.72 122.47 111.15 0.10 171.90 115.53 176.05 4.71 63.91 169.72 176.31 12.20 68.42 0.00 179.24

Phase

0.00159 0.00162 0.00715 0.00756 0.41422 0.03577 0.40108 0.36171 0.00930 0.00622 0.00696 1.00000 0.14246 0.98069 0.88442 0.02273

Modulus

p14

74.61 73.39 38.33 43.00 3.91 68.77 0.97 113.31 118.19 145.17 155.72 0.00 76.95 5.04 117.38 114.12

Phase

540

8 Small-Signal Stability Analysis of Power Systems

Similarly, for l7,8, the element with largest modulus in its participation vector is related to Do2. Hence it is an electromechanical oscillation mode. In addition, in its right eigenvector, components associated with Do2, Do3 have relatively large modulus (being 0.00261 and 0.00141, respectively) and the same direction (phase being 64.08 and 67.20 , respectively); component associated with Do1 has small modulus (0.00093) and opposite direction. Hence this mode will exhibit as electromechanical oscillation between generator 1, 2 and generator 3. It is also a local oscillation mode with oscillation frequency being 1.38007 Hz. Though this mode is stable, the damping ratio (0.01747) is not sufficient, exhibiting poor dynamic performance as far as oscillation decay is concerned. For mode l11,12, the element with largest modulus in its participation vector is related to D VR2; for l13,14, element with largest modulus in its participation vector is related to D E0 q3. Hence those modes are not electromechanical oscillation modes but control modes. [Example 8.3] We take the 39-node 10-machine simplified New England system as an example to demonstrate the procedure of power system oscillation analysis [221]. In the power system, ten machines are at nodes 30–39 and the machine at node 39 is an equivalent generator. Generators at nodes 30–38 have fast static excitation systems installed. [Solution] We obtain the system linearized equation by using the methods introduced above and then compute all eigenvalues of the system state matrix. Damping of nine modes associated with electromechanical oscillations is not sufficient and some of eigenvalues have positive real parts. For two eigenvalues 0.1022 j7.215 (mode 1) and 0.037 4.301 (mode 9), components associated with generator speed in their right eigenvectors are given in Table 8.5. From Table 8.5 we can see that in the eigenvector of the first mode, there are three components with large modulus (highlighted by bold figures), among these the direction of the first component (with phase being 0 ) is opposite to that of the Table 8.5 Components associated with generator speed in right eigenvectors of mode 1 and 2 Generator number

30 31 32 33 34 35 36 37 38 39

Mode 1

Mode 9

Modulus

Phase (degree)

Modulus

Phase (degree)

1.0 0.1408 0.0797 0.1851 0.4777 0.7935 0.7797 0.3468 0.1664 0.0170

0.0 44.5 241.9 152.3 32.1 170.2 170.5 10.1 111.4 191.3

0.5574 0.4757 0.5208 0.7601 1.0 0.7961 0.7977 0.5084 0.6694 0.4052

9.9 3.4 5.5 5.3 0.0 5.7 6.8 12.4 3.3 179.6

8.5 Oscillation Analysis of Power Systems

541

other two (with phase being about 170 ). This indicates that the mode mainly exhibits as an electromechanical oscillation between generator 30 and generators 35, 36; with oscillation frequency being 7.215/2p ¼ 1.148 Hz. This is a local oscillation mode. In the eigenvector of the second mode, except for the component associated with generator 39 with relatively small modulus, other components have similar values of modulus. Moreover, the first nine components have opposite direction (with phase being about 0 ) to that of the last one (with phase being about 180 ). This indicates that this mode exhibits mainly as an electromechanical oscillation between generators 30–38 and generator 39 (the equivalent generator that can be seen as a regional network). Hence this is an interarea oscillation with oscillation frequency being 4.301/2p ¼ 0.685 Hz. In addition to identifying electromechanical oscillation modes, participation vectors can also be used to estimate the relative effects of generator controls on specified oscillation modes. For example, a component associated with rotor speed in a participation vector gives eigenvalue sensitivity to the variation of damping applied on the associated generator. If it is zero, this indicates that installation of PSS on the generator will have no impact in improving oscillation damping. If it is a large positive number, this shows that the associated generator is a good candidate place to install PSS, to effectively increase damping of the relevant oscillation mode. In Table 8.6, components associated with generator rotor speed are given. From Table 8.6 we can see that for the local mode, a component in the participation vector associated with generator 30 has the largest value, about equal to the sum of components associated with generators 35 and 36. Hence applying damping control at generator 30 should almost be equivalent to similar applications at both generators 35 and 36 simultaneously. For the interarea oscillation mode, a component associated with generator 39 has the largest modulus. However, generator 39 is an equivalent machine and damping control cannot be applied there. The sum of components associated with generators 30–38 is about equal to the component related to generator 39. This means that damping control applied at generators 30– 38 will achieve a similar effect as that applied at generator 39. Moreover, we should note that although some generators have large participation factors, there will be little effect in applying damping control on those generators if their capacity is small. Applying damping control on generators with large capacity will be more effective than applying it on those with small capacity, as far as increased oscillation damping is concerned.

Table 8.6 Components in participation vector Generator number Mode 1 Mode 2

30

31

32

33

34

35

1.0 0.17

0.01 0.09

0.005 0.12

0.02 0.22

0.13 0.33

0.42 0.26

Values in the table are estimated from graphs presented in [221]

36

37

38

39

0.43 0.21

0.07 0.07

0.02 0.18

0.001 1.0

542

8 Small-Signal Stability Analysis of Power Systems

Thinking and Problem Solving 1. What are the purpose and significance of small-signal stability analysis for electrical power systems? 2. What are the basic principle and basic procedures of small-signal stability analysis? 3. What are the main methods to solve the eigenvalues of a linearized electrical power system? What are their advantages and disadvantages? 4. Why is the QR method not suitable for the eigenvalue analysis of large-scale electrical power systems? 5. What is the critical eigenvalue? What methods are there for calculating critical eigenvalues of large-scale electrical power systems? What advantages and disadvantages are there for each method? 6. How can we apply sparse matrix techniques to critical eigenvalue calculations for large-scale electrical power systems? Can the sparse matrix technique be used in the QR method? 7. How are the eigenvalue and corresponding left and right eigen vectors used to represent the modes of a linear system? 8. What is the participation factor? Why can the participation factor be used to represent both the observability and controllability of a system? 9. What are the main causes of increasing amplitude, low-frequency oscillation? 10. What are the major manifestations of low-frequency oscillation? Why is the oscillating frequency among local generators lower than that among generators in a plant, and the oscillating frequency among regional generators is lower than that among local generators? 11. What are the main measures to control low-frequency oscillation?

References

1. W.F. Tinney, I.W. Waiker, ‘‘Direct solutions of sparse network equation by optimal ordered triangular factorization,’’ Proceedings of IEEE 55(11), 1801–1809, 1967 2. W.F. Tinney, Some examples of sparse matrix methods for power system problems, Proceedings of Power Systems Computation Conference (PSCC), Rome, June 23–27, 1969 3. W.F. Tinney, V. Brandwajn, S.M. Chen, ‘‘Sparse vector methods,’’ IEEE Transactions on Power Apparatus and Systems, 104, 295–301, 1985 4. G.W. Stagg, A.H. El-Abiad, Computer Methods in Power Systems, McGraw Hill, New York, 1968 5. R.G. Andreich, H.E. Brown, H.H. Happ, C.E. Person, ‘‘The piecewise solution of the impedance matrix load flow,’’ IEEE Transactions on Power Apparatus and Systems, 87(10), 1877– 1882, 1968 6. W.F. Tinney, C.E. Hart, ‘‘Power flow solution by Newton’s method,’’ IEEE Transactions on Power Apparatus and Systems 86(4), 1449–1460, 1967 7. W.F. Tinney, ‘‘Compensation methods for network solutions by optimal ordered triangular factorization,’’ IEEE Transactions on Power Apparatus and Systems, 91(1), 123–127, 1972 8. B. Scott, O. Alsac, ‘‘Fast decoupled load flow,’’ IEEE Transactions on Power Apparatus and Systems, 93(3), 859–869, 1974 9. R. Van Amerongen, ‘‘A general-purpose version of the fast decoupled load flow,’’ IEEE Transactions on Power Systems, 4(2), 760–770, 1989 10. A. Monticeli, O.R. Savendra, ‘‘Fast decoupled load flow: Hypothesis, derivations, and testing,’’ IEEE Transactions on Power Systems, 5(4), 1425–1431, 1990 11. L. Wang, X. Rong Li, ‘‘Robust fast decoupled power flow,’’ IEEE Transactions on Power Systems, 15(1), 208–215, 2000 12. V.M. da Costa, N. Martins, J.L. Pereira, ‘‘Developments in the Newton Raphson power flow formulation based on current injections,’’ IEEE Transactions on Power Systems, 14(4), 1320–1326, 1999 13. A. Semlyen, F. de Leon, ‘‘Quasi-Newton power flow using partial Jacobian updates,’’ IEEE Transactions on Power Systems, 16(3), 332–339, 2001 14. V.H. Quintana, N. Muller, ‘‘Studies of load flow method in polar and rectangular coordinates,’’ Electric Power System Research, 20(1), 225–235, 1991 15. R.P. Klump, T. J. Overbye, ‘‘Techniques for improving power flow convergence,’’ Proceedings of PES Summer Meeting, Seattle, USA, vol. 1, July 2000 16. K.L. Lo, Y.J. Lin, W.H. Siew, ‘‘Fuzzy-logic method for adjustment of variable parameters in load flow calculation,’’ IEE Proceedings of Generation Transmission Distribution, 146(3), 276–282, 1999 17. W.L. Chan, A.T.P. So, L.L. Lai, ‘‘Initial applications of complex artificial neural networks to load-flow analysis,’’ IEE Proceedings of Generation Transmission Distribution, 147(6), 361–366, 2000

543

544

References

18. T. Nguyen, ‘‘Neural network load-flow,’’ IEE Proceedings of Generation Transmission Distribution, 142(1), 51–58, 1995 19. K.P. Wong, A. Li, T.M.Y. Law, ‘‘Advanced constrained genetic algorithm load flow method,’’ Proceedings of Generation Transmission Distribution, 146(6), 609–616, 1999 20. P.K. Mannava, L. Teeslink, A.R. Hasan, ‘‘Evaluation of efficiency of parallelization of power flow algorithms,’’ Proceedings of the 40th Midwest Symposium on Circuit and Systems, Sacramento, California, USA, August 1997, pp. 127–130 21. N. balu, T. Bertram, A. Bose, et al, ‘‘On-Line power system security analysis,’’ Proceedings of the IEEE, 80, 262–282, 1992 22. J. Carpentier, ‘‘Static security assessment and control: A short survey,’’ IEEE/NTUA Athens Power Tech Conference on Planning, Operation and Control of Today’s Electric Power Systems’’, Athens, Greece, September 5–8, 1993, pp. 1–9 23. B. Stott, O. Alsac, F.L. Alvarado, ‘‘Analytical and computational improvement in performance index raking algorithm for networks,’’ International Journal of Electrical Power and Energy Systems, 7(3), 154–160, 1985 24. T.A. Mikolinas, B.F. Wollenberg, ‘‘An advanced contingency selection algorithm,’’ IEEE Transactions on Power Apparatus and Systems, 100(2), 608–617, 1981 25. V. Brandwajn, ‘‘Efficient bounding method for linear contingency analysis,’’ IEEE Transactions on Power Systems, 3(2), 726–733, 1988 26. R. Bacher, W.F. Tinney, ‘‘Faster local power flow solutions: The zero mismatch approach,’’ IEEE Transactions on Power Systems, 4(4), 1345–1354, 1989 27. W.P. Luan, K.L. Lo, Y.X. Yu, ‘‘NN-based pattern recognition technique for power system security assessment,’’ Proceedings of the International Conference on Electric Utility Deregulation and Restructuring and Power Technologies, London, April 2000 28. Task Force on Probabilistic Aspects of Reliability Criteria of the IEEE PES Reliability, Risk and Probability Applications Subcommittee, ‘‘Probabilistic security assessment for power system operations’’, IEEE PES General Meeting, Denver, USA, June 6–10, 2004 29. A.M. Leite da Silva, L.C. Resende, L.A.F. Manso, et al, ‘‘Well-being analysis for composite generation and transmission systems’’, IEEE Transactions on Power Systems, 19(4), 1763–1770, 2004 30. R. Billinton, R.N. Allan, Reliability evaluation of Large Electric Power Systems, Kluwer, Boston, 1988 31. R. Billinton, R. Allan, Reliability Evaluation of Power Systems, Plenum Press, New York, 1996 32. X. Wang, J. McDonald, Modern Power System Planning, McGraw-Hill, London, 1994, pp. 108–110 33. R. Billinton, W. Li, Reliability Evaluation of Electric Power Systems using Monte Carlo Methods, Plenum Press, New York, 1994 34. A.M. Leite de Silva, S.M.P. Ribeiro, V.L. Arienti, et al., ‘‘Probabilistic load flow techniques applied to power system expansion planning’’, IEEE Transactions on Power Systems, 5(4), 1047–1053, 1990 35. P. Zhang, S.T. Lee, ‘‘Probabilistic load flow computation using the method of combined cumulants and Gram-Charlier expansion’’, IEEE Transactions on Power Systems, 19(1), 676–682, 2004 36. Z. Hu, X. Wang, ‘‘A probabilistic load flow method considering branch outages,’’ IEEE Transactions on Power Systems, 21(2), 507–514, 2006 37. K. Maurice, S. Alan, The Advanced Theory of Statistics, vol. 1, Macmillan, USA, 1977 38. J.C. Spall, ‘‘Estimation via Markov chain Monte Carlo,’’ IEEE Control System Magazine, 23 (2), 35–45, 2003 39. R. Chen, J.S. Liu, X. Wang, ‘‘Convergence analyses and comparisons of Markov chain Monte Carlo algorithms in digital communications,’’ IEEE Transactions on Signal Processing, 50(2), 255–270, 2002

References

545

40. Reliability Test System Task Force, ‘‘IEEE reliability test system,’’ IEEE Transactions on Power System, 14(3), 1010–1020, 1999 41. R. Von Mises, Mathematical Theory of Probability and Statistics, Academic Press, New York, 1964 42. P.A. Jensen, J.W. Barnes, Network-Flow Programming, Wiley, New York, 1980 43. Y.K. Lin, ‘‘Reliability of a stochastic flow network with unreliable branches and nodes, under budget constraints,’’ IEEE Transactions on Reliability, 53(3), 381–387, 2004 44. D. Maagee, A. Refsum, ‘‘RESIN, A desktop-computer program for finding cut set’’, IEEE Transactions on Reliability, 30(5), 407–410, 1981 45. K. Kobayashi, H. Yamamoto, ‘‘A new algorithm in enumerating all minimal paths in a sparse network’’, Reliability Engineering and System Safety, 65(1), 11–15, 1999 46. S. Ross, Introduction to Probability Models, Academic Press, NewYork, 2006 47. X. Wang, C. Pottle, ‘‘A concise frequency and duration approach to generating system reliability studies’’, IEEE Transactions on Power Amplifier Symposium, 102(8), 2521–2530, 1983 48. F.A. Rahimi, A. Vojdani, ‘‘Meet the emerging transmission market segments’’, IEEE Computer Application in Power, 12(1), 26–32, 1999 49. F.C. Schweppe, M.C. Caramanis, R.D. Tabors, et al, Spot Pricing of Electricity, Kluwer, Boston, 1988 50. J. Carpentier, ‘‘Contribution a’1’etude du dispatching economique,’’ Bulletin de la Societe Francaise des Electricients, 3, 431–447, 1962 51. H.W. Dommel, W.F. Tinney, ‘‘Optimal power flow solutions,’’ IEEE Transactions on Power Apparatus and Systems, 87(12), 1866–1876, 1968 52. A.M. Sasson, ‘‘Combined use of the parallel and fletcher-powell non-linear programming methods for optimal load flows,’’ IEEE Transactions on Power Apparatus and Systems, 88 (10), 1530–1537, 1969 53. A.M. Sasson, ‘‘Decomposition technique applied to the non-linear programming load flow method,’’ IEEE Transactions on Power Apparatus and Systems, 89(1), 78–82, 1970 54. R. Divi, H.K. Kesavan, ‘‘A shifted penalty function approach for optimal power flow,’’ IEEE Transactions on Power Apparatus and Systems, 101(9), 3502–3512, 1982 55. S.N. Talukdar, T.C. Giras, V.K. Kalyan, ‘‘Decompositions for optimal power flows,’’ IEEE Transactions on Power Apparatus and Systems, 102(12), 3877–3884, 1983 56. G.F. Reid, L. Hasdorf, ‘‘Economic dispatch using quadratic programming,’’ IEEE Transactions on Power Apparatus and Systems, 92, 2015–2023, 1973 57. R.C. Burchett, H.H. Happ, D.R. Vierath, ‘‘Quadratically convergent optimal power flow,’’ IEEE Transactions on Power Apparatus and Systems, 103, 3267–3276, 1984 58. D.W. Wells, ‘‘Method for economic secure loading of a power systems,’’ Proceedings of IEEE, 115(8), 606–614, 1968 59. C.M. Shen, M.A. Laughton, ‘‘Power system load scheduling with security constraints using dual linear programming,’’ Proceedings of IEEE, 117(1), 2117–2127, 1970 60. N. Nabona, L.L. Ferris, ‘‘Optimization of economic dispatch through quadratic and linear programming,’’ Proceedings of IEEE, 120(5), 1973 61. Z. Yan, N.D. Xiang, B.M. Zhang, et al, ‘‘A hybrid decoupled approach to optimal power flow,’’ IEEE on Power Systems, 11(2), 947–954, 1996 62. K.R. Frish, ‘‘Principles of linear programming: The double gradient form of the logarithmic potential method,’’ Memorandum, Institute of Economics, University of Oslo, Oslo, Norway, 1954 63. P. Huard, ‘‘Resolution of mathematical programming with nonlinear constraints by the method of centers,’’ Nonlinear Programming, 209–219, 1967 64. I.I. Dikin, ‘‘Iterative solution of problems of linear and quadratic programming,’’ Soviet Mathematics, 8, 674–675, 1967 65. N. Karmarkar, ‘‘A new polynomial-time algorithm for linear programming,’’ Combinatorica, 4, 373–395, 1984

546

References

66. P.E. Gill, W. Murray, M.A. Saunders, et al, ‘‘On the projected newton barrier methods for linear programming and an equivalence to karmarkar’s projective method,’’ Mathematical Programming, 36, 183–209, 1986 67. X. Guan, W.H. Edwin Liu, A.D. Papalexopoulos, ‘‘Application of a fuzzy set method in an optimal power flow,’’ Electric Power Systems Research, 34(1), 11–18, 1995 68. Y.T. Hsiao, C.C. Liu, H.D. Chiang, et al, ‘‘A new approach for optimal VAR sources planning in large scale electric power systems,’’ IEEE Transactions on Power Systems, 8(3), 988–996, 1993 69. K.R. Frisch, ‘‘The Logarithmic Potential Method for Convex Programming,’’ Memorandum, Institute of Economics, University of Oslo, Norway, May, 1955 70. A.V. Fiacco, G.P. MoCormic, Nonlinear Programming: Sequential Unconstrained Minimization Techniques, Wiley, New York, 1968 71. H. Wei, H. Sasaki, J. Kubokawa, et al, ‘‘An interior point nonlinear programming for optimal power flow problems with a novel data structure,’’ IEEE Transactions on Power Systems, 13 (3), 870–877, 1998 72. F.C. Schweppe, M.C. Caramanis, R.D. Tabors, et al, ‘‘Spot Pricing of Electricity,’’ Kluwer, Boston, 1988 73. M.C. Carmanis, R.E. Bohn, F.C. Schweppe, ‘‘Optimal spot pricing: Practice and theory,’’ IEEE Transactions on Power Apparatus and Systems, 101(9), 3234–3245, 1982 74. M.C. Carmanis, R.E. Bohn, F.C. Schweppe, ‘‘The costs of wheeling and optimal wheeling rates,’’ IEEE Transactions on Power Systems, 1(1), 63–73, 1986 75. D. Ray, F. Alvarado, ‘‘Use of an engineering model for economic analysis in the electricity utility industry,’’ The Advanced Workshop on Regulation and Public Utility Economics, Rutgers University, New Jersey, May 25–27, 1988 76. M.L. Baughman, S.N. Siddiqi, ‘‘Real time pricing of reactive power: theory and case study result,’’ IEEE Transactions on Power Systems, 6(1), 23–29, 1991 77. S.N. Siddiqi, M.L. Baughman, ‘‘Reliability differentiated pricing of spinning reserve,’’ IEEE Transactions on Power Systems, 10(3), 1211–1218, 1993 78. A. Zobian, M.D. llic, ‘‘Unbundling of transmission and ancillary services,’’ IEEE Transactions on Power systems, 12(2), 539–558, 1997 79. M.L. Baughman, S.N. Siddiqi, J.M. Zarnikau, ‘‘Advanced pricing in electrical system. Part 1: Theory,’’ IEEE Transactions on Power Systems, 12(1), 489–495, 1997 80. M.L. Baughman, S.N. Siddiqi, J.M. Zarnikau, ‘‘Advanced pricing in electrical system. Part 2: Implication,’’ IEEE Transactions on Power Systems, 12(1), 496–502, 1997 81. K. Xie, Y.H. Song, J. Stonham, et al, ‘‘Decomposition model and interior point methods for optimal spot pricing of electricity in deregulation environments,’’ IEEE Transactions on Power Systems, 15(1), 39–50, 2000 82. C.N. Yu, M.D. Ilic, ‘‘An algorithm for implementing transmission rights in competitive power industry,’’ IEEE Power Engineering Society Winter Meeting, 3, 1708–1714, 2000 83. X. Wang, Y.H. Song, Q. Lu, et al, ‘‘Series FACTS devices in financial transmission rights auction for congestion management,’’ IEEE Power Engineering Review, 21(11), 41–44, 2001 84. R. Bacher, H. Glavitsch, ‘‘Loss reduction by network switching,’’ IEEE Transactions on Power Systems, 3(2), 447–454, 1988 85. R. Baldick, E. Kahn, ‘‘Contract paths, phase shifters and efficient electricity trade,’’ IEEE Power Engineering Society Winter Meeting, 2, 968–974, 2000 86. S.Y. Ge, T.S. Chung, Y.K. Wong, ‘‘A new method to incorporate FACTS devices in optimal power flow,’’ Proceedings of International Conference on Energy Management and Power Delivery, 1, 122–271, 1998 87. X. Wang, Y.H. Song, Q. Lu, ‘‘Primal-dual interior point linear programming optimal power flow for real-time congestion management,’’ IEEE Power Engineering Society Winter Meeting, 3, 1643–1649, 2000 88. G. Hamoud, ‘‘Assessment of available transfer capability of transmission system,’’ IEEE Transactions on Power System, 15(1), 27–32, 2000

References

547

89. X. Luo, A.D. Patton, C. Singh, ‘‘Real power transfer capability calculations using multi-layer feed-forward neutral networks,’’ IEEE Transactions on Power Systems, 15(2), 903–908, 2000 90. M. Pavella, D. Ruiz-Vega, J. Giri, et al, ‘‘An integrated scheme for on-line static and transient stability constrained ATC calculations,’’ IEEE Power Engineering Society Summer Meeting, 1, 273–276, 1999 91. D.S. Kirschen, R.N. Allan, G. Strbac, ‘‘Contributions of individual generators to loads and flows’’, IEEE Transactions on Power Systems, 12(1), 52–60, 1997 92. X. Wang, X. Wang, ‘‘On current trace problem’’, Science in China (E), 30(3), 405–412, 2000 93. J. Bialek, ‘‘Topological generation and load distribution factors for supplement cost allocation in transmission open access’’, IEEE Transactions on Power Systems, 12(3), 1185–1193, 1997 94. D.S. Kirschen, G. Strbac, ‘‘Tracing active and reactive power between generators and loads using real and imaginary currents,’’ IEEE Transactions on Power Systems, 14(4), 1312–1319, 1999 95. Reliability Test System Task Force, ‘‘IEEE reliability test system-1996,’’ IEEE Transactions on Power Systems, 14(3), 1010–1020, 1999 96. Federal Energy Regulatory Commission, ‘‘Open access same-time information system (Formerly Real-time Information Networks) and standards of conduct,’’ Docket no. RM95-9-000, Order 889, 1996 97. North American Electric Reliability Council, ‘‘Available transfer capability definition and determination’’, NERC Planning Standards, June 1996 98. G. Hamoud, ‘‘Assessment of available transfer capability of transmission system,’’ IEEE Transactions on Power Systems, 15(1), 27–32, 2000 99. G.C. Ejebe, J.G. Waight, M. Sanots-Nieto, et al, ‘‘Available transfer capability calculations,’’ IEEE Transactions on Power Systems, 13(4), 1521–1527, 1998 100. X. Luo, A.D. Patton, C. Singh, ‘‘Real power transfer capability calculations using multi-layer feed-forward neural networks,’’ IEEE Transactions on Power Systems, 15(2), 903–908, 2000 101. G.C. Ejebe, J.G. Waight, M.S. Nieto, W.F. Tinney, ‘‘Fast calculation of linear available transfer capability,’’ IEEE Transactions on Power Systems, 15(3), 1112–1116, 2000 102. M. Shaaban, Y. Ni, F. Wu, ‘‘Total transfer capability calculations for competitive power networks using genetic algorithms,’’ Proceedings of International Conference on DRPT, City University, London, April 4–7, 2000 103. A.R. Vojdani, ‘‘Computing available transmission capability using trace,’’ EPRI Power System Planning and Operation News, 1, 1, 1995 104. Y. Xiao, Y.H. Song, ‘‘Available transfer capability (ATC) evaluation by stochastic programming,’’ IEEE Power Engineering Review, 20(9), 50–52, 2000 105. F. Xia, A.P.S. Meliopoulos, ‘‘A methodology for probabilistic simultaneous transfer capability analysis,’’ IEEE Transactions on Power Systems, 11(3), 1269–1278, 1996 106. A.B. Rodrigues, M.G. Da Silva, ‘‘Solution of simultaneous transfer capability problem by means of Monte Carlo simulation and primal-dual interior-point method,’’ Proceedings of PowerCon International Conference, 2, 1047–1052, 2000 107. X.F. Wang, C.J. Cao, Z.C. Zhou, Experiment on fractional frequency transmission system, IEEE Transactions on Power Systems, 21(1), 372–377, 2006 108. N.G. Higorani, ‘‘Power electronics in electric utilities: Role of power electronics in future power systems,’’ Proceedings of IEEE, 76(4), 481–482, 1988 109. L. Gyugyi, ‘‘Dynamic compensation of AC transmission lines by solid-state synchronous voltage source,’’ IEEE Transactions on Power Delivery, 9(2), 904–911, 1994 110. A.A. Edris, R. Aapa, M.H. Baker, et al, ‘‘Proposed terms and definitions for flexible AC transmission system (FACTS),’’ IEEE Transactions on Power Delivery, 12(4), 1848–1853, 1997 111. D.A. Braunagel, L.A. Kraft, J.L. Whysong, ‘‘Inclusion of DC converter and transmission equation directly in a Newton power flow,’’ IEEE Transactions on Power Apparatus and Systems, 95(1), 76–88, 1976

548

References

112. J. Arrillaga, P. Bodger, ‘‘Integration of HVDC links with fast decoupled load flow solutions,’’ Proceedings of IEE, 124(5), 463–468, 1977 113. J. Arrillaga, B. Smith, AC-DC Power System Analysis, The Institute of Electrical Engineers, UK, 1998 114. J. Reeve, G. Fahmy, B. Stott, ‘‘Versatile load flow method for multiterminal HVDC systems,’’ IEEE Transactions on Power Apparatus and Systems, 96(3), 925–932, 1977 115. H. Fudeh, C.M. Ong, ‘‘A simple and efficient AC-DC load flow method for miltiterminal DC systems,’’ IEEE Transactions on Power Apparatus and Systems, 100(11), 4389–4396, 1981 116. J. Arrillaga, C.P. Arnold, B.J. Harker, Computer Modeling of Electrical Power Systems, Wiley, New York, 1983 117. T. Smed, G. Andersson, G.B. Sheble´, L.L. Grigsby, ‘‘A new approach to AC/DC power flow,’’ IEEE Transactions on Power Systems, 6(3), 1238–1244, 1991 118. G.D. Breuer, J.F. Luini, C.C. Young, ‘‘Studies of large AC/DC systems on the digital computer,’’ IEEE Transactions on Power Apparatus and Systems, 85(11), 1107–1115, 1966 119. J.F. Clifford, A.H. Schmidt, ‘‘Digital representation of a DC transmission system and its control,’’ IEEE Transactions on Power Apparatus and Systems, 89(1), 97–105, 1970 120. N. Sato, N.V. David, S.M. Chan, A.L. Burn, J.J. Vithayathil, ‘‘Multiterminal HVDC system representation in a transient stability program,’’ IEEE Transactions on Power Apparatus and Systems, 99(5), 1927–1936, 1980 121. Working Group 38–01, Task Force no. 2 on SVC, CIGRE Report, Static Var Compensators, Ed. by I.A. Erimnez, CIGRE, UK, 1986 122. IEEE Special Stability Controls Working Group, ‘‘Static var compensator models for power flow and dynamic performance simulation’’, IEEE Transactions on Power Systems, 9(1), 229–240, 1994 123. L. Gyugyi, N.G. Hinggorani, P.R. Nannery, et al, ‘‘Advanced static var compensator using gate turn-off thyristors for utilities applications,’’ CIGRE Session, 1990, pp. 23–203 124. Y. Sumi, Y. Harumoto, T. Hasegawa, et al, ‘‘New static var control using force-commutated inverters,’’ IEEE Transactions on Power Apparatus and Systems, 100(9), 4216–4224, 1981 125. C.W. Edwards, P.R. Nannery, ‘‘Advanced static var generator employing GTO thyristors,’’ IEEE Transactions on Power Delivery, 3(4), 1622–1627, 1988 126. C. Schauder, M. Gernhardt, E. Stacey, et al, ‘‘Development of a 100Mvar static condensor for voltage control of transmission systems,’’ IEEE Transactions on Power Delivery, 10(3), 1486–1496, 1995 127. E.V. Larsen, K. Clark, S.A. Miske Jr., J. Urbanek, ‘‘Characteristics and rating consideration of thyristor controlled series compensation,’’ IEEE Transactions on Power Delivery, 9(2), 992–1000, 1994 128. G.G. Karady, T.H. Ortmeyer, B.R. Pilvelait, et al, ‘‘Continuously regulated series capacitor,’’ IEEE Transactions on Power Delivery, 8(3), 1348–1355, 1993 129. L. Gyugyi, C.D. Schauder, K.K. Sen, ‘‘Static synchronous series compensator: A solid-state approach to the series compensation of transmission lines,’’ IEEE Transactions on Power Delivery, 12(1), 406–417, 1997 130. Y.H. Song, A.T. Johns, Flexible AC transmission Systems (FACTS), IEE Press, London, 1999 131. S. Nyati, M. Eitzmann, J. Kappenmann, et al, ‘‘Design issues for a single core transformer thyristor controlled phase-angle regulator,’’ IEEE Transactions on Power Delivery, 10(4), 2013–2019, 1995 132. L. Gyugyi, ‘‘A unified power flow control concept for flexible AC transmission system,’’ Fifth International Conference on AC and DC Power Transmission, London, Sept’17–20, 1991 133. A. Nabavi-Niaki, M.R. Iravani, ‘‘Steady-state and dynamic models of unified power flow controller (UPFC) for power system studies’’, IEEE Transactions on Power Systems, 11(4), 1937–1943, 1996 134. Z.X. Han, ‘‘Phase shift and power flow control’’, IEEE Transactions on Power Apparatus and Systems, 101(10), 3790–3795, 1982

References

549

135. D.J. Gotham, G.T. Heydt, ‘‘Power flow control and studies for systems with FACTS devices,’’ IEEE Transactions on Power Systems, 13(1), 60–65, 1998 136. C.R. Fuerte-Esquivel, E. Acha, H. Ambriz-Prez, ‘‘A thyristor controlled series compensator model for the power flow solution of practical power networks,’’ IEEE Transactions on Power Systems, 15(1), 58–64, 2000 137. C.R. Fuerte-Esquivel, E. Acha, H. Ambriz-Prez, ‘‘A comprehensive Newton-Raphson UPFC for the quadratic power flow solution of practical power networks,’’ IEEE Transactions on Power Systems, 15(1), 102–109, 2000 138. H. Ambriz-Prez, E. Acha, C.R. Fuerte-Esquivel, ‘‘Advanced SVC models for NewtonRaphson load flow and Newton optimal power flow studies,’’ IEEE Transactions on Power Systems, 15(1), 129–946, 2000 139. C.R. Fuerte-Esquivel, E. Acha, ‘‘A Newton-type algorithm for the controlpower flow in electrical power networks,’’ IEEE Transactions on Power Systems, 12(4), 1474–1480, 1997 140. C.R. Fuerte-Esquivel, E. Acha, ‘‘Newton-Raphson algorithm for the reliable solution of large power networks with embedded FACTS devices,’’ IEE Proceedings of Generation, Transmission and Distribution, 143(5), 447–454, 1996 141. S. Arabi, P. Kundur, ‘‘A versatile FACTS device model for power flow and stability simulations,’’ IEEE Transactions on Power Systems, 11(4), 1944–1950, 1996 142. C.R. Fuerte-Esquivel, E. Acha, ‘‘United power flow controller: A critical comparison of Newton-Raphson UPFC algorithm in power flow studies,’’ IEE Proceedings of Generation, Transmission and Distribution, 144(5), 437–444, 1997 143. J.Y. Liu, Y.H. Song, ‘‘Strategies for handling UPFC constraints in steady-state power flow and voltage control,’’ IEEE Transactions on Power Systems, 15(2), 566–571, 2000 144. W. Fang, H.W. Ngan, ‘‘Control setting of unified power flow controllers through a robust load flow calculation’’, Proceedings of Generation, Transmission and Distribution, 146(4), 365–369, 1999 145. H. Sun, D.C. Yu, C. Luo, ‘‘A novel method of power flow analysis with Unified Power Flow Controller (UPFC),’’ IEEE Power Engineering Society Winter Meeting, 4, 2800–2805, 2000 146. A. Blondel, ‘‘The two-reaction method for study of oscillatory phenomena in coupled alternators’’, Revue Ge´ne´rale de Lelectricite´, 13, 235–251, February 1923; 515–531, March 1923 147. R.E. Doherty, C.A. Nickle, ‘‘Synchronous machines I and II’’, AIEE Transactions, 45, 912–942, 1926 148. R.H. Park, ‘‘Two-reaction theory of synchronous machines: Generalized method of analysis Part I’’, AIEE Transactions, 48, 716–727, 1929; Part II, 52, 352–355, 1933 149. C. Concordia, Synchronous Machine, Wiley, New york, 1951 150. G. Shackshaft, P.B. Henser, ‘‘Model of generator saturation for use in power system studies’’, Proceedings of IEE, 126(8), 759–763, 1979 151. G.R. Slemon, Magnetoelectric Devices, Wiley, New York, 1966 152. A.E. Fitzgerald, C. Kingsley, Electric Machinery, 2nd Edn., McGraw-Hill, New York, 1961 153. P. Kunder, Power System Stability and Control, McGraw-Hill, New York, 1994 154. D.W. Olive, ‘‘Digital simulation of synchronous machine transients’’, IEEE Transactions on Power Apparatus and Systems, 87(8), 1968 155. M.K. El-Sherbiny, A.M. El-Serafi, ‘‘Analysis of dynamic performance of saturated machine and analog simulation’’, IEEE Transactions on Power Apparatus and Systems, 101(7), 1899– 1906, 1982 156. D.W. Olive, ‘‘New techniques for the calculation of dynamic stability’’, IEEE Transactions on Power Apparatus and Systems, 85(7), 767–777, 1966 157. T.J. Hammons, D.J. Winning, ‘‘Comparisons of synchronous machine models in the study of the transient behaviour of electrical power systems’’, Proceedings of IEE, 118, 1442–1458, 1971 158. J. Arrillage, C.P. Arnold, B.J. Harker, Computer Modeling of Electrical Power Systems, Wiley, Chichester, 1983

550

References

159. IEEE Committee Report, ‘‘First benchmark model for computer simulation of subsynchronous resonance’’, IEEE Transactions on Power Apparatus and Systems, 96(5), 1565–1572, 1977 160. IEEE Power Engineering Society, IEEE Recommended Practice for Excitation System Models for Power System Stability Studies, IEEE Standards Board, New York, 1992 161. IEEE Working Group Report, ‘‘Hydraulic turbine and turbine control models for system dynamic studies’’, IEEE Transactions on Power Systems, PWRS-7(1), 167–179, 1992 162. D.G. Ramey, J.W. Skooglund, ‘‘Detailed hydrogovernor representation for system stability studies’’, IEEE Transactions on Power Apparatus and Systems, 89, 106–112, 1970 163. M. Leum, ‘‘The development and field experience of a transistor eletric governor for hydro turbines’’, IEEE Transactions on Power Apparatus and Systems, 85, 393–400, 1966 164. IEEE Working Group Report, ‘‘Dynamic models for fossil fueled steam units inpower system studies’’, IEEE Transactions on Power Systems, PWRS-6(2), 753–761, 1991 165. IEEE Committee Report, ‘‘Dynamic models for steam and hydro turbines in power system studies’’, IEEE Transactions on Power Apparatus and Systems, 92(6), 1904–1915, 1973 166. P. Kundur, D.C. Lee, J.P. Bayne, ‘‘Impact of turbine generator overspeed controls on unit performance under system disturbance conditions’’, IEEE Transactions on Power Apparatus and Systems, 104, 1262–1267, 1985 167. M.S. Baldwin, D.P. McFadden, ‘‘Power systems performance as affected by turbine-generator controls response during frequency disturbance’’, IEEE Transactions on Power Apparatus and Systems, 100, 2846–2494, 1981 168. IEEE Task Force on Load Representation for Dynamic Performance, ‘‘Standard load models for power flow and dynamic performance simulation’’, IEEE Transactions on Power Systems, 10(3), 1302–1313, 1995 169. IEEE Task Force on Load Representation for Dynamic Performance, ‘‘Load representation for dynamic performance analysis’’, IEEE Transactions on Power Systems, 8(2), 472–482, 1993 170. IEEE Task Force on Load Representation for Dynamic Performance System Dynamic Performance Subcommittee, Power System Engineering Committee, ‘‘Bibliography on load model for power flow and dynamic performance simulation’’, IEEE Transactions on Power Systems, 10(1), 523–538, 1995 171. T. Dovan, T.S. Dillon, C.S. Berger, K.E. Forward,‘‘A microcomputer based on-line identification approach to power system dynamic load modelling’’, IEEE Transactions on Power Systems, 2(3), 529–536, 1987 172. C.W. Talor, ‘‘Concepts of under-voltage load shedding for voltage stability’’, IEEE Transactions on Power Delivery, 7(2), 480–488, 1982 173. W.S. Kao, C.J. Lin, C.T. Huang, et al, ‘‘Comparison of simulated power system dynamics applying various load models with actual recorded data’’, IEEE Power Engineering Society Winter Meeting, 172–177, 1993 174. W.S. Kao, ‘‘The effect of load models on unstable low-frequency oscillation damping in taipower system experience w/wo power system stabilizers’’, IEEE Transactions on Power Systems, 16(3), 463–472, 2001 175. A. Borghetti, R. Caldon, A. Mari, et al, ‘‘On dynamic load models for voltage stability studies’’, IEEE Transactions on Power Systems, 12(1), 293–303, 1997 176. F. Nozari, M.D. Kankam, W.W. Price, ‘‘Aggregation of induction motors for transient stability load modeling’’, IEEE Transactions on Power systems, 2(4), 1096–1103, 1987 177. P. Kunder, Power System Stability and Control, McGraw-Hill, New York, 1994 178. P.M. Anderson, A.A. Fouad, Power System Control and Stability, The Iowa State University Press, Iowa, 1977 179. J. Arrillaga, C.P. Arnold, Computer Analysis of Power Systems, Wiley, New York, 1990 180. Y.H. Song, A.T. Johns, Flexible AC Transmission Systems (FACTS), The Institution of Electrical Engineers, London, 1999

References

551

181. J.A. Momoh, M.E. El-Hawary, Electric Systems, Dynamics, and Stability with Artificial Intelligence Applications, Marcel Dekker, New York, 2000 182. W.L. Brogan, Modern Control Theory, Prentice Hall, New Jersey, 1991 183. J.J.E. Slotine, W. Li, Applied Nonlinear Control, Prentice Hall, New Jersey, 1991 184. C.W. Gear, Numerical Initial Value Problems in Ordinary Differential Equations, Prentice Hall, New Jersey, 1971 185. L. Lapidus, J.H. Seinfeld, Numerical Solution of Ordinary Differential Equations, Academic Press, New York, 1971 186. J.D. Lambert, Computational Methods in Ordinary Differential Equations, Wiley, New York, 1973 187. J.H. Wilkinson, The Algebraic Eigenvalue Problem, Clarendon Press, Oxford, 1965 188. G.H. Golub, C.F. Van Loan, Matrix Computations, 3rd Edn., The Johns Hopkins University Press, 1996 189. G. Rogers, Power System Oscillations, Kluwer, Dordecht, 2000 190. P. Kundur, G.J. Rogers, D.Y. Wong, et al, ‘‘A comprehensive computer program package for small signal stability analysis of power systems’’, IEEE Transactions on Power Systems, 5 (4), 1076–1083, 1990 191. S. Aribi, G.J. Rogers, D.Y. Wong, et al, ‘‘Small signal stability analysis of SVC and HVDC in AC power systems’’, IEEE Transactions on Power Systems, 6(3), 1147–1153, 1991 192. I.J. Perez-Arriaga, G.C. Verghese, F.C. Schweppe, ‘‘Selective modal analysis with applications to electric power systems, Part I: Heuristic introduction, Part II: The dynamic stability problem’’, IEEE Transactions on Power Apparatus and Systems, 101(9), 3117–3134, 1982 193. J.L. Sancha, I.J. Perez-Arriaga, ‘‘Selective modal analysis of electric power system oscillatory instability’’, IEEE Transactions on Power Systems, 3(2), 429–438, 1988 194. R.T. Byerly, R.J. Bennon, D.E. Sherman, ‘‘Eigenvalue analysis of synchronizing power flow oscillations in large electric power systems’’, IEEE Transactions on Power Apparatus and Systems, 101(1), 235–243, 1982 195. N. Martins, ‘‘Efficient eigenvalue and frequency response methods applied to power system small-signal stability studies’’, IEEE Transactions on Power Systems, 1(1), 217–226, 1986 196. D.Y. Wong, G.J. Rogers, B. Porretta, P. Kundur, ‘‘Eigenvalue analysis of very large power systems’’, IEEE Transactions on Power Systems, 3(2), 472–480, 1988 197. P.W. Sauer, C. Rajagopalan, M.P. Pai, ‘‘An explanation and generalization of the AESOPS and PEALS algorithms’’, IEEE Transactions on Power Systems, 6(1), 293–299, 1991 198. N. Uchida, T. Nagao, ‘‘A new Eigen-analysis method of steady-state stability studies for large power systems: S matrix method’’, IEEE Transactions on Power Systems, 3(2), 706– 714, 1988 199. W.J. Stewart, A. Jennings, ‘‘A simultaneous iteration algorithm for real matrices’’, ACM Transactions on Mathematical Software, 7(2), 184–198, 1981 200. S. Duff, J.A. Scott, ‘‘Computing selected eigenvalues of sparse unsymmetric matrices using subspace iteration’’, ACM Transactions on Mathematical Software, 19(2), 137–159, 1993 201. J.A. Scott, ‘‘An Arnoldi code for computing selected Eigenvalues of sparse, real, unsymmetric matrices’’, ACM Transactions on Mathematical Software, 21(4), 432–475, 1995 202. A. Semlyen, L. Wang, ‘‘Sequential computation of the complete eigensystem for the study zone in small signal stability analysis of large power systems’’, IEEE Transactions on Power Systems, 3(2), 715–725, 1988 203. L. Wang, A. Semlyen, ‘‘Application of sparse eigenvalue techniques to the small signal stability analysis of large power systems’’, IEEE Transactions on Power Systems, 5(4), 635–642, 1990 204. D.J. Stadnicki, J.E. Van Ness, ‘‘Invariant subspace method for eigenvalue computation’’, IEEE Transactions on Power Systems, 8(2), 572–580, 1993 205. N. Mori, J. Kanno, S. Tsuzuki, ‘‘A sparsed–oriented techniques for power system small signal stability analysis with a precondition conjugate residual method’’, IEEE Transactions on Power Systems, 8(3), 1150–1158, 1993

552

References

206. G. Angelidis, A. Semlyen, ‘‘Efficient calculation of critical eigenvalue clusters in the small signal stability analysis of large power systems’’, IEEE Transactions on Power Systems, 10 (1), 427–432, 1995 207. L.T.G. Lima, L.H. Bezerra, C. Tomei, N. Martins, ‘‘New methods for fast small-signal stability assessment of large scale power systems’’, IEEE Transactions on Power Systems, 10(4), 1979–1985, 1995 208. G. Angelidis, A. Semlyen, ‘‘Improved methodologies for the calculation of critical eigenvalues in small signal stability analysis’’, IEEE Transactions on Power Systems, 11(3), 1209–1217, 1996 209. J.M. Campagnolo, N. Martins, D.M. Falcao, ‘‘Refactored bi-Iteration: A high performance eigensolution method for large power system matrices’’, IEEE Transactions on Power Systems, 11(3), 1228–1235, 1996 210. N. Martins, L.T.G. Lima, H.J.C.P. Pinto, ‘‘Computing dominant poles of power system transfer functions’’, IEEE Transactions on Power Systems, 11(1), 162–1170, 1996 211. N. Martins, ‘‘The dominant pole spectrum eigenslover’’, IEEE Transactions on Power Systems, 12(1), 245–254, 1997 212. J.M. Campagnolo, N. Martins, J.L.R. Pereira, et al, ‘‘Fast small-signal stability assessment using parallel processing’’, IEEE Transactions on Power Systems, 9(2), 949–956, 1994 213. J.M. Campagnolo, N. Martins, D.M. Falcao, ‘‘An efficient and robust eigenvalue method for small-signal stability assessment in parallel computers’’, IEEE Transactions on Power Systems, 10(1), 506–511, 1995 214. V. Ajjarapu, ‘‘Reducibility and eigenvalue sensitivity for identifying critical generations in multimachine power systems’’, IEEE Transactions on Power Systems, 5(3), 712–719, 1990 215. T. Smed, ‘‘Feasible eigenvalue sensitivity for large power systems’’, IEEE Transactions on Power Systems, 8(2), 555–563, 1993 216. H.K. Nam, Y.K. Kim, ‘‘A new eigen-sensitivity theory of augmented matrix and its applications to power system stability’’, IEEE Transactions on Power Systems, 15(1), 363–369, 2000 217. K.W. Wang, C.Y. Chung, ‘‘Multimachine eigenvalue sensitivities of power system parameters’’, IEEE Transactions on Power Systems, 15(2), 741–747, 2000 218. F.P. Demello, C. Concordia, ‘‘Concepts of synchronous machine stability as affected by excitation control’’, IEEE Transactions on Power Apparatus and Systems, 88(4), 316–329, 1969 219. P. Kunder, D.C. Lee, H.M. Zein-el-din, ‘‘Power system stabilizers for thermal units: Analytical techniques and on-site validation’’, IEEE Transactions on Power Apparatus and Systems, 100(1), 81–89, 1981 220. M. Klein, G.J. Rogers, P. Kundur, ‘‘A fundamental study of inter-area oscillations in power systems’’, IEEE Transactions on Power Systems, 6(3), 914–921, 1991 221. G. Rogers, ‘‘Demystifying power system oscillations’’, IEEE Computer Application in Power, 9(3), 30–35, 1996 222. J. Hauer, D. Trudnowski, G. Rogers, et al, ‘‘Keeping an eye on power system dynamics’’, IEEE Computer Application in Power, 10(4), 50–54, 1997 223. G. Rogers, ‘‘Power system structure and oscillations’’, IEEE Computer Application in Power, 12(2), 14,16,18,20,21, 1999 224. S.K. Starrett, A.A. Fouad, ‘‘Nonlinear measures of mode-machine participation’’, IEEE Transactions on Power Systems, 13(2), 389–394, 1998 225. F.L. Pagola, I.J.P. Arriaga, G.C. Verghese, ‘‘On sensitivity, residues and participations: applications to oscillatory stability analysis and control’’, IEEE Transactions on Power Systems, 4(1), 278–285, 1989 226. M. Klein, G.J. Rogers, S. Moorty, P. Kundur, ‘‘Analytical investigation of factors influencing power system stabilizers performance’’, IEEE Transactions on Energy Conversion, 7(3), 382–390, 1992

References

553

227. P. Kundur, M. Klein, G.J. Rogers, M.S. Zywno, ‘‘Application of power system stabilizers for enhancement of overall system stability’’, IEEE Transactions on Power Systems, 4(2), 614–626, 1989 228. L. Xu, S. Ahmed-Zaid, ‘‘Tuning of power system controllers using symbolic eigensensitivity analysis and linear programming’’, IEEE Transactions on Power Systems, 10(1), 314–322, 1995 229. J.F. Hauer, F. Vakili, ‘‘A oscillation detector used in the bpa power system disturbance monitor’’, IEEE Transactions on Power Systems, 5(1), 74–79, 1990

Index

AC exciter, 365, 369, 371–374, 379–380 Alternate solution method, 425–427 Ancillary services, 193, 195, 221, 224 Artificial intelligence methods, 201, 202 ASVG, 308–312 Asymptotically stable, 490, 492, 509 ATC1, 248–250, 252–253 ATC2, 248–250 Augmented matrix, 23–26, 67, 93, 98 Automatic generation control (AGC), 76, 250, 251, 406, 534 Automatic voltage regulator (AVR), 333, 352, 357, 363, 375, 377, 378, 534–536 Available transfer capabilities (ATC), 196, 222–224, 241–249, 253 Axiomatic definitions of probability, 131 Bayes’ formula, 132 Binomial distribution, 136, 244, 245 Bipolar system, 258 Blackout, 114, 129, 152, 178, 179, 191 Branch addition method, 56, 57, 64, 65 BX algorithm, 106 Cascading outages, 179–180 CBM, 242 Characteristic equation, 520 Characteristic polynomial, 520 Classical model, 351, 352, 384, 399, 425, 434, 454, 457, 459, 463, 511 Common mode failures, 114 Compensation method, 38, 113–119, 460 Conditional probability, 131, 139, 158 Congestion management, 196, 218, 222–223 Constant eq0 model, 351, 352

Contingency ranking, 123–124, 127 Continuation power flow (CPF), 243 Continuous random variable, 132–137, 162, 173, 244 Convergence characteristic, 73, 89, 106, 112, 216 Convergence condition, 89, 104, 109, 112 Convergence of the Monte Carlo simulation, 146 Convergence property, 100, 112, 113 Converter, 258–262, 264–270, 272, 273, 276–286, 289, 294–301, 326, 329, 331, 392, 479, 480, 505, 507 Converter basic equations, 261, 267, 278, 282, 283, 285, 289 Converter bridge, 259, 260, 280 Converter control, 260, 279, 285, 289, 297 Converter equivalent circuits, 273–276 Converter transformer, 259, 261, 267, 276–286, 294, 296–298 Convolution of random variable, 135 Coordinate transformation, 338, 432, 448, 494, 496, 499 Correction equations, 81, 83, 87, 89, 94, 103–106, 109 Correction equations of fast decoupled method, 104–107 Critically stable, 490, 492 Cumulant method, 162 Cumulative probability, 141–144, 154, 188 Current decomposition, 229, 230, 234 Damping ratio, 523, 536, 537, 540 Damping winding, 335–336, 351, 352, 354, 356, 370, 536

555

556

d axis open-circuit sub-transient time constant, 346, 347, 349, 350, 359, 465–467, 494 d axis open-circuit transient time constant, 346 d axis subtransient reactance, 344–351, 356, 359, 360, 363, 432, 465–467, 475, 485, 494 d axis synchronous reactance, 344–351, 356, 358, 359, 401, 431, 432, 453, 465 d axis transient reactance, 344, 345, 347–352, 356, 359, 370, 374, 401, 432–434, 448, 453, 454, 465–467, 475, 494 DC exciter, 365–367, 369, 370, 378, 379, 494 DC load flow, 114, 119–127, 150, 154 DC network equations, 285, 286, 297 Decaying mode, 523 Decremental bidding prices, 224 Difference equations, 419–423, 464, 466, 470, 472, 474, 475, 483, 484, 487 Differential-algebraic equations, 407, 425, 427–429, 454, 472, 492, 530 Discrete random variable, 132–135, 163, 173 Distribution factor, 229, 230, 235–238, 240, 241, 243 Distribution function, 132–135, 137, 138, 141, 149, 166, 172, 173, 178, 245 Dominant eigenvalue, 527–529 dq0 coordinate system, 335 dq0 transformation, 338 Dynamic load model, 397–403 Dynamic ordering scheme, 46–48 Edgeworth series, 162, 166, 167 Effect of saturation, 357 Eigensolution, 489, 492, 493, 510, 519, 521, 526, 527, 530, 531 Eigensolution analysis, 489, 492, 519 Eigenvalue sensitivity, 524–525, 531–534, 541 Eigenvalue sensitivity analysis, 533–534 Electromagnetic torque, 361–363 Electromechanical oscillations, 492, 493, 536, 540

Index

Energy management system (EMS), 243, 457 Enumeration method, 159, 183, 244 Equilibrium point, 490–492, 509 Equivalent circuits of transformer, 9–11 ETC, 242 Euler method, 409, 412–417, 419, 421, 422, 424, 425, 447, 450, 452 Excitation system, 334, 363–365, 375–381, 407, 430, 446, 463, 464, 469, 472, 473, 494, 495, 498, 511, 535, 536, 540 Expected energy not supplied (EENS), 152, 155, 156, 159 Extended OPF problems, 224 Factor table, 27–31, 34–37, 39, 40, 42–44, 105, 108–111, 114–116, 118, 292, 295 FACTS, 9, 222, 224, 255–258, 260, 301–302, 325, 407, 436, 463, 464, 475, 502, 510, 534 Fast decoupled method, 72, 77, 88, 101–113, 118, 120 Feasible solution, 182–189, 199, 202, 256 Field winding, 335, 336, 350–352, 359, 363, 364, 369–371, 398, 400 Financial transmission right (FTR), 222 Firing angle, 260, 262, 264–269, 271–274, 279, 280, 299–302, 304, 305, 314, 316, 318, 372, 374, 476 Flat start, 89, 289 Fractional frequency transmission system (FFTS), 255 Gauss elimination, 22–27, 31, 43, 44, 89, 93, 94, 434, 447, 448, 450 Genetic algorithm (GA), 73, 201, 243 Gibbs sampler, 157–159 Governing system, 361, 362, 381, 382, 386–389, 391–393, 407, 430, 446, 463, 464, 472, 473, 475, 496, 498 Gram-Charlier series expansion, 162, 166, 170, 173 Harmonic voltages, 259, 260 Hermite polynomial, 167, 168 Hessenberg matrix, 527 Homopolar connection, 258 Householder matrix, 528, 529

Index

HVDC, 255, 256, 258–261, 272, 276, 279, 281, 299, 300, 341, 372, 430, 436, 437, 464, 475, 479, 485–487, 493, 503, 510, 534, 535 HVDC dynamic mathematical models, 299–301 Hydraulic turbine, 381–388, 406, 496 Ideal synchronous generator, 336–338 Implicit integration methods, 419–424 Incidence matrix, 4–7, 9, 122, 247 Incremental bidding price, 224 Independent operator (ISO), 193–196, 222, 223, 226, 227 Independent power producers (IPPs), 228, 235 Initial condition, 143, 157, 234, 315, 354, 408 Initial value problem, 407, 420, 421, 425 Interarea oscillation, 535, 536, 541 Interior point method (IPM), 200–202, 207, 208, 216, 219 Inverse power method, 529–532 Inverter, 259, 260, 266, 272–274, 279, 280, 284–286, 289, 299, 301, 308–310, 319, 320, 326, 437, 438, 481, 483, 484, 503, 504, 506, 507 Jacobean, 72 Jacobean matrix, 72 Lagrangian function, 187, 204 Lagrangian multipliers, 202, 207, 216, 221 Linear optimal excitation controller (LOEC), 364 Linear programming, 151, 198–200, 223, 243 Load flow, 71–127, 130, 150, 151, 154, 155, 161–162, 168–173, 175, 176, 178, 189 Load model, 140, 141, 394–397, 399, 400, 403, 500, 501 Local-mode oscillation, 535, 536 Location marginal price (LMP), 222 Loop current method, 2 Loss allocation, 229, 231–235, 238, 240 Loss of load probability (LOLP), 145, 151, 155, 159–161

557

Lower boundary point, 186, 188 Lyapunov linearized method, 489–491 Market clearing price (MCP), 195 Markov chain, 139, 156, 157, 159, 160 Markov chain Monte Carlo (MCMC) simulation, 156–159 Markov process, 138–140 Mathematical expectation, 134, 173 Mean value, 134, 137 Memory requirement, 71, 72, 89, 149 Midterm and long-term stability, 405 Mixed programming, 198, 200 Modal analysis, 523, 537 Model of load curtailment, 150–151, 153 Modified Euler’s method, 412–417, 421, 425, 447, 450, 452 Monopolar system, 258 Monte Carlo simulation, 130, 145–149, 152, 153, 155, 157, 159, 244–248, 250 MSC, 306 Multi-fold outages, 114 Multiple bridge, 276 Multistep or multivalue algorithms, 411 Mutual admittance, 3, 4, 14, 19, 21 Mutual inductance, 337, 338, 340, 343, 357, 366 Newton-Raphson method, 22, 72, 76, 79, 81, 87–90, 93, 100, 172, 173, 175, 198 N-l checking, 123–124 Nodal admittance matrix, 1, 4, 7, 13–22, 48–50, 72, 104, 247 Nodal impedance matrix, 1, 48–64, 72, 122, 123 Nodal self-admittance, 3, 4 Node power equations, 76–78, 284 Node voltage method, 2 Nondamping winding model, 351, 352 Nonlinear algebraic equation, 71, 426 Nonlinear optimal excitation controller (NOEC), 364 Nonlinear programming, 196, 198–199 Nonoscillatory mode, 523 Non-periodic instability, 523 Normal distribution, 137–138, 140–141, 146, 167, 168, 244–246, 250

558

Numeral characteristics of random variable, 133–135 Numerical stability, 421, 424, 425, 430, 462 One-step transition probability, 139, 140 Open access same-time information system (OASIS), 243–244 Operation risk, 129 Optimal ordering schemes, 43 Optimal power flow (OPF), 1, 38, 196–203, 207, 209, 216–224, 226, 227, 243, 257 Ordinary differential equations, 299, 407, 408, 424 Orthogonal matrix, 499, 526, 527 Oscillation analysis, 493, 534, 537, 540 Outage analysis, 122, 123, 127 Outage table, 142–144, 150, 188 Parallel computing algorithms, 73 Park’s transformation, 335, 338–340 Participation factor, 525–526, 541 Participation matrix, 525 Per unit equations, 282, 340–341, 343, 401 Phase shifting transformer, 1, 11, 13, 16, 18, 20, 38, 301, 322, 325 Piecewise solution method, 72 Polar form of the nodal power equations, 77 Potier voltage, 357, 360 Power exchange (PX), 193–195, 252, 309, 535 Power flow tracing, 196, 228–241 Power market, 114, 129, 193–196, 222–224, 241, 242, 248, 253, 256 Power method, 527–532 Power rectifier, 371, 373, 379, 380 Power system stabilizers (PSS), 364, 365, 377, 407, 468, 469, 471, 495, 533, 535, 541 P-Q decoupled method, 72, 73, 107, 113, 290, 291, 294 PQ nodes, 76, 85, 90, 109, 111 Predictor-corrector method, 462 Prime mover, 142, 360–362, 381, 382, 406, 407, 430, 446–448, 463, 466, 472, 473, 496 Probabilistic load flow, 130, 161–178 Probabilistic model of load, 140–141

Index

Probabilistic models of transformer and generator, 142 Probability density function, 133–135, 137, 138, 171, 177–179 Probability of stochastic events, 130–132 Proportional-integral-differential (PID), 364, 388, 392 Pseudo-random numbers, 149 12-pulse converter, 277, 278 PV nodes, 76, 83, 85, 88, 105, 109, 110 q axis sub-transient reactance, 344 q axis synchronous reactance, 340, 344 q axis transient reactance, 344, 356 QR method, 511, 518, 526–527, 531 Quadratic programming, 198, 199 Quasi-steady state model, 480–484, 493 Random process, 138, 139, 159 Random variable, 130, 132–139, 143, 145–148, 151, 162–164, 166–168, 170–173, 244, 245 Random variable’s moment, 163 Real-time balancing market (RBM), 223, 224 Rectangular form of the nodal power equations, 78 Recursive formula, 143, 144, 410, 411, 419–421 Reliability of transmission system, 188–191 Right eigenvector, 520, 521, 523–525, 531, 533, 537–540 Rotor motion equation, 352, 360–362, 398–400, 464 Round-off error, 412, 424 Round rotor generator, 335, 336, 338, 340 Runge-Kutta method, 417–419, 421, 425, 464 Salient-pole generator, 335, 336, 340, 351 Sampling, 145, 147–150, 152, 155, 157–160, 253, 254 Saturation effect, 336, 337, 359, 366, 369 Saturation factor, 355, 366, 369, 370, 379 Schur decomposition, 526, 527 Self inductance, 337, 338, 340, 365, 366 Semi-dynamic ordering Scheme, 45, 46 Sensitivity method, 114, 246

Index

Simplified models for transient stability, 446 Simultaneous solution method, 425, 427 Single-step algorithm, 411 Slack node, 76, 83, 85, 90, 94, 174, 216, 247, 250 Small-signal stability, 489–493, 500, 506–518, 530, 536 Smoothing reactor, 259, 260, 264, 480 Sparse vector method, 22, 38–43, 460, 461 Sparsity techniques, 1 Spot pricing, 195, 196, 219–221 SPWM, 326 SSSC, 301, 302, 319–322, 326, 329, 331 STATCOM, 301, 302, 308–312, 319, 326, 329, 331 State matrix, 492, 509–511, 518, 519, 533 State space, 139, 145, 150 Static load model, 394–397, 399, 400, 500 Static ordering scheme, 44–46 Static security analysis, 73, 113–114, 189 Stationary exciter, 374, 375 Steady-state equations, 354, 356, 360, 437, 438 Steam turbine, 362, 381, 382, 389–393, 492, 493 Step-by-step integration, 408, 409 Step size, 143, 199, 408, 409, 411, 412, 420, 421, 423–426, 428, 454, 462–464, 466, 467 Stiff differential equations, 425 Stiff rotor, 360–362 Stochastic programming, 244 Stormer and Numerov integration formula, 462 Sub-synchronous oscillation, 493 Sub-transient parameter, 335, 344 SVR, 301, 302 Symmetric, 13, 16, 19, 32, 34, 38, 51, 102, 104, 105, 125, 261, 299, 308, 314 Synchronous generator, 334–340, 343, 344, 347–363, 365, 369–372, 398, 400–402, 493–496, 498, 500, 535 Synthesized impedance matrix, 440, 441, 444, 446, 486 System contingency, 124 System performance index, 124, 126, 127

559

Taylor series, 79, 81, 84, 86, 168, 169 TCPST, 301, 302, 322, 323, 325, 329 TCR. See Thyristor controlled reactors TCSC, 301, 302, 313, 314, 316–319, 322, 325, 436, 437, 464, 477–479, 485, 487, 502, 503, 507, 508, 510 Telegen’s theorem, 124 Three rotor winding model, 351 Three-winding transformer, 10, 11 Thyristor controlled reactors, 302–307, 313, 316, 476, 478 Thyristor switched capacitors, 302 Time domain, 522 Torsional oscillations, 365, 492, 493 Torsional torque, 493 Transformation matrix, 170, 335, 340 Transient parameter, 335, 344 Transient stability, 357, 362, 400, 405–407, 425, 427–431, 435, 444, 446, 447, 450, 453, 457, 459, 463, 464, 480, 484, 537 Transmission open access, 193, 221 Transmission right, 222, 224 Trapezoidal rule, 419–421, 425–427, 464, 466, 470, 471, 474, 477–479, 482, 483 Triangular decomposition, 27–34, 530–532 TRM, 242 TSC, 302, 303, 305, 306 TTC, 224, 242, 243 Two winding model, 351–352 Uniform distribution, 136–137, 148–150, 154, 158, 251 UPFC, 301, 302, 325–331 Usage sharing problem, 234 Variance, 134–137, 140, 141, 145–147, 155, 157, 159, 160, 168, 245, 246, 394 Voltage regulators (AVR), 333, 352, 357, 363, 375, 377, 378, 534–536 XB algorithm, 106

Xi-Fan Wang

l

Yonghua Song

l

Malcolm Irving

Modern Power Systems Analysis

123

Xi-Fan Wang Xi’an Jiaotong University Xi’an People’s Republic of China

Yonghua Song The University of Liverpool Liverpool United Kingdom

Malcolm Irving Brunel University Middlesex United Kingdom

ISBN 978-0-387-72852-0

e-ISBN 978-0-387-72853-7

Library of Congress Control Number: 2008924670 # 2008 Springer Science+Business Media, LLC All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. While the advice and information in this book are believed to be true and accurate at the date of going to press, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein. Printed on acid-free paper. 987654321 springer.com

Preface

The power industry, a capital and technology intensive industry, is a basic national infrastructure. Its security, reliability, and economy have enormous and far-reaching effects on a national economy. An electrical power system is a typical large-scale system. Questions such as how to reflect accurately the characteristics of modern electrical power systems, how to analyze effectively their operating features, and how to improve further the operating performance are always at the forefront of electrical power systems research. Electrical power system analysis is used as the basic and fundamental measure to study planning and operating problems. In the last century, electrical power researchers have undertaken a great deal of investigation and development in this area, have made great progress in theoretical analysis and numerical calculation, and have written excellent monographs and textbooks. Over the last 20 years, the changes in electrical power systems and other relevant technologies have had a profound influence on the techniques and methodologies of electrical power system analysis. First, the development of digital computer technology has significantly improved the performance of hardware and software. Now, we can easily deal with load flow issues with over ten thousand nodes. Optimal load flow and static security analysis, which were once considered hard problems, have attained online practical applications. Second, the applications of HVDC and AC flexible transmission technologies (FACTS) have added new control measures to electrical power systems, and have increased power transmission capacity, enhanced control capability, and improved operating characteristics. However, these technologies bring new challenges into the area of electrical power system analysis. We must build corresponding mathematical models for these new devices and develop algorithms for static and dynamic analysis of electrical power systems including these devices. In addition, the rapid development of communication technology has enabled online monitoring of electrical power systems. Therefore, the demand for online software for electrical power system analysis becomes more and more pressing. Furthermore, worldwide power industry restructuring and deregulation has separated the former vertically integrated system into various parts, and the once

v

vi

Preface

unified problem of power system dispatching is now conducted via complicated bilateral contracts and spot markets. New issues such as transmission ancillary service and transmission congestion have emerged. In recent years, several power blackouts have taken place worldwide, especially the ‘‘8.13’’ blackout on the eastern grid of USA and Canada and the blackouts that occurred successively in other countries have attracted a great deal of attention. All of these aspects require new theories, models, and algorithms for electrical power system analysis. It is within such an environment that this book has been developed. The book is written as a textbook for senior students and postgraduates as well as a reference book for power system researchers. We acknowledge the support from various research funding organizations, their colleagues, and students, especially, the special funds for Major State Basic Research Projects of China ‘‘Research on Power System Reliability under Deregulated Environment of Power Market’’ (2004CB217905). We express our special gratitude to Professor Wan-Liang Fang and Professor Zheng-Chun Du for providing the original materials of Chaps. 5 and 6, and 7 and 8, respectively. We also express our sincere gratitude to the following colleagues for their contributions to various chapters of the book: Professor Zhao-Hong Bie for Chaps. 1 and 3; Professor Xiu-Li Wang for Chaps. 2 and 4; Dr. Ze-Chun Hu for Chap. 3; Dr. Xiao-Ying Ding for Chap. 4; Dr. Lin Duan for Chaps. 5 and 6; Professor De-Chiang Gang for Chap. 7; and Professor Hai-Feng Wang for Chaps. 6 and 8. Xi’an, China Liverpool, UK London, UK

Xi-Fan Wang Yonghuna Song Malcolm Irving

Contents

1

Mathematical Model and Solution of Electric Network . . . . . . . . . . . . . . . . . 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.1 Node Equation and Loop Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2.2 Equivalent Circuit of Transformer and Phase Shift Transformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.3 Nodal Admittance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.1 Basic Concept of Nodal Admittance Matrix . . . . . . . . . . . . . . . . . . 13 1.3.2 Formulation and Modification of Nodal Admittance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.4 Solution to Electric Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.4.1 Gauss Elimination Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.4.2 Triangular Decomposition and Factor Table . . . . . . . . . . . . . . . . . 27 1.4.3 Sparse Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1.4.4 Sparse Vector Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 1.4.5 Optimal Ordering Schemes of Electric Network Nodes . . . . . . 43 1.5 Nodal Impedance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 1.5.1 Basic Concept of Nodal Impedance Matrix . . . . . . . . . . . . . . . . . . 48 1.5.2 Forming Nodal Impedance Matrix Using Admittance Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 1.5.3 Forming Nodal Impedance Matrix by Branch Addition Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2

Load Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Formulation of Load Flow Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Classification of Node Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Node Power Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Load Flow Solution by Newton Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Basic Concept of Newton Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Correction Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71 71 73 73 76 79 79 83

vii

viii

Contents

2.4

2.5

2.6

3

2.3.3 Solution Process of Newton Method . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2.3.4 Solution of Correction Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 Fast Decoupled Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 2.4.1 Introduction to Fast Decoupled Method . . . . . . . . . . . . . . . . . . . . . 101 2.4.2 Correction Equations of Fast Decoupled method . . . . . . . . . . . . 104 2.4.3 Flowchart of Fast Decoupled Method . . . . . . . . . . . . . . . . . . . . . . . 107 Static Security Analysis and Compensation Method . . . . . . . . . . . . . . . . 113 2.5.1 Survey of Static Security Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 2.5.2 Compensation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 DC Load Flow Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 2.6.1 Model of DC Load Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 2.6.2 Outage Analysis by DC Load Flow Method . . . . . . . . . . . . . . . . . 122 2.6.3 N-1 Checking and Contingency Ranking Method . . . . . . . . . . . 123

Stochastic Security Analysis of Electrical Power Systems . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Basic Concepts of Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Probability of Stochastic Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Random Variables and its Distribution . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Numeral Character of Random Variable . . . . . . . . . . . . . . . . . . . . 3.2.4 Convolution of Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.5 Several Usual Random Variable Distributions . . . . . . . . . . . . . . 3.2.6 Markov Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Probabilistic Model of Power Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Probabilistic Model of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Probabilistic Model of Power System Components . . . . . . . . . 3.3.3 Outage Table of Power System Components . . . . . . . . . . . . . . . . 3.4 Monte Carlo Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Fundamental Theory of Monte Carlo Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Sampling of System Operation State . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 State Evaluation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Indices of Reliability Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.5 Flowchart of Composite System Adequacy Evaluation . . . . . 3.4.6 Markov Chain Monte Carlo (MCMC) Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Probabilistic Load Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Cumulants of Random Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.2 Linearization of Load Flow Equation . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3 Computing Process of Probabilistic Load Flow . . . . . . . . . . . . . 3.6 Probabilistic Network-Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.2 Network-Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.3 Lower Boundary Points of Feasible Flow Solutions . . . . . . . . 3.6.4 Reliability of Transmission System . . . . . . . . . . . . . . . . . . . . . . . . . .

129 129 130 130 132 133 135 136 138 140 140 141 142 145 145 148 150 151 152 156 161 162 168 171 178 178 180 186 188

Contents

4

5

ix

Power Flow Analysis in Market Environment . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Transmission Owner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 Independent Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.3 Power Exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.4 Ancillary Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.5 Scheduling Coordinator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Optimal Power Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 General Formulation of OPF Problem . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Approaches to OPF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Interior Point Method (IPM) for OPF Problem . . . . . . . . . . . . . . 4.3 Application of Optimal Power Flow in Electricity Market . . . . . . . . . . 4.3.1 Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Congestion Management Method Based On OPF . . . . . . . . . . . 4.4 Power Flow Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Current Decomposition Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2 Mathematical Model of Loss Allocation . . . . . . . . . . . . . . . . . . . . 4.4.3 Usage Sharing Problem of Transmission Facilities . . . . . . . . . . 4.4.4 Methodology of Graph Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Available Transfer Capability of Transmission System . . . . . . . . . . . . . 4.5.1 Introduction To Available Transfer Capability . . . . . . . . . . . . . . 4.5.2 Application of Monte Carlo Simulation in ATC Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.3 ATC Calculation with Sensitivity Analysis Method . . . . . . . .

193 193 193 194 194 195 195 196 196 198 202 217 217 223 228 230 232 234 238 241 241

HVDC and FACTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 HVDC Basic Principles and Mathematical Models . . . . . . . . . . . . . . . . . 5.2.1 HVDC Basic Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Converter Basic Equations Neglecting Lc . . . . . . . . . . . . . . . . . . . 5.2.3 Converter Basic Equations Considering Lc . . . . . . . . . . . . . . . . . 5.2.4 Converter Equivalent Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.5 Multiple Bridge Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.6 Converter Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Power Flow Calculation of AC/DC Interconnected Systems . . . . . . . 5.3.1 Converter Basic Equations in per Unit System . . . . . . . . . . . . . . 5.3.2 Power Flow Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.3 Jacobian Matrix of Power Flow Equations . . . . . . . . . . . . . . . . . . 5.3.4 Integrated Iteration formula of AC/DC Interconnected Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.5 Alternating Iteration for AC/DC Interconnected Systems . . . 5.4 HVDC Dynamic Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Basic Principles and Mathematical Models of FACTS . . . . . . . . . . . . . 5.5.1 Basic Principle and Mathematical Model of SVC . . . . . . . . . . . 5.5.2 Basic Principle and Mathematical Model of STATCOM . . .

255 255 258 258 261 267 273 276 279 281 282 283 286

245 246

289 294 299 301 302 308

x

Contents

5.5.3 5.5.4 5.5.5 5.5.6 6

7

Basic Principle and Mathematical Model of TCSC . . . . . . . . . Basic Principle and Mathematical Model of SSSC . . . . . . . . . . Basic Principle and Mathematical Model of TCPST . . . . . . . . Basic Principle and Mathematical Model of UPFC . . . . . . . . .

313 319 322 325

Mathematical Model of Synchronous Generator and Load . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Mathematical Model of Synchronous Generator . . . . . . . . . . . . . . . . . . . . 6.2.1 Basic Mathematical Equations of Synchronous Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2 Mathematical Equations of Synchronous Generator Using Machine Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.3 Simplified Mathematical Model of Synchronous Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.4 Steady-State Equations and Phasor Diagram . . . . . . . . . . . . . . . . 6.2.5 Mathematical Equations Considering Effect of Saturation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.6 Rotor Motion Equation of Synchronous Generator . . . . . . . . . . 6.3 Mathematical Model of Generator Excitation Systems . . . . . . . . . . . . . 6.3.1 Mathematical Model of Exciter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.2 Voltage Measurement and Load Compensation Unit . . . . . . . 6.3.3 Limiters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.4 Mathematical Model of Power System Stabilizer . . . . . . . . . . . 6.3.5 Mathematical Model of Excitation Systems . . . . . . . . . . . . . . . . . 6.4 Mathematical Model of Prime Mover and Governing System . . . . . . 6.4.1 Mathematical Model of Hydro-Turbine and Governing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.2 Mathematical Model of Steam Turbine and Governing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Mathematical Model of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.1 Static Load Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.2 Dynamic Load Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

333 333 335

Power System Transient Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Numerical Methods for Transient Stability Analysis . . . . . . . . . . . . . . . . 7.2.1 Numerical Methods for Ordinary Differential Equations . . . 7.2.2 Numerical Methods for Differential-Algebraic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.3 General Procedure for Transient Stability Analysis . . . . . . . . . 7.3 Network Mathematical Model for Transient Stability Analysis . . . . . 7.3.1 The Relationship Between Network and Dynamic Devices . 7.3.2 Modeling Network Switching and Faults . . . . . . . . . . . . . . . . . . . .

336 343 351 354 357 360 363 365 375 376 377 377 381 382 389 393 395 397 405 405 407 408 425 427 430 431 439

Contents

7.4

Transient Stability Analysis with Simplified Model . . . . . . . . . . . . . . . . . 7.4.1 Computing Initial Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.2 Solving Network Equations with Direct Method . . . . . . . . . . . . 7.4.3 Solving Differential Equations by Modified Euler Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.4 Numerical Integration Methods for Transient Stability Analysis under Classical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transient Stability Analysis with FACTS Devices . . . . . . . . . . . . . . . . . . 7.5.1 Initial Values and Difference Equations of Generators . . . . . 7.5.2 Initial Values and Difference Equations of FACTS and HVDC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.3 Forming Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.4 Simultaneous Solution of Difference and Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

446 447 448

Small-Signal Stability Analysis of Power Systems . . . . . . . . . . . . . . . . . . . . . 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Linearized Equations of Power System Dynamic Components . . . . . 8.2.1 Linearized Equation of Synchronous Generator . . . . . . . . . . . . . 8.2.2 Linearized Equation of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.3 Linearized Equation of FACTS Components . . . . . . . . . . . . . . . . 8.2.4 Linearized Equation of HVDC Transmission System . . . . . . . 8.3 Steps in Small-Signal Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.1 Network Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.2 Linearized Differential Equations of Whole Power System . 8.3.3 Program Package for Small-Signal Stability Analysis . . . . . . 8.4 Eigenvalue Problem in Small-Signal Stability Analysis . . . . . . . . . . . . 8.4.1 Characteristics of State Matrix Given by Its Eigensolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.2 Modal Analysis of Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.3 Computation of Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.4 Eigensolution of Sparse Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.5 Application of Eigenvalue Sensitivity Analysis . . . . . . . . . . . . . 8.5 Oscillation Analysis of Power Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

489 489 493 493 500 502 503 506 506 508 510 519

7.5

8

xi

450 457 463 464 475 484 487

519 523 526 530 533 534

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555

Chapter 1

Mathematical Model and Solution of Electric Network

1.1

Introduction

The mathematical model of an electric network is the basis of modern power system analysis, which is to be used in studies of power flow, optimal power flow, fault analysis, and contingency analysis. The electric network is constituted by transmission lines, transformers, parallel/series capacitors, and other static elements. From the viewpoint of electrical theory, no matter how complicated the network is, we can always establish its equivalent circuit and then analyze it according to the AC circuit laws. In this chapter, the electric network is represented by the linear lumped parameter model that is suitable for studies at synchronous frequency. For electromagnetic transient analysis, the high frequency phenomena and wave processes should be considered. In that situation, it is necessary to apply equivalent circuits described by distributed parameters. Generally speaking, an electric network can be always represented by a nodal admittance matrix or a nodal impedance matrix. A modern power system usually involves thousands of nodes; therefore methods of describing and analyzing the electric network have a great influence on modern power system analysis. The nodal admittance matrix of a typical power system is large and sparse. To enhance the computational efficiency, sparsity techniques are extensively employed. The nodal admittance matrix and associated sparsity techniques will be thoroughly discussed in this chapter. The nodal impedance matrix is widely applied in the fault analysis of power systems and will be introduced in Sect. 1.5. The equivalent circuits of the transformer and phase-shifting transformer are also presented in Sect. 1.1 because they require special representation methods.

X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

1

2

1.2 1.2.1

1 Mathematical Model and Solution of Electric Network

Basic Concepts Node Equation and Loop Equation

There are two methods usually employed in analyzing AC circuits, i.e., the node voltage method and loop current method. Both methods require the solution of simultaneous equations. The difference between them is that the former applies node equations while the latter applies loop equations. At present, node equations are more widespread in analyzing power systems, and loop equations are used sometimes as an auxiliary tool. In the following, we use a simple electric network as an example to illustrate the principle and characteristics of the node equation method. As shown in Fig. 1.1, the sample system has two generators and an equivalent load, with five nodes and six branches whose admittances are y1 y6 . Assigning the ground as the reference node, we can write the nodal equations according to the Kirchoff’s current law, 9 y4 ðV_ 2 V_ 1 Þ þ y5 ðV_ 3 V_1 Þ y6 V_ 1 ¼ 0 > > > > > y1 ðV_ 4 V_ 2 Þ þ y3 ðV_ 3 V_2 Þ þ y4 ðV_ 1 V_2 Þ ¼ 0 > > = y2 ðV_ 5 V_ 3 Þ þ y3 ðV_ 2 V_3 Þ þ y5 ðV_ 1 V_ 3 Þ ¼ 0 ; > > > > y1 ðV_ 4 V_ 2 Þ ¼ I_1 > > > ; y2 ðV_ 5 V_ 3 Þ ¼ I_2

ð1:1Þ

where V_1 V_ 5 denote the node voltages. Combining the coefficients of node voltages, we obtain the following equations:

•

V4

y1

y3

•

V2

•

•

V3

•

•

I1

I2

I3 •

y4

y2

•

I4

I5

•

V1 •

I6

Fig. 1.1 Sample system for node voltage method

y6

y5

•

V5

1.2 Basic Concepts

3

9 ðy4 þ y5 þ y6 ÞV_1 y4 V_ 2 y5 V_ 3 ¼ 0 > > > > _ _ _ > y4 V1 þ ðy1 þ y3 þ y4 ÞV2 y3 V3 y1 V4 ¼ 0 > > = _ _ _ _ y5 V1 y3 V2 þ ðy2 þ y3 þ y5 ÞV:3 y2 V5 ¼ 0 : > > > > y1 V_2 þ y1 V_ 4 ¼ I_1 > > > ; _ _ _ y2 V3 þ y2 V5 ¼ I2

ð1:2Þ

In (1.2), the left-hand term is the current flowing from the node and the right-hand term is the current flowing into the node. The above equations can be rewritten in more general form as follows: 9 Y11 V_1 þ Y12 V_ 2 þ Y13 V_ 3 þ Y14 V_4 þ Y15 V_ 5 ¼ I_1 > > > > _ _ _ _ _ _ > Y21 V1 þ Y22 V2 þ Y23 V3 þ Y24 V4 þ Y25 V5 ¼ I2 > > = _ _ _ _ _ _ Y31 V1 þ Y32 V2 þ Y33 V3 þ Y34 V4 þ Y35 V5 ¼ I3 : > > > Y41 V_1 þ Y42 V_ 2 þ Y43 V_ 3 þ Y44 V_4 þ Y45 V_ 5 ¼ I_4 > > > > ; _ _ _ _ _ _ Y51 V1 þ Y52 V2 þ Y53 V3 þ Y54 V4 þ Y55 V5 ¼ I5

ð1:3Þ

Comparing (1.3) with (1.2), we can see Y11 ¼ y4 þ y5 þ y6 ; Y22 ¼ y1 þ y3 þ y4 ; Y33 ¼ y2 þ y3 þ y5 ; Y44 ¼ y1 ; Y55 ¼ y2 : These elements are known as nodal self-admittances. Y12 ¼ Y21 ¼ y4 ; Y13 ¼ Y31 ¼ y5 ; Y23 ¼ Y32 ¼ y3 ; Y24 ¼ Y42 ¼ y1 ; Y35 ¼ Y53 ¼ y2 : Similarly, the above elements are known as mutual admittances between the connected nodes. The mutual admittances of the pair of disconnected nodes are zero. Equation (1.3) is the node equation of the electric network. It reflects the relationship between node voltages and injection currents. Here I_1 I_5 are the nodal injection currents. In this example, except I_4 and I_5 , all other nodal injection currents are zero.

4

1 Mathematical Model and Solution of Electric Network

Equation (1.3) can be solved to get node voltages V_ 1 V_ 5 , then the branch currents can be obtained. Thus, we have obtained all the variables of the network. Generally, for a n node network, we can establish n linear node equations in (1.3) format. In matrix notation, we have I ¼ YV;

ð1:4Þ

where 2

3 I_1 6_ 7 6 I2 7 7 I¼6 6 .. 7; 4. 5 I_n

2

3 V_ 1 6 _ 7 6 V2 7 7 V¼6 6 .. 7: 4 . 5 V_ n

Here I is the vector of nodal injection currents and V is the vector of nodal voltages; Y is called the nodal admittance matrix 2

Y11 6 Y21 Y¼6 4 Yn1

Y12 Y22 Yn2

3 Y1n Y2n 7 7: 5 Ynn

As we have seen, its diagonal element Yii is the nodal self-admittance and the off diagonal element Yij is the mutual admittance between node i and node j. Now we introduce the incidence matrix that is very important in network representations. The incidence matrix represents the topology of an electric network. Different incidence matrices correspond to different networks configurations. The elements of the incidence matrix are only 0, þ1, or 1. They do not include the parameters of network branches. For example, there are five nodes and six branches in Fig. 1.1. Its incidence matrix is a matrix with five rows and six columns. 2

0 0 6 1 0 6 A¼6 6 0 1 4 1 0 0 1

0 1 1 0 0

1 1 0 0 0

3 1 1 0 07 7 1 07 7: 0 05 0 0

In the incidence matrix, the serial numbers of rows correspond to the node numbers and the serial numbers of columns correspond to the branch numbers. For example, the first row has three nonzero elements, which denotes node 1 is connected with three branches. These three nonzero elements are in the fourth, fifth, and sixth columns, which means the branches connected with node 1 are branches 4, 5, and 6.

1.2 Basic Concepts

5

If the branch current flows into the node, the nonzero element equals 1; if the branch current flows out of the node, the nonzero element equals 1. The positions of the nonzero elements in each column denote the two node numbers of the relevant branch. For example, in the fifth column the nonzero elements are in the first and third row, which means the fifth branch connects node 1 and 3. In the sixth column, there is only one nonzero element in the first row, which means the sixth branch is a grounded branch. From the above discussion we see that an incidence matrix can uniquely determine the topology of a network configuration. The incidence matrix has a close relationship with the network node equation. If there are n nodes and b branches in an electric network, the state equation for every branch is I_Bk ¼ yBk V_ Bk ;

ð1:5Þ

where yBk is the admittance of branch k; IBk the current flowing in branch k; and V_ Bk is the voltage difference of branch k, whose direction is determined by IBk : If branch k includes a voltage source, as shown in Fig. 1.2a, it should be transformed to the equivalent current source as shown in Fig. 1.2b. yBk ¼ 1=zBk a_ Bk ¼ e_Bk =zBk ¼ yBk e_ Bk

) :

The current source can be treated as current injecting into the electric network, thus the branch can also be represented by (1.5). In matrix notation, the equation of a b branch network is I B ¼ YB V B ;

ð1:6Þ

eBK •

zBK

a aBK •

yBK •

IBK

Fig. 1.2 Transformation from voltage source to current source

•

b

VBK

6

1 Mathematical Model and Solution of Electric Network

where I B is the vector of the currents in branches, V B the vector of the branch voltage differences, and YB is a diagonal matrix constituted by the branch admittances. According to Kirchoff’s current law, the injection current I_i of node i in an electric network can be expressed as follows I_i ¼

b X

aik I_Bk

ði ¼ 1; 2; . . . ; nÞ;

ð1:7Þ

k¼1

where aik is a coefficient. If branch current I_Bk directs toward node i, aik ¼ 1; if branch current I_Bk directs away from the node i, aik ¼ 1; and if branch k does not connect to node i, aik ¼ 0. It is easy to get the relationship between nodal current vector I_ and branch current vector I_B as follows, I ¼ AI B ;

ð1:8Þ

where A is the incidence matrix of the network. Assuming the power consumed in the whole network is S, we can obtain the following equation, S¼

b X

I^Bk V_ Bk ¼ I^B V_ B ;

i¼1

where I^Bk and I^B are the conjugate of the corresponding vector and * is the scalar product of the two vectors. From the viewpoint of the nodal input power, we have S¼

n X

^ V: _ I^i V_ i ¼ I

i¼1

Obviously, I^ V_ ¼ I^B V_ B : From (1.8), we see I^ ¼ I^B AT : Substituting it into (1.9), we obtain, I^B AT V_ ¼ I^B V_ B :

ð1:9Þ

1.2 Basic Concepts

7

Therefore, AT V_ ¼ V_ B :

ð1:10Þ

Substituting (1.6) and (1.10) into (1.8) sequentially, we can get _ I_ ¼ AYB AT V_ ¼ YV;

ð1:11Þ

where Y is the nodal admittance matrix of the electric network Y ¼ AYB AT :

ð1:12Þ

Thus the nodal equations of an electric network can be obtained from its incidence matrix. In the following, the network shown in Fig. 1.1 is used again to illustrate the basic principle of analyzing the electric network by the loop current equations. In the loop equation method, the network elements are often represented in impedance form. The equivalent circuit is shown in Fig. 1.3. There are three independent loops in the network and the loop currents are I_1 ; I_2 ; and I_3 , respectively. According to Kirchoff’s voltage law, the voltage equations of the loops are 9 V_ 4 ¼ ðz1 þ z4 þ z6 ÞI_1 þ z6 I_2 z4 I_3 > = : V_ 5 ¼ z6 I_1 þ ðz2 þ z5 þ z6 ÞI_2 þ z5 I_3 > ; _ _ _ 0 ¼ z4 I1 þ z5 I2 þ ðz3 þ z4 þ z5 ÞI3

ð1:13Þ

Rewrite the above equation into the normative form, 9 E_ 1 ¼ Z11 I_1 þ Z12 I_2 þ Z13 I_3 > = E_ 2 ¼ Z21 I_1 þ Z22 I_2 þ Z23 I_3 ; > ; E_ 3 ¼ Z31 I_1 þ Z32 I_2 þ Z33 I_3

•

V4 4

z1

ð1:14Þ

z3

2

z2

3

•

i2

•

i3

•

i1 •

•

I1

I3

z4

z5

•

i4

•

•

•

Fig. 1.3 Sample system with loop currents

1 V1 z6

i6

i5

•

I2

5

8

1 Mathematical Model and Solution of Electric Network

where E_ 1 ¼ V_4 ; E_ 2 ¼ V_ 5 ; E_ 1 ¼ 0 are voltage potentials of three loops, respectively, Z11 ¼ z1 þ z4 þ z6 ; Z22 ¼ z2 þ z5 þ z6 ; Z33 ¼ z3 þ z4 þ z5 are loop self-impedances, Z12 ¼ Z21 ¼ z6 ; Z13 ¼ Z31 ¼ z4 ; Z23 ¼ Z32 ¼ z5 are the loop mutual impedances. If we know loop voltage E_ 1 ; E_ 2 ; and E_ 3 , we can solve the loop current I_1 ; I_2 ; and I_3 from (1.14), and then obtain the branch current, i_1 ¼ I_1 ; i_2 ¼ I_2 ; i_3 ¼ I_3 ; i_4 ¼ I_1 I_3 ; i_5 ¼ I_2 þ I_3 ; i_6 ¼ I_1 þ I_2 : And the node voltages are V_ 1 ¼ z6 i_6 ;

V_ 2 ¼ V_ 4 z1 i_1 ; V_ 3 ¼ V_5 z2 i_2 :

Thus all the variables of the electric network are solved. Generally, an electric network with m independent loops can be formulated by m loop equations. In matrix notation, we have E1 ¼ Z1 I1 ;

ð1:15Þ

where 2

3 I_1 6 _ 7 6 I2 7 7 I1 ¼ 6 6 .. 7; 4 . 5 I_m

2

3 E_ 1 6 _ 7 6 E2 7 7 E1 ¼ 6 6 .. 7 4 . 5 E_ m

are vectors of the loop currents and voltage phasors, respectively; 2

Z11 6 Z21 Z1 ¼ 6 4 Zm1

Z12 Z22 Zm2

3 Z1m Z2m 7 7 5 Zmm

ð1:16Þ

is the loop impedance matrix, where Zii is the self-impedance of the loop i and equals the sum of the branch impedances in the loop; Zij is the mutual impedance between loop i and loop j, and equals the sum of the impedances of their common branches. The sign of Zij depends on the directions of loop currents of loop i and loop j. If their directions are identical, Zij is positive, and if their directions are different, Zij is negative.

1.2 Basic Concepts

9

For the example shown in Fig. 1.3 we can write the basic loop incidence matrix according to the three independent loops, 2 3 1 0 0 1 0 1 B ¼ 4 0 1 0 0 1 1 5: 0 0 1 1 1 0 The serial numbers of rows correspond to the loop numbers and the serial numbers of columns correspond to the branch numbers. For example, in the third row, there are three nonzero elements in the third, fourth, and fifth columns which means loop 3 includes branches 3, 4, and 5. If the branch current has the same direction as the basic loop current, the corresponding nonzero element equals þ1; if the directions of branch current and loop current are different the corresponding nonzero element equals 1. It should be noted that a basic loop incidence matrix cannot uniquely determine a network configuration. In other words, there may be different configurations corresponding to the same basic loop incidence matrix. Similarly to the discussion on the node incidence matrix above, we can get the basic loop equations of an electric network from its basic loop incidence matrix B, ZL ¼ BZB BT ;

ð1:17Þ

where ZB is a diagonal matrix composed of the branch impedances. The application of incidence matrices is quite extensive. If we have the above basic concepts, network analysis problems can be dealt with more flexibly. The details will be discussed in the relevant later sections.

1.2.2

Equivalent Circuit of Transformer and Phase-Shift Transformer

The equivalent circuit of an electric network is established by the equivalent circuits of its elements such as transmission lines and transformers. The AC transmission line is often described by the nominal P equivalent circuit which can be found in other textbooks. In this section, only the equivalent circuits of the transformer and the phase-shift transformer are discussed, especially the transformer with off-nominal turns ratios. Flexible AC Transmission Systems (FACTS) are increasingly involved in power systems, and we will discuss the equivalent circuit of FACTS elements in Chap. 5. When the exciting circuit is neglected or treated as a load (or an impedance), a transformer can be represented by its leakage impedance connected in series with an ideal transformer as shown in Fig. 1.4a. The relation between currents and voltages can be formulated as follows:

10

1 Mathematical Model and Solution of Electric Network 1:K •

i

Vi

a

zT •

Ii

•

Vj •

Ij

j

i I&i

•

Vj

KzT i Vi Ii KzT K 2zT K−1 1−K

•

•

j

•

Ij

b

•

Vi (K−1)yT K

c

yT K

•

Ij j •

(1−K)yT K2

Vj

Fig. 1.4 Transformer equivalent circuit

9 I_i þ K I_j ¼ 0 = V_j : V_ i zT I_i ¼ ; K Solving the above equation, we can obtain 1 1 _ I_i ¼ V_ i Vj ; zT KzT 1 _ 1 I_j ¼ Vi þ 2 V_j : KzT K zT

ð1:18Þ

9 K1 _ 1 _ > Vi þ ðVi V_ j Þ > = KzT KzT : 1K 1 _ > ; Ij ¼ 2 V_j þ ðVj V_ i Þ > K zT KzT

ð1:19Þ

Rewrite (1.18) as follows Ii ¼

According to (1.19), we can get the equivalent circuit as shown in Fig. 1.4b. If the parameters are expressed in terms of admittance, the equivalent circuit is shown in Fig. 1.4c, where yT ¼

1 : zT

It should be especially noted in Fig. 1.4a the leakage impedance zT is at the terminal where the ratio is 1. When the leakage impedance zT is at the terminal where ratio is K, we should transform it to z0T by using the following equation, so that the equivalent circuit shown in Fig. 1.4 also can be applied in this situation z0T ¼ zT =K 2 :

ð1:20Þ

The equivalent circuit of a two-winding transformer has been discussed above. A similar circuit can be used to represent a three-winding transformer. For example, Fig. 1.5 shows the equivalent circuit of a three-winding transformer that can be transformed into two two-winding transformers’ equivalent circuits.

1.2 Basic Concepts

11

1 : Ki k

Fig. 1.5 Three-winding transformer equivalent circuit

k zkh zih

i

h zjh

j 1 : Ki j

After obtaining the transformer equivalent circuit, we can establish the equivalent circuit for a multivoltage network. For example, an electric network shown in Fig. 1.6 can be represented by the equivalent circuit shown in Fig. 1.6b or c when the leakage impedances of transformer T1 and T2 have been normalized to side and side . It can be proved that the two representations have an identical ultimate equivalent circuit as shown in the Fig. 1.6d. When we analysis the operation of a power system, the per-unit system is extensively used. In this situation, all the parameters of an electric network are denoted in the per-unit system. For example, in the Fig. 1.6, if the voltage base at side is Vj1 , at sides and is Vj2 and at side is Vj4 , then the base ratio (nominal turns ratio) of transformer T1 and T2 are Kj1 ¼

Vj2 Vj2 ; Kj2 ¼ : Vj1 Vj4

ð1:21Þ

The ratios of transformer T1 and T2 on a per-unit base (off-nominal turns ratio) are K1 ¼

K1 ; Kj1

K2 ¼

K2 : Kj2

ð1:22Þ

Therefore, the ratio of the transformer should be K1 or K2 when its equivalent circuit is expressed in a per-unit system. In modern power systems, especially in the circumstances of deregulation, the power flow often needs to be controlled. Therefore the application of the phaseshifting transformer is increasing. As we know, a transformer just transforms the voltages of its two terminals and its turn ratio is a real number. The phase-shifting transformer can also change the phase angle between voltages of its two terminals. Thus its turn ratio is a complex number. When the exciting current is neglected or treated as a load (or an impedance), a phase-shifting transformer can be represented

12

1 Mathematical Model and Solution of Electric Network

T1

1

T2

l

1 : K1

2

K2 : 1

3

4

a zT1 1 : K1

K2 : 1

zl

zT2

yl 2

yl 2

b zT2

zl

zT1 1 :1 K1

yl 2

yl 2

1:

1 K2

c zl

K1zT1 K1zT1 K1−1

2

K 1 zT1 1−K1

yl 2

K2zT2 yl 2

2

K 2 z21 1−K2

K 2zT2 K2−1

d Fig. 1.6 Equivalent circuit of a multivoltage electric network

by its leakage impedance, which is connected in series with an ideal transformer having a complex turns ratio as shown in Fig. 1.7. From this figure, we can obtain the equations as follows, V_ i I_i zT ¼ V_ j0 I_i þ I_0 ¼ 0:

ð1:23Þ

j

Apparently, the two terminal voltages are related by _ V_ j0 ¼ V_ j =K: Since there is no power loss in an ideal autotransformer, V_ j0 I^j0 ¼ V_ j I^j ;

ð1:24Þ

1.3 Nodal Admittance Matrix

13

Fig. 1.7 Phase-shifting transformer representation

•

1:K •

zT V ′ j

i

Vj

•

•

Vi

•

•

Ii

Ij

j

•

Ij

where I^j0 and I^j are the conjugates of I^j0 and I^j , respectively. It follows from the above equations that ^_ I_ : I_j0 ¼ K j

ð1:25Þ

Substituting (1.24) and (1.25) into (1.23) V_ j V_i I_i ¼ ¼ Yii V_ i þ Yij V_ j _ T zT Kz V_ j V_ i I_j ¼ þ 2 ¼ Yji V_ i þ Yjj V_ j ; ^ T K zT Kz

ð1:26Þ

where Yii ¼

1 ; zT

Yij ¼

1 ; _ KzT

Yji ¼

1 ; ^ KzT

Yjj ¼

1 K 2 zT

:

Equation (1.26) is the mathematical model of the phase-shifting transformer. It is easy to be proved that (1.26) is the same as (1.18) when the turn ratio is a real number. This illustrates that the transformer is a particular case of the phaseshifting transformer. Because the ratio of a phase-shifting transformer is complex number, and Yij 6¼ Yji , it has no equivalent circuit and the admittance matrix of the electric network with the phase-shifting transformer is not symmetric.

1.3 1.3.1

Nodal Admittance Matrix Basic Concept of Nodal Admittance Matrix

As mentioned above, the node equation (1.3) is usually adopted in modern power system analysis. If the number of nodes in a network is n, we have the following general simultaneous equations:

14

1 Mathematical Model and Solution of Electric Network

9 I_1 ¼ Y11 V_ 1 þ Y12 V_ 1 þ þ Y1i V_ i þ þ Y1n V_ n > > > > _I2 ¼ Y21 V_ 1 þ Y22 V_ 2 þ þ Y2i V_ i þ þ Y2n V_ n > > > > > > > .. > = . : I_i ¼ Yi1 V_ 1 þ Yi2 V_ 2 þ þ Yii V_ i þ þ Yin V_n > > > > > > .. > > > . > > > ; I_n ¼ Yn1 V_ 1 þ Yn2 V_ 2 þ þ Yni V_ i þ þ Ynn V_ n

ð1:27Þ

The matrix constituted by the coefficients of (1.27) is the nodal admittance matrix 2Y

11

6 Y21 6 . 6 . 6 . Y¼6 6 Yi1 6 . 4 . . Yn1

Y12 Y22 .. . Yi2 .. .

Yn2

Y1i Y2i .. . Yii .. .

Yni

Yan 3 Y2n 7 .. 7 7 . 7 7: Yin 7 .. 7 5 . Ynn

ð1:28Þ

A nodal admittance matrix reflects the topology and parameters of an electric network, so it can be regarded as a mathematical abstraction of the electric network. The node equation based on the admittance matrix is a widely used mathematical model of electric networks. Next we will introduce some physical meaning of the matrix elements. If we set a unit voltage at node i and ground other nodes, i.e., V_i ¼ 1 V_j ¼ 0

ðj ¼ 1; 2; . . . ; n; j 6¼ iÞ;

then the following relationships hold according to (1.27), Ij ¼ Yji

j ¼ 1; 2; . . . ; n:

ð1:29Þ

From (1.29) we can see the physical meaning of the ith column elements in the admittance matrix: the diagonal element Yii in the ith column, the self-admittance of node i, is equal to the injection current of the node i; the off-diagonal elements Yij in the ith column, the mutual-admittance of node i and node j, is equal to the injection current of node j in this situation. We will further illustrate these concepts by a simple network shown in Fig. 1.8. The network has three nodes (plus ground), thus the dimension of its admittance matrix is 3 3,

1.3 Nodal Admittance Matrix 2

15

1

z12

3

•

I2

z13

z10

•

I12 z10

3

•

V1 = 1 I1

z12

a 2

1

2

•

I10

•

I3

•

I13

z13

b 1 •

I2

3

I1

•

I12 z12

•

I2

•

I3

•

•

I13 = 0

c

•

•

I1

•

I12 = 0 z12

z13

z10

1

2

V3 = 1 •

I31 z10

3 •

I3

z13

d 1

3

2

z12

z20

z23

e Fig. 1.8 Construction process of admittance matrix in simple electric network

2

Y 11 Y ¼ 4 Y 21 Y 31

Y 12 Y 22 Y 32

3 Y 13 Y 23 5: Y 33

According to the above discussion, we can get the elements of the first column: Y11 ; Y21 ; and Y31 , by setting a unit voltage on node 1 and grounding node 2 and node 3 as shown in Fig. 1.8b. Evidently, 1 1 1 I_1 ¼ I_12 þ I_13 þ I_10 ¼ þ þ ¼ Y11 ; z12 z10 z13 1 I_2 ¼ I_12 ¼ ¼ Y21 ; z12 1 I_3 ¼ I_13 ¼ ¼ Y31 : z13 Similarly, setting a unit voltage at node 2 and grounding node 1 and node 3 as shown in Fig. 1.8c, we can get the elements of the second column: 1 I_1 ¼ I_21 ¼ ¼ Y12 ; z12 1 I_2 ¼ I_21 ¼ ¼ Y22 ; z12 I_3 ¼ 0 ¼ Y32 :

16

1 Mathematical Model and Solution of Electric Network

For the elements of the third column we have (see Fig. 1.8d), 1 I_1 ¼ I_31 ¼ ¼ Y13 ; z31 I_2 ¼ 0 ¼ Y23 ; 1 I_3 ¼ I_31 ¼ ¼ Y33 : z13 Finally, the admittance matrix of the above simple network becomes 2

1 1 1 þ þ 6 z12 z10 z13 6 1 6 Y¼6 6 z12 4 1 z13

1 z12 1 z12

0

3 1 z13 7 7 7 0 7: 7 1 5 z13

ð1:30Þ

If we change the node numbers in Fig. 1.8a, e.g., exchange the number ordering of node 1 with node 2, as shown in Fig. 1.8e, then the admittance matrix becomes, 2

1 6 z12 6 6 1 6 0 Y ¼ 6 6 z12 6 4 0

1 z12 1 1 1 þ þ z12 z20 z23 1 z23

3 0

7 7 1 7 7 : z23 7 7 7 1 5 z23

The above matrix can be obtained through exchanging the first row with the second row, and at the same time exchanging the first column with the second column of the matrix shown in (1.30). The exchange of the rows and columns of the admittance matrix corresponds to the exchange of the sequence of node equations and their variables. The properties of the admittance matrix can be summarized as follows: 1. The admittance matrix is symmetric if there is no phase-shifting transformer in the network. From (1.30) we have Y12 ¼ Y21 ¼

1 1 ; Y13 ¼ Y31 ¼ ; Y23 ¼ Y32 ¼ 0: z12 z13

Generally, according to the reciprocity of the network, Yij ¼ Yji : Therefore, the admittance matrix is symmetric. We will discuss the networks with phase-shifting transformers later.

1.3 Nodal Admittance Matrix

17

2. The admittance matrix is sparse. From the discussion above, we know that Yij and Yji will be zero if node i does not directly connect with node j. For example, in Fig. 1.8a, node 2 does not directly connect with node 3, so both of Y23 and Y32 are zero. In general, the number of nonzero off-diagonal elements of each row is equal to the number of branches that are incident to the corresponding node. Usually, the number of branches connected to one node is 2–4, thus there are only 2–4 nonzero off-diagonal elements in each row. The property that only a few nonzero elements exist in a matrix is called sparsity. This phenomenon will be more remarkable with increase of the power system scale. For instance, for a network with 1,000 nodes, if each node directly connects three branches on average, the total number of nonzero elements for the network is 4,000, which is only 0.4% of the total elements in the admittance matrix. The symmetry and sparsity of an admittance matrix are very important features for large-scale power systems. If we make full use of these two properties, the computation speed will be accelerated and the computer memory will be saved dramatically.

1.3.2

Formulation and Modification of Nodal Admittance Matrix

Now we discuss formulation of an admittance matrix by inspection first. When an electric network is composed of only transmission lines, the principles of constructing its admittance matrix can be summarized as follows: 1. The order of the admittance matrix is equal to the number of the nodes of the electric network. 2. The number of the nonzero off-diagonal elements in each row is equal to the number of the ungrounded branches connected to the corresponding node. 3. The diagonal elements of the admittance matrix, i.e., the self-admittance of the node, is equal to the sum of all the admittances of the incident branches of the corresponding node. Thus X Yii ¼ yij ; ð1:31Þ j2i

where yij is the reciprocal of zij , which is the branch impedance between node i and node j, ‘‘j I’’ denotes that only the incident branches of node i (including the grounding branch) are included to the summation. For example, in Fig. 1.8, the self-admittance of node 1, i.e., Y11 , should be Y11 ¼

1 1 1 þ þ ¼ y12 þ y10 þ y13 : z12 z10 z13

The self-admittance of node 2, i.e., Y22 , should be Y22 ¼

1 ¼ y12 : z12

18

1 Mathematical Model and Solution of Electric Network

4. The off-diagonal element of the admittance matrix, Yij , is equal to the negative of the admittance between node i and node j Yij ¼

1 ¼ yij : zij

ð1:32Þ

For example, in Fig. 1.8a, 1 ¼ y12 ; z12 1 ¼ ¼ y13 : z13

Y12 ¼ Y13

Therefore, no matter how complicated the configuration of an electric network is, its admittance matrix can be established directly by inspection according to the parameters and the topology of the network. When the electric network involves transformers or phase-shifting transformers, they need special treatment. When branch ij is a transformer, the admittance matrix certainly can be formed following the above steps if the transformer is substituted beforehand by the P equivalent circuit as shown in Fig. 1.4a. However, in practical application the transformer is often treated directly in forming the admittance matrix. If branch ij is a transformer, as shown in Fig. 1.4a, the elements of the admittance matrix related to the branch can be obtained as follows: 1. Add two nonzero off-diagonal elements into the admittance matrix Yij ¼ Yji ¼

yT : K

ð1:33Þ

2. Add to the self-admittance of node i by, DYii ¼

K1 1 yT þ yT ¼ yT : K K

ð1:34Þ

3. Add to the self-admittance of node j by DYjj ¼

1 1K yT yT þ yT ¼ 2 : K K2 K

ð1:35Þ

When branch ij is a phase-shifting transformer, its equivalent circuit is Fig. 1.7. Then the corresponding matrix elements are obtained as follows: 1. Add two nonzero off-diagonal elements into the admittance matrix Yij ¼

1 ; _ KzT

ð1:36Þ

1.3 Nodal Admittance Matrix

19

Yji ¼

1 : ^ KzT

ð1:37Þ

1 : zT

ð1:38Þ

2. Add to the self-admittance of node i by DYii ¼ 3. Add to the self-admittance of node j by DYjj ¼

1 K2 z

:

ð1:39Þ

T

It can be seen from (1.36) and (1.37) that Yij 6¼ Yji , thus the admittance matrix is not symmetric any more although its structure is still symmetric. Studies of different system operation states, such as transformer or transmission line outages, play an important part in modern power system analysis. Because the outage of branch ij only affects the self and mutual admittance of node i and node j, we can obtain the new admittance matrix for the contingency state by modifying the original admittance matrix. The modification methods for different situations are introduced as follows: 1. To add a new node with a new branch for the original network as shown in Fig. 1.9a. Assume that i is a node of the original network and j is the new node; zij is the impedance of the new branch. The dimension of the admittance matrix becomes N þ 1 because of the new node. There is only one branch connected to node j, therefore, its self-admittance is, 1 ; zij The self-admittance of node i should be modified (added) by, Yjj ¼

DYii ¼

1 : zij

zij i

N

j

i

i

N

N

zij

j

a

b

i

N

−zij

j

c

Fig. 1.9 Four cases of modifying the electric network

j

d

−zij z′ij

20

1 Mathematical Model and Solution of Electric Network

Two off-diagonal elements should also be created 1 : zij 2. To add a new branch between node i and node j as shown in Fig. 1.9b. In this case, no new node is introduced and the dimension of the new admittance matrix is the same as the original one, while the following modifications should be made. 9 1 > > DYii ¼ > > zij > > > = 1 DYjj ¼ : ð1:40Þ zij > > > > 1> > DYij ¼ DYji ¼ > ; zij Yij ¼ Yji ¼

3. To remove a branch with impedance zij between node i and node j. In this case, it is equivalent to adding a new branch of impedance zij between node i and node j as shown in Fig. 1.9c. Therefore, the modifications of the admittance matrix are as follows: 9 > > > > > > > = : > > > > 1> > DYij ¼ DYji ¼ > ; zij 1 zij 1 DYjj ¼ zij DYii ¼

ð1:41Þ

4. To change branch impedance zij for z0ij . This case is equivalent to removing branch impedance zij first and then adding a branch of impedance z0ij between node i and node j as shown in Fig. 1.9d. Thus the modifications can be carried out according to (1.40) and (1.41). It should be noted that the above discussion is based on the assumption that the added or removed branch is a pure impedance branch. If the branch is a transformer or a phase-shifting transformer, the modifications should be carried out according to (1.33)–(1.35) or (1.36)–(1.39). [Example 1.1] Figure 1.10 shows an equivalent circuit of a simple electric network with two transformers. The branch impedance and grounding admittance in per unit are shown in the figure. Determine the nodal admittance matrix for the electric network. [Solution] According to the method introduced in Sect. 1.2.2, we can assemble the elements of the admittance matrix node by node.

1.3 Nodal Admittance Matrix

21

1:1.05

0.08 + j0.30

j 0.015

j0.25

1.05:1 j0.25

+ 04 0.

j0 .2 5

j0.25

1

0.

+

j0.03

5

.3

j0

j0.25

Fig. 1.10 Equivalent circuit for Example 1.1

In Fig. 1.10, parameters are in admittance for grounding branches and in impedance for other branches (branches in series connection). Using (1.31), we obtain the self-admittance of node 1 as follows: Y11 ¼ y10 þ y12 þ y13 ¼ j0:25 þ

1 1 þ 0:04 þ j0:25 0:1 þ j0:35

¼ 1:378742 j6:291665: The mutual admittances related to node 1 can be obtained according to (1.32), 1 ¼ 0:624025 þ j3:900156 0:04 þ j0:25 1 ¼ ¼ 0:754717 þ j2:641509: 0:1 þ j0:35

Y21 ¼ Y12 ¼ y12 ¼ Y31 ¼ Y13 ¼ y13

Because branch 2–4 is a transformer, the self-admittance of node 2 should be calculated according to (1.31) and (1.35) based on the equivalent circuit as shown in Fig. 1.4a Y22 ¼ y20 þ y12 þ y23 þ

y42 2 K42

1 1 1 1 þ þ 0:04 þ j0:25 0:08 þ j0:30 j0:015 1:052 ¼ 1:453909 j66:98082: ¼ ðj0:25 þ j0:25Þ þ

The mutual admittances related to node 2 are Y23 ¼ Y32 ¼

1 ¼ 0:829876 þ j3:112033: 0:08 þ j0:30

Using (1.33) we have Y24 ¼ Y42 ¼

y42 1 1 ¼ ¼ j63:49206: K42 j0:015 1:05

22

1 Mathematical Model and Solution of Electric Network

The other elements of the admittance matrix can be calculated in a similar way. The ultimate result is 2

1:378742 6 j6:291665 6 6 6 6 0:24024 6 6 þj3:900156 6 6 6 6 0:754717 Y¼6 6 þj2:641509 6 6 6 6 6 6 6 6 6 4

3

0:924024 0:754717 þj3:900156 þj2:641509 1:453909 j66:98082

0:829876 0:000000 þj3:112033 þj63:19206

0:929876 1:584596 þj3:112033 j35:73786 0:000000 þj63:49206

0:000000 j66:66667

0:000000 þj31:74603

7 7 7 7 7 7 7 7 7 7 0:000000 7 7; þj31:74603 7 7 7 7 7 7 7 7 7 7 0:000000 5 j33:33333

where the vacancies are zero elements.

1.4 1.4.1

Solution to Electric Network Equations Gauss Elimination Method

At present, Gauss elimination is the most popular method to solve the electric network equations. In the initial stage of computer application in power systems, iterative methods were also been used because of the limitation of computer memory. The fatal disadvantage of the iterative methods is the convergence problem. Therefore, the Gauss elimination method almost has substituted for iterative methods after successful application of the sparse techniques [1, 2]. The Gauss elimination method is introduced in this section, and the sparse technique and sparse vector method will be described successively. The Gauss elimination method in solving simultaneous linear equations consists of two steps, i.e., forward elimination and back substitution. Both forward elimination and back substitution can be carried out by either row or column orientation. Generally, the column-oriented forward elimination and row-oriented back substitution scheme are widely used. The related algorithm is introduced next, and other algorithms can be easily deduced similarly. A system of n simultaneous linear equations may be written in the matrix form as AX ¼ B in which elements in matrix A and vector B can be either real or complex numbers. For example, the coefficient matrix of (1.3) is complex, while that of the correction equation in the Newton–Raphson method (see (2.40) in Chap. 2) is real.

1.4 Solution to Electric Network Equations

23

Because the forward eliminations involve manipulations with matrix A and B, a n ðn þ 1Þ augmented matrix is formed by appending B as the ðn þ 1Þth column of A, 2

A ¼ ½ A

a11 6 a21 B ¼ 6 6 6 6 an1

a12 a22 an2

a1n a2n ann

3 2 a11 b1 6 a21 b2 7 7¼6 6 7 7 6 bn 7 6 an1

a12 a22 an2

a1n a2n ann

3 a1;nþ1 a2;nþ1 7 7: 7 7 an;nþ1 7

In the above equation, bj is substituted by aj;nþ1 ðj ¼ 1; 2; . . . ; nÞ to simplify the following representation. The process of the column-oriented forward eliminations is introduced first.

Step 1.

Eliminate the first column

First, normalize the first row of the augmented matrix A, 1

ð1Þ

ð1Þ

...

ð1Þ

a1;nþ1 ;

ð1:42Þ

a12

a13

a1j a11

ðj ¼ 2; 3; . . . ; n þ 1Þ:

where ð1Þ

a1j ¼

Then the derived row as shown in (1.42) is used to eliminate the elements and the remaining elements of the second to the nth row a21 ; a31 ; . . . ; an1 of A, can be calculated by ð1Þ

ð1Þ

aij ¼ aij ai1 a1j

ðj ¼ 2; 3; . . . ; n þ 1Þ; ði ¼ 2; 3; . . . ; nÞ;

where the superscript (1) denotes that the relative element is the result of the first manipulation. At this stage, matrix A is changed into A1 , 2 A1 ¼ ½ A1

6 6 6 B1 ¼ 6 6 6 4

1

ð1Þ

a12

ð1Þ

ð1Þ

.. .

a22 .. .

an2

ð1Þ

a1n

ð1Þ

a2n .. .

ð1Þ

ann

ð1Þ

a1;nþ1

3

7 ð1Þ a2;nþ1 7 7 7 .. 7: . 7 5 ð1Þ an;nþ1

The corresponding equation is A1 X ¼ B1 which has the same solution as the original equation. In the above matrix, the vacancies are zero elements.

24

1 Mathematical Model and Solution of Electric Network

Step 2.

Eliminate the second column

Normalize the second row of the augmented matrix A as the following ð2Þ

1 a23

0

ð2Þ

a2;nþ1 ;

...

ð1:43Þ

where ð2Þ

ð1Þ

ð1Þ

a2j ¼ a2j =a22

ðj ¼ 3; 4; . . . ; n þ 1Þ:

Then the derived row shown in (1.43) is used to eliminate the elements ð1Þ ð1Þ ð1Þ a32 ; a42 ; . . . ; a4n of A1 and the remaining elements of the third to the nth row can be calculated by, ð2Þ

ð1Þ

ð1Þ ð2Þ

aij ¼ aij ai2 a2j

ðj ¼ 3; 4; . . . ; n þ 1Þ; ði ¼ 3; 4; . . . ; nÞ;

where the superscript (2) denotes that the relative element is the result of the second manipulation. Now, matrix A1 has been transformed into A2 , 2

A2 ¼ ½ A2

ð1Þ

1 a12 6 1 6 6 6 B2 ¼ 6 6 4

ð1Þ

a13

ð2Þ

ð2Þ

a23

a33 ð2Þ an3

ð1Þ

a1n

ð2Þ

a2n

ð2Þ

a3n

að2Þ nn

ð1Þ

a1;nþ1

3

7 ð2Þ a2;nþ1 7 7 7: pgð2Þ a3;nþ1 7 7 5 ð2Þ an;nþ1

Generally, the following computation should be executed when eliminating the kth column ðkÞ

ðk1Þ

akj ¼ akj ðkÞ

ðk1Þ

aij ¼ aij

ðk1Þ ðkÞ akj

aik

ðk1Þ

=akk

ðj ¼ k þ 1; . . . ; n þ 1Þ;

ð1:44Þ

ðj ¼ k þ 1; . . . ; n þ 1Þ; ði ¼ k þ 1; . . . ; nÞ:

ð1:45Þ

After proceeding with the elimination n times in this manner, the elements below the diagonal of the matrix become zero, and the nth derived augmented matrix is obtained. 2

An ¼ ½ An

6 6 6 6 Bn ¼ 6 6 6 6 4

1 a12

ð1Þ

a13

ð1Þ

...

a1n

1

a23 1

ð2Þ

... ...

a2n

..

.

ð1Þ

ð1Þ

a1;nþ1

3

ð3Þ

7 ð2Þ a2;nþ1 7 7 7 ð3Þ a3;nþ1 7 7: .. 7 7 . 5

1

an;nþ1

ð2Þ

a3n .. .

ðnÞ

ð1:46Þ

1.4 Solution to Electric Network Equations

25

The corresponding equation becomes An X ¼ Bn , that is x1 þ

ð1Þ

ð1Þ

...

þa1n xn ¼

a23 x3 þ

ð2Þ

...

þa2n xn ¼

x3

... .. .

a12 x2 þ

a13 x3 þ

x2

þ

þ

ð1Þ

a1;nþ1

ð1Þ

ð2Þ

a2;nþ1

þa3n xn ¼ .. .

ð3Þ

a3;nþ1 .. .

xn ¼

an;nþ1

ð2Þ ð3Þ

ð1:47Þ

ðnÞ

Its solution is the same as the original equation AX ¼ B. For (1.47), back substitution is carried out in a bottom-up sequence. The value of xn is obtained directly from the nth equation, ðnÞ

xn ¼ an;nþ1 : Then substituting xn into the ðn 1Þth equation we get the solution of xn1 , ðn1Þ

ðn1Þ

xn1 ¼ an1;nþ1 an1;n xn : Substituting xn1 and xn into the ðn 2Þth equation, we obtain xn2 . Generally, xi can be obtained by substituting the solved variables xiþ1 ; xiþ2 ; . . . ; xn into the ith equation, ðiÞ

xi ¼ ai;nþ1

n X j¼iþ1

ðiÞ

aij xj

ði ¼ n; . . . ; 2; 1Þ:

ð1:48Þ

This is the general equation of the row-oriented back substitution. [Example 1.2] Solve the following simultaneous linear equations by using the Gauss elimination method. x1 þ 2x2 þ x3 þ x4 ¼ 5 2x1 þ x2 ¼ 3 x1 þ x3 ¼ 2 x1 þ x4 ¼ 2 [Solution] Write the augmented below. 2 6 ð1Þ 6 6 2 6 6 4 1 1

:

matrix according to the original equations as

2

1

1

1

0

0

0

1

0

0

0

1

.. . .. . .. . .. .

3 57 7 37 7: 7 25 2

26

1 Mathematical Model and Solution of Electric Network

As an initial step, normalize the first row of the augmented matrix according to (1.44), i.e., divide the first row by its diagonal element. 2 6 1 6 6 ð2Þ 6 6 6 ð1Þ 4

2

1

1

1

0

0

0

1

0

ð1Þ

0

0

1

.. . .. . .. . .. .

3 57 7 37 7: 7 27 5 2

Then eliminate the first column according to (1.45) 2 61 6 6 6 6 4

2

1

ð3Þ

2

2

0

2

1

3 .. . 57 . 7 2 ..7 7 : .. 7 7 1 .3 5 . 0 ..3 1

The next step is the elimination of the second column. When normalizing the second row, we divide the elements in the second row by the diagonal element –3 2 61 6 6 6 6 6 4

2

1

1

1

2 3

2 3

ð2Þ

0

1

ð2Þ 1

0

3 .. . 57 .. 7 7 . 37 7: 7 .. .3 7 5 .. .3

Then eliminate the second column in terms of (1.45) to obtain 2 61 6 6 6 6 6 4

3 . 1 .. 5 7 .. 7 7 2 1 23 . 37 3 7 4 1 .. 5 7: . 37 3 3 5 .. 5 1 4 .

2

1

3

3

3

Repeat the procedure for the third column. Normalize the third row through dividing the third row by the diagonal element 4/3. 2 61 6 6 6 6 6 4

3 .. 2 1 1 . 57 .. 7 7 2 1 23 . 37 3 7 .. 5 7: 1 1 4 . 47 5 1 4 .. 5 . 3 3 3

1.4 Solution to Electric Network Equations

27

Then eliminate the third column in terms of (1.45) to obtain 2 61 6 6 6 6 6 4

3 .. . 57 .. 7 7 2 . 37 3 7 .. 5 7: 1 . 47 4 5 5 .. 5 .4 4

2 1 1

1

2 3

1

The last step is normalizing the fourth row according to (1.44), that is, dividing the fourth row by the diagonal element 5/4. 2 61 6 6 6 6 6 4

2

1

1

1

2 3

2 3

1

1 4

1

3 .. . 57 .. 7 7 . 37 7 .. 5 7: . 47 5 .. .1

The transformed equations after elimination become x1 þ

2x2 þ x2 þ

x3 þ 2 3x 3 þ x3 þ

x4 ¼ 2 3x 4 ¼ 1 4x 4 ¼

5 7 3 5 4

x4 ¼

:

1

x4 ; x3 ; x2 ; x1 can be obtained through the back substitution according to (1.48). x4 ¼ 1 x3 ¼ 54 14x4 ¼ 1 x2 ¼ 73 23x3 23x4 ¼ 1 x1 ¼ 5 2x2 x3 x4 ¼ 1

1.4.2

:

Triangular Decomposition and Factor Table

In practical applications, the simultaneous equations often need to be solved repeatedly when only right-hand vector B changes while coefficient matrix A is a constant matrix. In such cases, the factor table method is often used to improve computation efficiency. The factor table records all the operations on right-hand vector B in the Gauss elimination process. As the discussion above, The Gauss elimination method involves forward elimination and back substitution. Back substitution is determined

28

1 Mathematical Model and Solution of Electric Network

by the upper triangular elements of the coefficient matrix after elimination operation as shown (1.46). In order to execute the elimination operation (forward elimination), the relative operation factors also need to be recorded in the elimination process. The forward elimination includes normalization and elimination operation. Take column-oriented elimination as an example, operations on the i th element of B (i.e., bi;nþ1 ) in the forward elimination are as follows (see (1.44) and (1.45)), ðiÞ

ði1Þ

ðkÞ

ðk1Þ

bi ¼ bi

bi ¼ bi

ði1Þ

=aii

ði ¼ 1; 2; . . . ; nÞ;

ðk1Þ ðkÞ bk

aik

ð1Þ

ð1:49Þ

ðk ¼ 1; 2; . . . ; i 1Þ:

ð1Þ

ði2Þ

ð1:50Þ

ði1Þ

The above operation factors ai1 ; ai2 ; ai2 ; . . . ; ai;i1 and aii are to be stored in the lower triangular matrix row by row and appended to the upper triangular elements of the (1.46). Thus, we obtain the factor table as the following a11

a12

ð1Þ

a13

ð1Þ

a14

ð1Þ

a1n

ð1Þ

a21

a22

ð1Þ

a23

ð2Þ

a24

ð2Þ

a2n

a31

a32

ð1Þ

a33

ð2Þ

a34

ð3Þ

a3n

a41 .. .

a42 .. .

ð1Þ

a43 .. .

ð2Þ

a44 .. .

ð3Þ

.. .

a4n .. .

an1

an2

ð1Þ

an3

ð2Þ

an4

ð3Þ

ð2Þ ð3Þ ð4Þ

:

ðn1Þ

ann

Where the lower triangular elements are used in elimination operations on B and the upper triangular elements are used in back substitution operations. The factor table also can be denoted in the following format d11 l21 l31 l41 .. . ln1

u12 d22 l32 l42 .. . ln2

u13 u23 d33 l43 .. . ln3

u14 u24 u34 d44 .. . ln4

.. .

where ði1Þ

;

ðiÞ

ði < jÞ;

dii ¼ aii uij ¼ aij lij ¼

ðj1Þ aij

ðj < iÞ:

u1n u2n u3n u4n ; .. . dnn

ð1:51Þ

1.4 Solution to Electric Network Equations

29

We can see that the lower triangular elements of the factor table are exactly the operation elements used in the elimination process. Therefore, if we retain them in the original position and take the reciprocals of the diagonal elements, the lower triangular elements of the factor table can be readily obtained. The upper triangular elements of the factor table are just the upper triangular part of the coefficient matrix after the elimination operations. If the simultaneous equations need to be solved repeatedly for different righthand vector B, we should first carry out the elimination operation on coefficient matrix A to obtain its factor table. Then the factor table can be used directly and repeatedly to solve the equations with different B. In this situation, we will carry out the elimination operation on the following equations instead of (1.49) and (1.50), ðiÞ

ði1Þ

=dii ;

ð1:52Þ

ðkÞ

ði ¼ k þ 1; . . . ; nÞ:

ð1:53Þ

bi ¼ bi ðkÞ

ðk1Þ

bi ¼ bi

lik bk

The back substitution will be carried out on the following equations instead of (1.48) xn ¼ bðnÞ n ; ðiÞ

x i ¼ bi

n X

uij xj :

ð1:54Þ

j¼iþ1

[Example 1.3] For the simultaneous linear equations of Example 1.2, find the factor table of its coefficient matrix A and solve the equation when B ¼ ½ 1 1 2 0 T . [Solution] Inspecting the solution process of Example 1.2, we can directly obtain the factor table of coefficient matrix A, 1 2 2 3 1 2 1 2

1

1

2 3 4 3 1 3

2 3 1 4 5 4

d11 l21 , l31 l41

u12 d22 l32 l42

u13 u23 d33 l43

u14 u24 : u34 d44

The lower triangular elements of the above factor table are just the operation factors in brackets which appeared in the elimination process, and the upper triangular elements are the upper triangular part of the coefficient matrix after elimination operation. Now we first use the lower triangular elements of the factor table to operate column-oriented elimination on B. Normalize b1 according to (1.52),

30

1 Mathematical Model and Solution of Electric Network ð1Þ

b1 ¼ b1 =d11 ¼ ð1Þ=1 ¼ 1: Then operations on b2 ; b3 ; b4 are carried out by using the elements of the factor table’s first column in the lower triangular part according to (1.53) ð1Þ

ð1Þ

ð1Þ

ð1Þ

ð1Þ

ð1Þ

b2 ¼ b2 l21 b1 ¼ 1 2 ð1Þ ¼ 3; b3 ¼ b3 l31 b1 ¼ 2 1 ð1Þ ¼ 3; b4 ¼ b4 l41 b1 ¼ 0 1 ð1Þ ¼ 1: Thus the elimination operation of the first column is completed, and we have, Bð1Þ ¼ ½ 1

3

3 1 T :

ð1Þ

Next, normalize b2 according to (1.52), ð2Þ

ð1Þ

b2 ¼ b2 =d22 ¼ 3=ð3Þ ¼ 1: ð1Þ

ð1Þ

The elimination operation on b3 ; b4 is followed by using the elements of the second column in the lower triangular part according to (1.53), ð2Þ

ð1Þ

ð2Þ

ð2Þ

ð1Þ

ð2Þ

b3 ¼ b3 l32 b2 ¼ 3 ð2Þ ð1Þ ¼ 1; b4 ¼ b4 l42 b2 ¼ 1 ð2Þ ð1Þ ¼ 1: Thus the elimination operation of the second column is finished, and we have Bð2Þ ¼ ½ 1 1

1 1 T :

ð2Þ

ð3Þ

Normalize b3 according to (1.52) and operate b4 according to (1.53) ð3Þ

ð2Þ

b3 ¼ b3 =d33 ¼ 1=43 ¼ 34: ð2Þ

Again, the elimination operation on b4 is followed by using the elements of the third column in the lower triangular part according to (1.53) ð3Þ

ð2Þ

ð3Þ

b4 ¼ b4 l43 b3 ¼ 1 13 34 ¼ 54: Thus the elimination operation on the third column is finished, and we have Bð3Þ ¼ 1

1

3 4

54

T

:

1.4 Solution to Electric Network Equations

31 ð3Þ

The last step of the elimination operation is to normalize b4 according to (1.52) ð4Þ

ð3Þ

b4 ¼ b4 =d44 ¼ 45=

4 5

¼ 1:

Now, all the elimination operations are fulfilled. Bð4Þ ¼ 1 1

3 4

1

T

:

Comparing with the factor table, we obtain the following identical solution equations x1 þ 2x2 þ x2 þ

x3 2 3x3 x3

þ þ þ

x4 ¼ 1 2 x ¼ 1 4 3 : 1 3 4x 4 ¼ 4 x4 ¼ 1

Now, the unknowns could be solved using the upper triangular part of the factor table according to (1.54). ð4Þ

x4 ¼ b4 ¼ 1 ð3Þ

x3 ¼ b3 u34 x4 ¼ 34 14 ð1Þ ¼ 1 ð2Þ

x2 ¼ b2 u23 x3 u24 x4 ¼ 1 23 1 23 ð1Þ ¼ 1 ð1Þ

x1 ¼ b1 u12 x2 u13 x3 u14 x4 ¼ 1 2 ðÞ 1 1 1 ð1Þ ¼ 1: It should be pointed out that the factor table as shown in (1.50) can be established not only by the Gauss elimination method but also by the triangular decomposition method. From the above example, we can verify that the following relationship between the factor table and its coefficient matrix holds, A ¼ L0 U;

ð1:55Þ

where 2

1 6 2 L0 ¼ 6 41 1

3 0 0 0 3 0 0 7 7 2 43 0 5 2

1 3

5 4

2

1 60 U¼6 40 0

2 1 0 0

1 1 2 3

2 3 1 4

1 0 1

3 7 7: 5

L0 can be decomposed further, L0 ¼ LD:

ð1:56Þ

32

1 Mathematical Model and Solution of Electric Network

In the above example, L can be obtained through dividing off-diagonal elements in each column of L0 by the corresponding diagonal element, 2

1 62 L¼6 41 1

0 1 2 3 2 3

3 0 0 0 07 7 1 05 1 1 4

2

1 0 6 0 3 D¼6 40 0 0 0

3 0 0 0 07 7: 4 05 3 0 54

Therefore the original coefficient matrix can be generally represented as follows A ¼ LDU:

ð1:57Þ

From the example, we can also see the following relationship LT ¼ U or U ¼ LT :

ð1:58Þ

This phenomenon is not specific to this example. The relationship in (1.58) can be proved when the coefficient matrix is symmetric. In the following, we deduce the recursion formulae of the triangular decomposition. Expand (1.55) 2

a11 6 a21 6 6 a31 6 6 .. 4 . an1

a12 a22 a32 .. . an2

a13 a23 a33 .. .

an3

.. .

3 2 l0 11 a1n 6 l0 a2n 7 21 7 6 6 l0 a3n 7 7¼6 6 31 .. 7 6 .. 5 4 . . ann l0n1 2

3 l022 l032 .. . 0 ln2

1 u12 6 1 6 6 6 6 4

l033 .. .

l0n3 u13 u23 1

7 7 7 7 7 7 5

..

.

l0nn

.. .

3 u1n u2n 7 7 u3n 7 7: .. 7 . 5 1

ð1:59Þ

Comparing two sides of the above equation, the diagonal element of the first row can be found l011 ¼ a11 : Comparing the first element of the second row and the first two elements of the second column in both sides, we can obtain l021 ¼ a21 ; l011 u12 ¼ a12 ; l021 u12 þ l022 ¼ a22 :

1.4 Solution to Electric Network Equations

33

Hence the recursion formulae are l021 ¼ a21 ; u12 ¼ a12 =l011 ; l022 ¼ a12 l021 u12 : The following decomposition equation can be obtained

a11 a21

a12 a22

¼

l011 l021

l022

u12 : 1

1

Similarly, if the first k 1 rows of L0 and the first k 1 columns of U have been obtained, the equation becomes 2

a11 a21 a31

6 6 6 6 6 4 ak1;1 2 6 6 6 6 ¼6 6 6 4

a12 a22 a32 .. .

ak1;2 l011 l021 l031

a13 a23 a33 .. .

.. .

ak1;3

3

a1;k1 a2;k1 a3;k1 .. .

7 7 7 7 7 5

ak1;k1

.. .

l022 l032 .. .

l033 .. .

l0k1;1

l0k1;2

l0k1;3

3

..

.

l0k1;k1

2 1 7 6 7 6 7 6 7 6 76 7 6 7 4 5

u12 1

u13 u23 1

.. .

3 u1;k1 u2;k1 7 7 u3;k1 7 7: 7 .. 7 . 5 1

All the elements of the two matrices in the right hand of the above equation have been solved. Comparing the first k 1 elements in the kth row and the first k elements in the kth column of the two sides element by element, we can get the corresponding elements by the following formulae 1 uik ¼ 0 lii

aik

l0kj ¼ akj

i1 X p¼1

j1 X p¼1

! l0ip upk

l0kp upj

ði ¼ 1; 2; . . . ; k 1Þ ð1:60Þ

ðj ¼ 1; 2; . . . ; kÞ:

The above are recursion formulae. Taking k from 1 to n in sequence, the triangular decomposition, A ¼ L0 U, will be achieved by using these formulae. Furthermore, dividing the off-diagonal elements by the corresponding diagonal element, L can be obtained: 1 lkj ¼ ljj

akj

j1 X p¼1

! l0kp upj

ðk ¼ j þ 1; . . . ; nÞ:

ð1:61Þ

34

1 Mathematical Model and Solution of Electric Network

The diagonal elements of L0 constitute D, i.e., dii ¼ l0ii ði ¼ 1; 2; . . . ; nÞ. Now, the coefficient matrix is decomposed into the format A ¼ LDU. It should be particularly noted that (1.58) will always be true if the coefficient matrix is symmetric.

1.4.3

Sparse Techniques

From the discussion of the above section, we know that the solution process of the electric network equation is the process of operating the right-hand constant vector successively using the elements of its factor table. In Example 1.3, there are 16 elements in its factor table: four diagonal elements, six lower triangular elements, and six upper triangular elements. Therefore the solution involves 16 multiplication operations. According to (1.53) and (1.54), if elements in the factor table are zero, the corresponding multiplication operations can be avoided (since the product will be zero) and significant computational effort can be saved. Based on this idea, socalled sparse technique is widely used in power system analysis to improve solution efficiency. The concept of the sparse technique is illustrated by an example in the following. [Example 1.4] Solve the simultaneous linear equations in Example 1.2 by using the sparse method. [Solution] In Example 1.2, the simultaneous linear equations are x1 2x1 x1 x1

2x2 þx2

x3 þx3

þx4 þx4

¼5 ¼3 : ¼2 ¼2

ð1:62Þ

In order to make full use of the sparsity advantages of the equations, the following transformation should be made first, x1 ¼ y4 ; x2 ¼ y2 ; x3 ¼ y3 ; x4 ¼ y1 :

ð1:63Þ

Then, the original equations are transformed into y1 y1

y2 þ2y2

y3 þy3

þy4 þ2y4 þy4 þy4

¼2 ¼3 : ¼2 ¼5

ð1:64Þ

We will solve the equations by using its factor table. The coefficient matrix is 2

ð1Þ 6 0 6 4 0 ð1Þ

3 0 0 1 1 0 27 7: 0 1 15 2 1 1

1.4 Solution to Electric Network Equations

35

First, we normalize the first row and eliminate the first column. There are only two operations: one normalization operation and one elimination operation in this step. The elements in brackets are the computing factors. For a 4 4 coefficient matrix, the elimination of the first column should include one normalization operation and three elimination operations. However, because both a21 and a31 are zero, two corresponding operations are avoided. After the above operations, we obtain 2

1 0 6 0 ð1Þ 6 40 0 0 ð2Þ

0 0 1 1

3 1 27 7: 15 0

The next step is the normalization of the second row and elimination of the second column. There are also only two operations, one normalization operation and one elimination operation in this step. The figures in the brackets of the above matrix are the computing factors. For a 4 4 coefficient matrix, the elimination of the second column should include one normalization operation and two elimination operations. ð1Þ Because a32 is zero, the corresponding operation is avoided. After these operations, we obtain 2

1 60 6 40 0

0 0 1 0 0 ð1Þ 0 ð1Þ

3 1 2 7 7: 1 5 4

To normalize the third row and eliminate the third column, we also need two operations, one normalization operation and one elimination operation. The computing factors are the elements in the brackets of the above matrix. After these operations, we obtain 2

1 60 6 40 0

0 1 0 0

0 0 1 0

3 1 2 7 7: 1 5 ð5Þ

Here, the factor table of the coefficient matrix can be readily written, 1 0 0 1

0 1 0 2

0 1 0 2 : 1 1 1 5

The above factor table can also be found using (1.60) and (1.61). Because there are only six zero off-diagonal elements in the above factor table, six multiply–add

36

1 Mathematical Model and Solution of Electric Network

operations are avoided. In the following, we will use this factor table to obtain the solution to the constant vector: B ¼ ½2 3

2

5 T :

First, eliminating B column by column is executed by using the lower triangular part of the factor table. According to (1.52), b1 is normalized, ð1Þ

b1 ¼ b1 =d11 ¼ 2=1 ¼ 2: Then the operations on b2 ; b3 ; b4 are continued by using the elements of the first column in the lower triangular part according to (1.53). Because l21 and l31 are zero, we have ð1Þ

ð1Þ

ð1Þ

ð1Þ

b2 ¼ b2 l21 b1 ¼ b2 ¼ 3; b3 ¼ b3 l31 b1 ¼ b3 ¼ 2: The above two steps should be avoided and only the following operation is needed ð1Þ

ð1Þ

b4 ¼ b4 l41 b1 ¼ 5 1 2 ¼ 3: After the elimination operation of the first column, we obtain Bð1Þ ¼ ½ 2

3

2

3 T :

ð1Þ

Then normalize b2 according to (1.52) ð2Þ

ð1Þ

b2 ¼ b2 =d22 ¼ 3=1 ¼ 3: ð1Þ

ð1Þ

Now, the operation on b3 ; b4 should use the elements of the second column in the lower triangular part according to (1.53). Because l32 is zero, only the operation related to l42 will be performed. Thus, ð2Þ

ð1Þ

ð2Þ

b4 ¼ b4 l42 b2 ¼ 3 2 3 ¼ 3: After finishing elimination operation of the second column, we have Bð2Þ ¼ ½ 2

3

2

3 T :

ð2Þ

Next, we normalize b3 according to (1.52) ð3Þ

ð2Þ

b3 ¼ b3 =d33 ¼ 2=1 ¼ 2:

1.4 Solution to Electric Network Equations

37

ð3Þ

And then compute b4 according to (1.53) ð3Þ

ð2Þ

ð3Þ

b4 ¼ b4 l43 b3 ¼ 3 1 2 ¼ 5: After finishing the elimination operation of the third column, we obtain Bð3Þ ¼ ½ 2

3

2

5 T : ð3Þ

The last step of the elimination operation is to normalize b4 according to (1.52) ð4Þ

ð3Þ

b4 ¼ b4 =d44 ¼ 5=ð5Þ ¼ 1: At this stage, all of the elimination operation have been completed, the right-hand vector becomes Bð4Þ ¼ ½ 2

3

2

1 T :

Comparing with the factor table, we obtain the following identical solution equations of (1.64) y1

y2

y3

þy4 ¼ þ2y4 ¼ þy4 ¼ y4 ¼

2 3 : 2 1

Now, the unknowns can be solved using the upper triangular part of the factor table according to (1.54). Because u12 ; u13 ; and u23 are zero, corresponding operations are avoided in back substitution. ð4Þ

y4 ¼ b4 ¼ 1 ð3Þ

y3 ¼ b3 u34 y4 ¼ 2 1 1 ¼ 1 ð2Þ

y2 ¼ b2 u24 y4 ¼ 3 2 1 ¼ 1

:

ð1Þ

y1 ¼ b1 u14 y4 ¼ 2 1 1 ¼ 1 Substituting the above results into (1.63), the solutions to original equation (1.62) can be obtained. From the above example, we can see that the computation effort can be saved not only in the formation of the factor table but also in the forward and back substitution. The amount of computation saved by the sparse technique depends on the number of zero elements in the factor table. Therefore, the key point of improving computation efficiency is to keep the number of zero elements in the factor table as high as possible.

38

1.4.4

1 Mathematical Model and Solution of Electric Network

Sparse Vector Method

Nowadays, the sparse matrix techniques are adopted to solve almost all large-scale power network problems. In this section, the sparse vector method, which can further improve the computation efficiency, will be introduced [3]. Sparse vector methods are useful for solving a system of simultaneous linear equations when the independent (right-hand) vector is sparse, or only few elements in the unknown vector are wanted. To take advantage of vector sparsity is relatively simple, but the results of improving computational efficiency and saving memory can be quite dramatic. Therefore sparse vector methods are often used in the compensation method, fault analysis, optimal power flow problem and contingency analysis. In principle, the sparse vector method can be applied to both full- and sparsematrix equations. This section focuses only on the implementation of sparse vector methods in the sparse-matrix situation. According to the above discussion, the admittance matrix Y of an electric network without phase-shifting transformers is symmetric. If there are phase-shifting transformers in the network the sparse admittance matrix is only symmetric in its structure. Nodal voltage equations can be written as YV ¼ I:

ð1:65Þ

For generality, we assume Y is an incidence-symmetric square matrix of order n and can be factorized as Y ¼ LDU;

ð1:66Þ

where L and U are lower and upper triangular matrices with unity diagonals, respectively, and D is a diagonal matrix. It is easy to solve the nodal equations using the above expressions. For example, the simultaneous equations can be written in the following form LDUV ¼ I:

ð1:67Þ

The above formulae can be decomposed as LX ¼ I;

ð1:68Þ

DW ¼ X;

ð1:69Þ

UV ¼ W:

ð1:70Þ

V can be obtained when (1.68)–(1.70) are solved in sequence. If Y is symmetric, matrix U is the transpose of L. If Y is incidence symmetric, matrix U is not the transpose of L, but they are identical in the sparsity structure.

1.4 Solution to Electric Network Equations

39

The forward substitution operations can be expressed as W ¼ D1 L1 I:

ð1:71Þ

The back substitution operations can be expressed as V ¼ U1 W:

ð1:72Þ

Generally, these operations can be performed either by rows or by columns. However, for the sparse vector method, the forward elimination (1.71) must be performed by columns, while the back substitution (1.72) by rows. Many different schemes can be used for storing and accessing L and U. For the sparse vector method, the lowest-numbered, nonzero, off-diagonal element in each column of L or in each row of U must be directly accessed without search. This requirement is satisfied by most storage schemes for L and U. The independent vector I is sparse in many applications. However, the solution vector V is not sparse in general. The term ‘‘sparse vector’’ in the following refers to either a sparse vector I or a subset of vector V containing the elements of interest. The exact meaning is always clear from the context. If the vector I is sparse, only a subset of the columns of L is needed for the forward elimination. This is called the fast forward (FF) process. If only certain elements of vector V are actually wanted, only a subset of the rows of U is needed for the backward substitution. This is called the fast backward (FB) process. [Example 1.5] Solve the following simultaneous linear equations V1

V1

V2 þ2V2

V3 þV3

þV4 þ2V4 þV4 þV4

¼0 ¼1 : ¼0 ¼0

[Solution] The coefficient matrix of above simultaneous linear equations is the same as in (1.64) of Example 1.4. The only difference is that the right-hand vector is sparse. I ¼ B ¼ ½0

1 0

0 T :

Therefore, the factor table of these simultaneous linear equations is the same as that of (1.64). 1 0 0 1

0 1 0 2

0 1 0 2 : 1 1 1 5

40

1 Mathematical Model and Solution of Electric Network

Decomposing the factor table, we obtain 2

1 60 L¼6 40 1

1 0 2

3

2

7 7; 5 1 1 1

6 D¼6 4

1 1 1

3

2

7 7; 5

6 U¼6 4

1

5

From (1.53), we can see that all the operations related with lik ðkÞ can be avoided if bk is equal to zero: ðkÞ

ðk1Þ

bi ¼ bi

ðkÞ

lik bk

3 0 0 1 1 0 27 7: 1 15 1

ði ¼ k þ 1; . . . ; nÞ

ði ¼ k þ 1; . . . ; nÞ:

In other words, the kth column in the lower triangular matrix can be ignored. In this example, b1 is equal to zero, so we can skip the first column of L. For this sparse vector, the elimination should begin from the second column. The elimination also includes the normalization and elimination operations. After this, the right-hand sparse vector is transformed into B0 ¼ ½ 0

1 T :

1 0

The next step is elimination of the third column. Because b03 is zero, the operations related to the third column of L are skipped, thus the elimination of the fourth column is performed directly. Here, we use d44 to normalize b04 , and the ultimate result vector after the elimination operation is B00 ¼ 0 1

0

1 T : 5

As we know, the backward substitution operations must be performed by rows. If only V3 is wanted, the operations with the first and second rows of U can be neglected. If only V2 is wanted, the operations with the first row of U can be avoided. Furthermore, the operations with the third row of U also can be omitted because b03 ¼ 0. Therefore, the back substitution is only needed to perform on the second row of U. Therefore, we have V2 ¼ b002 u24 b004 ¼ 1 2 15 ¼ 35: From the above example, we can see that the key task of sparse vector methods is to identify the active subsets of L and U for FF and FB operations. The active subset of columns for FF depends on the sparsity structure of L and I while the active subset of rows for FB depends on the sparsity structures of U and V. In order to find the active subset of FF and improve the computation efficiency, the following simple algorithm can be summarized according to the above example

1.4 Solution to Electric Network Equations

41

1. Zero all locations in I, and enter the given nonzero elements in I. 2. Search the nonzero elements in I and let k be the location number of the lowestnumbered nonzero element. 3. Perform the forward eliminations defined by column k of L on I. 4. If k ¼ n, exit. Else, return to Step 2. This algorithm ensures that only the necessary nonzero operations of FF are performed, but it is wasteful because of zeroing and searching. A similar algorithm can be used to FB, but it is even more wasteful. In the following we introduce a more efficient algorithm based on the concept of the factorization path. A factorization path for a sparse vector is represented by an ordered column list of L for FF operations. A path is executed in forward order for FF and in reverse order for FB. The same or different paths may be used for FF and FB depending on the application. The path for a singleton is basic to the path concept. A singleton is a vector with only one nonzero element. Assume that the nonzero element is in location k. The following algorithm determines the path of the singleton: 1. Let k be the first number in the path. 2. Get the number of the lowest-numbered nonzero element in column k of L (or row k of U). Replace k with this number, and list it in the path. 3. If k ¼ n, exit. Else, return to Step 2. The path for a singleton can be determined directly from the indexing arrays without searching or testing. A general sparse vector is the sum of singleton vectors, and its path is the union of the paths of its composite singleton vectors. For any sparse system, a path can be always associated with a given sparse vector. [Example 1.6] Find the factorization path of the electric network shown in Fig. 1.11. [Solution] Figure 1.12 shows the sparsity structure of the incidence symmetric admittance matrix of the network as shown in Fig. 1.11 (only the lower triangular part of the matrix is labeled). Because there are 21 branches in the network, 21 l 12

13 3

1

7

9

8

11 4

2 14

10

6

Fig. 1.11 Example electric network

15

5

42

1 Mathematical Model and Solution of Electric Network

represent the off-diagonal elements of the matrix. After triangular factorization, 10 fill-in elements (labeled as *) are added. Therefore there are altogether 31 nonzero elements in the factor table. The factorization path of any singleton can be directly obtained from the structure of the factor table. For example When k ¼ 1, the singleton path is 1 ! 2 ! 7 ! 12 ! 13 ! 14 ! 15 When k ¼ 5, the singleton path is 5 ! 11 ! 13 ! 14 ! 15 When k ¼ 6, the singleton path is 6 ! 9 ! 10 ! 12 ! 13 ! 14 ! 15 When a sparse vector is not a singleton, its path is the union of the paths of its composite singletons. For a sparse vector as follows I ¼ ½1 0

0

0

1 0

0

0

0

0 0

0

0

0 0 T

we have its path as the union of the paths of its composite singletons when k ¼ 1 and k ¼ 5, 1 ! 2 ! 7 ! 12 ! 5 ! 11 ! 13 ! 14 ! 15: In Table 1.1 we list the entire factorization paths for the network shown in Fig. 1.12. A pictorial view of the path table is provided by the path graph shown in Fig. 1.13. Utilizing this path graph, highly efficient algorithms for the sparse vector can be obtained. For example, assume the injected current at node 5 is I5 while the injected currents of other nodes are zero, and the voltage at node 1 is wanted. To do so, we carry out FF operations according to the following active column sequence: 5 ! 11 ! 13 ! 14 ! 15: And then carry out FB operations according to the following active row sequence: 15 ! 14 ! 13 ! 12 ! 7 ! 2 ! 1 In the above solution process, only the elements of five columns in lower triangular and seven rows in upper triangular elements are employed, the computation efficiency is improved dramatically. For sparse vector methods, the above path graph Table 1.1 Path table Node Next node 1 2 2 7 3 4 4 8 5 11 6 9 7 12

Node 8 9 10 11 12 13 14

Next node 10 10 12 13 13 14 15

1.4 Solution to Electric Network Equations

43

I = [1 0 0 0 1 0 0 0 0 0 0 0 0 0 1

2

3

4

5

6

7

8

0] 9

T

10

11

12

13

14

15

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Fig. 1.12 Sparse structure of a network’s factor table Fig. 1.13 The path graph

6 9

3 4

1

8

10 2 7

12 13

5 11

14 15

should be determined in advance and then be utilized directly, thus unnecessary zeroing and searching can be skipped.

1.4.5

Optimal Ordering Schemes of Electric Network Nodes

At present, the Gauss elimination method introduced in Sect. 1.3.1 is applied to solve the node equations I ¼ YV in most power system analysis programs. In order to solve the network equation repeatedly, the admittance matrix is usually

44

1 Mathematical Model and Solution of Electric Network

Fig. 1.14 Relationship between Gauss elimination and Y-D transformation

i l

1 j

factorized first, and then the factor table can be directly used to solve the equations with different right-hand vectors. As we know, the admittance matrix is sparse and the triangular matrices after factorization are also sparse. Generally, the distributions of nonzero elements in the admittance matrix are different from those in the factorized triangular matrix, because some new nonzero elements, i.e., the fill-in elements, may occur in the elimination or LU factorization process. The addition of fill-ins in the elimination process can be explained intuitively by Y-D transformation. As shown in Fig. 1.14, node l does not directly connect with nodes i and j in the initial network, thus corresponding elements Yil and Ylj in its admittance matrix are zero while Yij is nonzero. It can be proved that eliminating the first column of the admittance matrix in Gauss elimination is equivalent to eliminating node 1 by Y-D transformation as shown in Fig. 1.14. New branches connecting node pairs ij, il, and lj are created. Therefore, in the new admittance matrix, Yil ; Ylj ; and Yij are all nonzero elements, thus two fill-ins occur in eliminating the first column. Generally, eliminating node k which is the central point of a star network will create a mesh network whose vertexes are nodes connecting directly with node k. If the number of nodes connecting directly with k is Jk , the branches in the mesh network should be combinations of any two nodes of Jk nodes, which is equal to ð1=2ÞJk ðJk 1Þ. Assuming that there already exist Dk branches connecting these Jk nodes, the number of new branches (the number of fill-ins) after the elimination of node k is 1 Dbk ¼ JK ðJK 1Þ Dk : 2

ð1:73Þ

The number of fill-ins highly depends on the elimination sequence or the ordering number of the nodes. In Fig. 1.15, four number ordering schemes and the corresponding fill-ins in the triangular matrix are denoted. Apparently, different number ordering schemes will result in different fill-ins. An optimal ordering minimizes the fill-ins in the factor table during the LU factorization process. Different number ordering schemes should be compared according to the number of fill-ins. At present, several effective schemes have been developed. Among them the following three ordering schemes are widely employed: 1. Static ordering scheme: This scheme numbers the nodes according to the number of branches connected to them. It means that the nodes are ordered

1.4 Solution to Electric Network Equations Different ordering schemes

2

1

3

4 3 2

1

2

4

5 2

3

3

1

4

5 2 5

1

4

4

•

Admittance matrix

• • • • • • •

3

Lower triangular matrix

• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •

• • • • • •

•

•

•

1

5

45

•

•

•

•

• • • • • • • • • • •

Non-zero element

×

6

•

• • • • × • • × × •

3

• • • • • • • × •

1

•

•

•

× • × × • × × ×

Fillins

•

•

•

• • • • • •

0

Non-zero fill-in

Fig. 1.15 Illustration of number ordering

from the node with fewest branches to the node with most branches. If the numbers of connected branches for more than one node are the same, any one of them can be ordered first. Before ordering, the number of the branches connected to each node needs be counted. The scheme can be explained intuitively as follows: in the admittance matrix, the node with the fewest connected branches corresponds to the row which has the fewest nonzero elements, so the fill-ins will be generated with less possibility in the elimination operation. This scheme is very simple and suitable to be applied to small networks with fewer loops. 2. Semidynamic ordering scheme: In the above scheme, the number of branches connected to each node is counted based on the initial network and is constant in the ordering process. In fact, in the process of node elimination, the number of branches connected to each node will change according to D Y transformation. Therefore, the number of branches of the remaining nodes should be updated

46

1 Mathematical Model and Solution of Electric Network

after each elimination and then they should be ordered according to new data. This ordering scheme might be expected to result in better fill-in reduction, because it considers the changing number of incident branches during the elimination process. 3. Dynamic ordering scheme: The above two schemes are only suboptimal, which cannot guarantee minimizing fill-in number. The more rigorous scheme numbers the node according to the principle that introduces the fewest new branches. The ordering process is as follows l

l

l

According to D Y transformation, count the number of new branches (the number of fill-ins) added after the elimination on each node, and the node with the fewest branches (including fill-ins) is numbered first. Update the new number of incident branches connected to the remaining nodes. It is clear that the computation complexity of this scheme is much more than the other two.

[Example 1.7] Ordering the nodes of the network as shown in Fig. 1.16. [Solution] The above three ordering schemes are performed and compared as follows for the network as shown in Fig. 1.16. 1. Static optimal ordering scheme: There are eight nodes and 14 branches in this network. The number of incident branches on each node is listed in Table 1.2. The ordering results according to the static ordering scheme are shown in Fig. 1.17a. There are four new branches added in the process of node eliminations. When node 1 is eliminated, branch 2–7 and branch 2–8 are generated and when node 2 is eliminated, branch 3–7 and branch 4–7 are added. Factorizing the corresponding admittance matrix, we get the structure of the lower triangular matrix as shown in Fig. 1.17b. Four fill-ins, l72 ; l73 ; l74 ; and l82 correspond with the four new added branches. 2. Semidynamic ordering scheme: The process of numbering is shown in Table 1.3 and the result in Fig. 1.18a.

O

P Q

N M

T S

Fig. 1.16 Example of the node ordering

R

1.4 Solution to Electric Network Equations

47

Table 1.2 Number of branches at each node for network shown in Fig. 1.16 Node M N O P Q R Number of incident branches 4 3 3 3 3 3

3

2

4

×

3 8

7 5

4

× × ×

5 6

8

× ×

b

1

7 6

a

T 6

1 2

1

S 3

o

×

× × × × × × × × o

o

2 3

4

o

5

6

7

Fig. 1.17 Results of static optimal ordering Table 1.3 Process of semidynamic ordering scheme Node Process of numbering

M 4 4 4 4 (3)

N (3)

O 3 4 3 (2)

P 3 (3)

Q 3 3 (2)

R 3 3 3 3 3 (2)

S 3 3 3 3 3 2 (1)

T 6 6 5 4 3 2 1 (0)

Node ordered N P Q O M R S T

Node number 1 2 3 4 5 6 7 8

In this scheme, two new branches are introduced in the elimination process, that is, when node 1 is eliminated, branch 4–5 and branch 4–8 are added. 3. Dynamic ordering scheme: In order to number the nodes, we need to count the number of new branches (the number of fill-ins) added after eliminating each node. The result is listed in Table 1.4. From this table, we can see that node R or S should be numbered first. Suppose that node R is selected as node 1. After this node is eliminated we count the new branch numbers when eliminating other nodes. The results are shown in Table 1.5. From Table 1.5, node S should be numbered as node 2. The computation is repeated until the last node has been numbered. The results are shown in Fig. 1.18b. Only one new branch is added by this scheme. Therefore, for complex networks, the dynamic ordering scheme can obtain more satisfactory results.

48

1 Mathematical Model and Solution of Electric Network

Fig. 1.18 Result of semidynamic and dynamic optimal ordering

4

5

2

7

3

1

4 8

5

Table 1.4 First step of dynamic ordering scheme Node eliminated M N O Number of new branches 2 2 2

Table 1.5 Second step of dynamic ordering scheme Node eliminated M N O Number of new branches 1 2 2

1.5 1.5.1

8

3 1

6

a

6

7

P 1

Q 1

P 1

Q 1

b

2

R 0

S 0

T 10

S 0

T 7

Nodal Impedance Matrix Basic Concept of Nodal Impedance Matrix

As described above, the nodal equation of electric network can be generally represented as I ¼ YV; where I is the column vector of the nodal injection currents. Generally, it is the known variable in power system analysis; V is the column vector of the nodal voltages. Generally, it is unknown variable in power system analysis; and Y is the nodal admittance matrix. The above linear simultaneous equations can be solved by various methods, such as the direct method by inverting the admittance matrix. Suppose Z ¼ Y1 :

ð1:74Þ

Then, the above nodal equation can be written as V ¼ ZI

ð1:75Þ

1.5 Nodal Impedance Matrix

49

or in the expansion 9 V_ 1 ¼ Z11 I_1 þ Z12 I_2 þ þ Z1i I_i þ þ Z1n I_n > > > > _ _ _ _ _ > V2 ¼ Z21 I1 þ Z22 I2 þ þ Z2i Ii þ þ Z2n In > > = : V_ i ¼ Zi1 I_1 þ Zi2 I_2 þ þ Zii I_i þ þ Zin I_n > > > > > > > ; V_ n ¼ Zn1 I_1 þ Zn2 I_2 þ þ Zni I_i þ þ Znn I_n

ð1:76Þ

Comparing (1.75) with (1.76), we can see that 2

Z11 6 Z21 6 6 6 Z¼6 6 Zi1 6 4

Z12 Z22

Zn1

Zn2

Zi2

Z1i Z2i .. .

Zii

Zni

.. .

3 Z1n Z2n 7 7 7 7 7: Zin 7 7 5

ð1:77Þ

Znn

This is the nodal impedance matrix corresponding to the nodal admittance matrix Y, and they have the same order. The diagonal element Zii is called the self-impedance or the input impedance, and the off-diagonal element Zij is called the mutual impedance or the transfer impedance between the node i and node j. When the injection currents are known, the nodal voltages of the network can be solved directly through (1.75) or (1.76). The physical meaning of the elements in the nodal impedance matrix can be explained as follows: If a unit current is injected into node i, and all other nodes are open, i.e., I_i ¼ 1 I_j ¼ 0 ðj ¼ 1; 2; . . . ; n; j 6¼ iÞ: Then from (1.76), we can get V_ 1 ¼ Zi1 V_ 2 ¼ Zi2 V_ i ¼ Zii _ Vn ¼ Zin : Thus, we know that the elements in the ith column of the impedance matrix have the following physical meaning:

50

1 Mathematical Model and Solution of Electric Network

1. The diagonal element Zii of the impedance matrix is equal in value to the voltage of node i, when a unit current is injected into node i and all the other nodes are open. Therefore, Zii can be also regarded as the equivalent impedance between node i and the ground when all other nodes are open. If the network has some grounding branches and node i is connected to the network, Zii must be a nonzero element. 2. The off-diagonal element Zij is the mutual impedance between node i and j. When a unit current is injected into node i and all the other nodes are open, Zij is equal in value to the voltage of node j. Because there are always some electromagnetic connections (including indirect connections) among the nodes of a power network, the voltage of every node should be nonzero when node i is injected with a unit current and the other nodes are open. That is to say, all the mutual impedance elements Zij are nonzero elements. Therefore, the impedance matrix is a full matrix without zero elements. The impedance matrix method for directly solving network voltage used to be very popular in the early stages of power system analysis by computer. But the impedance matrix is a full matrix, more memory and operations are required, which limits its applications especially for large-scale networks. Nevertheless, it is conceptually very useful in many aspects of power system analysis. This will be introduced in later chapters.

1.5.2

Forming Nodal Impedance Matrix by Using Nodal Admittance Matrix

Comparing with the admittance matrix, it is more difficult to formulae the nodal impedance matrix of an electric network. Two general methods of constructing the impedance matrix will be introduced in the next sections. According to the discussion in Sect. 1.2.2, the admittance matrix of an electric network can be obtained directly from its configuration and parameters. So we can get the impedance matrix by inverting the admittance matrix. Several methods can be used to invert a matrix. In the following, we will illustrate one of them – inversion of a matrix through solving linear equations. Consider an admittance matrix Y and its corresponding impedance matrix Z. Solving the linear equation YZj ¼ Bj

ð1:78Þ

we can get the element Zj of the column j in the impedance matrix, where Bj is a column vector: Bj ¼ ½ 0

0 1 0 j

0 t :

1.5 Nodal Impedance Matrix

51

Solving (1.78) successively for j ¼ 1; 2; . . . ; n, we can obtain all elements of the impedance matrix. When the elements are solved column by column, only the righthand vector Bj is changed in (1.78). Therefore, the triangular factorization algorithm is very efficient to solve (1.78) (refer to Sect. 1.3.2 for details). Since the admittance matrix is symmetric, it can be factorized as: Y ¼ LDLT : The elements of the unit lower triangular matrix L and the diagonal matrix D can be obtained from (1.61). Therefore, (1.78) can be rewritten as: LDLT Zj ¼ Bj :

ð1:79Þ

LT Zj ¼ W j ;

ð1:80Þ

DW j ¼ X j :

ð1:81Þ

LX j ¼ Bj :

ð1:82Þ

Let

Then according to (1.79), we have

Thus the whole process of solving (1.79) can be decomposed into three steps: 1. Solve Xj from (1.82) Expand (1.82) as 2

1 6 l21 6 6 l31 6 6 .. 6 . 6 6 lj1 6 6 .. 4 . ln1

32

1 l32 .. . lj2 .. .

ln2

1 .. .

..

. lj;j1 .. .

lnj

1

..

. ln;n1

3 2 3 x1 0 7 6 x2 7 6 0 7 76 7 6 . 7 7 6 x3 7 6 . 7 76 7 6 . 7 76 .. 7 6 . 7 76 . 7 ¼ 6 .. 7: 76 7 6 7 7 6 xj 7 6 1 7 76 7 6 7 76 .. 7 6 0 7 54 . 5 4 5 .. 1 . xn

ð1:83Þ

Then we can get x1 , x2 , . . ., xn sequentially from the above equation. This is the forward substitution. 2. Obtain W j from (1.81) Expand (1.81) as

52

1 Mathematical Model and Solution of Electric Network

2 6 6 6 6 6 6 6 6 6 6 4

d1

32

d2

d3

..

.

dj

3 2 3 w1 x1 76 w2 7 6 x2 7 76 7 6 7 76 w3 7 6 x3 7 76 7 6 7 76 .. 7 6 .. 7 76 . 7 ¼ 6 . 7: 76 7 6 7 76 wj 7 6 xj 7 76 7 6 7 76 .. 7 6 .. 7 .. 54 . 5 4 . 5 . dn wn xn

ð1:84Þ

Then we can get w1 ; w2 ; . . . ; wn sequentially from the above equation. This is the normalization. wi ¼ xi =di

i ¼ 1; 2; . . . ; n:

ð1:85Þ

3. Obtain Zj from (1.80) Expand (1.80) as 2

1 l21 6 1 6 6 6 6 6 6 6 6 6 4

l31 l32 .. .

1

lj1 lj2

.. . 1

ln1 ln2 .. .

32

Z1j Z2j .. .

3

2

W1 W2 .. .

3

76 7 6 7 76 7 6 7 76 7 6 7 76 7 6 7 76 7 6 7 6 Zjj 7 ¼ 6 Wj 7: lnj 7 76 7 6 7 6 . 7 6 .. 7 .. 7 . 7 6 7 6 . 76 . 7 6 . 7 7 ln;n1 54 Zn1;j 5 4 Wn1 5 Wn Znj 1

ð1:86Þ

Then we can solve Znj ; Zn1;j ; . . . ; Zjj ; . . . ; Z2j ; Z1j one by one from bottom to top sequence. This is the backward substitution. [Example 1.8] Form the impedance matrix of the electric network shown in the Example 1.1 from its admittance matrix by applying the factorization algorithm. [Solution] The admittance matrix can be factorized by using (1.61), d1 ¼ Y11 ¼ 1:378742 j6:291665 Y21 0:6242024 þ j3:900156 l21 ¼ ¼ ¼ 0:612227 þ j0:034979 d1 1:378742 j64:57121 d2 ¼ Y22 l221 d1 ¼ ð1:453909 j66:98082Þ ð0:61227 þ j0:031979Þ2 ð1:378742 j6:291665Þ ¼ 1:208288 j64:57121 Similarly, other elements can be found through using the recursion formulae. Then the admittance matrix is factorized as

:

1.5 Nodal Impedance Matrix

2 6 6 6 6 6 6 6 L¼6 6 6 6 6 6 4

1 0:612227 þj0:034979 0:425687 j0:026671

53

3

0:073971 j0:017193 0:982943 j0:018393

1 0:137743 j0:027718 0:924654 þj0:027559

1 1:189287 þj0:048151

2

1:378742 6 j6:291665 6 6 6 6 6 6 D¼6 6 6 6 6 6 4

7 7 7 7 7 7 7 7; 7 7 7 7 7 5

1

1 3

1:208288 j64:57121 1:022377 j34:30237 0:887283 j3:640902

7 7 7 7 7 7 7 7: 7 7 7 7 7 0:038964 5 þj1:263678

The first step is to get the first column elements Z1 of the impedance matrix. In this situation, (1.83) should be written as 2 6 6 6 6 6 6 6 6 6 6 6 6 6 4

1 0:612227 þj0:034979 0:425687 j0:026671

3 7 72 3 2 3 7 1 7 x1 76 7 6 7 76 x2 7 6 0 7 76 7 6 7 76 x3 7 ¼ 6 0 7: 74 5 4 5 0 7 x4 7 0 7 x5 7 5

1 0:073971 j0:017193 0:982943 j0:018393

1 0:137743 j0:027718 0:924654 þj0:027559

1 1:189287 þj0:048151

1

Therefore, x1 ¼ 1 x2 ¼ 0 l21 x1 ¼ 0:612227 j0:034979 x3 ¼ 0 l31 x1 l32 x2 ¼ 0:471576 þ j0:034609 x4 ¼ 0 l42 x2 l43 x3 ¼ 0:665138 j0:027805 x5 ¼ 0 l53 x3 l54 x4 ¼ 1:226700 j0:046890:

54

1 Mathematical Model and Solution of Electric Network

From (1.85), we obtain x1 d1 x2 ¼ d2 x4 ¼ d4 x3 ¼ d3 x5 ¼ d5

w1 ¼ w2 w4 w3 w5

1 ¼ 0:033234 þ j0:151658 1:378742 j63291665 0:612227 j0:034979 ¼ ¼ 0:000719 þ j0:009468; 1:208288 j64:57121 0:665138 j0:027805 ¼ ¼ 0:049233 þ j0:170687 0:887283 j3:640902 0:471576 þ j0:034609 ¼ ¼ 0:000599 þ j0:013765; 1:022377 j34:30237 1:226700 j0:046890 ¼ ¼ 0:006535 j0:970940: 0:03894 þ j1:263678 ¼

Back substitution is executed using the following equation 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4

1

3

0:612227 þj0:034979

0:425687 j0:026671

1

0:073971 j0:017193

0:982943 j0:018393

1

0:137743 j0:027718 1

3 0:033234 7 6 þj0:151658 7 7 6 7 72 7 3 6 7 Z11 6 7 7 6 0:000719 7 76 7 6 7 76 7 6 7 76 Z21 7 6 þj0:009468 7 76 7 6 7 76 7 6 7 76 7 6 0:000599 7 7 0:924654 76 Z31 7 ¼ 6 76 7 6 þj0:013765 7: 7 6 7 þj0:027559 76 76 7 6 7 76 Z41 7 6 7 76 7 6 0:049233 7 4 5 6 7 1:189287 7 7 6 þj0:170687 7 7 6 7 þj0:048151 7 Z51 6 7 7 6 7 5 4 0:006535 5 1

2

j0:970940

Then the first column elements of the impedance matrix are Z51 ¼ 0:006535 j0:970940 Z41 ¼ 0:005290 j0:983725 Z31 ¼ 0:006862 j1:019487 Z21 ¼ 0:005555 j1:032911 Z11 ¼ 0:017972 j0:914690: The computation can be performed in a similar way and the whole impedance matrix can be obtained column by column. Thus we finally have

1.5 Nodal Impedance Matrix

2 6 6 6 6 6 6 6 6 Z¼6 6 6 6 6 6 6 4

0:017972 j0:914690 0:0055555 j1:032911 0:006862 j1:019487 0:005290 j0:983725 0:006535 j0:970940

0:005555 j1:032911 0:007781 j0:961291 0:010007 j1:037907 0:007410 j0:918658 0:009530 j0:988482

55

0:006862 j1:019487 0:010007 j1:037907 0:026875 j0:90470 0:007410 j0:918658 0:009530 j0:988482

0:005290 j0:983725 0:007410 j0:918658 0:009530 j0:988482 0:007057 j0:859912 0:009076 j0:941412

0:006535 j0:970940 0:009530 j0:988482 0:025596 j0:861619 0:009076 j0:941412 0:024377 j0:790589

3 7 7 7 7 7 7 7 7 7: 7 7 7 7 7 7 5

As described above, the elements of the jth column in the impedance matrix are equal to the nodal voltages in value when a per-unit current is injected into node j and other nodes are open. Therefore, finding the elements of the jth column from (1.78) is equivalent to solving the following nodal equation YV ¼ I j ;

ð1:87Þ

where all the elements of the current column vector I j are zero except the jth element equals 1. Obviously, V obtained from this equation is equal to Zj in value. It is worth noting that the computation burden of this method is a little too heavy in some situations; for example, if we want to derive the impedance matrix of a network with n nodes, n linear equations must be solved n times. Hence this method is only suitable for the case in which only a few elements are of interest. In power flow and short circuit analysis, the input impedance of one pair of nodes and the transfer impedance between two node pairs are often calculated using the above method. In Fig. 1.19, in order to get the input impedance of node i and j and the

•

V1 •

V2 •

•

N

Vi Ii = 1 •

•

Vj Ij =−1 •

Vk Vl •

Fig. 1.19 Solving node pair’s input and transfer impedance

56

1 Mathematical Model and Solution of Electric Network

transfer impedance between ij and kl, a unit current is injected between node i and node j, while other nodes are open. That is I_i ¼ 1;

I_j ¼ 1:

In this case, solve the network equation YV ¼ Fij ;

ð1:88Þ

where 2

3 0 6 .. 7 6 . 7 6 7 6 1 7 i 6 7 6 0 7 6 . 7 . 7 Fij ¼ 6 : 6 . 7 6 1 7 j 6 7 6 0 7 6 7 6 . 7 4 .. 5 0 The nodal voltage can be obtained and the input impedance of node pair ij is Zijij ¼ V_i V_ j :

ð1:89Þ

The transfer impedance between ij and kl is Zklij ¼ V_ k V_ l :

1.5.3

ð1:90Þ

Forming Nodal Impedance Matrix by Branch Addition Method

In the above section, we have described a method of forming the impedance matrix by using the admittance matrix. An alternative method is to form the impedance matrix directly by the branch addition method. The method is straightforward in computation and allows easy impedance matrix modification for changes in the network. Therefore it is applied widely. The forming process is illustrated by Fig. 1.20. We start to form the impedance matrix from a grounded branch and a matrix of order 1 is formed. In Fig. 1.20, z10 first is used to form this matrix. Then branch z12 is added and the new branch creates a new node . We call it adding a tree branch if a new node is generated when adding a branch. At the same time, the order of the

1.5 Nodal Impedance Matrix

57 3

z23

z13 4

z14

z12

1

z10

z25

2

5

z20

Fig. 1.20 Process of branch addition method in forming impedance matrix

corresponding matrix increases by 1. Thus after adding tree branch z12 , we obtain a 2 2 impedance matrix. We next add branch z20 . In this situation, there is no new node generated. The order of the impedance matrix does not change. This is called adding a link branch. All the elements of the impedance matrix must be updated when a link branch is added. Repeat the operations in a similar way: after adding tree branch z13 , node 3 is created. Then the order of the impedance matrix becomes three. After adding tree branch z14 , node 4 is created and the order of the impedance matrix becomes four. After adding tree branch z25 , node 5 is generated. The order of the impedance matrix becomes five. When adding link z25 , no new node is generated and the order of the impedance matrix is still five. The impedance matrix is formed after all the branches have been added to the electric network. It should be noted that the sequence of adding the branches is not unique. An alternative sequence is as follows: Tree branch z10 ! tree branch z20 ! link z12 ! tree branch z13 ! link z23 ! tree branch z14 ! tree branch z25 . Of course, there are some other schemes besides these two schemes. And it can be proved that whatever the branch adding sequence is, the impedance matrix is the same when the node number ordering is the same. However, the computation efforts under the different adding sequences are quite different. The effects of adding a tree branch or a link branch on the impedance matrix will be discussed in the following: 1. Adding a tree branch Assume that the m m impedance matrix of an electric network has been formed for the first m nodes. 2

Z11 6 Z21 6 6 ZN ¼ 6 6 Zi1 6 4 Zm1

Z12 Z22 Zi2 Zm2

Z1i Z2i Zii Zmi

3 Z1m Z2m 7 7 7 7: Zim 7 7 5 Zmm

ð1:91Þ

58

1 Mathematical Model and Solution of Electric Network

When a tree branch Zij is added at node i, a new node j is created and the order of the impedance matrix becomes m þ 1 (see Fig. 1.21). Suppose the new impedance matrix is 2 0 Z11

6 6 6 0 6 Z21 6 6 6 6 6 0 6Z 0 ZN ¼ 6 i1 6 6 6 6 0 6 Zm1 6 6 6 4 0 Zj1

0 Z12 0 Z22

0 Zi2

0 Z1i

0 Z2i

Zii0

0 Z1m

0 Z2m

0 Zim

0 Zm2

0 Zmi

0 Zmm

0 Zj2

Zji0

.. . .. . .. . .. . .. . .. . .. . .. .

0 Zjm

3 0 Z1j

7 7 7 7 7 7 7 7 0 7 Zmj 7 7: 7 7 7 0 7 7 Zmj 7 7 7 5 Zjj0 0 Z2j

ð1:92Þ

We first solve the m m matrix inside the dashed lines of (1.92). In order to obtain 0 0 0 0 the values of the first column Z11 Z21 Zi1 Zm1 ; a unit current is injected in node 1 and the other nodes are open as shown in Fig. 1.21a. In this case, voltages of the node 1, 2, . . ., m have nothing to do with branch zij , therefore, 0 0 0 0 Z11 ¼ Z11 ; Z21 ¼ Z21 ; . . . ; Zi1 ¼ Zi1 ; . . . ; Zm1 ¼ Zm1 :

It means that the first column of Z0N is the same as the first column of ZN . Similarly, the second column of Z0N is the same as the second column of ZN . Therefore we can deduce that the m m matrix inside the dashed lines of (1.92) is the original impedance matrix before adding the branch zij .

•

V1

•

V1 V2

•

I2 = 1

•

•

N

V2

Vi zij •

N

j

•

Vm

a Fig. 1.21 Adding tree branch

Vi zij •

•

Vm

b

j Ij = 1 •

1.5 Nodal Impedance Matrix

59

We now solve the jth column of Z0N . Similarly, j and other nodes are open as shown in Fig. 1.21b. In this situation, voltages of the node 1; 2; . . . ; i; . . . ; m are the same as those when a unit current is injected in node i, so we have, 0 0 0 Z1j ¼ Z1i ; Z2j ¼ Z2i ; . . . ; Zij0 ¼ Zii ; . . . ; Zmj ¼ Zmi :

ð1:93Þ

The voltage of node j is V_ j ¼ V_i þ zij 1: According to the physical meaning of the impedance matrix, we obtain Zjj ¼ Zii þ zij :

ð1:94Þ

Due to the symmetry of the impedance matrix, the off-diagonal elements of the jth row in Z0N can be obtained as follows, 0 0 0 Zj1 ¼ Z1j ; Zj2 ¼ Z2j ; . . . ; Zji0 ¼ Zij ; . . . ; Zjm ¼ Zmj :

ð1:95Þ

Hence all the elements in the impedance matrix after adding tree branch zij are found. Additionally, although the order of the new impedance matrix increases by 1, the computation to form it is relatively simple. 1. Adding a link The impedance matrix of the initial network is denoted as ZN . When link zij is added between nodes i and j, the impedance matrix becomes Z0N . The orders of these two matrices are the same because no new node is generated in the network. We now consider how to calculate the elements of new impedance matrix Z0N . As shown in Fig. 1.22, suppose the injection current vector of the new network is I, I ¼ I_1

I_2

I_i

I_j

V_i

V_ j

I_m

t

and the nodal voltage vector is V V ¼ V_ 1

V_2

t V_ m :

Thus the following relationship holds V ¼ Z0N I:

ð1:96Þ

60

1 Mathematical Model and Solution of Electric Network

Fig. 1.22 Adding a link V1

•

I1

•

•

I2

•

V2 •

N

Vi •

Iij

zij

•

Vj •

Vm

•

Ij •

Ij •

Im

From Fig. 1.22, the nodal current injected into the initial network is 2

3 I_1 6 I_2 7 6 7 6 . 7 6 .. 7 6 7 6 I_i I_ij 7 6 7 7 I0 ¼ 6 6 ... 7 ¼ I AM I ij ; 6 7 6 I_ þ I_ 7 6 j ij 7 6 . 7 6 . 7 4 . 5 I_m

ð1:97Þ

where AM is a column vector related to the added link branch, 3 0 6 .. 7 6 . 7 6 7 6 1 7 6 7 6 0 7 6 . 7 . 7 AM ¼ 6 6 . 7 6 1 7 6 7 6 0 7 6 7 6 . 7 4 .. 5 0 2

i ð1:98Þ j

According to the nodal equation of the original network, V ¼ ZN I0 ¼ ZN I ZN AM I_ij :

ð1:99Þ

ZN AM ¼ ZL :

ð1:100Þ

Assume

1.5 Nodal Impedance Matrix

61

We know that ZL is a column vector 2

Z1i Z1j Z2i Z2j .. .

3

6 7 6 7 6 7 6 7 6 7 6 Zii Zij 7 6 7 ZL ¼ 6 7: .. 6 7 . 6 7 6 Zji Zjj 7 6 7 6 7 .. 4 5 . Zmi Zmj

ð1:101Þ

Rewrite (1.99) as, V ¼ ZN I ZL I_ij :

ð1:102Þ

The voltage difference between nodes i and j is equal to V_ i V_ j ¼ zij I_ij ¼ ATM V;

ð1:103Þ

where ATM is the transpose of AM . Substituting (1.102) into (1.103), we obtain zij I_ij ¼ ATM ZN I ATM ZL I_ij : I_ij can be solved as follows 1 T I_ij ¼ Z I; ZLL L

ð1:104Þ

ZLL ¼ ATM ZL þ zij ¼ Zii þ Zjj 2Zij þ zij ;

ð1:105Þ

where

ZTL ¼ ATM ZN ¼ ðZN AM ÞT : Substituting (1.104) into (1.102), we have

1 T V ¼ ZN ZL ZL I: ZLL

ð1:106Þ

Comparing (1.96) with (1.106), we obtain the new impedance matrix Z0N , Z0N ¼ ZN

1 ZL ZTL : ZLL

ð1:107Þ

62

1 Mathematical Model and Solution of Electric Network

Expanding the above equation, we have the following formulae of the elements in Z0N Zkl0

ZLk ZLl ¼ Zkl ZLL

k ¼ 1; 2; . . . ; m : l ¼ 1; 2; . . . ; m

ð1:108Þ

In contrast with adding a tree branch, the computation of adding a link is quite heavy and complicated in which each element of the impedance matrix must be recalculated according to (1.108). The speed of forming the impedance matrix mainly depends on computations for adding links. Therefore the sequence of adding branches affects the computation speed dramatically. For example of the network in Fig. 1.20, the computations of adding link z23 according to the first sequence are performed on a 5 5 matrix, but the recalculations are just executed on a 3 3 matrix according to the second sequence. Hence the more reasonable sequence of adding branches is to add links as early as possible. If the transformer branch is involved, the P equivalent circuit as shown in Fig. 1.4 can be used in forming the impedance matrix. Comparing with a transmission line, two more branches must be added for each transformer and in most circumstances both of them are links. Therefore the computation burden increases notably. Now a direct method of adding a transformer branch is introduced in the following, which need not use the P equivalent circuit. First, we discuss the situation that the transformer added is a tree branch. In Fig. 1.23a, the leakage impedance is put at the nominal turn ratio side of the transformer. If the leakage impedance is put at the off nominal side, the formulae can be derived in a similar way. The impedance matrix of the original network is denoted as ZN (see (1.91)). When the transformer is added as a tree branch, the order of the new impedance matrix Z0N increases by 1 (see (1.92)). It can be proved that the m m block matrix in the top-left of Z0N is just ZN . As shown in Fig. 1.23b, the transformer is substituted for its equivalent circuit. When node j is open, the transformer’s P equivalent circuit is also opened as viewed from node i. This can be explained as follows. •

V1

1 1:K Ij = 1 2 Ii = K •

•

•

V2

•

N

2

N

i z ij

Fig. 1.23 Adding a transformer

i Kzij

Kzij

m

a

•

I1 I2

1

K−1

b

N

j

•

Iij =K

•

Vi

•

zi j

Ii

1:K Ij Im m

K 2zij

•

•

−Vj

1−K

c

•

1.5 Nodal Impedance Matrix

63

The impedance of the loop constituted by nodes i, j and the ground is zij0 ¼ Kzij þ

K2 K zij ¼ zij : 1K 1K

And the impedance between node i and the ground is zi0 ¼ ðK=ð1 KÞÞzij0 . The value of the parallel impedance of zi0 and zij0 becomes infinity. When a unit current is injected at each node of the original network, the current distribution of the original network is unchanged after adding a transformer as a tree branch. Hence the nodal voltages are also unchanged. The issue now is how to solve the new elements of Z0N . Focus on this question, a unit current is injected at node j and the other nodes are open as shown in Fig. 1.23b. This is just like the injecting current K into the original network at node i. Thus the nodal voltages are V_ 1 ¼ KZ1i ; V_ 2 ¼ KZ2i ; . . . ; V_ i ¼ KZii ; V_m ¼ KZmi : The voltage of node j is V_ j ¼ KðV_i þ Kzij Þ ¼ K 2 ðZii þ zij Þ: Thus, we obtain 0 0 0 Z1j ¼ Kz1i ; Z2j ¼ Kz2i ; . . . ; Zij0 ¼ KZii ; . . . ; Zmj ¼ KZmi ;

Zjj0 ¼ K 2 ðZii þ zij Þ:

ð1:109Þ ð1:110Þ

Obviously, (1.109) and (1.110) will be changed into (1.93) and (1.94) when the turn ratio K ¼ 1. The situation when the transformer added is a link branch is shown in Fig. 1.23c. Assume that the current injected into the network after adding the transformer branch is a column vector I, thus the current injected into the original network 2

3 I_1 6 I_ 7 2 6 7 6 7 .. 6 7 . 6 7 6 7 6 I_i K I_ij 7 6 7 I0 ¼ 6 7 ¼ I A0M Iij ; .. 6 7 . 6 7 6 _ 7 6 Ij þ I_ij 7 6 7 6 7 .. 5 4 . I_m

ð1:111Þ

64

1 Mathematical Model and Solution of Electric Network

where A0M is a column vector. 3 0 6 .. 7 6 . 7 6 7 6 K 7 6 7 6 0 7 6 7 . . 7 A0M ¼ 6 6 . 7 6 1 7 6 7 6 0 7 6 7 6 . 7 4 .. 5 0 2

i j

:

The following steps are similar to that of a simple impedance link branches (see (1.99)–(1.108)). The only difference is to substitute the original AM for A0M . Therefore (1.101) should be changed as follows: 2

KZ1i Z1j KZ2i Z2j .. . KZii Zij .. .

3

6 7 6 7 6 7 6 7 6 7 6 7 6 7 ZL ¼ 6 7: 6 7 6 7 6 KZji Zjj 7 6 7 6 7 .. 4 5 . KZmi Zmj

ð1:112Þ

Equation (1.103) should be rewritten as K V_ i V_j ¼ K 2 zij I_ij ¼ A0T M V:

ð1:113Þ

Accordingly, (1.105) is changed as ZLL ¼ KZLi ZLj þ K 2 zij :

ð1:114Þ

After calculating ZL and ZLL , the elements of Z0N can be calculated according to (1.108). Briefly, the process of forming an impedance matrix by using the branch addition method is a process of adding branches one by one. If the configuration of a network is changed or a branch needs to be added, the impedance matrix can be modified directly according to the above formulae. For instance, if a branch zij needs to be removed, the equivalent operation is to add a branch zij into the network.

1.5 Nodal Impedance Matrix

65

[Example 1.9] Form the impedance matrix of the electric network shown in Fig. 1.10 by using the branch addition method. [Solution] For convenience of the computation, line-to-ground capacitances at both ends of the transmission lines are lumped to the corresponding node and denoted in the format of line-to-ground reactance. The equivalent circuit is shown in Fig. 1.24. According to the node ordering, we can make the sequence table of branch adding as follows.

Sequence of branches added

Terminal nodes of branch

(1)

0L 1

- j4

(2)

0L 2

- j2

(3)

1L 2

0.04 + j 0.25

(4)

0L 3

- j4

(5)

1L 3

0.1 + j 0.35

(6)

2L 3

0.08 + j 0.30

(7)

2L 4

j 0.015

(8)

3L 5

j 0.03

Impedance of branch

iL j

Then label the branch adding sequence on the figure as shown in Fig. 1.24. 1:1.05

1.05:1 0.08 + j 0.30

2 4

3

j 0.015 (7)

j 0.03

(6)

(2) −j 2

−j4

(3)

(5)

1 (1)

− j4

Fig. 1.24 Impedance matrix formed by using branch addition method

(4) (8)

5

66

1 Mathematical Model and Solution of Electric Network

The procedure for forming the impedance matrix is shown as follows: 1. Start from the grounded branch z01 to form a 1 1 matrix. Its element is j4 2. Add branch (2): z02 is a tree branch, i ¼ 0; j ¼ 2. According to (1.93) and (1.94), the new elements are Z12 ¼ Z21 ¼ Z10 ; Z22 ¼ Z00 þ Z02 : According to the physical meaning of the impedance matrix element, we have Z10 ¼ Z00 ¼ 0: Then Z12 ¼ Z21 ¼ 0; Z22 ¼ Z02 ¼ j2 and the 2 2 matrix is

1 1 2

2

− j4 − j2

3. Add branch (3): z12 is a link branch. The elements of ZL can be obtained according to (1.101) and (1.105), ZL1 ¼ Z11 Z12 ¼ j4 ZL2 ¼ Z12 Z22 ¼ j2: From (1.105) we know, ZLL ¼ ZL1 ZL2 þ z12 ¼ j4 j2 þ 0:04 þ j0:25 ¼ 0:04 j5:75: Modify the elements of the 2 2 matrix according to (1.108) ZL1 ZL1 ðj4Þ2 ¼ j4 ¼ 0:019356 j1:217526 ZLL 0:04 j5:75 ZL2 ZL1 j2 ðj4Þ ¼ Z12 ¼0 ¼ 0:096782 j1:301237 ZLL 0:04 j5:75

0 Z11 ¼ Z11 0 0 Z12 ¼ Z21

0 Z22 ¼ Z22

ZL2 ZL2 ðj2Þ2 ¼ j2 ¼ 0:004839 j1:304381: ZLL 0:04 j5:75

1.5 Nodal Impedance Matrix

67

Thus we obtain the impedance matrix constituted by branches 1, 2, and 3

1

2

1

2

0.019356

− 0.096282

− j1.1217526

− j1.391237

− 0.096282

0.004839

− j 1.391237

− j1.304381

4. Add branch (4): z03 is a grounded tree branch. The computation process is the same as that in Step 2. The augmented matrix 3 3 is

1

2

1

2

3

0.019356

−0.096282

− j1.1217526

−j 1.391237

− 0.096282

0.004839

− j 1.391237

− j 1.304381

3

−j 4

5. Add branch (5) z13 and branch (6) z23 . Because both of these are links, the matrix order is unchanged. The computation process is the same as that in Step 3. The augmented matrix 3 3 is

1

2

3

1

2

3

0.017972

− 0.005555

− 0.006862

− j0.914690

− j 1.032911

− j1.019487

− 0.005555

0.007781

− 0.010007

− j 1.0329111

− j 0.964591

− j1.037907

− 0.006862

− 0. 010007

0.026875

− j1.019487

− j1.037907

− j0.904700

6. Add branch (7): z24 is a transformer tree branch. In this network, the off normal turns ratio of the transformer is at node i. The computation cannot be performed

68

1 Mathematical Model and Solution of Electric Network

1:1.05

1 :1 1.05

2

j 0.015×(1.05)2

j 0.015

4

2

4

Fig. 1.25 Equivalent circuit of the transformer

directly using (1.109) and (1.110). We should transfer the off normal turns ratio to the other terminal node of the transformer. It is illustrated in the Fig. 1.25. Then the elements of the fourth row and column can be calculated according to (1.109) and (1.110). 1 ð0:005555 j1:032911Þ ¼ 0:005290 j0:983725 1:05 1 ¼ Z24 ¼ K 0 Z22 ¼ ð0:007781 j0:964591Þ ¼ 0:007410 j0:918658 1:05 1 ¼ Z34 ¼ K 0 Z23 ¼ ð0:010007 j1:037907Þ ¼ 0:009530 j0:988482 1:05 1 ¼ K 02 ðZ22 þ z024 Þ ¼ ð0:007781 j0:964591Þ þ j0:015 1:052 ¼ 0:007057 j0:859912:

Z41 ¼ Z14 ¼ K 0 Z21 ¼ Z42 Z43 Z44

We now have a 4 4 matrix

1

1

2

3

4

0.017972

− 0.005555

− 0.006862

− 0.005290

− j0.914690

−j

− j1.019487

− j0.983725

1.0329111 2

3

4

− 0.005555

0.007781

− 0.010007

0.007410

−j

−j

− j1.037907

− j0.918658

1.0329111

0.964591

− 0.006862

− 0.010007

0.026875

− 0.009530

− j1.019487

− j1.037907

− j0.904700

− j0.988482

− 0.005290

0.007410

− 0.009530

0.007057

− j0.983725

− j0.918658

− j0.988482

− j0.859912

Thinking and Problem Solving

69

7. Add branch (8): z35 is also a transformer tree branch. Its off normal turns ratio is also at node i. The computation process is the same as that in Step 6. The final impedance matrix is 2 6 6 6 6 6 6 6 6 Z¼6 6 6 6 6 6 6 4

0:017972 j0:914690 0:0055555 j1:032911 0:006862 j1:019487 0:005290 j0:983725 0:006535 j0:970940

0:0055555 j1:032911 0:007781 j0:964591 0:010007 j1:037907 0:007410 j0:918658 0:009530 j0:988482

0:006862 j1:019487 0:010007 j1:037907 0:026875 j0:904700 0:009530 j988482 0:025596 j0:861619

0:005290 j0:983725 0:007410 j0:918658 0:009530 j988482 0:007057 j0:859912 0:009076 j0:941412

0:006535 j0:970940 0:009530 j0:988482 0:025596 j0:861619 0:009076 j0:941412 0:024377 j0:790589

3 7 7 7 7 7 7 7 7 7: 7 7 7 7 7 7 5

Thinking and Problem Solving 1. Prove that the incidence matrix of an electrical power network is a singular matrix 2. Is the admittance matrix generally a singular matrix? In what condition can the admittance matrix be a singular matrix? 3. What simplifications can be made to the equivalent circuit of the transformer in Fig. 1.4? 4. Why is the admittance matrix including phase shifter(s) not a symmetric matrix? 5. How many elements are there in the admittance matrix of an electrical power network with N nodes and B branches? 6. What changes will occur in the admittance matrix when the turn ratio of a transformer varies? 7. What changes will occur in the admittance matrix when a line is out of service? 8. What characteristics does the electrical power network equation have? And what requirements are there for its solution algorithm? 9. Why is the method of Gauss elimination often adopted to solve network equations? 10. How is the factor table formed? Compare the features between two methods of forming the factor tables. 11. What is the key idea behind sparse technique? 12. What fields can the sparse vector method be applied to? 13. Compare the features and application areas of three kinds of node optimal ordering methods.

70

1 Mathematical Model and Solution of Electric Network

14. State the significance of self-impedance, input impedance, mutual impedance, and transfer impedance. 15. How can an admittance matrix be used to find self-impedance Zii and mutual impedance Zij ? Give a detailed program flowchart. 16. Describe the storage scheme of a sparse admittance matrix.

Chapter 2

Load Flow Analysis

2.1

Introduction

Load flow analysis is the most important and essential approach to investigating problems in power system operating and planning. Based on a specified generating state and transmission network structure, load flow analysis solves the steady operation state with node voltages and branch power flow in the power system. Load flow analysis can provide a balanced steady operation state of the power system, without considering system transient processes. Hence, the mathematic model of load flow problem is a nonlinear algebraic equation system without differential equations. Power system dynamic analysis (see Chaps. 5 and 6) investigates system stability under some given disturbances. Its mathematic model includes differential equations. It should be pointed out that dynamic analysis is based on load flow analysis and the algorithm of load flow analysis is also the base for dynamic analysis methods. Therefore, familiarity with the theory and algorithms of load flow analysis is essential to understanding the methodology of modern power system analysis. Using digital computers to calculate load flow started from the middle of the 1950s. Since then, a variety of methods has been used in load flow calculation. The development of these methods is mainly led by the basic requirements of load flow calculation, which can be summed up as: 1. The convergence properties 2. The computing efficiency and memory requirements 3. The convenience and flexibility of the implementation Mathematically, the load flow problem is a problem of solving a system of nonlinear algebraic equations. Its solution usually cannot avoid some iteration process. Thus reliable convergence becomes the prime criterion for a load flow calculation method. With the scale of power system continually expanding, the dimension of load flow equations now becomes very high (several thousands to tens of thousands). For the equations with such high dimensions, we cannot ensure that any mathematical method can converge to a correct solution. This situation requires the researchers and scholars in the power system analysis field to seek more reliable methods. X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

71

72

2 Load Flow Analysis

In the early stages of using digital computers to solve power system load flow problems, the widely used method was the Gauss–Seidel iterative method based on a nodal admittance matrix (it will be simply called the admittance method below) [4]. The principle of this method is rather simple and its memory requirement is relatively small. These properties made it suit the level of computer and power system theory at that time. However, its convergence is not satisfactory. When the system scale becomes larger, the number of iteration increases sharply, and sometimes the iteration process cannot converge. This problem led to the use of the sequential substitution method based on the nodal impedance matrix (also called the impedance method). At the beginning of the 1960s, the digital computer had developed to the second generation. The memory and computing speed of computers were improved significantly, providing suitable conditions for the application of the impedance method. As mentioned in Chap. 1, the impedance matrix is a full matrix. The impedance method requires the computer to store the impedance matrix that represents the topology and parameters of the power network. Thus it needs a great amount of computer memory. Furthermore, in each iteration, every element in the impedance matrix must be operated with, so the computing burden is very heavy. The impedance method improved convergence and solved some load flow problems that the admittance method could not solve. Therefore, the impedance method was widely applied from then on and made a great contribution to power system design, operation, and research. The main disadvantage of the impedance method is its high memory requirement and computing burden. The larger the system is, the more serious these defects are. To overcome the disadvantage, the piecewise solution method based on impedance matrix was developed [5]. This method divides a large system up into several small local systems and only the impedance matrixes of local systems and the impedances of tie lines between these local systems are to be stored in the computer. In this way, the memory requirement and computing burden are greatly alleviated. The other approach to overcoming the disadvantages of the impedance method is to apply the Newton–Raphson method (also called the Newton method) [6]. The Newton method is a typical method used to solve nonlinear equations in mathematics with very favorable convergence. As long as the sparsity of the Jacobean matrix is utilized in the iterative process, the computing efficiency of the Newton method can be greatly improved. Since the optimal order eliminating method [7] began to be employed in the middle of the 1960s, the Newton method has surpassed the impedance method in the aspects of convergence, memory demand, and computing speed. It is still the favored method, and is widely used in load flow calculation today. Since the 1970s, the load flow calculating method continues to develop in various ways. Among them the most successful is the fast decoupled method, also called the P Q decoupled method [8]. Comparing with the Newton method, this method is much simpler and more efficient algorithmically, and therefore more popular in many applications.

2.2 Formulation of Load Flow Problem

73

In the recent 20 years, research on load flow calculation is still very active. Many contributions seek to improve the convergence characteristics of the Newton method and the P Q decoupled method [9–15]. Along with the development of artificial intelligent theory, the genetic algorithm, artificial neural network algorithm, and fuzzy algorithm have also been introduced to load flow analysis [16–19]. However, until now these new models and new algorithms still cannot replace the Newton method and P Q decoupled method. Because the scales of power systems continue to expand and the requirements for online calculation become more and more urgent, the parallel computing algorithms are also studied intensively now and may become an important research field [20]. This chapter mainly discusses the currently widely used Newton method and P Q decoupled method. The degree of flexibility and convenience of load flow calculation are also very important to computer application. In practice, load flow analysis is usually part of an interactive environment, rather than a pure calculation problem. Therefore, the human–computer interface should be friendly, allowing users to monitor and control the calculation process. To obtain an ideal operation scheme, it is usually necessary to modify the original data according to the computing results. Thus, the computing method should be flexible, permitting users to readily modify and adjust their operation scheme. Input and output processes should also receive careful attention. Power system steady state analysis includes load flow analysis and static security analysis. Load flow analysis is mainly used in analyzing the normal operation state, while static security analysis is used when some elements are out of service. Its purpose is to check whether the system can operate safely, i.e., if there are equipment overloads, or some node voltages are too low or too high. In principle, static security analysis can be replaced by a series of load flow analyses. However, usually there are very many contingency states to be checked and the computation burden is quite large if a rigorous load flow calculation method is used. Hence special methods have to be developed to meet the requirement of efficient calculation. In the first part of this chapter, the models and algorithms of load flow calculation are introduced. In the second part, the problems related to static security analysis are discussed.

2.2 2.2.1

Formulation of Load Flow Problem Classification of Node Types

An electric power system is composed of generators, transformers, transmission lines and loads, etc. A simple power system is illustrated in Fig. 2.1. In the process of power system analysis, the static components, such as transformers, transmission lines, shunt capacitors and reactors, are represented by their equivalent circuits

74

2 Load Flow Analysis PH2 + jQH2

PF 2 + jQF 2

6

PF1 + jQF1

5

3

4

2

1

PH1 + jQH1

(a) −PH1 + jQH1

1

V1

•

I1

•

•

•

2

0

V2

I2

•

3

0

V3

I3

V4

•

I4

•

I5

PF1 + jQF1

4

−PH2 − jQH2

5

PF2 − jQF 2

6

•

Linear Network

•

•

V5 •

•

V6

I6

(It can be described by admittance matrix or impedance matrix)

(b) Fig. 2.1 Simple power system

consisting of R, L, C elements. Therefore, the network formed by these static components can be considered as a linear network and represented by the corresponding admittance matrix or impedance matrix. In load flow calculation, the generators and loads are treated as nonlinear components. They cannot be embodied in the linear network, see Fig. 2.1b. The connecting nodes with zero injected power also represent boundary conditions on the network. In Fig. 2.1b, the relationship between node current and voltage in the linear network can be described by the following node equation: I ¼ YV

ð2:1Þ

or I_i ¼

n X

Yij V_ j

ði ¼ 1; 2; . . . ; nÞ

ð2:2Þ

j¼1

where I_i and V_ j are the injected current at bus i and voltage at bus j, respectively, Yij is an element of the admittance matrix, n is the total number of nodes in the system.

2.2 Formulation of Load Flow Problem

75

To solve the load flow equation, the relation of node power with current should be used Pi jQi I_i ¼ V^i

ði ¼ 1; 2; . . . ; nÞ

ð2:3Þ

where Pi , Qi are the injected active and reactive power at node i, respectively. If node i is a load node, then Pi and Qi should take negative values. In (2.3), V^i is the conjugate of the voltage vector at node i. Substituting (2.3) to (2.2), we have, n Pi jQi X ¼ Yij V_ j V^i j¼1

ði ¼ 1; 2; . . . ; nÞ

n Pi þ jQi X ¼ Y^ij V^j V_ i

ði ¼ 1; 2; . . . ; nÞ

or ð2:4Þ

j¼1

There are n nonlinear complex equations in (2.4). They are the principal equations in load flow calculation. Based on different methods to solve (2.4), various load flow algorithms can be formed. In the power system load flow problem, the variables are nodal complex voltages and complex powers: V, y, P, Q. If there are n nodes in a power system, then the total number of variables is 4 n. As mentioned above, there are n complex equations or 2n real equations defined in principal by (2.4), thus only 2n variables can be solved from these equations, while the other 2n variables should be specified as original data. Usually, two variables at each node are assumed known, while the other two variables are treated as state variables to be resolved. According to the original data, the nodes in power systems can be classified into three types: 1. PQ Nodes: For PQ nodes, the active and reactive power (P; Q) are specified as known parameters, and the complex voltage (V; y) is to be resolved. Usually, substation nodes are taken as PQ nodes where the load powers are given constants. When output P and Q are fixed in some power plants, these nodes can also be taken as PQ node. Most nodes in power systems belong to the PQ type in load flow calculation. 2. PV Nodes: For PV nodes, active power P and voltage magnitude V are specified as known variables, while reactive power Q and voltage angle y are to be resolved. Usually, PV nodes should have some controllable reactive power resources and can thus maintain node voltage magnitude at a desirable value. Generally speaking, the buses of power plants can be taken as PV nodes, because voltages at these buses can be controlled with reactive power capacity of their generators. Some substations can also be considered as PV nodes when they have enough reactive power compensation devices to control the voltage.

76

2 Load Flow Analysis

3. Slack Node: In load flow studies, there should be one and only one slack node specified in the power system, which is specified by a voltage, constant in magnitude and phase angle. Therefore, V and y are given as known variables at the slack node, while the active power P and reactive power Q are the variables to be solved. The effective generator at this node supplies the losses to the network. This is necessary because the magnitude of losses will not be known until the calculation of currents is complete, and this cannot be achieved unless one node has no power constraint and can feed the required losses into the system. The location of the slack node can influence the complexity of the calculations; the node most closely approaching a large AGC power station should be used. We will employ different methods to treat the above three kinds of nodes in power flow calculations.

2.2.2

Node Power Equations

As described above, power system load flow calculations can be roughly considered as the problem of solving the node voltage phasor for each node when the injecting complex power is specified. If the complex power can be represented by equations of complex voltages, then a nonlinear equation solving method, such as the Newton–Raphson method, can be used to solve the node voltage phasors. In this section, node power equations are deduced first. The complex node voltage has two representation forms – the polar form and the rectangular form. Accordingly, the node power equations also have two forms. From (2.4), the node power equations can be expressed as Pi þ jQi ¼ V_ i

X

Y^ij V^j

ði ¼ 1; 2; . . . ; nÞ

ð2:5Þ

j2i

where j 2 i means the node j should be directly connected with node i, including j ¼ i. As discussed in Chap.1, the admittance matrix is a sparse matrix, and the terms in S are correspondingly few. If the voltage vector of (2.5) adopts polar form, V_i ¼ Vi ejyi

ð2:6Þ

where Vi ,yi are the magnitude and phase angle of voltage at node i. The elements of admittance matrix can be expressed as Yij ¼ Gij þ jBij

2.2 Formulation of Load Flow Problem

77

Hence (2.5) can be rewritten as Pi þ jQi ¼ Vi ejyi

X

ðGij jBij ÞVj ejyj

ði ¼ 1; 2; . . . ; nÞ

ð2:7Þ

j2i

Combining the exponential items of above equation and using the relationship ejy ¼ cos y þ j sin y we have, Pi þ jQi ¼ Vi

X

Vj ðGij jBij Þðcos yij þ j sin yij Þ

ði ¼ 1; 2; . . . ; nÞ

ð2:8Þ

j2i

where yij ¼ yi yj , is the voltage phase angle difference between node i and j. Dividing above equations into real and imaginary parts, Pi ¼ Vi

X j2i

Qi ¼ V i

X j2i

9 Vj ðGij cos yij þ Bij sin yij Þ > > = Vj ðGij sin yij Bij cos yij Þ > > ;

ði ¼ 1; 2; ; nÞ

ð2:9Þ

This is the polar form of the nodal power equations. It is not only very important in the Newton–Raphson calculation process, but also essential to establish the fast decoupled method. When the voltage vector is expressed in rectangular form, V_i ¼ ei þ jfi where ei ¼ Vi cos yi

fi ¼ Vi sin yi

We can obtain from (2.5), P i ¼ ei

X

ðGij ej Bij fj Þ þ fi

j2i

Qi ¼ fi

X j2i

X j2i

ðGij ej Bij fj Þ ei

X j2i

9 ðGij fj þ Bij ej Þ > > = ðGij fj þ Bij ej Þ > > ;

ði ¼ 1; 2; . . . ; nÞ

ð2:10Þ

78

2 Load Flow Analysis

Let X j2i

X j2i

9 ðGij ej Bij fj Þ ¼ ai > > =

ð2:11Þ

ðGij fj þ Bij ej Þ ¼ bi > > ;

Obviously, ai and bi are the real and imaginary parts of injected current at node i and (2.10) can be simplified as, Pi ¼ ei ai þ fi bi Q i ¼ f i ai e i bi

) ði ¼ 1; 2; . . . ; nÞ

ð2:12Þ

This is the rectangular form of the nodal power equations. Both (2.9) and (2.10) are the simultaneous nonlinear equations of node voltage phasors. They are usually expressed as the following forms as mathematical models of the load flow problem: X

9 Vj ðGij cos yij þ Bij sin yij Þ ¼ 0 > > = j2i X DQi ¼ Qis Vi Vj ðGij sin yij Bij cos yij Þ ¼ 0 > > ; DPi ¼ Pis Vi

ði ¼ 1; 2; . . . ; nÞ ð2:13Þ

j2i

and X

X

9 ðGij fj þ Bij ej Þ ¼ 0 > > = j2i j2i X X DQi ¼ Qis fi ðGij ej Bij fj Þ þ ei ðGij fj þ Bij ej Þ ¼ 0 > > ; DPi ¼ Pis ei

j2i

ðGij ej Bij fj Þ fi

ð2:14Þ

j2i

ði ¼ 1; 2; . . . ; nÞ where Pis , Qis are the specified active and reactive powers at node i. Based on the above two simultaneous equations, the load flow problem can be roughly summarized as: for specified Pis ,Qis ði ¼ 1; 2; . . . ; nÞ, find voltage vector Vi , yi or ei , fi ði ¼ 1; 2; . . . ; nÞ, such that the magnitudes of the power errors DPi , DQi , ði ¼ 1; 2; . . . ; nÞ of (2.13) or (2.14) are less then an acceptable tolerance.

2.3 2.3.1

Load Flow Solution by Newton Method Basic Concept of Newton Method

The Newton–Raphson method is an efficient algorithm to solve nonlinear equations. It transforms the procedure of solving nonlinear equations into the procedure

2.3 Load Flow Solution by Newton Method

79

of repeatedly solving linear equations. This sequential linearization process is the core of the Newton–Raphson method. We now introduce the Newton–Raphson method by the following nonlinear equation example, f ðxÞ ¼ 0

ð2:15Þ

Let xð0Þ be the initial guess value of the above equation solution. Assume the real solution x is close to xð0Þ , x ¼ xð0Þ Dxð0Þ

ð2:16Þ

where Dxð0Þ is a modification value of xð0Þ . The following equation holds, f ðxð0Þ Dxð0Þ Þ ¼ 0

ð2:17Þ

When Dxð0Þ is known, the solution x can be calculated by (2.16). Expanding this function in a Taylor series expansion about point xð0Þ yields: ð0Þ 2

f ðxð0Þ Dxð0Þ Þ ¼ f ðxð0Þ Þ f 0 ðxð0Þ ÞDxð0Þ þ f 00 ðxð0Þ Þ ðDx2! ð0Þ n

þ ð1Þn f ðnÞ ðxð0Þ Þ ðDxn!

Þ

þ ¼ 0

Þ

ð2:18Þ

where f 0 ðxð0Þ Þ,. . ., f ðnÞ ðxð0Þ Þ are the different order partial derivatives of f ðxÞ at xð0Þ . If the initial guess is sufficiently close to the actual solution, the higher order terms of the Taylor series expansion could be neglected. Equation (2.18) becomes, f ðxð0Þ Þ f 0 ðxð0Þ ÞDxð0Þ ¼ 0

ð2:19Þ

This is a linear equation in Dxð0Þ and can be easily solved. Using Dxð0Þ to modify xð0Þ , we can get xð1Þ : xð1Þ ¼ xð0Þ Dxð0Þ

ð2:20Þ

xð1Þ may be more close to the actual solution. Then using xð1Þ as the new guess value, we solve the following equation similar to (2.19), f ðxð1Þ Þ f 0 ðxð1Þ ÞDxð1Þ ¼ 0 Thus xð2Þ is obtained: xð2Þ ¼ xð1Þ Dxð1Þ

ð2:21Þ

Repeating this procedure, we establish the correction equation in the tth iteration: f ðxðtÞ Þ f 0 ðxðtÞ ÞDxðtÞ ¼ 0

ð2:22Þ

80

2 Load Flow Analysis y

Fig. 2.2 Geometric interpretation of Newton method

y = f (x)

f (x(t))

f (x(t+1)) α(t)

x

0

x(t+1)

x(t)

x

Δx(t+1) Δx(t)

or f ðxðtÞ Þ ¼ f 0 ðxðtÞ ÞDxðtÞ

ð2:23Þ

The left hand of the above equation can be considered as the error produced by approximate solution xðtÞ . When f ðxðtÞ Þ ! 0, (2.15) is satisfied, so xðtÞ is the solution of the equation. In (2.22), f 0 ðxðtÞ Þ is the first-order partial derivative of function f ðxÞ at point xðtÞ . It is also the slope of the curve at point xðtÞ , as shown in Fig. 2.2, tan aðtÞ ¼ f 0 ðxðtÞ Þ

ð2:24Þ

The correction value DxðtÞ is determined by the intersection of the tangent line at xðtÞ with the abscissa. We can comprehend the iterative process more intuitively from Fig. 2.2. Now we will extend the Newton method to simultaneous nonlinear equations. Assume the nonlinear equations with variables x1 ; x2 ; . . . ; xn ; 9 f1 ðx1 ; x2 ; . . . ; xn Þ ¼ 0 > > > > = f2 ðx1 ; x2 ; . . . ; xn Þ ¼ 0 > .. > > . > > > ; fn ðx1 ; x2 ; . . . ; xn Þ ¼ 0

ð2:25Þ

ð0Þ

ð0Þ

ð0Þ

ð0Þ

Specify the initial guess values of all variables x1 ; x2 ; . . . ; xn , let Dx1 ; ð0Þ

ð0Þ

Dx2 ; . . . ; Dxn be the correction values to satisfy the following equations, 9 ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ f1 ðx1 Dx1 ; x2 Dx2 ; . . . ; xð0Þ > n Dxn Þ ¼ 0 > > > > ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ = f ðx Dx ; x Dx ; . . . ; x Dx Þ ¼ 0 > 2

1

1

2

2

n

n

.. . ð0Þ

ð0Þ

ð0Þ

ð0Þ

ð0Þ fn ðx1 Dx1 ; x2 Dx2 ; . . . ; xð0Þ n Dxn Þ ¼ 0

> > > > > > ;

ð2:26Þ

2.3 Load Flow Solution by Newton Method

81

Expanding the above n equations via the multivariate Taylor series and neglecting the higher order terms, we have the following equations, 9 @f1 @f1 @f1 ð0Þ ð0Þ > ð0Þ > Dx ¼ 0 0 Dx1 þ 0 Dx2 þ; . . . ; þ > 0 n > @x1 @x2 @xn > > > > > @f @f @f > 2 2 2 ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ = f2 ðx1 ; x2 ; . . . ; xn Þ ¼ 0> 0 Dx1 þ 0 Dx2 þ; . . . ; þ 0 Dxn @x1 @x2 @xn ð2:27Þ > .. > > > . > > > > > @fn @fn @fn > ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ > ; fn ðx1 ; x2 ; . . . ; xn Þ Dx ¼ 0 0 Dx1 þ 0 Dx2 þ; . . . ; þ 0 n @x1 @x2 @xn ð0Þ

ð0Þ

f1 ðx1 ; x2 ; . . . ; xð0Þ n Þ

here

@fi @xj j0

is the partial derivative of function fi ðx1 ; x2 ; . . . ; xn Þ over independent ð0Þ

ð0Þ

ð0Þ

variable xj at point ðx1 ; x2 ; . . . ; xn Þ. Rewrite the above equation in matrix form, 3 2 3 ð0Þ Dx1 7 76 6 7 6 @f2 @f2 . . . @f2 76 ð0Þ 7 6 @x1 0 @x2 0 @xn 0 76 Dx2 7 6 7¼6 76 7 6. 7 6 76 . 7 6.. 7 6 .. 76 .. 7 4 5 6. 74 5 4 5 ð0Þ ð0Þ ð0Þ ð0Þ @fn @fn @fn fn ðx1 ; x2 ; . . . ; xn Þ Dx n @x1 0 @x2 0 . . . @xn 0 2

2

@f1 @f1 @f 0 0 . . . @x1n 0 6 @x1 @x2

3

ð0Þ ð0Þ ð0Þ f1 ðx1 ; x2 ; . . . ; xn Þ 6 7 6 ð0Þ ð0Þ ð0Þ 7 6 f2 ðx1 ; x2 ; . . . ; xn Þ 7

ð2:28Þ

ð0Þ

ð0Þ

This is a set of simultaneous linear equations in the variables Dx1 ; Dx2 ; . . . ; ð0Þ Dxn ;

usually called the correction equations of the Newton–Raphson method. ð0Þ

ð0Þ

ð0Þ

After solving Dx1 ; Dx2 ; . . . ; Dxn ; we can get, 9 ð1Þ ð0Þ ð0Þ > x1 ¼ x1 Dx1 > > > ð1Þ ð0Þ ð0Þ > > x2 ¼ x2 Dx2 = .. .. .. > > > . . . > > > ð1Þ ð0Þ ð0Þ ; xn ¼ xn Dxn ð1Þ

ð1Þ

ð2:29Þ

ð1Þ

x1 ; x2 ; . . . ; xn will approach the actual solution more closely. The updated values are used as the new guess to solve the correction equation (2.28) and to further correct the variables. In this way the iterative process of the Newton– Raphson method is formed. Generally, the correction equation in the tth iteration can be written as, 2

ðtÞ

ðtÞ

3

2 @f

32

3 ðtÞ Dx 1 7 6 7 6 7 @f2 @f2 @f2 76 ðtÞ 7 6 f ðxðtÞ ; xðtÞ ; . . . ; xðtÞ 7 6 . . . j j j 6 76 t t t Þ n 2 @x @x @x 6 7 6 1 6 Dx2 7 2 n 1 2 7 6 7¼6 6 7 6 .. 7 6 .. 7 .. .. 7 .. .. .. 76 6 . 7 6 7 . . . 74 . . . 5 4 5 6 4 5 ðtÞ ðtÞ ðtÞ @fn @fn @fn ðtÞ fn ðx1 ; x2 ; . . . ; xn Þ Dxn @x1 jt @x2 jt . . . @xn jt ðtÞ

f1 ðx1 ; x2 ; . . . ; xn Þ

@f1 @f1 @x1 jt @x2 jt . . . @xn jt 1

ð2:30Þ

82

2 Load Flow Analysis

or expressed in matrix form, FðXðtÞ Þ ¼ JðtÞ DXðtÞ

ð2:31Þ

where 2

ðtÞ

ðtÞ

ðtÞ

f1 ðx1 ; x2 ; . . . ; xn Þ

3

6 7 6 f2 ðxðtÞ ; xðtÞ ; . . . ; xðtÞ 7 n Þ7 6 1 2 7 FðXðtÞ Þ ¼ 6 6 7 .. 6 7 . 4 5 ðtÞ

ðtÞ

ð2:32Þ

ðtÞ

fn ðx1 ; x2 ; . . . ; xn Þ is the error vector in the tth iteration; 2 @f1

@f1 @f1 @x1 jt @x2 jt ::: @xn jt

JðtÞ

3

6 @f @f 7 6 2 jt 2 jt ::: @f2 jt 7 6 @x1 @x2 @xn 7 7 ¼6 7 6 .. 6 7 . 4 5

ð2:33Þ

@fn @fn @fn @x1 jt @x2 jt ::: @xn jt

is the Jacobian matrix of tth iteration; 2 DXðtÞ

ðtÞ

Dx1

3

6 7 6 DxðtÞ 7 6 2 7 7 ¼6 6 . 7 6 .. 7 4 5

ð2:34Þ

ðtÞ

Dxn

is the correction value vector in the tth iteration. We also have the equation similar to (2.29), Xðtþ1Þ ¼ XðtÞ DXðtÞ

ð2:35Þ

With (2.31) and (2.35) solved alternately in each iteration, Xðtþ1Þ gradually approaches the actual solution. Convergence can be evaluated by the norm of the correction value, ðtÞ DX < e1

ð2:36Þ

FðXðtÞ Þ < e2

ð2:37Þ

or by the norm of the function,

Here e1 and e2 are very small positive numbers specified beforehand.

2.3 Load Flow Solution by Newton Method

2.3.2

83

Correction Equations

In Section 2.3.1, we derived two forms of the nodal power equations. Either can be applied in the load flow calculation model. When the polar form (2.13) is used, the node voltage magnitudes and angles Vi ,yi ði ¼ 1; 2; . . . ; nÞ are the variables to be solved. For a PV node, the magnitude of the voltage is specified. At the same time, its reactive power Qis cannot be fixed beforehand as a constraint. Therefore, the reactive equations relative to PV nodes should not be considered in the iterative process. These equations will be used only to calculate the reactive power of each PV node after the iterative process is over and all node voltages have been calculated. Similarly, the voltage magnitude and angle of the slack node are specified, hence the related power equations do not appear in the iterative process. When the iteration has converged, the active and reactive power of the slack node can be calculated by using these power equations. Assume that total number of system nodes is n, the number of PV nodes is r. For convenience, let the slack bus be the last node, i.e., node n.Therefore, we have n 1 active power equations, DP1 ¼ P1s V1

X

9 > > > > > > > > > > > =

Vj ðG1j cos y1j þ B1j sin y1j Þ ¼ 0

j21

DP2 ¼ P2s V2

X

Vj ðG2j cos y2j þ B2j sin y2j Þ ¼ 0

j22

.. . DPn1 ¼ Pn1;s Vn1

> > > > > > X > > > Vj ðGn1;j cos yn1; j þ Bn1; j sin yn1; j Þ ¼ 0 > > ;

j2ðn1Þ

ð2:38Þ and n r 1 reactive power equations. DQ1 ¼ Q1s V1

X

Vj ðG1j sin y1j B1j cos y1j Þ ¼ 0

j21

DQ2 ¼ Q2s V2

X

Vj ðG2j sin y2j B2j cos y2j Þ ¼ 0

j22

.. . DQn1 ¼ Qn1;s Vn1

9 > > > > > > > > > > > =

> > > > > > X > > > Vj ðGn1; j sin yn1; j Bn1; j cos yn1; j Þ ¼ 0 > > ;

ð2:39Þ

j2ðn1Þ

In the above equations, node voltage angle yi and magnitude Vi are the variables to be resolved. Here the number of yi is n 1 and the number of Vi is n r 1. There

84

2 Load Flow Analysis

are 2n r 2 unknown variables in total and they can be solved by the above 2n r 2 equations. Expanding (2.38) and (2.39) in a Taylor series, neglecting the high-order terms, the correction equation can be written as, 2

DP1 DP2 .. .

3 2

H12 6 7 6 H11 6 7 6 H21 H22 6 7 6 6 7 6 6 7 6 6 DPn1 7 6 H 6 7 6 n1;1 Hn1;2 6 7 ¼ 6 ......... ......... 6 7 6 6 DQ1 7 6 J12 6 7 6 J 6 DQ2 7 6 11 6 7 6 J J22 6 .. 7 6 21 4 . 5 4 DQn1 Jn1;1 Jn1;2

.. . N11 N12 .. . N21 N22 .. .. Hn1;n1 .. Nn1;1 Nn1;2 ......... .. ......... ......... .. ... J1;n1 .. L11 L12 .. ... J2;n1 . L21 L22 .. ... .. ... Jn1;n1 .. Jn1;1 Jn1;2 ... ... ... ... ...

H1;n1 H2;n1

3 2 3 Dy1 N1;n1 7 6 7 Dy2 6 7 N2;n1 7 .. 7 6 7 7 6 7 . 7 6 7 6 Dyn1 7 Nn1;n1 7 7 6 7 6 7 ......... 7 7 6 ......... 7 ð2:40Þ 7 6 DV1 =V1 7 ... L1;n1 7 6 7 7 6 DV2 =V2 7 6 7 ... L2;n1 7 7 6 7 .. 5 4 5 . ... DV =V n1 n1 ... Jn1;n1 ... ... ... ... ...

The form of the voltage magnitude correction values represented here, DV1 =V1 ; DV2 =V2 ; . . . ; DVn1 =Vn1 ; allow the elements in the Jacobian matrix to have similar expressions. Taking partial derivations of (2.38), or (2.39), and noting that both Pis , Qis are constants, we can obtain the elements of the Jacobian matrix as, Hij ¼

@DPi ¼ Vi Vj ðGij sin yij Bij cos yij Þ @yj

Hii ¼

X @DPi ¼ Vi Vj ðGij sin yij Bij cos yij Þ @yi j2i

j 6¼ i

ð2:41Þ ð2:42Þ

j6¼i

or Hii ¼ Vi2 Bii þ Qi Nij ¼ Nii ¼

ð2:43Þ

@DPi Vj ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ @Vj

j 6¼ i

ð2:44Þ

X @DPi Vi ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ 2Vi2 Gii ¼ Vi2 Gii Pi ð2:45Þ @Vi j2i j6¼i

Jij ¼ Jii ¼

@DPi ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ @yj

j 6¼ i

X @DPi ¼ Vi Vj ðGij cos yij þ Bij sin yij Þ ¼ Vi2 Gii Pi @yj j2i j6¼i

ð2:46Þ ð2:47Þ

2.3 Load Flow Solution by Newton Method

85

Lij ¼

@DQi Vj ¼ Vi Vj ðGij sin yij Bij cos yij Þ @Vj

Lii ¼

X @DQi Vi ¼ Vi Vj ðGij sin yij Bij cos yij Þ þ 2Vi2 Bii ¼ Vi2 Bii Qi ð2:49Þ @Vi j2i

j 6¼ i

ð2:48Þ

j6¼i

The concise form of (2.40) is

DP H ¼ DQ J

N L

Dy DV=V

ð2:50Þ

Comparing (2.50) with (2.40), the meaning of elements is obvious. The correction equation can be rearranged into the following form for convenience, 2

3 2 H11 DP1 6 DQ1 7 6 J11 6 7 6 6 DP2 7 6 H21 6 7 6 6 DQ2 7 6 J21 6 7¼6 6 .. 7 6 .. 6 . 7 6 . 6 7 6 4 DPn1 5 4 Hn1;1 DQn1 Jn1;1

N11 L11 N21 L21 .. .

Nn1;1 Ln1;1

H12 J12 H22 J22 .. .

Hn1;2 Jn1;2

N12 L12 N22 L22 .. .

Nn1;2 Ln1;2

... ... ... ... .. .

H1;n1 J1;n1 H2;n1 J2;n1 .. .

. . . Hn1;n1 . . . Jn1;n1

3 Dy1 76 DV1 =V1 7 76 7 76 7 Dy2 76 7 76 DV2 =V2 7 76 7 ð2:51Þ 76 7 .. 76 7 . 76 7 5 4 Nn1;n1 Dyn1 5 Ln1;n1 DVn1 =Vn1 N1;n1 L1;n1 N2;n1 L2;n1 .. .

32

When the rectangular form is adopted in the load flow model, the state variables to be solved are the real and imaginary parts of voltages, i.e., e1 ; f1 ; e2; f2 ; . . . ; en ; fn . Since the voltage phasor of the slack node is specified, the number of state variables is 2ðn 1Þ. We need 2ðn 1Þ equations to solve these variables. In fact, every node has two equations except the slack bus. For PQ nodes, Pis , Qis are given, so the equations are X

X

9 ðGij fj þ Bij ej Þ ¼ 0 > > = j2i j2i X X DQi ¼ Qis fi ðGij ej Bij fj Þ þ ei ðGij fj þ Bij ej Þ ¼ 0 > > ; DPi ¼ Pis ei

ðGij ej Bij fj Þ fi

j2i

ð2:52Þ

j2i

For PV nodes, Pis , Vis are given, so the equations are DPi ¼ Pis ei

X

ðGij ej Bij fj Þ fi

j2i

DVi2

¼

Vis2

ðe2i

þ

X j2i

fi2 Þ

¼0

9 ðGij fj þ Bij ej Þ ¼ 0 = ;

ð2:53Þ

86

2 Load Flow Analysis

There are 2ðn 1Þ equations included in (2.52) and (2.53). Expanding them in a Taylor series expansion, neglecting the higher order terms, we can obtain the correction equation as follows, 2

DP1

3

2

@DP1 @e1

6 7 6 6 7 6 @DQ1 6 DQ1 7 6 @e1 6 7 6 6 7 6 @DP 6 7 6 @e1 2 6 DP2 7 6 6 7 6 @DQ2 6 7 6 @e1 6 7 6 6 6 DQ2 7 ¼ 6 6 7 6 .. 6 .. 7 6 . 6. 7 6 6 7 @DPi 6 DPi 7 6 @e1 6 7 6 6 6 7 6 6 7 0 6 DV 2 7 6 4 i 5 6 4 .. .. . .

@DP1 @f1

@DP1 @e2

@DP1 @f2

@DP1 @ei

@DP1 @fi

@DQ1 @f1

@DQ1 @e2

@DQ1 @f2

@DQ1 @ei

@DQ1 @fi

@DP2 @f1

@DP2 @e2

@DP2 @f2

@DP2 @ei

@DP2 @fi

@DQ2 @f1

@DQ2 @e2

@DQ2 @f2

@DQ2 @ei

@DQ2 @fi

.. .

.. .

.. .

.. .

.. .

.. .

@DPi @f1

@DPi @e2

@DPi @f2

@DPi @ei

@DPi @fi

0

0

0

@DVi2 @ei

@DVi2 @fi

.. .

.. .

.. .

.. .

.. .

.. .

32

De1

3

76 7 76 7 76 Df1 7 76 7 76 7 6 7 7 76 De2 7 76 7 6 7 7 76 7 76 Df2 7 7 6 7 ð2:54Þ .. 76 7 . 76 . 7 76 .. 7 6 7 7 76 De 7 76 i 7 76 7 6 7 7 76 7 54 Dfi 5 .. .. . .

By differentiating (2.52) and (2.53), we can obtain elements of the Jacobian matrix. The off-diagonal elements of the Jacobian matrix for j 6¼ i can be expressed as, 9 @DPi @DQi > ¼ ¼ ðGij ei þ Bij fi Þ > > > @ej @fj > > > = @DPi @DQi ¼ ¼ Bij ei Gij fi @fj @ej > > > > > @DVi2 @DVi2 > > ; ¼ ¼0 @ej @fj The diagonal elements of the Jacobian matrix for j ¼ i, X @DPi ¼ ðGij ej Bij fj Þ Gii ei Bii fi @ei j2i Using (2.11), we can rewrite the above expression as @DPi ¼ ai Gii ei Bii fi @ei and can obtain the following elements similarly,

ð2:55Þ

2.3 Load Flow Solution by Newton Method

87

9 X @DQi ¼ ðGij ej Bij fj Þ þ Gii ei þ Bii fi ¼ ai þ Gii ei þ Bii fi > > > > @fi > j2i > > > > X > @DPi > ¼ ðGij fj þ Bij ej Þ þ Bii ei Gii fi ¼ bi þ Bii ei Gii fi > > > @fi > > j2i > > = X @DQi ¼ ðGij fj þ Bij ej Þ þ Bii ei Gii fi ¼ bi þ Bii ei Gii fi > @ei > j2i > > > > 2 > @DVi > > > ¼ 2ei > > @ei > > > > 2 > @DVi > ; ¼ 2fi @fi

ð2:56Þ

The correction equations, in either polar form or rectangular form, are the basic equations that need repeatedly solving in Newton–Raphson load flow calculation. Investigating these equations, we can observe the following properties: 1. Equations (2.54) and (2.40) include 2ðn 1Þ and 2ðn 1Þ r equations respectively. 2. From the expression of the off-diagonal elements of the Jacobian matrix either in polar form or in rectangular form, i.e., (2.41), (2.44), (2.46), (2.48), and (2.55), we can see that each of them is related to only one element of the admittance matrix. Therefore, if the element Yij in the admittance matrix is zero, the corresponding element in the Jacobian matrix of the correction equation is also zero. It means the Jacobian matrix is a sparse matrix, and has the same structure as the admittance matrix. 3. From the expression of the elements of the Jacobian matrix we can see that the Jacobian matrix is not symmetrical in either coordinate form. For example, @DPi @DPj 6¼ ; @yj @yi @DPi @DPj 6¼ ; @ej @ei

@DQi @DQj 6¼ @Vj @Vi @DQi @DQj 6¼ ; etc: @fj @fi

4. The elements in the Jacobian matrix are a function of node voltage phasors. Therefore, they will vary with node voltages during the iterative process. The Jacobian matrix must not only be updated but also be triangularized in each iteration. This has a major effect on the calculation efficiency of the Newton– Raphson method. Many improvements of the Newton–Raphson method have focused on this problem. For instance, when the rectangular coordinate is adopted and the injected current (see (2.4)) is used to form the load flow equations [12], the off-diagonal elements of

88

2 Load Flow Analysis

the Jacobian matrix become constant. This property can certainly be used to improve the solution efficiency. Semlyen and de Leon [13] suggest that the Jacobian matrix elements can be updated partially to alleviate the computing burden. Both the above two forms of coordinate system are widely used in Newton– Raphson load flow algorithms. When the polar form is used, PV nodes can be conveniently treated. When the rectangular form is used, the calculation of trigonometric functions is avoided. Generally speaking, the difference is not very significant. A comparison between the two coordinate systems is carried out in [14]. The fast decoupled method is derived from the Newton–Raphson method in polar form. It will be discussed in Sect. 2.4. In the next section, we mainly introduce the Newton–Raphson method based on the correction equation of (2.54) in rectangular form.

2.3.3

Solution Process of Newton Method

In the Newton–Raphson method, the electric network is described by its admittance matrix. From (2.52), (2.53), (2.55), and (2.56) we know that all operations are relative to the admittance matrix. Therefore, forming the admittance matrix is the first step in the algorithm. The solving process of the Newton method roughly consists of the following steps. 1. Specify the initial guess values of node voltage, eð0Þ , f ð0Þ ; 2. Substituting eð0Þ , f ð0Þ into (2.52) and (2.53), obtain the left-hand term of the correction equation, DPð0Þ , DQð0Þ , and ðDV 2 Þð0Þ ; 3. Substituting eð0Þ , f ð0Þ into (2.55) and (2.56), obtain the coefficient matrix (Jacobian matrix) of the correction equation; 4. Solving (2.54), obtain the correction variables, Deð0Þ and Df ð0Þ ; 5. Modify voltages; eð1Þ ¼ eð0Þ Deð0Þ f ð1Þ ¼ f ð0Þ Df ð0Þ

) ð2:57Þ

6. Substituting eð1Þ and f ð1Þ into (2.52) and (2.53), obtain DPð1Þ , DQð1Þ , and ðDV 2 Þð1Þ ; 7. Check whether the iteration has converged. When it has converged, calculate branch load flow and output the results; otherwise take eð1Þ and f ð1Þ as the new guess value, return to step (3) and start the next iteration. The main flowchart of the Newton–Raphson method is shown in Fig. 2.3. The above steps introduce the main principles of the solution process. There are still many details to be clarified. As mentioned above, the solution procedure of the

2.3 Load Flow Solution by Newton Method

89

Newton–Raphson method is essentially the process of iteratively forming and solving the correction equations. Dealing with the correction equation has a crucial influence over the memory requirement and computing burden. This problem will be presented in the next section. First, we discuss some other important issues. The convergence characteristic of the Newton–Raphson method is excellent. Generally, it can converge in 6–7 iterations, and the number of iteration does not depend on the scale of the power system. Theoretically speaking, the Newton– Raphson method has a quadratic convergence characteristic if the initial guess values are close to the solution. If the initial guess values are not good enough, the iterative process may not converge or may converge to a solution at which the power system cannot operate. This property stems from the Newton method itself. As described above, the substance of the Newton method is sequential linearization of nonlinear equations. It is established on the assumption that De and Df are very small so that their high-order terms can be neglected. Therefore, a good initial guess value is crucial because the Newton method is very sensitive to it. Under normal operation states of power systems, the node voltage magnitudes are usually close to their nominal voltages, and the phase angle differences between the nodes of a branch are not very large. Therefore, a ‘‘flat start’’ initial guess value, i.e., ð0Þ

ei

¼ 1:0

ð0Þ

fi

¼ 0:0

ði ¼ 1; 2; . . . ; nÞ

ð2:58Þ

can give satisfactory results. In Fig. 2.3, the convergence condition is ðtÞ DP ; DQðtÞ < e

ð2:59Þ

where DPðtÞ ; DQðtÞ is a norm representing the maximal modulus elements in vectors DPðtÞ ; DQðtÞ . This convergence criterion is very intuitive, and can be used to directly control the power errors. When the calculation is based on the per unit system, we can set e ¼ 104 or 103 . If the base value is 100 MVA, the maximum error corresponds to 0.01 MVA or 0.1 MVA. From Fig. 2.3 we know that in the Newton–Raphson load flow calculation, the Jacobian matrix must be formed and triangularized in each iteration. Hence the computing burden in each iteration is quite heavy. From the expressions of Jacobian elements one can see that in the iteration procedure, especially when it is near convergence, the change of the elements caused by voltage variation is not significant (see Example 2.1). Therefore, to decrease the computing effort, once a Jacobian matrix is formed, it could be used in several successive iterations.

2.3.4

Solution of Correction Equations

The Newton–Raphson method, with Gauss elimination solving the correction equation, has been used in load flow calculation since the 1950s.

90

2 Load Flow Analysis Input data Form admittance matrix

Give voltage initial value e(0) and f (0) t=0 Calculate ΔP (t), ΔQ(t) and ΔV 2 (t) according to (2.52) and (2.53)

Is convergent?

Yes

Output results

No Solve the elements of Jacobian matrix according to (2.55) and (2.56)

Solve modified equation (2.54) to obtain Δe(t) and Δf (t) Modify voltage on each node according to (2.57)

t = t+1

Fig. 2.3 Flowchart of Newton method

In the 1960s, the sparsity of the correction equation was fully investigated and employed in the iteration procedure. In this way, the storage and operation for zero elements in the Jacobian are avoided. When the technology of optimal node ordering is adopted, it can minimize the number of the fill-in nonzero elements in factorizing the Jacobian of the correction equation. This greatly reduces memory and computing requirements to almost proportional to the node number of the power system. Based on this sparsity technology, the Newton–Raphson method has become one of the most popular methods in power system load flow calculation [7]. With a simple system as shown in Fig. 2.4, we now illustrate some algorithmic tricks in solving the correction equation of the Newton–Raphson method. In Fig. 2.4, both node 3 and node 6 are generator nodes. We set node 3 as a PV node while node 6 the slack node; other nodes are all PQ nodes. The structure of the network admittance matrix is shown in Fig. 2.5. The correction equation is given as (2.60). It does not include the equation related to node 6, the slack node.

2.3 Load Flow Solution by Newton Method Fig. 2.4 Example of simple system

91

3

2 1

4

5

6

Y11 Y12 Y13 Y14

Fig. 2.5 Structure of admittance matrix

Y21 Y22

Y26

Y31

Y33 Y34

Y41

Y43 Y44 Y45 Y54 Y55 Y56 Y62

2

3 2 DP1 H11 6 DQ1 7 6 J11 6 7 6 6 DP2 7 6 H21 6 7 6 6 DQ2 7 6 J21 6 7 6 6 DP3 7 6 H31 6 27¼6 6 DV 7 6 0 6 37 6 6 DP4 7 6 H41 6 7 6 6 DQ4 7 6 J41 6 7 6 4 DP5 5 4 DQ5

N11 L11 N21 L21 N31 0 N41 L41

H12 J12 H22 J22

N12 H13 L12 J13 N22 L22 H33 R33 H43 J43

32

N13 H14 N14 L13 J14 L14 N33 H34 S33 0 N43 H44 L43 J44 H54 J54

N34 0 N44 L44 N54 L54

Y65 Y66

H45 J45 H55 J55

3 De1 76 Df1 7 76 7 76 De2 7 76 7 76 Df2 7 76 7 76 De3 7 76 7 76 Df3 7 ð2:60Þ 76 7 6 7 N45 7 76 De4 7 7 6 L45 76 Df4 7 7 N55 54 De5 5 L55 Df5

where the constant terms DPi , DQi can be obtained by (2.52), X

X

9 ðGij fj þ Bij ej Þ > > = j2i j2i X X DQi ¼ Qis fi ðGij ej Bij fj Þ þ ei ðGij fj þ Bij ej Þ > > ; DPi ¼ Pis ei

j2i

ðGij ej Bij fj Þ fi

j2i

or can be written as DPi ¼ Pis ðei ai þ fi bi Þ DQi ¼ Qis ðfi ai ei bi Þ

) ð2:61Þ

92

2 Load Flow Analysis

From (2.56) we know the diagonal elements of the Jacobian are 9 @DPi > Hii ¼ ¼ ai ðGii ei þ Bii fi Þ > > > @ei > > > > @DPi > > Nii ¼ ¼ bi þ ðBii ei Gii fi Þ > = @fi > @DQi > Jii ¼ ¼ bi þ ðBii ei Gii fi Þ > > > @ei > > > > > @DQi ; Lii ¼ ¼ ai þ ðGii ei þ Bii fi Þ > @fi

ð2:62Þ

Both (2.61) and (2.62) include components of the injected current at node i, ai and bi . To calculate DPi , DQi , and the diagonal elements of Jacobian Hii , Nii , Jii , Lii , we must first compute ai and bi . From (2.11) we can see, the injected current components ai and bi at node i only depends on the i th row elements of the admittance matrix and voltage components of corresponding nodes. Therefore, ai and bi can be accumulated by sequentially taking the two terms and performing multiplication plus operation. After ai , bi are known, DPi and DQi can be easily obtained according to (2.61). The nondiagonal elements of the Jacobian in (2.60) can be expressed by: 9 @DPi > > ¼ ðGij ei þ Bij fi Þ > > @ej > > > > > @DPi > > Nij ¼ ¼ Bij ei Gij fi > = @fj > @DQi > Jij ¼ ¼ Bij ei Gij fi ¼ Nij > > > @ej > > > > > @DQi > Lij ¼ ¼ Gij ei þ Bij fi ¼ Hij > ; @fj Hij ¼

ð2:63Þ

Obviously, the off-diagonal elements are only related to the corresponding admittance elements and voltage components. From (2.62), the ith diagonal element consists of, besides the injecting current components at node i(ai and bi ), only the arithmetic operation results of the diagonal elements of admittance matrix Gii þ jBii and voltage components ei þ jfi . In brief, the whole correction equation can be formed by sequentially taking and arithmetically operating the elements of the admittance matrix and corresponding voltage components. If node i is PV node, the equation of DQi should be replaced by the equation of DVi2 . The constant term DVi2 on the left hand and elements Rii and Sii of the Jacobian can be easily obtained from (2.53) and (2.56), 9 @DVi2 > ¼ 2ei > = @ei > @DVi2 ; Sii ¼ ¼ 2fi > @fi Rii ¼

ð2:64Þ

2.3 Load Flow Solution by Newton Method

93

Forming the correction equation is a very important step in the Newton–Raphson method which remarkably affects the efficiency of the whole algorithm. Therefore, we should investigate the above equations carefully in coding the program. When Gauss elimination is used to solve the correction equation, we usually eliminate the correction equation row by row. The augmented matrix corresponding to (2.60) is 2

H11 6 J11 6 6 H21 6 6 J21 6 6 H31 6 6 0 6 6 H41 6 6 J41 6 4

N11 L11 N21 L21 N31 0 N41 L41

H12 J12 H22 J22

N12 L12 N22 L22

H13 J13

N13 L13

H14 J14

N14 L14

H33 R33 H43 J43

N33 S33 N43 L43

H34 0 H44 J44 H54 J54

N34 0 N44 L44 N54 L54

H45 J45 H55 J55

N45 L45 N55 L55

DP1 DQ1 DP2 DQ2 DP3 DV32 DP4 DQ4 DP5 DQ5

3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

After the equations related to node 1 and 2 are eliminated, the augmented matrix is converted as shown in Fig. 2.6. This figure tell us when the equations related to node 2 are eliminated (row 3 and row 4), all operations are independent of equations related to node 3, 4, . . ., N. Therefore, in the eliminating procedure, we can eliminate the rows related to a node immediately after forming them. 00 00 In Fig. 2.6, elements such as H23 ; N23 ; . . . ; L0024 , etc. are fill-in nonzero elements created in the elimination process. To decrease the number of injected elements, we should optimize the node number ordering before load flow calculation (see Section 1.3.5). The element with superscript (00 ) represents that it has been manipulated. We need not save memory for the fill-in element in advance using this elimination procedure and thus the algorithm is simplified. When the whole elimination procedure finished, the augmented matrix of correction equation becomes, 1

N11′ 1

H12′ J12′ 1

N12′ L12′ ′ N22 1

H31

N31

H41 J41

H13′ J13′ ′′ H23 ′′ J23

N13′ H14′ L13′ J14′ ′′ H24 ′′ N23 ′′ J24 ′′ L23

N14′ L14′ ′′ N 24 ′′ L24

ΔP1′ ΔQ1′ ΔP2′ ΔQ2′ ΔP3 ΔV32 N 45 ΔP4 L45 ΔQ4 N55 ΔP5 L55 ΔQ5

H34

N34

N41

H33 N33 R 33 S33 H43 N43

H44

N44

H45

L41

J43

J44 H54 J54

L44 N54 L54

J45 H55 J55

Fig. 2.6 Diagram of eliminating row by row

L43

94

2 Load Flow Analysis

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4

1

0 N11 1

0 H12 0 J12 1

0 N12 0 L12 0 N22 1

0 H13 0 J13 00 H23 00 J23 1

0 N13 0 L13 00 N23 00 L23 0 N33 1

0 H14 0 J14 00 H24 00 J24 0 H34 00 J34 1

0 N14 0 L14 00 N24 00 L24 0 N34 00 L34 0 N44 1

0 H45 0 J45 1

3 DP01 DQ01 7 7 DP02 7 7 DQ02 7 7 DP03 7 0 7 DV32 7 7 0 DP0 7 N45 4 7 L045 DQ04 7 7 0 DP0 5 N55 5 1 DQ05

Finally, using a normal backward substitution, one can get De1 ; Df1 ; . . . ; De5 ; Df5 from DP01 ; DQ01 ; . . . ; DQ05 . Following to the above discussion, we can summarize the algorithm via flowchart shown in Fig. 2.7, where R represents the slack node. The correction equation can be solved by the common Gauss elimination method. The above procedure adopts the strategy of eliminating the rows related to a node immediately after forming them. At the same time, the corresponding constant terms of the correction equation are also accumulated and eliminated. Thus the operation count per iteration is significantly reduced. [Example 2.1] Calculate the load flow of the power system shown in Fig. 2.8. [Solution] The load flow is calculated according to the procedures of the flowchart. The first step includes forming the admittance matrix and specifying the initial voltage values. From Example 1.1 we know the admittance matrix of this system is 2

1:37874 6 j6:29166 6 6 0:62402 6 6 þj3:90015 6 6 0:75471 Y¼6 6 þj2:64150 6 6 6 6 6 4

0:62402 þj3:90015 1:45390 j66:98082 0:82987 þj3:11203 0:00000 þj63:49206

0:75471 þj2:64150 0:82987 0:00000 þj3:11203 þj63:49206 1:58459 0:00000 j35:73786 þj31:74603 0:00000 j66:66667 0:00000 0:00000 þj31:74603 j33:33333

3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

The initial values of node voltages are given in Table 2.1. According to (2.52) and (2.53), we can establish the expression of the constant terms (mismatch terms) of the correction equations as

2.3 Load Flow Solution by Newton Method

95

Input Optimize node number Form admittance matrix Give initial value, and iterate by using successive iteration method t=1

i=1

t=t+1

>

i>n

>

Substituted backward and modify voltage

i=R

Form two-row equation relative to node i

Is convergent?

Eliminate the (2i 1)th and (2i) th equations by using the 1st to the 2(i 1)th equations

Output

No

Yes

i = i+1

Fig. 2.7 Flowchart of Newton Method 4

1:1.05

0.08+j0.30

2

j0.015

3

1.05:1 j0.25

j0.25

j0.03

P4 = 5 V4 =1.05

0.1+j0.35 0.04+j0.25

2+j1 j0.25

j0.25

3.7+j1.3

1

1.6+j0.8

Fig. 2.8 Simple power system

5

V5=1.05 θ5 = 0

96

2 Load Flow Analysis Table 2.1 Voltage initial values Node 1 2 1.00000 1.00000 eð0Þ 0.00000 0.00000 f ð0Þ

3 1.00000 0.00000

4 1.05000 0.00000

5 1.05000 0.00000

DP1 ¼ P1s e1 ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ f1 ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ DQ1 ¼ Q1s f1 ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ þ e1 ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ DP4 ¼ P4s e4 ½ðG42 e2 B42 f2 Þ þ ðG44 e4 B44 f4 Þ f4 ½ðG42 f2 þ B42 e2 Þ þ ðG44 f4 þ B44 e4 Þ 2 DV42 ¼ V4s ðe24 þ f42 Þ

Using (2.55) and (2.56), we can obtain the expressions of Jacobian matrix elements: @DP1 @e1 @DP1 @f1 @DP1 @e2 @DP1 @e3 @DQ1 @e1 @DQ1 @f1 @DQ1 @e2 @DQ1 @e3 @DP4 @e4 @DP4 @f4

¼ ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ G11 e1 B11 f1 ¼ ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ þ B11 e1 G11 f1 ¼ ðG12 e1 þ B12 f1 Þ; ¼ ðG13 e1 þ B13 f1 Þ;

@DP1 ¼ B12 e1 G12 f1 @f2 @DP1 ¼ B13 e1 G13 f1 @f3

¼ ½ðG11 f1 þ B11 e1 Þ þ ðG12 f2 þ B12 e2 Þ þ ðG13 f3 þ B13 e3 Þ þ B11 e1 G11 f1 ¼ ½ðG11 e1 B11 f1 Þ þ ðG12 e2 B12 f2 Þ þ ðG13 e3 B13 f3 Þ þ G11 e1 þ B11 f1 @DP1 ; @f2 @DP1 ¼ ; @f3 ¼

@DQ1 @DP1 ¼ @f2 @e2 @DQ1 @DP1 ¼ @f3 @e3

¼ ½ðG42 e2 B42 f2 Þ þ ðG44 e4 B44 f4 Þ G44 e4 B44 f4 ¼ ½ðG42 f2 þ B42 e2 Þ þ ðG44 f4 þ B44 e4 Þ þ B44 e4 G44 f4

2.3 Load Flow Solution by Newton Method

97

@DV42 ¼ 2e4 @e4 @DV42 ¼ 2f4 @f4 Thus according to (2.60), the correction equation of the first iteration can be written as 2

1:37874 6 6:04166 6 6 0:62402 6 6 3:90015 6 6 0:75471 6 6 2:64150 6 4 2

6:54166 1:37874 3:90015 0:62402 2:64150 0:75471

3 1:60000 6 0:55000 7 6 7 6 2:00000 7 6 7 6 5:69799 7 7 ¼6 6 3:70000 7 6 7 6 2:04901 7 6 7 4 5:00000 5 0:00000

0:62402 3:90015 1:45390 60:28283 0:82897 3:11203 0:00000 0:00000

3:90015 0:62402 73:67881 1:45390 3:11203 0:82897 66:66666 0:00000

0:75471 2:64150 0:82897 3:11203 1:58459 32:38884

2:64150 0:75471 3:11203 0:00000 0:82897 63:49206 39:98688 1:58459 0:00000 2:10000

32

3 De1 76 Df1 7 76 7 6 7 63:49206 7 76 De2 7 6 Df2 7 0:00000 7 76 7 76 De3 7 76 7 76 Df3 7 76 7 63:49206 54 De4 5 0:00000 Df4

the above equation, the elements in italic have maximal absolute value in each row of the Jacobian matrix. Obviously, if elements are arranged this way, the maximal elements do not appear at the diagonal positions. It should be noted that this situation is not accidental. From the above equation @DQi i we can conclude that the maximal element of each row is @DP @fi or @ei . This is because the active power is mainly related to the vertical component of voltage while the reactive power is mainly related to the horizontal component of voltage in high voltage power systems. To reduce the rounding error of the calculations, the maximal elements should be located in diagonal positions. There are two methods to satisfy this requirement: the first is to exchange positions of the equations relative to DQ and DP, i.e., to exchange odd numbered rows with even numbered rows; the second method is to exchange the variables De and Df , i.e., to exchange odd numbered columns with even numbered columns of the Jacobian matrix. We now introduce the first approach. Thus the above equation will be rearranged as,

98

2 Load Flow Analysis

2

6:04166 6 1:37874 6 6 3:90015 6 6 0:62402 6 6 2:64150 6 6 0:75471 6 4 2

1:37874 6:54166 0:62402 3:90015 0:75471 2:64150

3 0:55000 6 1:60000 7 6 7 6 5:69799 7 6 7 6 2:00000 7 7 ¼6 6 2:04901 7 6 7 6 3:70000 7 6 7 4 0:00000 5 5:00000

3:90015 0:62402 60:28283 1:45390 3:11203 0:82897

0:62402 3:90015 1:45390 73:67881 0:82897 3:11203

0:00000

66:66666

2:64150 0:75471 3:11203 0:82897 32:38884 1:58459

32 3 0:75471 De1 76 Df1 7 2:64150 76 7 6 7 0:82897 63:49206 0:00000 7 76 De2 7 6 Df2 7 3:11203 0:00000 63:49206 7 76 7 76 De3 7 1:58459 76 7 7 6 7 39:98688 76 Df3 7 2:10000 0:00000 54 De4 5 0:00000 63:49206 Df4

We can see the maximal element of each row appears in the diagonal position except for row 8. As described in Section 2.3.4, the iteration procedure adopts the strategy of immediately eliminating the rows related to a node after forming them (see Fig. 2.7). The equations related to node 1 are formed as 2

3 .. 4 6:04166 1:37874 3:90015 0:62402 2:64150 0:75471 0 0 . 0:55000 5 . 1:37874 6:54166 0:62402 3:90015 0:75471 2:64150 0 0 .. 1:60000 After the elimination operation is executed, the first and second row of the upper triangular matrix can be obtained: 2

3 .. 4 1:00000 0:22820 0:64554 0:10328 0:43721 0:12491 0 0 . 0:09103 5 . 1:00000 0:03879 0:58961 0:02215 0:41038 0 0 .. 0:21505 Then we establish the equations related to node 2, the corresponding augmented matrix is

3:90015 0:62402 60:28283 1:45390 3:11203 0:82987 63:49206 0:62402 3:90015 1:45390 73:67881 0:82987 3:11203 0:0

3 .. . 5:69799 5 . 63:49206 .. 2:0 0:0

Executing the elimination operation, the third and forth rows of the upper triangular matrix become:

2.3 Load Flow Solution by Newton Method

99

2

3 .. 4 1:00000 0:02090 0:08348 0:02090 1:09894 0:00000 . 0:09184 5 . 1:00000 0:01528 0:06609 0:01859 0:88943 .. 0:04253 Continuing this procedure until the eliminating operation procedure is finished, we have the upper triangular matrix: 2

.. . .. . .. . .. . .. . .. . .. . . 1:00000 ..

6 1:00000 0:22820 0:64554 0:10328 0:43721 0:12491 6 6 1:00000 0:03879 0:58961 0:02215 0:41038 6 6 1:00000 0:02090 0:08348 0:02090 1:09894 0:00000 6 6 6 1:00000 0:01528 0:06609 0:01850 0:88943 6 6 6 1:00000 0:03303 0:17246 0:03146 6 6 6 1:00000 0:02816 0:11194 6 6 1:00000 0:00000 4

3 0:09103 7 7 0:21505 7 7 7 0:09148 7 7 0:04253 7 7 7 0:07548 7 7 7 0:12021 7 7 7 0:00000 5 0:45748

After the backward substitution operation, the correcting increments of node voltages can be obtained, 2

3 2 3 De1 0:03356 6 Df1 7 6 0:03348 7 6 7 6 7 6 De2 7 6 0:10538 7 6 7 6 7 6 Df2 7 6 0:36070 7 6 7¼6 7 6 De3 7 6 0:05881 7 6 7 6 7 6 Df3 7 6 0:06900 7 6 7 6 7 4 De4 5 4 0:00000 5 Df4 0:45748 Modifying the node voltage, the voltage vector becomes: 2

3 2 3 e1 0:96643 6 f1 7 6 0:33481 7 6 7 6 7 6 e2 7 6 1:10533 7 6 7 6 7 6 f2 7 6 0:36070 7 6 7¼6 7 6 e3 7 6 1:05881 7 6 7 6 7 6 f3 7 6 0:66900 7 6 7 6 7 4 e4 5 4 1:05000 5 f4 0:45748 Using this voltage vector as the initial voltage value, we can repeat above operations. If the tolerance is set to e ¼ 106 , the calculation converges after five iterations. The evolution process of node voltages and power mismatches is shown in Tables 2.2 and 2.3.

100 Table 2.2 Iterating No. 1 2 3 4 5

Table 2.3 Iterating No. 1 2 3 4 5

2 Load Flow Analysis Node voltages in iterative process e1

f1

e2

f2

e3

f3

e4

f4

0.96643 0.87365 0.85947 0.85915 0.85915

0.33481 0.07006 0.07176 0.07182 0.07182

1.10538 1.03350 1.02608 1.02600 1.02600

0.36074 0.32886 0.33047 0.33047 0.33047

1.05881 1.03564 1.03355 1.03351 1.03351

0.06900 0.07694 0.07737 0.07738 0.07738

1.05000 0.97694 0.97464 0.97461 0.97461

0.45748 0.38919 0.39061 0.39067 0.39067

Node power mismatches in iterative process DQ1 0.55000 0.07263 0.02569 0.00078 0.00000

DP1 DQ2 DP2 DQ3 1.60000 5.69799# 2.00000 2.04901 0.03473 6.00881# 2.10426 0.37144 0.06011 0.41159# 0.15764 0.00924 0.00032 0.0030# 0.00054 0.00002 0.00000 0.00000 0.00000 0.00000

DP3 3.70000 0.04904 0.00329 0.00000 0.00000

DP4 5.00000 2.39001 0.16193 0.00069 0.00000

Power error

101 100 10−1 10−2 10−3 10−4

Fig. 2.9 Convergence property of Newton–Raphson method

1

2

3

4

5

6

7 Iterations

To reveal the convergence property, the maximal power mismatches (with # in Table 2.3) in the iterative process are shown in Fig. 2.9. In the iteration process, especially when it approaches convergence, the changes of the diagonal elements in the Jacobian are not very significant. To illustrate this point, the changes of the diagonal elements are given in Table 2.4. The calculation results of node voltages are shown in Table 2.5.

2.4 Fast Decoupled Method

101

Table 2.4 Diagonal elements of Jacobian matrix in iterative process @DP1 @DP2 @DP3 @DQ3 @DQ1 @DQ2 Iterating no. @f1 @f2 @f3 @e1 @e2 @e3 1 6.04166 6.54166 60.28283 73.67881 32.38884 39.08688 2 5.22590 6.84268 79.81886 69.30868 36.62734 38.83341 3 4.37415 6.42613 69.78933 69.61682 35.38612 38.39351 4 4.23077 6.38634 68.89682 69.52026 35.29706 38.33158 5 4.22720 6.38577 68.88900 69.51747 35.29572 38.33048

Table 2.5 Node voltage vectors Node Magnitude 1 0.86215 2 1.07791 3 1.03641 4 1.05000 5 1.05000

2.4 2.4.1

@DV42 @e4 1.05000 0.96259 0.97528 0.97463 0.97461

@DP4 @f4 63.49206 70.18293 65.61929 65.14834 65.14332

Angle ( ) 4.77851 17.85353 4.28193 21.84332 0.00000

Fast Decoupled Method Introduction to Fast Decoupled Method

The basic idea of the fast decoupled method is expressing the nodal power as a function of voltages in polar form; separately solving the active and reactive power equations [9] by using active power mismatch to modify voltage angle and using reactive power mismatch to modify voltage magnitude. In this way, the computing burden of load flow calculation is alleviated significantly. In the following, the derivation of the fast decoupled method from the Newton method is discussed. As described previously, the core of the Newton load flow approach is to solve the correction equation. When the nodal power equation is expressed in polar form, the correction equation is (see (2.50)),

DP H ¼ DQ J

N L

Du DV=V

ð2:65Þ

or can be written as, DP ¼ HDu þ NDV=V DQ ¼ JDu þ LDV=V

ð2:66Þ

This equation is derived strictly from the mathematical viewpoint. It does not take the characteristics of power systems into consideration. We know that in high voltage power system the active power flow is mainly related to the angle of the nodal voltage phasor while reactive power flow is mainly

102

2 Load Flow Analysis

related to its magnitude. The experiences of many load flow calculations tell us that the element values of matrix N and J in (2.66) are usually relatively small. Therefore, the first step to simplify the Newton method is to neglect N and J, and (2.66) is simplified to )

DP ¼ HDu

ð2:67Þ

DQ ¼ LDV=V

Thus a simultaneous linear equation of dimension 2n is simplified to two simultaneous linear equations of dimension n. The second important step to simplify the Newton method is to approximate the coefficient matrices of (2.67) as constant and symmetric matrices. As the phase angle difference across a transmission line usually is not very large (does not exceed 10 20 ), so the following relations hold, )

cos yij 1

ð2:68Þ

Gij sin yij Bij

Furthermore, the admittance BLi corresponding to the node reactive power is certainly far smaller than the imaginary part of the node self-admittance, i.e., BLi ¼

Qi Bii Vi2

Accordingly, Qi Vi2 Bii

ð2:69Þ

Based on the above relationships, the element expressions of coefficient matrix in (2.67) can be represented as (see (2.41), (2.42), (2.48), and (2.49)): 9 Hii ¼ Vi2 Bii > > > > Hij ¼ Vi Vj Bij = Lii ¼ Vi2 Bii Lij ¼ Vi Vj Bij

ð2:70Þ

> > > > ;

Therefore, the coefficient matrix in (2.67) can be written as 2

V12 B11 6 V2 V1 B21 6 H¼L¼6 4 Vn V1 Bn1

V1 V2 B12 V22 B22 .. .

Vn V2 Bn2

3 . . . V1 Vn B1n . . . V2 Vn B2n 7 7 7 5 ...

Vn2 Bnn

ð2:71Þ

2.4 Fast Decoupled Method

103

It can be further represented as the product of the following matrices: 2 6 6 H¼L¼6 4

V1

32

32 3 B11 B12 . . . B1n V1 76 B21 B22 . . . B2n 76 7 0 V2 0 76 76 7 76 76 7 ð2:72Þ .. .. .. .. 5 4 5 4 5 . 0 . . . Vn Bn1 Bn2 . . . Bnn Vn

V2 0

Substituting (2.72) into (2.67), we can rewrite the correction equations as follows: 2

3 2 DP1 V1 6 DP2 7 6 6 7 6 6 .. 7 ¼ 6 4 . 5 4

32

V2

0 ..

0

DPn

B11 76 B21 76 76 54 Vn Bn1

.

B12 B22 .. . Bn2

... ... .. . ...

32 3 B1n V1 Dy1 6 7 B2n 7 76 V2 Dy2 7 76 .. 7 54 . 5

ð2:73Þ

Vn Dyn

Bnn

and 2

3 2 DQ1 V1 6 DQ2 7 6 6 7 6 6 .. 7 ¼ 6 4 . 5 4

32

V2 0

DQn

0 ..

B11 76 B21 76 76 54 Vn Bn1

.

B12 B22 .. .

Bn2

32 3 B1n DV1 6 7 B2n 7 76 DV2 7 76 .. 7 54 . 5

... ... .. . ...

ð2:74Þ

DVn

Bnn

Multiplying both sides of the above equation with matrix, 2 6 6 6 4

V1

31 V2

..

.

Vn

2

6 7 6 7 7 ¼6 6 5 4

1 V1

3 1 V2

..

7 7 7 7 5

. 1 Vn

one can obtain 2

3 2 DP1 =V1 B11 6 DP2 =V2 7 6 B21 6 7 6 6 7¼6 .. 4 5 4 . DPn =Vn

Bn1

B12 B22 .. .

Bn2

... ... .. . ...

32 3 B1n V1 Dy1 6 7 B2n 7 76 V2 Dy2 7 76 .. 7 54 . 5 Vn Dyn Bnn

ð2:75Þ

and 2

3 2 DQ1 =V1 B11 6 DQ2 =V2 7 6 B21 6 7 6 6 7¼6 .. 4 5 4 . DQn =Vn

Bn1

B12 B22 .. .

Bn2

... ... .. . ...

32 3 B1n DV1 6 7 B2n 7 76 DV2 7 76 .. 7 54 . 5 Bnn

DVn

ð2:76Þ

104

2 Load Flow Analysis

The above two equations are the correction equations of the fast decoupled load flow method. The coefficient matrix is merely the imaginary part of the nodal admittance matrix of the system, and is thus a symmetric, constant matrix. Combining with the power mismatch equation (2.13), we obtain the basic equations of the fast decoupled load flow model DPi ¼ Pis Vi

X

Vj ðGij cos yij þ Bij sin yij Þ

ði ¼ 1; 2; . . . ; nÞ

ð2:77Þ

Vj ðGij sin yij Bij cos yij Þ ði ¼ 1; 2; . . . ; nÞ

ð2:78Þ

j2i

DQi ¼ Qis Vi

X j2i

The iterative process can be briefly summarized in the following steps: ð0Þ 1. Specify node voltage vector initial value yð0Þ i , Vi 2. Calculate the node active power mismatch DPi according to (2.77), and then calculate DPi =Vi 3. Solving correction equation (2.75), calculate the node voltage angle correction Dyi 4. Modify the node voltage angle yi ðtÞ

ðt1Þ

yi ¼ yi

ðt1Þ

Dyi

ð2:79Þ

5. Calculate node reactive power mismatch DQi according to (2.78), and then calculate DQi =Vi 6. Solving correction equation (2.76), calculate the node voltage magnitude correction DVi , 7. Modify the node voltage magnitude Vi ; ðtÞ

ðt1Þ

Vi ¼ Vi

ðt1Þ

DVi

ð2:80Þ

8. Back to step (2) to continue the iterative process, until all node power mismatches DPi and DQi satisfy convergence conditions.

2.4.2

Correction Equations of Fast Decoupled Method

The main difference between the fast decoupled method and the Newton method stems from their correction equations. Comparing with correction (2.40) or (2.54) of the Newton method, the two correction equations of the fast decoupled method have the following features: 1. Equations (2.75) and (2.76) are two simultaneous linear equations of dimension n instead of a simultaneous linear equation of dimension 2n

2.4 Fast Decoupled Method

105

2. In (2.75) and (2.76), all elements of the coefficient matrix remain constant during the iterative process 3. In (2.75) and (2.76), the coefficient matrix is symmetric. The benefit of the first feature for computing speed and storage is obvious. The second feature alleviates the computing burden in forming and eliminating the Jacobian within the iterative process. We can first form the factor table for the coefficient matrix of the correction equation (see (2.76)) by triangularization. Then we can carry out elimination and backward substitution operations for different constant terms DP=V and DQ=V through repeatedly using the factor table. In this way, the correction equation can be solved very quickly. The third feature can further improve efficiency in forming and storing the factor table. All the simplifications adopted by the fast decoupled method only affect the structure of the correction equation. In other words, they only affect the iteration process, but do not affect the final results. The fast decoupled method and the Newton method use the same mathematical model of (2.13), if adopting the same convergence criteria we should expect the same accuracy of results. It seems that (2.75) and (2.76) derived above have the same coefficient matrix, but in practice the coefficient matrixes of the two correction equations in the fast decoupled algorithms are different. We can simply write them as DP=V ¼ B0 VDu DQ=V ¼ B00 DV

ð2:81Þ ð2:82Þ

Here V is a diagonal matrix with the diagonal elements being the node voltage magnitudes. First, we should point out that the dimensions of B0 and B00 are different. The dimension of B0 is n 1 while the dimension of B00 is lower than n 1. This is because (2.82) dose not include the equations related to PV nodes. Hence if the system has r PV nodes, then the dimension of B00 should be n r 1. To improve the convergence, we use different methods to treat B0 and B00 , and how we treat B0 and B00 will result in different fast decoupled methods, are not merely the imaginary part of the admittance matrix. As described above, (2.81) and (2.82) are the correction equations based on a series of simplifications. Equation (2.81) modifies the voltage phase angles according to the active power mismatch; (2.82) modifies the voltage magnitudes according to the reactive power mismatch. To speed up convergence, the factors that have no or less effect on the voltage angle should be removed from B0 . Therefore, we use the imaginary part of admittance to form B0 without considering the effects of shunt capacitor and transformer’s off-nominal taps. To be specific, the off-diagonal and diagonal elements of B0 can be calculated according to following equations: B0ij ¼

xij ; rij2 þ x2ij

B0ii ¼

X j2i

X xij ¼ B0ij rij2 þ x2ij j2i

where rij and xij is the resistance and reactance of branch ij, respectively.

ð2:83Þ

106

2 Load Flow Analysis

Theoretically, the factors that have less effect on voltage magnitude should be removed from B00 . For example, the effect of line resistance to B00 should be removed. Therefore, the off-diagonal and diagonal elements of B00 can be calculated according to the following equations: B00ij ¼

X1 X 1 1 ; B00ii ¼ bio B00ii ¼ bio xij x xij j2i ij j2i

ð2:84Þ

where bio is the shunt admittance of the grounding branch of node i. If B0 and B00 are formed according to (2.83) and (2.84), the fast decoupled method is usually called the BX algorithm. Another algorithm opposite to BX method is called the XB algorithm in which B0 used in the DP Dy iteration is formed according to (2.84), while B00 used in the DQ DV iteration is formed according to (2.83). Although these two algorithms have different correction equations, their convergence rates are almost the same. Several IEEE standard test systems have been calculated to compare the convergence of these algorithms. Table 2.6 shows the number of iterations needed to converge for these test systems. Many load flow calculations indicate that BX and XB methods can converge for most load flow problems for which the Newton method can converge. The authors of [9, 10] explain the implications of the simplifications made in the fast decoupled method. Wong et al. [19] propose a robust fast decoupled algorithm to especially treat the possible convergence problem caused by high r=x networks. Bacher and Tinney [26] adopt the sparse vector technique to improve the efficiency of the fast decoupled method. From the above discussion we know that the fast decoupled method uses different correction equations to the Newton method, hence the convergence properties are also different. Mathematically speaking, the iteration method based on a fixed coefficient matrix to solve a nonlinear equation belongs to ‘‘the constant slope method.’’ Its convergence process has the characteristic of the geometric series. If the iteration procedure is plotted on a logarithmic coordinate, the convergence characteristic is nearly a straight line. In contrast, convergence of the Newton method has a quadratic property and is quite similar to a parabola. Fig. 2.10 shows the typical convergence properties of the two methods. Figure 2.10 illustrates that the Newton method converges slower at the early stages, but once converged to some degree its convergence speed becomes very fast. The fast decoupled method converges almost at the same speed throughout the iteration procedure. If the specified convergence criterion is smaller than the errors Table 2.6 Convergence comparison of BX method and XB method Systems Newton BX XB IEEE-5 bus 4 10 10 IEEE-30 bus 3 5 5 IEEE-57 bus 3 6 6 IEEE-118 bus 3 6 7

2.4 Fast Decoupled Method

107

Power error 1

Newton Method

1e-1 P Q Decoupled Method 1e-2

A 1e-3

1e-4

1e-5

5

10

15 Iterations

Fig. 2.10 Convergence properties of fast decoupled method and Newton method

at point A in Fig. 2.10, the iteration number of the fast decoupled method is larger than that of the Newton method. It can be roughly considered that a linear relation exists between the iteration number and the required precision when using the fast decoupled method. Although the iteration number of the fast decoupled method is larger, its computing requirement in each iteration is far less than that of the Newton method. So the computing speed of the fast decoupled method is much higher than the Newton method.

2.4.3

Flowchart of Fast Decoupled Method

The principle flowchart of the fast decoupled method is shown in Fig. 2.11 which illustrates the main procedure and logical structure of the load flow calculation. The symbols used in Fig. 2.11 are first introduced below: t: counter for the iteration number K01 a flag with ‘‘0’’ and ‘‘1’’ states, ‘‘0’’ indicates the active power iteration; while ‘‘1’’ the reactive power iteration. A whole iteration includes an active power iteration and a reactive power iteration.

108

2 Load Flow Analysis

Input information and original data, and deal with original data

1 2

Form admittance matrix

Calculate coefficient matrix B⬘, and form the first factor table

3 4

Calculate coefficient matrix B⬙, and form the second factor table

5 Give voltage initial value on each node

6

t = 0, k 01= 0

Calculate [ DW(K01 ) /V] ;ERM (K01)

Solve modified equation (2.81)or(2. 82) , and modify V K 01

No

false

Yes

0: PðBÞ

ð3:7Þ

Several important formulas can be deduced according to conditional probability. 1. Multiplication probability theorem. Let A1, A2, . . ., An be n arbitrary events, the probability of their intersection set is PðA1 \ A2 \ \ An Þ ¼ PðA1 ÞPðA2 jA1 ÞP½A3 jðA1 \ A2 Þ P½An jðA1 \ A2 \ \ An1 Þ:

ð3:8Þ

However, when A1, A2, . . ., An are independent, we have PðA1 \ A2 \ \ An Þ ¼ PðA1 ÞPðA2 Þ PðAn Þ:

ð3:9Þ

2. Formula of total probability. Let event A occur according to the given condition of events B1, B2, . . ., Bn. A can only occur at the same time as one of B1, B2, . . .,

132

3 Stochastic Security Analysis of Electrical Power Systems

Bn occurs, and any two of Bi are mutually exclusive, but Ptheir union sets consist of the sample space of one event, that is, Bi Bj ¼ ’ði 6¼ jÞ; ni¼1 Bi ¼ O; PðBi Þ > 0, then the total probability of event A, P(A), is PðAÞ ¼

n X

PðBi ÞPðA=Bi Þ:

ð3:10Þ

i¼1

3. Bayes’ Formula. Assume the occurring condition of event Bi (i = 1,2,. . ., n) is same as that in (2), then the probability of occurrence of event Bi after the event A occurred, is denoted by PðBi =AÞ ¼

PðBi ÞPðA=Bi Þ ði ¼ 1; 2; . . .Þ: n P PðBi ÞPðA=Bi Þ

ð3:11Þ

i¼1

Equation (3.11) is Bayes’ Formula. It means that once event A occurred in experiment, (3.11) is used to reassess the cause Bi, so the probability P(Bi/A) is called posterior probability.

3.2.2

Random Variable and its Distribution

If the outcome of a random experiment can be described by one numerical variable, and this numerical value is determined by a certain probability, then the variable is named a random variable. In mathematical terms, it can be described that the set O of all sample points e is one sample space in a random experiment, and X is a realvalued function defined on the sample space, that is, e 2 O; XðeÞ 2 R: If there exist real values a < b, such that the set of sample points satisfies feja XðeÞ bg; then this set is an event, and the function X(e) is referred to as a random variable. If a = 1, event {e| ‐ 1 X(e) b} can be described by {X b} for short. Its probability measurement, FðxÞ ¼ PðX xÞ

ð3:12Þ

is defined as the distribution function of random variable X. x can be any given real value. The general random variable X can be classified into a discrete random variable and a continuous random variable according to its different possible values.

3.2 Basic Concepts of Probability Theory

133

For continuous random variables, another function to express its probability is the probability density function f(x), which is defined by, f ðxÞ ¼ lim

Dx!0

1 Pðx < X < x þ DxÞ; Dx

ð3:13Þ

which can also written in incremental format, Pðx < X < x þ DxÞ f ðxÞDx:

ð3:14Þ

Formula (3.14) can be interpreted as the probability under the condition that random variable X is in the interval (x, x + Dx) and Dx ! 0. Obviously, the probability of random variable X between a and b is, Zb Pða < X bÞ ¼

f ðxÞdx

ð3:15Þ

a

and the relationship between (3.15) and distribution function F(x) in formula (3.12) can be written as, Zx FðxÞ ¼

f ðxÞdx

ð3:16Þ

1

and f ðxÞ ¼

dFðxÞ : dx

ð3:17Þ

For a discrete random variable (as shown in Fig. 3.1), X may be xi (i = 1, 2, . . ., n), then its probability density function is defined as pðxÞ ¼

PðX ¼ xi Þ 0

x ¼ xi x 6¼ xi

ð3:18Þ

and the distribution function is FðxÞ ¼

X

pðxi Þ:

ð3:19Þ

xi x

3.2.3

Numeral Characteristics of Random Variable

In many practical problems, we can specify the characteristics of random variables by finding the average value of random variables and the degree of value dispersion. The two most commonly used methods are introduced as follows.

134

3 Stochastic Security Analysis of Electrical Power Systems

1.0 0.25 0.8 0.6

0.15

F(x)

P(x)

0.20

0.10

0.4

0.05

0.2

0.00

0.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13

a

0 1 2 3 4 5 6 7 8 9 10 1112 13

b

x

x

Fig. 3.1 The relative function of discrete random variable (a) probability density function; (b) distribution function

3.2.3.1

Mathematical Expectation (Mean Value)

Discrete random variable X can be x1, x2, . . ., and its corresponding probability is PðX ¼ xi Þ ¼ pi

i ¼ 1; 2; . . .

Then mathematical expectation or expectation, E(X), is defined as EðXÞ ¼

1 X

xi pi :

ð3:20Þ

i¼1

For a continuous random variable X, when its density function is f(x), we have Z1 EðXÞ ¼

xf ðxÞdx:

ð3:21Þ

1

For the mathematical expectation of a set of random variables Xi (i = 1, 2, . . ., n), there are characteristics such as described as follows E

n X

! Xi

i¼1

3.2.3.2

¼

n X

EðXi Þ:

ð3:22Þ

i¼1

Variance

Discrete random variable X is denoted as s2, which is defined by, s2 ¼

n X i¼1

ðxi mÞ2 pi ;

ð3:23Þ

3.2 Basic Concepts of Probability Theory

135

where m = E(X), that is average value. Obviously, s2 represents the degree of dispersion of its value deviating from the average value m. For a continuous random variable X, we get, Z1 ðx mÞ2 f ðxÞdx:

s ¼ 2

ð3:24Þ

1

Some properties and applications related to other numerical characteristics of random variables will be discussed in Sect. 3.5.1.

3.2.4

Convolution of Random Variable

Suppose two random variables X and Y are independent, and they have probability density functions f1(x) and f2(y), respectively, then Z = X + Y is still a random variable. The probability density function of Z is Z1 fðzÞ ¼

f ðx; z xÞdx 1 Z1

¼

f1 ðxÞf2 ðz xÞdx:

ð3:25Þ

1

Its distribution function is Zz

Z1

FðzÞ ¼

f1 ðxÞf2 ðz xÞdx dz: 1

ð3:26Þ

1

If X and Y are discrete random variables, then the distribution function is FðzÞ ¼

X

PðX ¼ xi ; Y ¼ yj Þ ¼

xi þyj C0i4 yq2 ¼ C0i4 ll yq2 ; > = l

i¼

l

1; 2;

l ¼ 1; 2; . . . ; nc :

ð5:102Þ

> > ;

We can see that 2nc among the 5nc elements of C20 yq are nonzero. 4. Building the vectors A1DX in (5.98). From (5.85) we have A1 DX ¼ ¼

0

0

0

0

A21

A22 0

0

0

A21 DVd þ A22 DId

0 0

:

DVTd

DITd

DKTT

DWT

DFT

T ð5:103Þ

294

5 HVDC and FACTS

Equations (5.66) and (5.67) yield ðA21 DVd þ A22 DId Þk ¼ ðIdk DVdk þ Vdk DIdk Þ;

k ¼ 1; 2; . . . ; nc :

ð5:104Þ

We have introduced the integrated iteration power flow calculations of interconnected systems.

5.3.5

Alternating Iteration for AC/DC Interconnected Systems

The alternating iteration method is a further simplification to the P–Q decoupled method in the integrated iteration power flow calculations. Based on the basic converter (5.52) and (5.53), the impact of AC systems on DC systems relies on the primary voltages Vt of converter transformers. If the AC voltages Vt of all converters in a multiterminal DC systems are known, the DC system will have (5.59)–(5.61), (5.63), and (5.64), a total of 5nc equations and 5nc unknown variables. We can obtain the 5nc unknown DC variables by solving only the DC system equations. The power taken out of, or injected into, AC systems, Pidc þ jQidc from converter transformers, represents the impact of DC systems on AC systems. If the power withdrawn from or injected into AC systems is known, power flow calculations of AC systems are not affected by DC systems. The ideal process is to designate the primary voltages of nc converter transformers ð0Þ

Vt

h Þ ¼ Vnð0a þ1

ð0 Þ

Vna þ2

...

i ð0Þ Vna þnc :

Obtain solution of DC variables X(0). Substituting X(0) into (5.57) yields the power of all converters Pdc(0) and Qdc(0). Using converter power in AC system equations forh conventional i power flow calculations gives rise to convergent solution Vð1Þ ¼ Vða1Þ ð1Þ

ð1Þ

Vt

ð1Þ

. Ideally the calculation completes if Vt ð0Þ

ð0Þ

equals Vt .

Generally Vt is not the same as Vt . The calculation is an iteration process. Based on the above, AC and DC system equations are separately solved in alternating iterations. When solving AC system equations, we use the known power at the DC buses to represent the corresponding DC systems. While solving DC system equations, we model AC systems as constant voltages at the AC buses of converters. At each iteration, the solution of the AC systems provides the converter AC bus voltages for the next DC iteration; the solution of DC systems in turn produces the equivalent real and reactive power of converters for the further AC iteration. The iteration goes on and on until convergence is achieved. We must point out that the convergence of this method is mathematically related to the Gauss– Seidel iteration. In fact, the alternating iteration is not a complete Gauss–Seidel iteration. For the AC system equations in the alternating iteration method, we usually use the Newton algorithm or P–Q decoupled algorithm. For DC system equations, the Newton algorithm is the most common approach [114]. The Gauss– Seidel algorithm applies only to the coupling between AC and DC equations.

5.3 Power Flow Calculation of AC/DC Interconnected Systems

295

Start Input data Build factor tables of B⬘ and B ⬙ Set initial values Compute DD from Eqs. 5.59

5.61, 5.63 and 5.64; Y

|| DD || kTmax, take Vds kTmax/kTworst as new voltage control value; otherwise take Vds kTmin/kTworst as the new voltage control value. Return to Step 1. The above steps are the main steps of this method. Its basic characteristics are the simplicity in theory and in programming. Comparing with Newton’s iteration for DC equations, it saves considerable memory. When assuming the converter transformer ratio is a continuous variable and with no over-limits, DC and AC system power flow solution can be attained in one computation. With necessary amendment to the above method, it can be applied to power flow calculations for fixed control angle control, discrete transformer ratio changes, etc. In which case, iteration is required. The details can be found in [115]. In the above we have discussed the two major types of power flow calculation methods for AC/DC interconnected systems. Integrated iteration takes into consideration the complete coupling between AC and DC systems, and has good convergence for various conditions of network and system operation. The Jacobian matrix has a higher order than for pure AC systems. The approach requires more programming, uses more memory, and needs longer computation time. Alternating iteration can be accomplished by adding DC modules to the existing power flow programs, due to its separated solution of AC and DC equations. It is easier to take into consideration the constraints on DC variables and the adjustment of operation modes. However, the convergence of alternating iteration is not as good as integrated iteration. The computational practice indicates that its convergence is good when the AC system is strong. If the AC system is weak, its convergence deteriorates, requiring more iterations or even becoming nonconvergent. This is the shortcoming of the alternating iteration method. The strength of AC systems is related to the rated capacity of converters. Taking the converter rated power PdcN as the base, the reciprocal of per unit equivalent reactance of AC system, as viewed from the AC bus of converters, is called the short-circuit ratio (SCR). The larger the SCR, the higher is the system strength. A weak AC system (SCR less than 3) has a larger equivalent reactance, making the AC bus voltage of the converter very sensitive to variation of the reactive power injection. Alternating iteration separates the solution of AC and DC equations, assuming constant Vt and Qtdc at the boundary between AC and DC systems to neglect their coupling. If the AC system is weak, the variation of Qtdc can bring potential change to Vt. This results in computational oscillation between Qtdc and Vt in alternating iterations and worsening convergence [116]. There have been some improved calculation methods [117] for alternating iteration to make it applicable to weak AC systems. We are not going to discuss them here for brevity.

5.4 HVDC Dynamic Mathematical Models

5.4

299

HVDC Dynamic Mathematical Models

We have introduced the steady-state models of HVDC systems in the previous section. The HVDC transients are quite complicated. The main causes of the complexity are as follows. (1) The firing pulses of bridge valves are triggered at discrete time points. In transients, the firing angle is regulated by the controller to make the corresponding time unevenly distributed. Thus the firing angle is a discrete variable with regards to computation. (2) We assume that AC systems are symmetric in steady-state analysis. From the steady-state analysis of converter valves we know that the valve on/off states are closely related to the commutating voltages, the time of firing, and the magnitudes of commutation angles. When firing angles or commutation angles are too large, commutation may fail. In transient states, AC systems are actually unsymmetrical. Some valves could have negative valve voltage and could not been turned on when firing pulses occur if commutation voltages are severe unsymmetrical. For HVDC systems under transient states, we need to establish derivative equations to take into consideration the variations of commutation voltages and firing angles as well as other exceptional conditions. The solution of these derivative equations reveals the time of valve state changes. (3) We should consider the distributed characteristics of DC lines for long distance transmission. Under such circumstances, the variations of voltage and current on DC lines become wave processes. Due to the above conditions, we need to solve ordinary differential equations and partial differential equations with both continuous and discrete variables to calculate accurately the transients of HVDC transmission systems. From the mathematical point of view, it is not difficult to solve these equations. Many previous works [118– 120] used detailed mathematical models resulting in huge computational requirements. We should simplify the transient DC models as much as possible without losing engineering accuracy. In general we can take a simple DC model for stability analysis, if AC systems are relatively strong; otherwise a detailed DC model is required. The general assumptions that we make in deriving DC steady-state models still apply to most analysis of power system stability. Thus we can use the steadystate mathematical models of converters (5.52), (5.54) as their dynamic models. Here we are going to introduce the mathematical models of control systems. The controllers in HVDC systems consist of electronic circuits. Their basic working principles are as follows: receiving control inputs, sending outputs to phase-control circuits, and pulse generation device to set converter firing angles in order to control converter operation. Different control signals and different control strategies result in different controller structures and control characteristics, as well as the dynamics of DC systems or even the whole power system. To achieve better operation characteristics, the adjustments of rectifiers and inverters should be coordinated. As stated before, the basic control mode is fixed current or fixed power for rectifiers, and fixed voltage or fixed extinction angle for inverters. The transformer ratio adjustment is slow and is a discrete variable. The ratio is not changed

300

5 HVDC and FACTS Idref Id

1 1 + sTc3

x1

−

kc1

+

+

PI regulator

−

aref +

a

+ a2

kc2

Measuring unit

a1

sTc2

a Fixed Current Control Pref 1 − 1 + sTp2 x2

Pd

+

Idref kp1 1+ sTp1

+ x3

kc1

a1

+

+

− Id

+ a2

kc2 sTc2

−

aref +

α

b Fixed Power Control Vdref Vd

1 1 + sTv3

− x4

k v1

+

b1

+ + kv2

−

bref +

b

b2

sTv2

c Fixed Voltage Control mref μ

1 1 + sTm3

x5

−

km1

+

+ km2

b1

+

bref +

b

+ b 2

sTm2

d Fixed Extinction Angle Control

Fig. 5.18 Transfer function block diagrams of HVDC control systems

very often and is an ancillary means to optimize the converter operation point. Figure 5.18 shows the transfer function of the four basic control modes. The transfer function of fixed current control is shown in Fig. 5.18a. It compares Id, the output of DC current, and given current Idref. The difference is amplified and goes through a proportional plus integral process. Then the signal is passed to the phase-shift control circuits to change the converter firing angle and to enforce the fixed current function. The transfer function of fixed power control is shown in Fig. 5.18b. HVDC systems are usually required to transport power as planned. Fixed power is a basic control method. When the variations of AC voltages on both terminals are not large,

5.5 Basic Principles and Mathematical Models of FACTS

301

using fixed current and fixed extinction angle can actually achieve fixed power control. When taking into consideration AC voltage fluctuations, using fixed current and fixed voltage can obtain exact fixed power control. These two control methods are to determine the DC current setting based on the given power and DC operation voltage at the control terminal. However DC voltage is related to DC current, so it is very difficult to set DC current beforehand. To overcome this problem, special control devices are set up for fixed power control. Fixed current control has high response speed, is capable of quickly constraining overcurrent to prevent converter overload, and is easy to set up. Power control devices are usually based on fixed current control and receive additional inputs rather than directly acting on phase control circuits. In the diagram, the DC power is compared with its target value. The difference is amplified and sent to the input of the fixed current controller. This works by changing the current setting of the fixed current control dynamically. The transfer functions of fixed voltage and fixed extinction angle controls are shown in Fig. 5.18c, d. They share the same structure as Fig. 5.18a with different parameters. We need to point out that extinction angles cannot be directly measured. They are indirectly obtained by measuring the time interval between valve voltage and current zero-crossing points. Although we do not show the quantity limitation block in the above diagrams, attention has to be paid to the constraints on various physical variables. There are minimum firing angle constraints for rectifier fixed current control, minimum extinction angle constraints for inverter fixed voltage control, etc. We need to notice that the controllers here are all for DC internal adjustments. DC systems can be used to affect AC system operation through these DC internal adjustments. The inputs of controllers may include AC system operation parameters, line power, the velocity of some generators, system frequency, and so on. This kind of control is the integrated control of AC/DC systems, also called external adjustments. The control strategy and control signals in these cases are an important field of power system research.

5.5

Basic Principles and Mathematical Models of FACTS

After the introduction of the FACTS concept, many FACTS devices have been proposed. We can classify them into three groups based on the maturity of the technology. The first group has been applied in the power industry, such as static VAr compensators (SVR), thyristor controlled series capacitor (TCSC), and static synchronous compensators (STATCOM). The second group has industrial sample machines and is still under investigation, such as unified power flow controller (UPFC). The third group has only a theoretical design without any industrial application, such as static synchronous series compensator (SSSC), thyristor controlled phase shifting transformer (TCPST). We will introduce their basic principles and mathematical models in this section. The power flow calculation for systems having these devices will be discussed in the next section.

302

5 HVDC and FACTS

FACTS devices can be classified based on their connection types as series, shunt, and combined types. SVC and STATCOM are shunt type. TCSC and SSSC are series type. TCPST and UPFC are combined type. Designed by US Electrical Power Research Institute (EPRI), manufactured by Westinghouse, and installed at AEP power system in USA for industrial testing operation, UPFC is the most powerful FACTS device proposed as of today. Its control strategy is presently under further research.

5.5.1

Basic Principle and Mathematical Model of SVC

A common practice of system voltage adjustment is shunt reactive power compensation. The synchronous condenser was historically an important tool of shunt reactive power compensation. Since it is a rotating machine, its operation and maintenance are quite complicated. New synchronous condensers are now seldom installed. The static shunt reactive power compensation, as opposed to the rotating synchronous condenser, has wide industrial application due to its low cost and simple operation and maintenance. Conventional static shunt reactive power compensation is to install capacitors, reactors, or their combination, at the compensated buses to inject or extract reactive power from the system. Mechanical switches are used to put the shunt capacitor/reactors into or out of operation. There are three disadvantages in this type of compensation. First, their adjustment is discrete. Second, their control actions are slow and cannot meet system dynamic requirements. Third, they have negative voltage characteristics. When system voltages drop (rise), the reactive power injection of shunt capacitors decreases (increases). However, they are widely applied in power systems due to their economic advantages and easy maintenance. Modern SVR with FACTS technology integrate power electronic elements into conventional static shunt reactive power compensation devices to achieve fast and continuously smooth adjustment. Ideal SVCs can maintain nearly constant voltages at the compensated buses. The good steady and dynamic characteristics render them widely applicable. Their basic elements are thyristor controlled reactors (TCRs) and thyristor switched capacitors. It is not difficult to understand other types of SVCs if we know the working principles of these two. Figure 5.19 shows their basic diagrams. To save cost, most SVCs connect to systems through step-down transformers. The valve control of the SVC produces harmonics. Filters are installed with SVCs to reduce harmonic contamination. They are capacitive as regards to fundamental frequency and inject reactive power into systems. Figure 5.20a, b shows TCR and TSC branches. Below we will analyze the control theory of TCR and TSC. TCR branch consists of reactors connected with two back-to-back thyristors as control elements. The system voltage on the branch is sinusoidal and shown in Fig. 5.21a. The valve delayed firing angle is a 2 [p/2, p]. The firing time is ot ¼ a þ kp

k ¼ 0; 1; 2; . . . :

5.5 Basic Principles and Mathematical Models of FACTS

303

High-voltage bus Step-down transformer

L

C1

TCR

C2

C3 TSC

Filter

Fig. 5.19 SVC basic diagram iC

iL

C

L VmSin wt

VmSin wt

TCR

TSC

Fig. 5.20 TCR and TSC branches

Apparently the inductor current is zero when the two valves are off. When the valve conducts, neglecting the resistance in the reactor, the inductor current is L

diL ¼ Vm sin ot; dt

ð5:113Þ

where L is the inductance of the reactor, Vm is the magnitude of the system voltage. Its general solution is iL ¼ K

Vm cos ot; oL

ð5:114Þ

where K is the integral constant. Since the inductor current is zero at firing, the above equation yields iL ¼ K

Vm cosða þ kpÞ ¼ 0: oL

304

5 HVDC and FACTS

Substituting the solution of K into (5.114) gives rise to the inductor current. iL ¼

Vm ½cosða þ kpÞ cos ot oL

k ¼ 0; 1; 2; . . . :

ð5:115Þ

Based on the above equation, inductor current returns to zero at ot ¼ (k þ 2)p a. Thus the valve conducting period is ot 2 ½kp þ a; ðk þ 2Þp a k ¼ 0; 1; 2; . . . : The waveform of inductor current is shown in Fig. 5.21b. The width of a single ripple of inductor current is ðk þ 2Þp a ðkp þ aÞ ¼ 2ðp aÞ ¼ 2b: b ¼ p a is called the conducting angle. To make sure that there is always one valve conducting at any moment, we should have ðk þ 2Þp a ¼ ðk þ 1Þp þ a;

k ¼ 0; 1; 2; . . . :

One valve should conduct the moment and another one is turned off, so a ¼ p/2. This operation mode corresponds to connecting the shunt reactor directly to the system. From the waveforms we can see that the valve conducting period decreases from p to zero as the firing angle rises from p/2 to p. Now the two valves are turned off at all times, corresponding to reactors out of service. Besides when a is less than V VmSinωt ωt α

a iL iL π- α α

b

π

2π-α

3π-α 2π

3π 2π+α

3π+α 4π-α

Fundamental component

Fig. 5.21 (a) TCR voltage waveforms (b) TCR current waveforms

ωt

5π-α 4π

iL1

5.5 Basic Principles and Mathematical Models of FACTS

305

p/2, the moment at which the current of a conducting valve returns to zero is later than the firing moment of the off valve as ðk þ 2Þp a > ðk þ 1Þp þ a: In this case, the conducting valve has not been turned off when the other valve receives a firing pulse. The off valve cannot be triggered on due to zero valve voltage. One of the two valves is off at any moment. Thus the main component of inductor current is DC. The normal operating ranges of TCR firing angles are a 2 [p/2, p]. Based on (5.115) and the waveforms, the current passing through the reactor is irregular and no longer sinusoidal due to valve control. The adjustment of firing angles changes the current peak values and conducting periods. Applying Fourier analysis to the current yields the magnitude of the fundamental frequency component

IL1

2 ¼ p

2pa Z

a

Vm Vm ðcos a cos yÞ cos ydy ¼ ½2ða pÞ sin 2a: oL poL

And the instantaneous value of fundamental frequency component is iL1 ¼ IL1 cos ot ¼

Vm ð2b sin 2bÞ sinðot p=2Þ: poL

ð5:116Þ

The equivalent fundamental frequency reactance of the TCR branch is XL ðbÞ ¼

h pi poL b 2 0; : 2b sin 2b 2

ð5:117Þ

Thus the TCR equivalent reactance of fundamental frequency components is the function of conducting angle b or the firing angle a. The control of firing angle a can smoothly adjust the equivalent shunt reactance. The reactive power consumed by TCR is V2 2b sin 2b 2 QL ¼ V_ I_L1 ¼ ¼ V : XL ðbÞ poL

ð5:118Þ

As shown in Fig. 5.20b, the TSC branch consists of a capacitor connected in series with two thyristors connected in parallel and in opposite directions. The TSC source voltage is the same as TCR. Its waveforms are in Fig. 5.21a. The TSC creates two operating states for the capacitors through valve control: shunt capacitors in service or out of service. Stopping the firing can simply put the capacitor out of service. Note that the natural switch-off from conduction happens when the capacitor

306

5 HVDC and FACTS

current is zero and its voltage at the peak of source voltage. Neglecting the capacitor leakage current, capacitor voltage maintains the peak value if firing stops after the natural switch-off. We need to pay attention to the timing of putting the capacitor into service. The principle is to reduce the impulse current in capacitors at the moment of in-service operation. We should use the correct valve based on the sign of the capacitor initial voltage, and put the capacitor into service at the moment when source voltage equals capacitor initial voltage. So the transient component of capacitor current is zero when put into service. After capacitors are in service, we need a ¼ p/2 to keep one valve conducting at all times. Ideally the capacitor voltage is the peak of source voltage. Using a ¼ p/2 makes no transients for the inservice operation. In reality, the source voltage and the capacitor initial voltage cannot be exactly the same. There is a small inductor in the TSC branch to reduce the possible impulse current. From the above analysis, we can see that the main difference between TSC and mechanically switched capacitors (MSC) is the fast control of in-service or out-of-service operation by valves in TSC. TSC dynamic characteristics can meet system control demands. The reactive power injection of the capacitors is QC ¼ oCV 2 ;

ð5:119Þ

where C is the capacitance of the capacitor. From (5.118) and (5.119) we have the reactive power injection from the SVC is QSVC ¼ QC QL ¼

2b sin 2b 2 oC V : poL

ð5:120Þ

The SVC reactive power injection can be smoothly adjusted when b 2 [0,p/2]. To expand the regulation ranges of SVC, we can have many TSC branches in one SVC, based on the compensation requirements. Figure 5.19 shows an SVC with three TSCs. When all three TSCs are in service, the C in (5.120) is C1 þ C2 þ C3. To guarantee a continuous adjustment, the TCR capacity should be slightly larger than a group of TSCs, that is, oC1 < 1/oL. Based on (5.120), the equivalent reactance of SVC is

XSVC

2b sin 2b 1 poL ¼ oC ¼ : poL 2b sin 2b po2 LC

ð5:121Þ

The SVC equivalent voltage–current characteristics are the combination of TCR and TSC. As b increases from zero to p/2, XSVC will change from capacitive maximum to inductive maximum. Generally, the control signal of SVC is derived from the voltage of the bus to which they are connected. Figure 5.22 shows that as the voltage V changes, the SVC equivalent reactance varies with b. In Fig. 5.22, there is a straight line going through the original corresponding to every b. The slope of the straight line is XSVC. Suppose that the system voltage

5.5 Basic Principles and Mathematical Models of FACTS Fig. 5.22 Equivalent reactance variation with b as voltage changes

307 b5

b6 = 0

b4 V

b3

b2 b = p/2 1

V2

A

V4 V6

V1 V3

B

V5

C

ISVC

Fig. 5.23 Voltage–current characteristic

V XSVGmin = −1/wC Vref B

b1 =

p/2

(

XSVGma = (w L) 1 − w 2LC

)

b6 = 0

ISVC

characteristic is V1. The control scheme is to make the TCR conducting angle b1 ¼ p/2, corresponding to maximum equivalent inductive reactance. The SVC operating point is the crossover point A between system voltage characteristic V1 and the straight line b1. With system voltage characteristic V2 and TCR conduction angle b2 < b1, XSVC decreases and the SVC operating point shifts accordingly. Until system voltage characteristic is V6 and conduction angle b6 ¼ 0, SVC equivalent reactance is maximum capacitive with operating point B. Apparently, voltage at B is higher than at C. When voltage changes between V1 and V6, the adjustment of b puts voltage under control. All the operating points constitute the straight line AB. The slope of AB and the crossover point with voltage axis Vref is determined by the control scheme of b. From voltage control point of view, the slope of AB is zero at best, without steady-state error. To maintain the control stability, SVC should have a small steady-state error and the slope of AB is around 0.05. Taking into consideration the steady-state control scheme, the SVC voltage– current characteristics are shown in Fig. 5.23. When system voltage varies within the SVC control range, SVC can be seen as a synchronous condenser having source voltage of Vref and internal reactance of Xe. V ¼ Vref þ Xe ISVC ;

ð5:122Þ

where Xe is the slope of the straight line AB in Fig. 5.23, V and ISVC are the SVC terminal voltage and current. When system voltage is out of the SVC control range, SVC becomes a fixed reactor, XSVCmin or XSVCmax.

308

5 HVDC and FACTS

SVC is considered as a variable shunt reactor in system stability and control analysis. SVC controller determines its admittance. Reference [122] provided the controller block diagram. We have introduced SVC basic principles. Special attention needs to be paid in industrial applications of SVC to capacity settings of reactors and capacitors, control strategy, flexibility of adjustments, protection, elimination of harmonics, etc. For example, in practical operation of an SVC, the range of the control angle is slightly less than [p/2, p] to make sure that valves can be triggered on and turned off securely.

5.5.2

Basic Principle and Mathematical Model of STATCOM

A STATCOM is also called an advanced static Var generator (ASVG). Its function is basically same as SVC with wider operation ranges and faster responses. As stated before, the control element of SVC is a thyristor, a semi-controllable element that can only be turned off when valve current crosses zero. STATCOM is made of fully controllable elements. Gyugyi et al. [123] presented the basic principles of using gate turn off thyristors (GTOs) to build a STATCOM. As yet, there have been several samples STATCOM operated in real systems [124–126]. The basic connection of a STATCOM is shown in Fig. 5.24. Its control element is the fully controlled valve (GTO). The ideal GTO switch characteristic is that the valve is turned on under positive valve voltage with positive control current on its gate; valve is turned off with negative control current on its gate. Valve resistor is zero when it conducts, and is infinity when it is turned off. A GTO can manage the switch-off by gate control in comparison with the thyristor where switch-off is only possible at current zero-crossing. STATCOM in Fig. 5.24 is a voltage type selfcommutation full-bridge inverter according to power electronic theory. The capacitor DC voltage acts as an ideal DC voltage source to support the inverter. The regular diode connected in the opposite direction and parallel with the GTO is a path for continuous current, providing route for the feedback energy from the AC side. The inverter normal operation is to transfer the DC voltage into AC voltage having controllable magnitude and phase angle at the same frequency as the AC system. The sum of instantaneous power of a symmetric three-phase system is a

vc +

. .

−

Fig. 5.24 Circuit of STATCOM

. .

. .

ia ib V ic ASVC •

5.5 Basic Principles and Mathematical Models of FACTS

309

constant. Thus the reactive power exchanges periodically within phases instead of between source and load. There is no need to have an energy storage element on the DC side, if considering the inverter as a load. However, the interacting power among harmonics produces a small amount of reactive power exchange between the inverter and the system. The capacitor on the inverter DC side will provide both DC voltage and energy storage. The electrical energy stored in the capacitor is 1 W ¼ CVC2 : 2 If the above energy is not considered as energy support for AC systems during power system dynamic events, the value of capacitor C can be small while the reactive capacity provided by the STATCOM is much more than the stored energy. We will see later that the maximum reactive power capacity of a STATCOM depends on the inverter capacity. The STATCOM does not need large size reactors and capacitors as the SVC does. Generally, there are three output voltage control modes for voltage-type inverters: phase-shift adjustment, pulse-width modulation, and direct DC source voltage control. The DC voltage of STATCOM is the charged voltage of the capacitor, not a DC source. So phase-shift adjustment and pulse-width modulation, instead of direct DC source voltage control, are usually used in STATCOMs. For brevity, we are not going to discuss the inverter working principles in detail. The width of the output voltage square waves y is controlled by the GTO gates (the magnitude of voltage square waves is the DC voltage on the capacitor). By Fourier analysis, we have the fundamental frequency voltage on the AC side y VASVG ¼ KVC sin ; 2

ð5:123Þ

where K is a constant related to inverter structure; VC is the capacitor DC voltage; y is the control variable. STATCOM connection to the systems is shown in Fig. 5.25. It must connect to systems through reactors or transformers because the use of voltage-bridge circuits. The connection reactor is needed to link the two unequal voltage sources, STATCOM and AC system. Its other function is to suppress the high-order harmonics in the current. Its inductance does not need to be large. The reactor in the figure is the transformer equivalent leakage reactance or the connection reactor. The resistor is the equivalent copper loss of the transformer or STATCOM loss. STATCOM is represented as an ideal synchronous condenser. Using the system voltage as the

•

VASVG

•

•

Fig. 5.25 STATCOM connection to systems

C

I P

r + jx

QASVG

VS

310

5 HVDC and FACTS

reference vector, the fundamental frequency component of the inverter output pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ voltage is VASVG and lagged phase angle is d. With y ¼ 1= r2 þ x2 , a ¼ arctg r/x, we have the real power consumed by the inverter as 2 P ¼ Vs VASVG y sinðd þ aÞ VASVG y sin a:

ð5:124Þ

The reactive power injection from STATCOM is QASVG

VASVG ﬀ d Vs _ _ ¼ Im Vs I ¼ Im Vs r jx

¼ Vs VASVG y cosðd aÞ Vs2 y cos a:

ð5:125Þ

In steady state, the inverter neither consumes nor generates real power. Based on (5.124), making P zero yields VASVG ¼ Vs

sinðd þ aÞ : sin a

ð5:126Þ

Taking (5.126) into (5.125) and (5.123) yields QASVG ¼ VC ¼

Vs2 sin 2d; 2r

Vs sinðd þ aÞ : K sin a sinðy=2Þ

ð5:127Þ ð5:128Þ

From the above two equations, we know that the adjustment of phase angle d while maintaining constant pulse width y can change the output reactive power as well as the capacitor voltage. The simultaneous adjustment of y and d can maintain capacitor voltage and change the reactive power output. The vector diagram of STATCOM steady-state operation is shown in Fig. 5.26. We use the equivalent resistance r to represent the inverter real power loss so that the inverter model neither consumes nor generates real power. In the diagram, compensation current I_ is perpendicular to inverter output voltage V_ ASVG . The inverter injects reactive power into the system when I_ leads V_ASVG . Otherwise it consumes reactive power. While an SVC changes its equivalent inductance through adjusting the timing of its connection to the system, the STATCOM controls the magnitude and phase-angle of its output voltage. As shown in the vector diagram, the reactive power provided by STATCOM is QASVG ¼ IVs cos d:

ð5:129Þ

5.5 Basic Principles and Mathematical Models of FACTS

311 •

•

I

jxI

•

VASVG •

d

•

d

jxI

•

VS

rI •

•

VASVG

•

rI

a STATCOM Inject Reactive

VS

•

I

b STATCOM Consumes Re-active Power from System

Power into System

Fig. 5.26 STATCOM steady-state vector diagram

Fig. 5.27 STATCOM voltage adjustment

VASVG1

I1r I0r

jI1x

VASVG0

jI0x Vs1

I2r

Vs0

Vref

jI2x

Vs2 I1

I0

I2

I

Note that d is the angle by which vector V_ASVG lags V_ s . The positive sign corresponds to a greater than zero d; the negative sign to a less than zero d. Substituting the above into (5.127) yields the magnitude of compensation current as I¼

Vs sin d: r

ð5:130Þ

The phase angle of the compensation current is (p/2d) as shown in Fig. 5.26. The real and reactive power components of the compensation current are IP ¼ I cos

p V Vs s d ¼ sin2 d ¼ ð1 cos 2dÞ; 2 r 2r

p V s IQ ¼ I sin d ¼ sin 2d: 2 2r

ð5:131Þ ð5:132Þ

Figure 5.27 shows the system voltage adjusted by STATCOM. Voltage Vs0 is the voltage setting value Vref under STATCOM output voltage of VASVG0 and compensation current of I0. When system operating conditions vary and the bus voltage

312

5 HVDC and FACTS V

Fig. 5.28 STATCOM volt– ampere characteristics

Vref

I ICmax

ILmax

decreases, STATCOM increases d to inject more reactive power. The compensation current is I1 while the voltage is maintained as Vref. The STATCOM keeps system voltage constant through the adjustment of its control parameters. A practical STATCOM usually implements bus voltage mismatch control. From the above analysis, the operation characteristics of STATCOM are shown in Fig. 5.28, and approach rectangular. The constraints of maximum voltage and current are determined by the STATCOM capacity. Voltage setting is determined by the control scheme. Comparing with SVC inverse triangular operational characteristics, STATCOM has wider operation ranges. We must notice that only one of the two control variables of the STATCOM is independent. The adjustment of d will change both the magnitude and phase angle of the compensation current. The control variable y is constrained by (5.128). As d changes, y should vary accordingly to maintain a constant capacitor voltage. The range of d variation is very limited. When a STATCOM consumes reactive power from the system, V_ s lags V_ ASVG by d. We can see from the vector diagram in Fig. 5.26b that d is always less than a. The equivalent resistance r is much less than equivalent reactance x so that a is very small. When a STATCOM injects reactive power into the system, V_ s leads V_ ASVG by d. As seen in (5.130) a small r makes d less constrained by compensation current. Hence (5.132) indicates an approximately linear relationship between reactive compensation current and d. To neglect resistance r for approximate analysis, setting a and d to zero in (5.124) and (5.125) yields P ¼ 0;

QASVG ¼ Vs

VASVG Vs : x

Now the free control variable of STATCOM is y, and VASVG is determined by (5.123). If VASVG is greater than VS, the STATCOM injects reactive power into the system, otherwise it consumes reactive power. A STATCOM can be represented as a shunt connected, controllable current source as noted in (5.130) for power system stability and control analysis. The magnitude and phase angle are determined by the STATCOM controller.

5.5 Basic Principles and Mathematical Models of FACTS

5.5.3

313

Basic Principle and Mathematical Model of TCSC

TCSC can rapidly and continuously change the equivalent reactance of the compensated line, to maintain a constant power flow on the line within certain operating conditions. In system transients, the TCSC can increase system stability through its fast variation of line reactance. The earliest TCSC was first put into operation in USA in 1991. TCSC have many different structures. One of its basic formations is shown in Fig. 5.29. Figure 5.29 shows a fixed capacitor and a parallel connected TCR. Its control element is the thyristor. We have seen TCR utilization in the above analysis of SVC. Since SVC is shunt-connected, the voltage on TCR is considered to be sinusoidal. However, the current flowing through TCR is irregular due to valve control, as shown in Fig. 5.21b. The TCR in a TCSC operates in different conditions as compared to those in an SVC. Note that the TCSC is series connected in the transmission line. The current flowing through the TCSC, the line current, is sinusoidal, due to harmonic filtering requirements and to physical operating constraints. Hence the irregular current in the TCR due to valve control will generate a nonsinusoidal capacitor voltage. This is the main difference between the two. Below we introduce the TCSC equivalent reactance at fundamental frequency to understand its working and control principles. The reference directions of various physical variables are shown in Fig. 5.29. The line current is sinusoidal with the waveforms shown in Fig. 5.30a. i ¼ Im sin ot

ð5:133Þ

Suppose that the circuits are in steady state. When the valve conducts, we have the following equations based on circuit theory i ¼ iL þ i C ;

v¼L

diL ; dt

iC ¼ C

dv : dt

ð5:134Þ

From the above we have iL þ LC

d 2 iL ¼ Im sin ot: dt2

ð5:135Þ

This is a nonhomogeneous differential equation of the inductor current. Its particular solution is the steady-state solution of the second-order circuit as follows: i

v

iC C

Fig. 5.29 Basic structure of TCSC

iL

L

314

5 HVDC and FACTS

V

i = Imsinωt π 2

0

π

3π 2

5π 2

wt

a iL

0 α−π 2

π 2

π 2 +α

3π 2

5π − α 2 wt

3π − α 2

b Fig. 5.30 (a) TCSC line current and capacitor voltage waveforms (b) inductor current waveforms

9 > l2 ¼ D sin ot; D ¼ 2 Im = : lp ﬃﬃﬃﬃﬃﬃ1 > ; l ¼ o0 =o; o0 ¼ 1= LC

isL

ð5:136Þ

The complementary solution to homogeneous equation is ifL ¼ A cos o0 t þ B sin o0 t;

ð5:137Þ

where A and B are the undetermined coefficients. The general solution to (5.135) is iL ¼ A cos o0 t þ B sin o0 t þ D sin ot:

ð5:138Þ

Denote a as the firing angle and assume its value in [p/2, p]. A is the electrical angle from capacitor voltage crossing zero to the time of firing. Under steady state, the waveform of inductor current is symmetric to the time point of capacitor voltage crossing zero. The capacitor voltages at the moments of valve turning on and off are equal in magnitude and opposite in direction. Supposing that capacitor voltage magnitude is V0 when valves turn on and off, the corresponding electrical angles are p yk ¼ a þ kp; 2

dk ¼

3p a þ kp; 2

k ¼ 0; 1; 2; . . . :

ð5:139Þ

5.5 Basic Principles and Mathematical Models of FACTS

315

Based on the initial conditions of inductor current and capacitor voltage: inductor current of zero and capacitor voltage of V0 at turning on (refer to Fig. 5.30a), we have the following equations A cos lyk þ B sin lyk þ D sin yk ¼ 0; Lðo0 A sin lyk þ o0 B cos lyk þ oD cos yk Þ ¼ ð1Þk V0 :

ð5:140Þ ð5:141Þ

The capacitor voltage is V0 at the turning off time of the valve Lðo0 A sin ldk þ o0 B cos ldk þ oD cos dk Þ ¼ ð1Þkþ1 V0 :

ð5:142Þ

The solution to the above three equations yields sin yk 2k þ 1 cos l p ; cos lb 2

ð5:143Þ

sin yk 2k þ 1 B ¼ D sin l p ; cos lb 2

ð5:144Þ

V0 ¼ DLðo sin a þ o0 cos a tg lbÞ;

ð5:145Þ

A ¼ D

where b ¼ p a, is called the conducting angle having a value within [0, p/2]. Substituting A and B in (5.138) yields the inductor current when the valve conducts as

cos a p iL ¼ D sin ot þ ð1Þ cos l ot kp : cos lb 2 k

ð5:146Þ

We can obtain the capacitor voltage from (5.134) as cos a p v ¼ DL o cos ot ð1Þk o0 sin l ot kp : cos lb 2

ð5:147Þ

The conducting period is 3p p ot 2 a þ kp ; a þ kp ; 2 2

k ¼ 0; 1; 2; . . . :

The capacitor current iC ¼ i þ (iL). There are two components in capacitor current, one is the line current; the other has the same magnitude and opposite direction to the inductor current.

316

5 HVDC and FACTS

We have assumed that the firing angle is within [p/2, p]. The reason for such an assumption is the same as for the TCR in SVC. As seen from the waveforms of inductor current, to make one valve conduct at any time we have 3p p a þ kp ¼ þ a þ kp: 2 2 There is one valve turned on when the other is turned off, so a ¼ p/2. The inductor current, as indicated in (5.146), is iL ¼ D sin ot: This is the inductor current when the inductor connects directly with the capacitor in parallel. We usually call this bypass mode. When a increases from p/2 to p, the valve conducting period decreases from p to zero, corresponding to the turning-off of two valves at any moment. This is as if the inductor is not in operation, called off mode. Besides, if a is less than p/2, the time at which the current of the conducting valve crosses zero is later than the firing time of the other nonconducting valve. Thus 3p p a þ kp > a þ kp þ p: 2 2 In this case the nonconducting valve cannot be triggered on with a zero voltage across it at the time of firing since the conducting valve is not turned off. Thus one of the valves is always nonconducting at any time, making DC current the main component in inductor current. Under normal operation, the firing angle of TCR in TCSC has an operating range of [p/2, p]. When both valves are turned off ot 2

hp

i p a þ kp ; a þ kp ; 2 2

k ¼ 0; 1; 2; . . . :

The inductor current is zero while the capacitor current is the line current. The capacitor voltage is C

dv ¼ Im sin ot: dt

The solution is v¼K

Im cos ot: oC

ð5:148Þ

5.5 Basic Principles and Mathematical Models of FACTS

317

We know from (5.147) that the absolute value of capacitor voltage when the valve conducts is V0. Take ot ¼ a p/2 þ kp as the moment of valve turns on in the above, so ð1Þk V0 ¼ K

Im p cos a þ kp : oC 2

Obtaining integral constant K and substituting into (5.148) yields capacitor voltage as v ¼ ð1Þk

Im Im sin a þ V0 cos ot: oC oC

ð5:149Þ

Equations (5.147) and (5.149) give the capacitor voltages when valves are turned on and off, respectively. Apparently, the capacitor voltage is not sinusoidal when a 6¼ p/2. Figure 5.30a, b shows the waveforms of capacitor voltage and inductor current. The Fourier analysis of nonsinusoidal capacitor voltage provides the fundamental frequency component V1 ¼

Im Im sin a þ V0 cos y cos y dy oC oC 0

Z 3p=2a 2 cos a p þ DL o cos ot o0 sin l ot cos y dy : ð5:150Þ p ap=2 cos lb 2 Z 2 p Im Im þ V0 sin a cos y cos y dy p 3p=2a oC oC

2 p

Z

ap=2

The integral of the first item above equals the integral of the third. The sum of the two is F1 þ F3 ¼

4 Im V0 cos a ð2a p þ sin 2aÞ : p 4oC

ð5:151Þ

Taking into account (5.136), the integral of the second item is

2 o 2o0 2 F2 ¼ DL ob þ sin 2a 2 ðl tg a þ tg lbÞ cos a : p 2 l 1

ð5:152Þ

Substituting (5.145) into (5.151) and rearranging yields the fundamental frequency reactance of TCSC XTCSC ¼

V1 F1 þ F3 þ F2 ¼ ¼ Kb XC ; Im Im

ð5:153Þ

318

5 HVDC and FACTS

Fig. 5.31 Kb b curve

where XC ¼ 1=oC;

2 l2 2 cos2 b sin 2b Kb ¼ 1 þ ðltg lb tg bÞ b : p l2 1 l2 1 2

ð5:154Þ ð5:155Þ

As shown in (5.155), the adjustment of the firing angle changes the reactance XTCSC that is series connected in the line, rendering a controllable equivalent line reactance. The valve control scheme is predefined. TCSC ideal dynamic responses can allow the transmission line capacity to reach its thermal limit. TCSC usually has oL < 1/oC and l2 around 7 to reduce its cost. Figure 5.31 shows the Kb b curve at l ¼ 3. When b 2 [0, p/2l], Kb is greater than zero and TCSC is capacitive. When b 2 [p/2l, p/2], Kb is less than zero and TCSC is inductive. In the off mode, b ¼ 0, Kb ¼ 1. In by-pass mode, b ! p/2, Kb ! 1/ (1 l2). When b ! p/2l, Kb ! 1 due to tglb ! 1, corresponding to parallel LC resonance. To prevent TCSC resonance over voltage, b is prohibited from being operated near p/2l. TCSC shown in Fig. 5.29 is a single module. A practical TCSC usually consists of many modules connected in series. Each module has its independent firing angle. The firing angle combination of different modules gives the TCSC equivalent reactance a wider range of variation and smoother adjustment. To protect the TCSC from damage due to overvoltages and overcurrents, there are various protection devices installed and corresponding operation constraints [127]. For power system stability and control analysis, a TCSC can be represented as a variable reactor series connected in the transmission line. The reactance is determined by the TCSC controller.

5.5 Basic Principles and Mathematical Models of FACTS

5.5.4

319

Basic Principle and Mathematical Model of SSSC

TCSC is a series compensation device using semi-controllable power electronic elements. There are many types of series compensation with fully controllable elements. Here we are going to introduce the SSSC built with GTO voltage-type inverters. The STATCOM discussed before uses voltage-type inverters. Connected to systems in parallel via reactors or transformers; the SSSC employs voltage-type inverters connected in series in a transmission line through transformers. Neglecting the line-ground branches, the basic connection is shown in Fig. 5.32, where r þ jx is the line impedance. Note that the inverter is different to STATCOM as a DC source may be present on the DC side. With a DC source, SSSC can provide reactive power compensation as well as real power compensation to AC systems. When an SSSC only supplies or consumes reactive power, the capacity of its DC source can be small or even zero (the SSSC loss being provided by the AC system). We know (from the introduction of the STATCOM) that the magnitude and phase angle of inverter output AC voltages are controllable. Hence we can consider the voltage of an SSSC, connected in series on a line, as an approximately ideal voltage source, as shown in Fig. 5.33a. Denote VSSSC the voltage magnitude of the ideal voltage source and d the voltage leading phase angle regarding voltage at bus l. The vector diagram is shown in Fig. 5.33b, where ’ is the leading phase angle of voltage at bus l with regards to line current. Apparently r + jx

Fig. 5.32 SSSC basic connection

i

Voltage source inverter

DC power source Vl ′ l

a

I

VSSSC

Vl ′

r+jx

j b

I

Fig. 5.33 (a) SSSC equivalent circuit (b) SSSC vector diagram

VSSSC d

Vl

j

320

5 HVDC and FACTS

V_ l0 ¼ V_ l þ V_ SSSC :

ð5:156Þ

For pure reactive power compensation, the inverter vector V_ SSSC is perpendicular to line current I_ d þ ’ ¼ p=2:

ð5:157Þ

In this way, SSSC corresponds to a reactor connected in series on a transmission line, denoting XSSSC its equivalent reactance, so V_ l V_l0 ¼ jXSSSC I_ ¼ V_ SSSC ;

ð5:158Þ

XSSSC ¼ VSSSC =I:

When V_ SSSC leads I,_ it is capacitive with negative sign; otherwise it is inductive with positive sign. Note that in the above equation, VSSSC is not related to line current and is controlled by the inverter. Hence the adjustment of VSSSC can change the equivalent reactance. In system analysis, once XSSSC is given, the line current can be determined by V_ l V_ m : r þ jðx þ XSSSC Þ

ð5:159Þ

9 VSSSC ¼ IjXSSSC j = : p d ¼ ’; 2

ð5:160Þ

I_ ¼ Thus

When XSSSC is less than zero, use positive sign; otherwise use negative sign. Generally, the source branch in Fig. 5.33a can be represented as a current source and impedance connected in parallel as shown in Fig. 5.34a by Norton’s theorem. The current source is I_c ¼ V_SSSC =ðr þ jxÞ:

r+jx

l

Ic a

l

Plc+jQlc b

ð5:161Þ

r+jx

m

Pmc+jQmc

Fig. 5.34 (a) Use equivalent current source (b) use equivalent power injection

5.5 Basic Principles and Mathematical Models of FACTS

321

In power system analysis, the bus power injection is used in most cases, further simplifying Fig. 5.34a, b. As indicated in (5.161) 9 _ SSSC > V > > Plc þ jQlc ¼ ¼ V_l = r þ jx : V_ SSSC > > > Pmc þ jQmc ¼ V_ m I_c ¼ V_ m ; r þ jx V_ l I_c

Note that the phase angle of V_SSSC is yl þ d, so Plc ¼ Vl VSSSC ðb sin d g cos dÞ

)

Qlc ¼ Vl VSSSC ðg sin d þ b cos dÞ

;

Pmc ¼ Vm VSSSC ½g cosðylm þ dÞ b sinðylm þ dÞ Qmc ¼ Vm VSSSC ½b cosðylm þ dÞ þ g sinðylm þ dÞ

ð5:162Þ ) ;

ð5:163Þ

where 9 r > > > r 2 þ x2 > = x : b¼ 2 > r þ x2 > > > ylm ¼ yl ylm ; g¼

ð5:164Þ

The power generated by SSSC is PSSSC þ jQSSSC ¼ V_ SSSC I_ ¼ V_SSSC

_ VSSSC þ V_l V_m ; r þ jx

2 PSSSC ¼ gVSSSC þ gVSSSC ½Vl cos d Vm cosðylm þ dÞ þ bVSSSC ½Vl sin d Vm sinðylm þ dÞ,

ð5:165Þ 2 QSSSC ¼ bVSSSC þ gVSSSC ½Vl sin d Vm sinðylm þ dÞ

bVSSSC ½Vl cos d Vm cosðylm þ dÞ:

ð5:166Þ

Apparently PSSSC is zero for pure reactive power compensation. Neglecting line resistance, d satisfies the following for pure reactive power compensation Vl sin d ¼ Vm sinðylm þ dÞ:

ð5:167Þ

322

5 HVDC and FACTS

In (5.166), the reactive power compensated by SSSC is not related to the line current directly since the adjustment of VSSSC is not related to the line current. The power flows from bus l to bus m is: Plm þ jQlm ¼ V_ l I_ ¼ V_ l

V_SSSC þ V_l V_ m r þ jx

;

Plm ¼ gVl2 þ gVl ½VSSSC cos d Vm cos ylm bVl ½VSSSC sin d þ Vm sin ylm Qlm ¼ bVl2 gVl ½VSSSC sin d þ Vm sin ylm bVl ½VSSSC cos d Vm cos ylm

) :

ð5:168Þ So the power on the line is controlled by two parameters. For pure reactive power compensation, SSSC has only one independent control variable due to the constraint of (5.167) and has one control objective. In power system stability and control analysis, SSSC can also be represented as a voltage source connected in series in the line. The controller determines the magnitude and phase angle. The voltage vector is always perpendicular to line current for pure reactive power compensation.

5.5.5

Basic Principle and Mathematical Model of TCPST

Thyristor controlled phase shifting transformer is abridged as TCPST. Phase shifters using mechanical switches to change the tap positions have been utilized in power systems for a long time. It is also called a series voltage booster. Since the response speeds of mechanical switches are slow in tap changing, this type of shifter can only be used in power system steady-state adjustment. Furthermore, the short operational life is a major drawback of this type of shifter. Substituting mechanical switches with thyristors can provide the phase shifter with faster responses and wider application. There are many types of implementation [130, 131]. We are going to use a relative simple type to introduce the working principles and mathematical models. Figure 5.35 shows a basic connection of TCPST. Phase shifters consist of parallel transformer (ET), series transformer (BT), and switches. Parallel and series transformers are also called excitation transformer and boosting transformer. Figure 5.35 shows only phase c of the secondary side of parallel transformer and secondary and the primary side of the series transformer. The other two phases have the same structure. Switch S is made up of a pair of thyristors connected in parallel in opposite directions, having the same working principles as discussed in TCSC. S1–S5 can only have one conducting and all others are turned off under all circumstances. We can see that the ratio of the parallel transformer varies with the conducting conditions of S1–S4. When S1–S4 are all turned off, S5 must conduct to short-circuit the primary of the series transformer.

5.5 Basic Principles and Mathematical Models of FACTS

a b c

VE

323

I2a

*

I1b

I2b

*

I1c

I2c

I1a

I3b *a VEa1

x

VEb1

*

I3c

y

*

b VEc1

I3a c

z

s1 s2

IBC1

*

BT

*

IEC2

s3

ET

VP

s5 VBC1

s4 VEC2

Fig. 5.35 TCPST basic configuration

This is to prevent series connection of the series transformer excitation reactance into the transmission line. Notice the relationship between the primary voltage of the parallel transformer and the line phase voltage. Phase a, b, and c on the primary of the parallel transformer correspond to phase b, c, and a of the line voltage, respectively. Since the parallel transformer has D connection, the relationship between the primary voltages of the parallel transformer and the line phase voltages are 9 V_ Ea1 ¼ V_ Eb V_ Ec > = ð5:169Þ V_ Eb1 ¼ V_ Ec V_ Ea : > ; _ _ _ VEc1 ¼ VEa VEb Supposing that the ratios of parallel and series transformers are kE and kB, respectively, and neglecting the voltage loss of transformers, the phase voltages on the secondary of the parallel transformer have the following relationship with the line phase voltages 9 pﬃﬃﬃ V_Ea2 ¼ kE V_ Eb V_Ec = 3 ¼ jkE V_ Ea > > = pﬃﬃﬃ _ _ _ _ VEb2 ¼ kE VEc VEa = 3 ¼ jkE VEb : > pﬃﬃﬃ > V_ Ec2 ¼ kE V_ Ea V_ Eb = 3 ¼ jkE V_Ec ;

ð5:170Þ

The phase voltages on the secondary of the series transformer are 9 V_ Ba2 ¼ kB V_Ba1 ¼ kB V_ Ea2 ¼ jkB kE V_ Ea > = V_ Bb2 ¼ kB V_Bb1 ¼ kB V_ Eb2 ¼ jkB kE V_ Eb : > ; V_ Bc2 ¼ kB V_ Bc1 ¼ kB V_ Ec2 ¼ jkB kE V_Ec

ð5:171Þ

324

5 HVDC and FACTS VBa VPa

I 3a

VEa I2a

I1a

j

j

VPb VEc VBc

j VPc

j

VBb VEb

I1c I3c

j

j

I2c

I1b

I2b I3b

Fig. 5.36 Phase shifter vector diagram

Using a single phase expression to replace the above three-phase we have V_ B ¼ jkB kE V_ E ;

ð5:172Þ

where V_ E and V_B are input voltage of the parallel transformer and output voltage of the series transformer. Similarly we can obtain the expression for the currents I_3 ¼ jkB kE I_2 :

ð5:173Þ

The vector diagrams are shown in Fig. 5.36. From (5.172), (5.173), and Fig. 5.35 we can obtain V_ P ¼ V_ E þ V_B ¼ ð1 þ jkB kE ÞV_ E ;

ð5:174Þ

I_1 ¼ I_2 þ I_3 ¼ ð1 jkB kE ÞI_2 :

ð5:175Þ

Hence we can represent phase shifter as a transformer having complex ratio as follows: 9 _P > V > K_ P ¼ ¼ 1 þ jkB kE ¼ KP ﬀ ’ > > > = V_ E ’ ¼tg1 kB kE qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ KP ¼ 1 þ ðkB kE Þ2 ¼ sec ’

> > > > > ;

:

ð5:176Þ

Since the ratio of the parallel transformer, kE is related to the on/off state of switches S1–S5, we can change ’ by controlling switch states. Apparently ’ is a discrete variable. Note that the product of kE and kB is much less than 1. VP is a little larger than VE. The main function of phase shifters is to change the phase angle ’ of VE. Based on (5.174) and (5.175), we can use the phase shifter equivalent circuits shown in Fig. 5.37 for power system stability and control analysis.

5.5 Basic Principles and Mathematical Models of FACTS

325

Fig. 5.37 Phase shifter equivalent circuit

The above phase shifter is also called a quadrature boosting transformer (QBT) since its output voltage V_ B is always perpendicular to V_ E . A new type of phase shifter has been proposed to use a series voltage source V_ B , and its voltage magnitude and phase angle can both be continuously adjusted to provide easier implementation. Generally, this type of phase shifting transformer (PST) has the mathematical model shown in Fig. 5.37. The control variables are voltage V_ B magnitude and phase angle. Note that the phase shifter is an inactive element. Neglecting its loss, phase shifter output complex power equals its input complex power. So V_B I_2 ¼ V_ E I_3 :

ð5:177Þ

The power generated from the series transformer is consumed by the shunt current source. Thus V_ B I_3 ¼ ¼ k ﬀ ’; V_E I_B V_ B where k ¼ : V_E The adjustment of V_ B magnitude and phase angle can control k and ’. Both phase angle and magnitude of V_ P are controllable, distinguishing this from QBT. There are two independent control variables. This type of phase shifter is similar to a UPFC and we are not going to discuss it in detail.

5.5.6

Basic Principle and Mathematical Model of UPFC

The FACTS devices that we discussed above manipulate only one of the three parameters affecting power transmission. TCSC and SSC compensate a line parameter. SVC and STATCOM control a bus voltage magnitude. TCPST adjusts bus voltage phase angle. The UPFC [132] is a combination of the above FACTS devices and can adjust the three parameters at the same time. In June 1998, the first UPFC was put into trial operation at AEP in the United States. Its application and control strategy are still under investigation. The basic structure of UPFC is shown in Fig. 5.38.

326

5 HVDC and FACTS VEt

VBt I1

I3 = It + Iq

I2

VP

TB

TE Series Converter

Parallel Converter

IB

IE VE

VEa

VEb

Cdc VEc

Idc vdc

+

VBc

VBb

VBa

VB

Variable Reference

Parameter setting

Controller Measurements

Fig. 5.38 UPFC basic configuration

UPFC is like a combination of SATCOM and SSSC. The two GTO voltage inverters share a capacitor to couple the STATCOM and the SSSC. Nabavi-Niaki and Iravani [133] presented a dynamic model of UPFC at fundamental frequency and for symmetrical operation. The converters utilize sinusoid pulse width modulation (SPWM). Here we are going to introduce this model. The control variables of SPWM are modulation ratio and phase angle of sinusoidal control signal. As shown in Fig. 5.38, we can separate UPFC into an AC part and a DC part using transformers as the delimiters. The output voltages of the two converters are 1 V_ E ¼ pﬃﬃﬃ mE vdc ﬀ dE 2 2

mE 2 ½0; 1;

ð5:178Þ

1 V_ B ¼ pﬃﬃﬃ mB vdc ﬀ dB 2 2

mB 2 ½0; 1;

ð5:179Þ

where mE and mB are parallel and series converter modulation ratios; dE and dB are the phase angles of sinusoidal control signals; vdc is the instantaneous voltage on the DC capacitor. It is not difficult to understand that there is the following relation between the variation rate of electrical energy on the capacitor and the real power of the converter Cdc vdc

dvdc ¼ Re V_ E I_E V_ B I_B ; dt

ð5:180Þ

where I_E and I_B are AC currents on the parallel and series converters. In power system stability analysis, we use a sub-steady-state model for power networks.

5.5 Basic Principles and Mathematical Models of FACTS

327

Correspondingly, the AC currents of the converters and the voltages on AC side have the following relation using the reference direction in Fig. 5.38 ðrE þ jolE ÞI_E ¼ V_ Et V_ E ;

ð5:181Þ

ðrB þ jolB ÞI_B ¼ V_B V_Bt ;

ð5:182Þ

where impedances ZB and ZE are equivalent impedances of parallel and series transformers and the converter losses; V_Et and V_ Bt are the voltages transferred from UPFC terminal to converter side. Below we will convert (5.178)–(5.182) into the per unit system. To determine the voltage base, we generally assume that the converter output voltages are rated values when vdc reaches its rated value vdcN and modulation ratios approach 1. In the design of UPFC physical parameters, we have 0 VEN

¼ TE VEN

0 VBN

¼ TB VBN

9 pﬃﬃﬃ > 2 > ¼ TE 1 vdcN ¼ kE VN > = 4 pﬃﬃﬃ ; > 2 > > ¼ TB 1 vdcN ¼ kB VN ; 4

ð5:183Þ

where TE and TB are parallel and series transformer ratios; VN is the AC network 0 0 rated voltage; VEN and VBN are the AC side voltages transformed from rated converter output voltages; kE and kB are the two parameters of UPFC. Due to voltage static security constraints, kE and kB must not be too large, for instance, 1.2 and 0.3, respectively. As seen from the above two equations, ratios of parallel and series transformers are different. Having VN as the voltage base of the AC network, we use the following voltage bases for converters to work with the network per unit system while considering (5.183): VEB ¼

VN vdcN ¼ pﬃﬃﬃ ; TE 2 2 k E

ð5:184Þ

VBB ¼

VN vdcN ¼ pﬃﬃﬃ : TB 2 2 k B

ð5:185Þ

We now have the corresponding current base and impedance base of converters from the above voltage base. DC voltage base is vdcN. The expressions of (5.178) and (5.179) in the per unit system are V_ E ¼ kE mE vdc ﬀ dE ;

mE 2 ½0; 1;

ð5:186Þ

V_B ¼ kB mB vdc ﬀ dB ;

mB 2 ½0; 1:

ð5:187Þ

328

5 HVDC and FACTS

In the per unit system, (5.181) and (5.182) are unchanged. On dividing the two sides of (5.180) by the power base leaves, the right side is unchanged and the left side is Cdc vdc SB Cdc vdc SB

dvdc 2 1 vdc dvdc =vdcN 2 ¼ Cdc vdcN ; dt SB 2 vdcN dt dvdc 2 1 vdc dvdc =vdcN dvdc ¼ Cdc v2dcN ¼ vdc Tu dt SB 2 vdcN dt dt

where Tu is the UPFC time constant with the following value Tu ¼

2W 2 1 ¼ Cdc v2dcN : SB SB 2

ð5:188Þ

The time constant of a UPFC is related to the rated electrical energy stored in the DC capacitor. Equation (5.180) in the per unit system becomes vdc Tu

dvdc ¼ Re V_ E I_E V_B I_B : dt

ð5:189Þ

For the convenience of expression, we remove the subscripts for per unit system. In power system stability analysis, we need to use the two algebraic equations together with network equations. Substituting (5.186) and (5.187) into (5.181), (5.182), and (5.189), and separating real and imaginary parts we have

rE xE

xE rE

rB xB

xB rB Tu

IEx IEy IBx IBy

VEtx kE mE vdc cos dE ¼ ; VEty kE mE vdc sin dE

VBtx kB mB vdc cos dB ¼ ; VBty kB mB vdc sin dB

ð5:190Þ

dvdc ¼ kE mE IEx cos dE þ IEy sin dE dt kB mB IBx cos dB þ IBy sin dBy :

ð5:191Þ

ð5:192Þ

Equations (5.190)–(5.192) constitute UPFC dynamic models in the per unit system. In steady-state operation, UPFC is an inactive device and has constant capacitor voltage, so Re V_ E I_E V_B I_B ¼ 0:

ð5:193Þ

5.5 Basic Principles and Mathematical Models of FACTS VEt

I1 I3

VB

ZB I2

VP

VEt

Pc + jQc

VB

I1 It

VE

329

ZB I2

Iq

VP

Pc + jQc

PE = PB

ZE

a

b

Fig. 5.39 Equivalent circuit of UPFC

Hence, UPFC can be represented as two branches having impedance, connected in series with ideal voltage sources, as shown in Fig. 5.39. V_ B and V_ E are adjusted by GTO gate control signals from the parallel and series converters. Parallel branch current I_3 can be separated into two components I_t and I_q as shown in Fig. 5.39. V_ Et V_ E I_3 ¼ ¼ I_t þ I_q ; ZE

ð5:194Þ

where I_t and I_q components are in phase and perpendicular to bus voltage V_Et . The parallel branch power is PE ¼ Re V_Et I_3 ¼ V_ Et I_t ¼ VEt It ;

ð5:195Þ

jQE ¼ jIm V_Et I_3 ¼ V_Et I_q ¼ jVEt Iq :

ð5:196Þ

In (5.195), we use a negative sign when current is in opposite phase with voltage. In (5.196), we use a positive sign when current leads voltage; otherwise a use negative sign. For the parallel branch, the magnitude and phase of V_ E determine the magnitudes of It and Iq from (5.194). We can see from the above two equations that Iq is the reactive power component of the parallel branch to provide parallel reactive power compensation as in the STATCOM; It is the real power component to consume or inject real power into the AC system. This is to maintain constant DC voltage Vdc and make the phase of the series voltage source V_ B to be 360 controllable. The power generated from the series voltage source is SB ¼ PB þ jQB ¼ V_ B I_2 :

ð5:197Þ

If we control the phase of V_B to make it perpendicular to line current, the function of the series voltage source is like the SSSC series compensation. Generally, the phase and magnitude of V_ B are fully controllable. TCPST and SSSC do not have this kind of capability. Such a function of UPFC comes from the fact that the real power in

330

5 HVDC and FACTS

(5.197) is provided by the parallel branch. The control of two voltage sources under the conditions of (5.193) means that the real power generated or consumed by the series voltage source equals to that consumed or generated by the parallel voltage source. Apparently in this case, the electric field energy stored in the DC capacitor does not change and DC voltage is constant. This is the steady state of the UPFC. On the basis of the above analysis, the relationships for the variables in Fig. 5.39 under steady-state conditions are given:

arg I_t ¼

V_ p ¼ V_ Et þ V_B I_2 ZB ;

ð5:198Þ

I_2 ¼ I_1 I_t I_q ;

ð5:199Þ

V_ Et V_ E I_t þ I_q ¼ ; ZE

ð5:200Þ

It ¼ Re V_B I_2 =VEt ;

ð5:201Þ

(

arg V_ Et þ 0 arg V_ Et þ p

Re V_ B I_2 0 ; Re V_ B I_ > 0

arg I_q ¼ arg V_ Et p=2;

ð5:202Þ

2

ð5:203Þ

where arg represents the phase angle of the vector. Equations (5.201) and (5.202) correspond to (5.193). Although the magnitudes and phases of voltage sources V_ B and V_E can be continuously adjusted, the constraint of (5.193) reduces the number of independent variables from four to three 9 0 VB VB max > = 0 ’B 2p ; > ; 0 Iq Iq max

ð5:204Þ

where ’B is the phase angle of parallel voltage source; VBmax and Iqmax are constants related to UPFC rated capacity. The phase vector diagram for UPFC steady-state operation is shown in Fig. 5.40. For the convenience of analysis, ZB is ignored in the vector diagram. The series connected V_ B changes the bus voltage from V_ Et to V_ p . The change of V_ B makes V_ p vary within the circle centered at V_ Et to control the real and reactive power on the line directly. Note that the compensation of I_q makes the magnitude of V_ Et controllable by the UPFC. One UPFC has three independent control variables to manipulate three operation variables Pc, Qc, and VEt. The steady-state equivalent circuit of UPFC can also be represented as shown in Fig. 5.39b.

Thinking and Problem Solving

331 Vp VB VEt

ϕB

Iq VB max It

o

I1 −Iq

Iq max

I2

Reference phase vector

−It

Fig. 5.40 Phase vector diagram for UPFC

We know from the previous analysis of STATCOM and SSSC that both of them require DC voltages VC to be constant in steady-state operation. The converter AC voltage VASVG of a STATCOM is perpendicular to the AC current flowing out of the system; the VSSSC of SSSC is perpendicular to the AC current on the transmission line. VASVG is to satisfy (5.126) and VSSSC to satisfy (5.160). Their phase angles cannot be freely adjusted. Although UPFC still needs to maintain a constant DC voltage, the coupling between the two converters through the DC capacitor allows the real power consumed by the STATCOM to be sent back through the SSSC or vice versa. The magnitude and phase angle of series transformer output V_ B can then be freely adjusted. The parallel transformer can provide not only reactive power compensation but also the real power transfer between the system and the series transformer. The functional difference between UPFC and PST is due to (5.193) and (5.177). Iq in UPFC is a free variable. For PST, the real power and reactive power taken from the system by the parallel branch are injected into the system by the series branch due to the constraint (5.177). Hence UPFC has STATCOM function while PST does not.

Thinking and Problem Solving 1. What are the factors that limit power transmission distance and capacity? 2. What are the advantages and disadvantages of AC transmission? 3. Discuss the advantages and disadvantages of DC transmission and the applications for which DC transmission is more suitable. 4. What are the characteristics of other new power transmission modes being studied at present? 5. What is the free load flow?

332

5 HVDC and FACTS

6. Why should flexible electrical power systems be introduced? 7. Can we contemplate Id < 0 when V2d > V1d in (5.1)? Why? 8. Why can DC transmission lines only transmit active power, but the converters absorb reactive power from the AC system? 9. What is the physical significance of phase-shifter resistance Rg in (5.21)? 10. Distinguish trigger delay angle, phase-shifter angle, extinguish angle, trigger lead angle, and extinguish lead angle, paying attention to their operating areas. 11. Discuss the steady load flow control method of DC transmission. 12. Compare load flow calculation models with and without DC transmission lines. 13. Give an appropriate value L, C, and draw the curve of SVC equivalent reactance XSVC–b denoted in (5.121). 14. Draw V–b curve according to (5.121), in which V 2 [0.9,1.1], when per unit value Vref ¼ 1.05, Xe ¼ 0.05 in (5.122). 15. Draw the equivalent circuit diagram when S5 is also tripped with S1–S4 tripping in Fig. 5.35. 16. In steady state, UPFC can be regarded equivalently as two voltage source converters (VSC), with voltage amplitude values VB and VE, respectively, and phase angles dB and dE, respectively. Analyze why UPFC can only control three operational variables (active power Pc, reactive power Qc, and nodal VEt on a transmission line) in the steady state? 17. Discuss the capacitive and inductive value range of XTCSC when conduction angle b should avoid the resonance region according to (5.153)–(5.155). Then discuss whether line transmission active power Pc ¼ Plp controlled by TCSC can vary continuously.

Chapter 6

Mathematical Model of Synchronous Generator and Load

6.1

Introduction

The continuous increase of power system complexity and installation of more and more new equipment in power systems has demanded better methods for power system analysis, planning, and control. At present, analysis of modern power systems is generally based on digital computers. Hence, establishment of a mathematical model, describing the physical processes of a power system, is the foundation for the analysis and investigation of various power system problems. Correct and accurate computation for power system analysis requires a correct and accurate mathematical model of the power system. Transient processes of the power system are very fast. This is why power system operation heavily relies on the applications of automatic control. With the installation of many different automatic control devices, the operation of which largely depends on the application of electronic and computing technology, modern power system operation has reached a very high level of automation. For such large-scale and complex systems, the mathematical description is nonlinear and high dimensional, consisting of a large number of nonlinear equations. Hence it is both appropriate and practical that the analysis and computation of such a system ought to start from simple local devices and be completed finally for the complex overall system. Therefore, in modeling of large-scale and complex power systems, these systems are first decomposed into independent basic components, such as synchronous generators, transformers, transmission lines, governors and automatic voltage regulators (AVR), etc. Then those components are modeled separately according to circuit theory or other related principles. Models of those components are building-bricks to construct the mathematical model of whole power systems. For the study of different problems on the same system, different models are required. Mathematically, a power system is a nonlinear dynamic system. When the steady-state operation of a dynamic system is studied, the mathematical description of the system is in the form of algebraic equations. Differential equations (sometimes partial differential equations) give a mathematical description of system

X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

333

334

6 Mathematical Model of Synchronous Generator and Load

dynamics. For the study of some specific problems, model parameters could be time-variable and variables may not be continuous. In addition, to meet the requirement of different computing accuracy, different models could be used. Obviously, a mathematical model for qualitative analysis could be simpler than that for quantitative analysis. Computing accuracy and speed are always two conflicting factors which need to be considered carefully when a power system model is established. The more accurate the computation is, the more the computing work and hence the longer the computing time. On the other hand, to sacrifice some computing accuracy will be compensated with high computing speed, which has been a common practice in modeling power systems and developing computing algorithms. In this aspect, the effort has been to develop a mathematical model of a power system and the associated solution methods, such that the need for both computing accuracy and computing speed is met. Often, the result is a compromise between those two requirements based on available computing tools. There are two major issues in mathematical modeling. The first one is to describe a subject under investigation mathematically in the form of equations. There are two methods to establish those mathematical equations. The first method, the analytical method, is to derive those mathematical descriptions by using special knowledge and theory about the subject; the second is to identify them by carrying out experiments or using data obtained from its operation. That is the method of system identification in control theory. The second major issue in mathematical modeling is to obtain parameters of the mathematical description of the subject. No matter whether the plant is described by algebraic or differential equations, various parameters in those equations need to be obtained. Generally, for simple components of the subject, model parameters can be derived from design parameters according to certain physical (such as mechanical or electrical) principles. For example, four parameters of an overhead line, i.e., resistance, inductance, capacitance, and conductance to earth, can be obtained by applying electromagnetic theory to the way the line is arranged in space, the materials of the line, and the natural environment where the line is located. That is a typical analytical method. However, for complex components or systems, usually there is certain difference between the actual parameters and design parameters. A typical example is the generator parameter which could be affected by variations of power system operating conditions, saturation, and a series of complex conversion processes among mechanical, electrical, magnetic, and thermal energy. Therefore, in addition to the method of theoretical derivation, there is another important way to obtain model parameters of complex components and systems. This is the method of parameter estimation which is one of the methods in system identification. Parameter estimation and system identification is a research field which will not be discussed in this book. In Chap. 1, the mathematical model of a power network has been introduced. Mathematical models of HVDC and FACTS are discussed in Chap. 5. Hence, in this chapter the focus is the introduction to mathematical models of generator and load, including the mathematical models of synchronous generator, excitation systems, and governing systems.

6.2 Mathematical Model of Synchronous Generator

6.2

335

Mathematical Model of Synchronous Generator

The dynamics of a synchronous generator is the basis for the study of the dynamic behavior of the power system. In the history of developing the mathematical model of a synchronous generator, two milestones are the establishment of two-reaction theory in 1920s [146, 147] and the proposal of Park’s transformation [148]. Under the ideally assumed conditions and by using two-reaction principle, Park derived the basic mathematical equations of a synchronous generator in dq0 coordinate system. Since then, mathematical models of synchronous generators have been based on Park’s contribution with further major development regarding the number of equivalent windings to model the generator rotor winding, different assumptions about when a synchronous generator should be described by transient or subtransient parameters, different ways to describe magnetic saturation, etc. Details of all these points above can be found in [149–152]. In this section, we shall focus on those mathematical descriptions of the synchronous generator which have been widely used. Readers should note that in other references, different symbols, defined positive directions of physical variables, form of transformation matrix, and selection of base values may be used. From the structure of a synchronous generator we know that on the rotor, the field winding is a physical winding; while damping windings may just be electrically equivalent windings. For a salient-pole generator, damping windings represent the damping function of damping rods distributed on the rotor. While for a round rotor generator, they simulate the damping function produced by the eddy current inside the whole rotor. Since they are just equivalent windings, the damping function can be represented by a single or multiple damping windings. In theory, the more the equivalent damping windings, the more accurate the representation can be. However, if more equivalent damping windings are used, there could be two problems. The first is the increase of the order of differential equations in the mathematical model, adding computational burden for their solution. The second problem is that it is more difficult to obtain the relative electrical parameters accurately. Hence in the commonly used mathematical model of a synchronous generator, the number of equivalent damping windings is usually not more than three. Since the damping rods on the rotor of a salient-pole generator are more like real windings than the whole rotor of a round rotor generator and the magnetic circuit of the salient-pole generator is different in d and q directions, the damping function of the salient-pole generator is usually represented by two damping windings, one in the direction of direct axis (d), denoted as D damping winding and another in that of the quadrature axis (q), denoted as Q damping winding; For the round rotor generator, in addition to D and Q damping windings, one more equivalent damping winding in the quadrature direction (g winding) is used. Q and g winding represents the weaker and stronger eddy current effect, respectively. According to the theory of electric machines, the ideal assumptions about the synchronous generators are that the magnetic circuits are symmetrical, saturation is negligible, and flux waveforms have sinusoidal space distribution. In the following,

336

6 Mathematical Model of Synchronous Generator and Load

we shall first derive the mathematical model of an ideal synchronous generator with D, g, and Q damping winding, followed by introduction of a method considering magnetic saturation effects.

6.2.1

Basic Mathematical Equations of Synchronous Generator

6.2.1.1

Three-Phase Mathematical Equations

Figure 6.1a, b shows the structure of a synchronous generator and winding circuit diagram. We consider the general case of a salient-pole generator with D, g, Q three damping windings and treat a round rotor generator as a special case since it has only D, Q two damping windings. In the figures, the defined positive direction of voltage, current, and magnetic flux is related to the three-phase armature windings abc, field winding f and damping winding D, g, Q. It must be pointed out that the positive direction of magnetic flux related to the three-phase armature windings is opposite to that induced by the armature current of each winding in the positive direction; while magnetic flux associated with rotor windings is defined in the same direction as that induced by the current in each winding in the positive direction; q-axis leads d-axis by 90 in the rotational direction of the generator rotor. In addition, the positive direction of all flux axes is chosen to be those of the corresponding magnetic flux.

α

ia

L aa

b

+

D

−

ω

−

−

f b

a

−

q

c

−

if R f Lff

f ⫻

−

D

⫻

Rb

Lbb ib

c

⫻

Vf +−

R

−

z

⫻

g − Q − − ⫻

D

−

a

c

x

⫻

D

⫻ Q X g ⫻ X

Ra

−

iD R D

Lc

d

ic

LDD

⫻

vc Lgg

y

c

Rg

ig

vb va

LQQ RQ iQ

b

Fig. 6.1 Structure of synchronous generator and winding circuit. (a) Structure of synchronous generator (b) winding circuit

6.2 Mathematical Model of Synchronous Generator

337

From Fig. 6.1b, the following voltage equation for all the windings can be obtained 2

3 2 Ra va 6 vb 7 6 0 6 7 6 6 vc 7 6 0 6 7 6 6 7 6 6 7 6 6 vf 7 ¼ 6 0 6 7 6 6 0 7 6 0 6 7 6 4 0 5 4 0 0 0

0 Ra 0

0 0 Ra

0 0 0

0 0 0

0 0 0

0 0 0 0

0 0 0 0

Rf 0 0 0

0 RD 0 0

0 0 Rg 0

3 2 3 ’a ia 76 ib 7 6 ’b 7 76 7 6 7 76 ic 7 6 ’c 7 76 7 6 7 76 7 6 7 76 7 þ p6 7; 6 if 7 6 ’f 7 0 7 76 7 6 7 6 7 6 ’D 7 0 7 76 iD 7 6 7 4 ’g 5 0 54 ig 5 ’Q RQ iQ 0 0 0

32

ð6:1Þ

where p ¼ dtd denotes the differentiation operator. For an ideal synchronous generator, magnetic saturation effects can be ignored. Hence magnetic flux linkage of each winding can be written in the form of selfinductance and mutual inductance as shown by the following flux linkage equation 2 3 2 32 3 ’a Laa Mab Mac Maf MaD Mag MaQ ia 6 ’b 7 6 Mba Lbb Mbc 6 7 Mbf MbD Mbg MbQ 7 6 7 6 76 ib 7 6 ’c 7 6 Mca Mcb Lcc 76 ic 7 M M M M cf cD cg cQ 6 7 6 76 7 6 7 6 76 7 6 7¼6 76 7: ð6:2Þ 6 ’f 7 6 Mfa Mfb Mfc 6 7 Lff MfD Mfg MfQ 7 6 7 6 76 if 7 6 ’D 7 6 MDa MDb MDc 7 6 MDf LDD MDg MDQ 76 iD 7 6 7 6 7 4 ’g 5 4 Mga Mgb Mgc Mgf MgD Lgg MgQ 54 ig 5 ’Q MQa MQb MQc MQf MQD MQg LQQ iQ From circuit theory we know that the above coefficient matrix is symmetrical. From Fig. 6.1a we can see that due to the rotor rotation, the reluctance of the magnetic circuit of some windings changes periodically with the variation of rotor position. Hence the self-inductance and mutual inductance of those windings are a function of rotor position. According to the assumptions of an ideal synchronous generator, both the magnetomotive force (mmf) induced by armature current and mutual flux between armature windings and rotor windings have sinusoidal space distribution. Rotor position can be described by the angle between d-axis and flux axis of phase a armature winding y ¼ y0 þ ot. Hence the self-inductance and mutual inductance of each winding can be expressed as follows [153]. 1. Self-inductance and mutual inductance of armature windings 9 Laa ¼ l0 þ l2 cos 2y > = Lbb ¼ l0 þ l2 cos 2ðy 2p=3Þ ; > ; Lcc ¼ l0 þ l2 cos 2ðy þ 2p=3Þ

9 Mab ¼ ½m0 þ m2 cos 2ðy þ p=6Þ > = Mbc ¼ ½m0 þ m2 cos 2ðy p=2Þ : > ; Mca ¼ ½m0 þ m2 cos 2ðy þ 5p=6Þ

ð6:3Þ

ð6:4Þ

338

6 Mathematical Model of Synchronous Generator and Load

Under the assumptions of an ideal synchronous generator, it can be proved that l2 ¼ m2. Furthermore, for a round rotor generator, reluctance of magnetic circuits related to the self-inductance and mutual inductance of armature windings does not vary with rotor rotation, and we have l2 ¼ m2 ¼ 0. Hence those selfinductance and mutual inductance above are constant. 2. Mutual inductance between armature and rotor windings 9 9 Maf ¼ maf cos y > > = MaD ¼ maD cos y = Mbf ¼ maf cosðy 2p=3Þ ; MbD ¼ maD cosðy 2p=3Þ ; > > ; ; Mcf ¼ maf cosðy þ 2p=3Þ McD ¼ maD cosðy þ 2p=3Þ 9 9 Mag ¼ mag sin y > > = Mag ¼ maQ sin y = Mbg ¼ mag sinðy 2p=3Þ ; Mbg ¼ maQ sinðy 2p=3Þ : > > ; ; Mcg ¼ mag sinðy þ 2p=3Þ Mcg ¼ maQ sinðy þ 2p=3Þ

ð6:5Þ

ð6:6Þ

3. Self-inductance and mutual inductance of rotor windings Since rotor windings rotate with the generator rotor, for salient-pole or round rotor generator, reluctance of magnetic circuits does not vary with the change of rotor position. Hence self-inductance and mutual inductance of rotor windings are constant. D, f winding on direct axis (d) is vertical to g, Q winding on quadrature axis (q). Hence mutual inductance between them is zero, that is Mfg ¼ MfQ ¼ MDg ¼ MDQ ¼ 0: 6.2.1.2

ð6:7Þ

Basic Equations in dq0 Coordinate

From the discussion above we know that the self-inductance and mutual inductance of generator windings are not constant and some of them vary with the position of the generator rotor. Hence (6.1) and (6.2) are time-variant differential equations which are difficult to solve. To transfer these into time-invariant differential equations, some of methods of coordinate transformation have been proposed, among which the dq0 transformation proposed by Park [148] has been most widely used. In dq0 coordinate, flux linkage equations become time invariant. Hence the mathematical model of a synchronous generator is presented as a group of timeinvariant differential equations. In the following, we shall discuss the details of Park’s transformation. Park’s transformation converts three-phase flux linkage, armature current, and voltage into d, q, 0 components in the dq0 coordinates, through an equivalent coordinate transformation. It can be written as 2

3 2 Ad cos y 2 4 Aq 5 ¼ 4 sin y 3 A0 1=2

32 3 cosðy 2p=3Þ cosðy þ 2p=3Þ Aa sinðy 2p=3Þ sinðy þ 2p=3Þ 54 Ab 5: 1=2 1=2 Ac

ð6:8Þ

6.2 Mathematical Model of Synchronous Generator

339

For simplicity of expression, the equation above can be written in the compact form as follows Adq0 ¼ PAabc :

ð6:9Þ

The inverse Park’s transformation is 2

3 2 Aa cos y sin y 4 Ab 5 ¼ 4 cosðy 2p=3Þ sinðy 2p=3Þ Ac cosðy þ 2p=3Þ sinðy þ 2p=3Þ

32 3 1 Ad 1 54 Aq 5 1 A0

or Aabc ¼ P1 Adq0 :

ð6:10Þ ð6:11Þ

In (6.8)–(6.11), symbol A represents current, voltage, or flux linkage, i.e., idq0 ¼ Piabc ; vdq0 ¼ Pvabc ; Cdq0 ¼ PCabc ;

ð6:12Þ

iabc ¼ P1 idq0 ; vabc ¼ P1 vdq0 ; Cabc ¼ P1 Cdq0 :

ð6:13Þ

Applying the transformations of (6.12) and (6.13) as well as (6.3)–(6.7), (6.1) and (6.2) can be converted into the following equations in dq0 coordinates 2

3 2 Ra vd 6 vq 7 6 0 6 7 6 6 v0 7 6 0 6 7 6 6 vf 7 ¼ 6 0 6 7 6 607 60 6 7 6 405 40 0 0 2

0 Ra 0 0 0 0 0

0 0 Ra 0 0 0 0

0 0 0 0 0 0 Rf 0 0 RD 0 0 0 0

3 2 ’d Ld 0 6 ’q 7 6 0 L q 6 7 6 6 ’0 7 6 0 0 6 7 6 6 7 6 6 7¼6 6 ’f 7 6 3maf =2 0 6 7 6 6 ’D 7 6 3maD =2 0 6 7 6 4 ’g 5 4 0 3mag =2 ’Q 0 3maQ =2

0 0 0 0 0 Rg 0 0 0 L0 0 0 0 0

32 3 2 3 2 3 ’d 0 id o’q 6 7 6 ’q 7 6 o’d 7 0 7 76 iq 7 6 7 6 7 6 i0 7 6 ’0 7 6 0 7 0 7 76 7 6 7 6 7 6 7 6 7 6 7 0 7 76 if 7 þ p6 ’f 7 6 0 7; ð6:14Þ 6 iD 7 6 ’D 7 6 0 7 0 7 76 7 6 7 6 7 4 ’g 5 4 0 5 0 54 ig 5 ’Q RQ iQ 0 32 3 maf maD 0 0 id 6 7 0 0 mag maQ 7 76 iq 7 7 6 0 0 0 0 76 i0 7 7 76 7 76 7; 7 6 Lf mfD 0 0 76 i f 7 7 6 7 mfD LD 0 0 7 7 6 iD 7 5 4 0 0 Lg mgQ ig 5 0 0 mgQ LQ iQ

ð6:15Þ

where Ld ¼ l0 þ m0 þ 3l2 =2; Lq ¼ l0 þ m0 þ 3l2 =2; L0 ¼ l0 2m0 ; Lf ¼ Lff ; LD ¼ LDD ; Lg ¼ Lgg ; LQ ¼ LQQ ; mgQ ¼ MfD ; mgQ ¼ MgQ and o ¼

dy is the angular speed of the synchronous generator. dt

ð6:16Þ

340

6 Mathematical Model of Synchronous Generator and Load

Park’s transformation, in fact, replaces three-phase armature windings by their three structurally equivalent windings – d winding, q winding, and 0 winding. The difference is that the magnetic flux axis of three-phase armature windings is stationary in space; while that of dq0 windings rotates in space at rotor speed. The positive direction of magnetic flux axis of d winding and q winding is as same as that of d- and q-axis of generator rotor, respectively, to describe the behavior of electrical variables in the direction of d- and q-axis; while 0 winding represents the zero-sequence component in the three-phase armature current, voltage and flux linkage. Ld, Lq, and L0 in (6.16) is the self-inductance of equivalent d, q, and 0 winding, corresponding to d, q, and 0 synchronous reactance, respectively. From (6.16) we can see that the coefficient matrix in (6.15) is a constant matrix. Hence the mathematical model of (6.14) of synchronous generator has been transformed into a set of time-invariant differential equations. Equation (6.14) indicates that the phase voltage of the synchronous generator consists of three parts. The first part is the voltage drop across the resistance of armature windings; the second is the EMF induced from the variation of flux linking the armature windings, which is usually called the transformer voltage of a synchronous generator; the third part is the EMF due to the rotation of the synchronous generator which is termed speed voltage. The value of speed voltage is much greater than that of transformer voltage. The coefficient matrix in (6.15) is nonsymmetrical, i.e., the mutual inductance between windings on generator rotor and d, q, and 0 winding is not reciprocal. That is caused by the transformation. If the current in rotor windings is multiplied by 3/2, or an orthogonal transformation matrix is adopted, these mutual inductance will become reciprocal. From (6.16) we can see that for a salient-pole generator Ld > Lq and round rotor generator Ld ¼ Lq because l2 ¼ 0. This difference makes it applicable to represent round rotor generators by the mathematical model of the salient-pole generator. According to the reference direction of current and voltage given in Fig. 6.1b, the total output power from the three-phase armature windings is po ¼ va ia þ vb ib þ vc ic ¼ vTabc iabc :

ð6:17Þ

Applying Park’s transformation to the equation above, from (6.13) we can obtain the output power from armature windings in dq0 coordinates to be 3 po ¼ ðP1 vdq0 ÞT ðP1 idq0 Þ ¼ ðvd id þ vq iq þ 2v0 i0 Þ: 2

6.2.1.3

ð6:18Þ

Per Unit Equations of the Synchronous Generator

The per unit system is commonly used in power system analysis and calculation due to its many advantages. Parameters of synchronous generator are also usually given in per unit. Hence we need to convert the mathematical model of synchronous generator of (6.14) and (6.15) using actual values of various variables into the per unit equations

6.2 Mathematical Model of Synchronous Generator

341

where variables are described by per unit values. When we introduce the per unit equations used in HVDC in 4.3.1, we have mentioned the principle that, in a per unit system, the base values of different physical variables must have the same relationship that they have when using actual values. Hence in a per unit system, some base values are defined by users and others are derived from the physical relationships among variables. Obviously, the difference in defining those base values by users will lead to different per unit systems. This book will adopt a widely used per unit system – ‘‘unit excitation voltage/unit stator voltage’’ base value system. Subscript B is still used to denote base values of various physical variables and ‘‘*’’ to represent per unit variables. Firstly we define the base value for generator speed to be the synchronous angular speed os. Because ot ¼ y and y is dimensionless (without base value), oBtB ¼ 1 which can lead to the base value for time t. Hence oB ¼ os tB ¼ 1=os

) :

ð6:19Þ

On the generator stator side, we define the magnitude of armature current and voltage as their base values. From the definition, we can derive the base values for power, impedance, and flux linkage as follows VB IB 3 SB ¼ 3 pﬃﬃﬃ pﬃﬃﬃ ¼ VB IB ; 2 2 2

ð6:20Þ

ZB ¼

VB 3 VB2 ¼ ; IB 2 SB

ð6:21Þ

’B ¼

ZB IB ¼ ZB IB tB ¼ VB tB : oB

ð6:22Þ

In a per unit system, there should be only one base value for power. Hence for f, D, g, and Q, four rotor windings, we have 3 Vf B If B ¼ VDB IDB ¼ VgB IgB ¼ VQB IQB ¼ VB IB ¼ SB : 2

ð6:23Þ

Due to the constraint of above equation, we can only define one base value between current and voltage, for each rotor winding, and then derive the other. After the base values for the voltage and current of rotor windings are obtained, base values for impedance and flux linkage can be found from the following equations Zf B ¼ Vf B =If B ; ZDB ¼ VDB =IDB ; ZgB ¼ VgB =IgB ; ZQB ¼ VQB =IQB ; ’f B ¼ Vf B tB ; ’DB ¼ VDB tB ;

’gB ¼ VgB tB ; ’QB ¼ VQB tB :

ð6:24Þ

342

6 Mathematical Model of Synchronous Generator and Load

With base values introduced, we can convert the mathematical equations of (6.14) and (6.15) into the description in per unit as follows. Dividing both sides of each voltage equations of (6.14) by corresponding base voltage value and using the relationships among various base values as given in (6.19)–(6.24), we can obtain 2

3 2 Ra 0 0 vd 6 vq 7 6 0 Ra 0 6 7 6 6 v0 7 6 0 0 Ra 6 7 6 6 7 6 6 7¼6 6 vf 7 6 0 0 0 6 7 6 6 0 7 6 0 0 0 6 7 6 4 0 5 4 0 0 0 0 0 0 0

0 0 0

0 0 0

Rf 0 0 RD 0 0 0 0

3 2 3 2 3 ’d id o ’q 76 iq 7 6 ’q 7 6 o ’d 7 76 7 6 7 6 7 76 i0 7 6 ’0 7 6 0 7 76 7 6 7 6 7 76 7 6 7 6 7 76 7 þ p 6 76 7 6 if 7 6 ’f 7 6 0 7; ð6:25Þ 0 0 7 76 7 6 7 6 7 6 7 6 ’D 7 6 0 7 0 0 7 76 iD 7 6 7 6 7 4 ’g 5 4 0 5 Rg 0 54 ig 5 ’Q 0 RQ iQ 0 0 0 0

0 0 0

32

where p* is differentiation operator in per unit: p d d ¼ tB ¼ ; oB dt dt Ra ¼ Ra =ZB 2 Rf If B 2 Rf Rf ¼ ¼ Zf B 3 ZB IB 2 RD 2 RD IDB : RD ¼ ¼ ZDB 3 ZB IB 2 Rg IgB 2 Rg Rg ¼ ¼ ZgB 3 ZB IB 2 RQ 2 RQ IQB RQ ¼ ¼ ZQB 3 ZB IB p ¼

ð6:26Þ

Using the similar procedure for (6.15), we can obtain 2

3 2 ’d Xd 6 ’q 7 6 0 6 7 6 6 ’0 7 6 0 6 7 6 6 7 6 6 7 6 6 ’f 7 ¼ 6 Xaf 6 7 6 6 ’D 7 6 XaD 6 7 6 4 ’g 5 4 0 ’Q 0

0 Xq 0

0 0 X0

Xaf 0 0

XaD 0 0

0 Xag 0

0 0 Xag XaQ

0 0 0 0

Xf XfD 0 0

XfD XD 0 0

0 0 Xg XgQ

3 id 6 7 XaQ 7 76 iq 7 6 i0 7 0 7 76 7 76 7 76 7 6 if 7; ð6:27Þ 0 7 76 7 7 6 0 7 76 iD 7 XgQ 54 ig 5 XQ iQ 0

32

6.2 Mathematical Model of Synchronous Generator

343

where 9 > > > > > > > > > > > > > > > > > > > > > =

Xd ¼ oB Ld =ZB Xq ¼ oB Lq =ZB X0 ¼ oB L0 =ZB

2 oB Lf 2 oB Lf If B Xf ¼ ¼ Zf B 3 ZB IB 2 oB LD 2 oB LD IDB > XD ¼ ¼ > > > ZDB 3 ZB IB > > 2 > > > oB Lg 2 oB Lg IgB > > > Xg ¼ ¼ > > ZgB 3 ZB IB > > 2 > > > oB LQ 2 oB LQ IQB > > ; XQ ¼ ¼ ZQB 3 ZB IB Xaf Xag XfD

oB maf If B ¼ ; ZB IB oB mag IgB ¼ ; ZB IB 2 oB mfD If B IDB ¼ ; 3 ZB IB2

XaD XaQ XgQ

oB maD IDB ¼ ZB IB oB maQ IQB ¼ : ZB IB 2 oB mgQ IgB IQB ¼ 3 ZB IB2

ð6:28aÞ

ð6:28bÞ

In addition, dividing both sides of (6.18) by SB, from (6.20) we can obtain the output power from synchronous generator in per unit to be po ¼ vd id þ vq iq þ 2v0 i0 :

ð6:29Þ

We should note that the per unit equations of the synchronous generator of (6.25) have the similar form to those using actual values of (6.14). However, in the per unit equations, the coefficient matrix in flux linkage equation of (6.27) is symmetrical, i.e., the mutual inductance between stator and rotor windings becomes reciprocal. Furthermore, with proper choice of base values for inductance, we can make the per unit values of inductance to be equal to that of reactance. Hence the coefficient matrix in the flux linkage equation can also be expressed by use of per unit reactance.

6.2.2

Mathematical Equations of Synchronous Generator Using Machine Parameters

For simplicity of expression, in the following discussion we shall use per unit system and omit the subscript ‘‘*’’ to express per unit variables.

344

6 Mathematical Model of Synchronous Generator and Load

In the mathematical equations of the synchronous generator of (6.25) and (6.27), a total of 18 parameters are presented in (6.26) and (6.28). We regard those 18 parameters as basic parameters of a synchronous generator which are decided by physical design and materials used. Strictly speaking, for two generators of the same type and same model, the parameters will not be exactly the same. Usually it is extremely difficult to obtain the values of those parameters through analytical calculation. Therefore, in practice we convert those 18 basic parameters of a synchronous generator into a group of 11 steady-state, transient, and subtransient parameters. These 11 parameters are called machine parameters and can be obtained directly from machine experiments. They are resistance of stator winding (Ra), q- and d-axis synchronous reactance (Xd, Xq), transient reactance (Xd0 ; Xq0 ), and 0 0 00 00 subtransient reactance (Xd00 ; Xq00 ) as well as the four time constants (Td0 ; Tq0 ; Td0 ; Tq0 ). Because machine parameters are fewer than the basic parameters, certain assumptions are needed for the conversion between these two sets of parameters. Firstly, from the basic (6.25) and (6.27) of a synchronous generator we can see that the magnetic field in space generated by zero-sequence component, i0, is zero and hence it has no impact on any electrical variables associated with generator rotor. Therefore, the zero-sequence equation in (6.25) and (6.27) and the parameter X0 can be ignored. Equation (6.25) now becomes 2

3 2 vd Ra 4 vf 5 ¼ 4 0 0 0 2

3 2 Ra vq 405¼4 0 0 0

0 Rf 0

32 3 2 3 2 3 0 id ’d o’q 0 54 if 5 þ p4 ’f 5 4 0 5; RD iD ’D 0

ð6:30Þ

0 Rg 0

32 3 2 3 2 3 ’q 0 iq o’d 0 54 ig 5 þ p4 ’g 5 þ 4 0 5: ’Q RQ iQ 0

ð6:31Þ

Equation (6.27) can be written as 2

3 2 Xd ’d 4 ’f 5 ¼ 4 Xaf XaD ’D 2

3 2 ’q Xq 4 ’g 5 ¼ 4 Xag ’Q XaQ

Xaf Xf XfD

32 3 XaD id XfD 54 if 5; XD iD

ð6:32Þ

Xag Xg XgQ

32 3 XaQ iq XgQ 54 ig 5: XQ iQ

ð6:33Þ

We can assume that there exist relationships as shown in the following (6.34) among basic parameters in (6.32) and (6.33) [154] Xaf XD ¼ XaD XfD Xag XQ ¼ XaQ XgQ

) :

ð6:34Þ

6.2 Mathematical Model of Synchronous Generator

345

d-axis machine parameters are related to basic parameters as follows: 1. The definition of d-axis synchronous reactance Xd is that when f and D winding are open-circuited and there exists only the d-axis component of current in the armature winding, the measured armature reactance is Xd. From the definition we know that in (6.32) when if ¼ iD ¼ 0, we have ’d ¼ Xd id ; i.e., basic parameter Xd is machine parameter Xd. 2. d-axis transient reactance Xd0 is defined such that when f winding is shortcircuited, D winding open-circuited and only a d-axis component of current suddenly flows through the armature winding, the measured armature reactance is Xd0 . From the definition we know that with D winding being open-circuited, iD ¼ 0; and with f winding being short-circuited, at the moment of sudden flow of current through the armature winding, ff ¼ 0. Hence in (6.32) we have ’d ¼ Xd id þ Xaf if

)

’f ¼ Xaf id þ Xf if ¼ 0

:

Canceling if in the above equations we obtain ! 2 Xaf ’d ¼ Xd id : Xf Therefore Xd0 ¼

2 Xaf ’d ¼ Xd : id Xf

ð6:35Þ

3. The definition of d-axis subtransient reactance Xd00 is that when f and D winding are short-circuited and only a d component of current suddenly flows through the armature winding, the measured armature reactance is Xd00 . According to the definition, with ff ¼ fD ¼ 0 in (6.32), we have 9 ’d ¼ Xd id þ Xaf if þ XaD iD > = ’f ¼ Xaf id þ Xf if þ XfD iD ¼ 0 : > ; ’D ¼ Xad id þ XfD if þ XD iD ¼ 0 By canceling if and iD in the above equation, we obtain ’d ¼ Xd

2 2 XD Xaf 2Xaf XfD XaD þ Xf XaD 2 XD Xf XfD

! id :

346

6 Mathematical Model of Synchronous Generator and Load

That is Xd00 ¼

2 2 XD Xaf 2Xaf XfD XaD þ Xf XaD ’d ¼ Xd : 2 id XD Xf XfD

ð6:36Þ

From the first equation, on the previous assumption of (6.34), we can find XfD. By substituting it into (6.36) we have Xd00 ¼ Xd

2 XaD : XD

ð6:37Þ

4. The definition of d-axis open-circuit transient time constant is the decaying time constant of if when d and D winding are open-circuited. This means that in (6.30) and (6.32), we have id ¼ iD ¼ 0, fd ¼ fD ¼ 0. Hence ) vf ¼ Rf if þ p’f : ’f ¼ Xf if In per unit we have Xf ¼ Lf. From the equation above we can obtain v f ¼ R f if þ Lf

dif : dt

Hence 0 Td0 ¼ Lf =Rf ¼ Xf =Rf :

ð6:38Þ

In fact, when d and D winding are open-circuited, f winding becomes an isolated winding. Hence the decaying time constant of the winding current is the time constant of f winding itself. 00 5. d-axis open-circuit subtransient time constant Td0 is defined to be the decaying time constant of D winding when d winding is open-circuited and f winding short-circuited. From the definition we have id ¼ 0, vf ¼ 0 in (6.30) and (6.32). Hence 9 Rf if þ p’f ¼ 0 > > > > RD iD þ p’D ¼ 0 = : > ’f ¼ Xf if þ XfD iD > > > ; ’D ¼ XfD if þ XD iD That is

Xf XfD

XfD i Rf p f ¼ XD iD 0

0 RD

if : iD

6.2 Mathematical Model of Synchronous Generator

347

This is obviously a second-order electrical circuit and hence there are two time constants. Because usually Rf is very small we can assume Rf ¼ 0. By canceling if in the above equation we have ! 2 XfD XD piD ¼ RD iD : Xf Hence 00 Td0

¼

2 XfD XD Xf

!, RD :

ð6:39Þ

So far we have established the relationship between five d-axis machine parameters and basic parameters. In the similar way, from the definition of various q-axis machine parameters, q-axis voltage equation of (6.31), flux linkage equation of (6.33), and the assumption of (6.34), we can also obtain the relationship between five q-axis machine parameters and basic parameters. In total, the relationship between 11 machine parameters and 18 basic parameters can be listed as follows (on the left side of equations are the machine parameters and the right side the basic parameters). Ra ¼ Ra ; Xd ¼ Xd ; Xq ¼ Xq ; 2 Xd0 ¼ Xd Xaf =Xf 2 Xq0 ¼ Xq Xag =Xg 2 Xd00 ¼ Xd XaD =XD 2 Xq00 ¼ Xq XaQ =XQ 0 Td0 ¼ Xf =Rf 0 Tq0 ¼ Xg =Rg

ð6:40aÞ

) ;

ð6:40bÞ

) ;

ð6:40cÞ

) ;

9 00 2 = Td0 ¼ XD XfD =Xf =RD > : 00 2 ; Tq0 ¼ XQ XgQ =Xg =RQ >

ð6:40dÞ

ð6:40eÞ

Eleven machine parameters can be obtained through experiment. We should point out that the relationship between machine and basic parameters of (6.40) depends on the assumption given in (6.34). Different assumption may be made that will lead to different relationship between the machine and basic parameters such as that given in; while different relationships will result in different mathematical equations of a synchronous generator represented by using machine parameters. However, values of machine parameters are only affected by their definitions, irrelevant of the initial assumptions.

348

6 Mathematical Model of Synchronous Generator and Load

In the following, we will establish the mathematical equations of synchronous generators represented by machine parameters. To do so, we first introduce the noload voltage that is proportional to the current of various rotor windings, and transient and subtransient excitation voltages that are proportional to the flux linkage of rotor windings as follows. No-load voltage: 9 eq1 ¼ Xaf if > > > ed1 ¼ Xag ig = : ð6:41Þ eq2 ¼ XaD iD > > > ; ed2 ¼ XaQ iQ Transient and subtransient voltage: e0q ¼ ðXaf =Xf Þ’f e0d ¼ ðXag =Xg Þ’g

9 > > > > =

e00q ¼ ðXaD =XD Þ’D > > > > ; e00d ¼ ðXaQ =XQ Þ’Q

:

ð6:42Þ

In the mathematical equations of a synchronous generator represented by basic parameters of (6.30)–(6.33), we can express current and flux linkage of all rotor windings by the associated voltage defined in (6.40)–(6.42). By using the relationship between basic and machine parameters of (6.40) and the assumption of (6.34), we will obtain the following mathematical equations of a synchronous generator represented by machine parameters. Flux linkage equation of armature windings ’d ¼ Xd id þ eq1 þ eq2 ’q ¼ Xq iq ed1 ed2

) :

ð6:43Þ

Flux linkage equation of rotor windings 9 Xd Xd0 > ¼ ðXd þ eq1 þ eq2 > > > > Xd Xd00 > > > > 00 00 = eq ¼ ðXd Xd Þid þ eq1 þ eq2 : Xq Xq0 > > > e0d ¼ ðXq Xq0 Þiq þ ed1 þ e d2 > > Xq Xq00 > > > > ; 00 00 ed ¼ ðXq Xq Þiq þ ed1 þ ed2 e0q

Xd0 Þid

ð6:44Þ

Voltage equation of armature windings vd ¼ r’d o’q Ra id vq ¼ r’q þ o’d Ra iq

) :

ð6:45Þ

6.2 Mathematical Model of Synchronous Generator

349

Voltage equation of rotor windings 9 > > > > 0 00 > > X X > 00 00 d d > Td0 req ¼ e > q2 00 = Xd X 0 Td0 re0q ¼ Efq eq1

d

0 Tq0 re0d ¼ ed1

> > > > > 0 00 > X X > q q 00 00 > Tq0 red ¼ e d2 > ; 00 X X q

;

ð6:46Þ

q

where Efq ¼

Xaf uf : Rf

ð6:47Þ

Efq is the voltage across the armature winding when synchronous generator is connected to no load at the steady-state operation. In fact, vf/Rf is an imaginary field current due to vf at steady state. During the transient process, it is not equal to the actual if. From the definition of (6.41) we can see that the product of this steadystate field current and Xaf gives the no-load voltage. Hence Efq is called the imaginary voltage. To express eq1, eq2, ed1, and ed2 directly from the flux linkage equation of rotor windings of (6.44), we have

eq1 eq2 ed1 ed2

9 Xd Xd00 0 Xd Xd0 00 > > ¼ 0 e e > > Xd Xd00 q Xd0 Xd00 q > > > > 00 00 > Xd Xd 0 Xd Xd 00 > 00 > ¼ 0 eq þ 0 eq þ ðXd Xd Þid > > 00 00 = Xd Xd Xd Xd : 00 0 Xq Xq 0 Xq Xq 00 > > > ¼ 0 e e > > Xq Xq00 d Xq0 Xq00 d > > > > 00 00 > Xq Xq 0 Xq Xq 00 > 00 > ¼ 0 ed þ 0 ed ðXq X Þiq > ; 00 00 Xq Xq Xq Xq

ð6:48Þ

Substituting (6.48) into (6.43) and (6.46) we can obtain the flux linkage equation of the armature windings ’d ¼ e00q Xd00 id ’q ¼ e00d Xq00 iq and the voltage equation of rotor windings

) ð6:49Þ

350

6 Mathematical Model of Synchronous Generator and Load

9 Xd Xd00 0 Xd Xd0 00 > ¼ 0 e þ e þ Efq > > > > Xd Xd00 q Xd0 Xd00 q > > > > 00 00 0 00 0 00 = Td0 req ¼ eq eq ðXd Xd Þid 0 Td0 re0q

0 Tq0 re0d ¼

Xq Xq00 0 Xq Xq0 00 e þ e Xq0 Xq00 d Xq0 Xq00 d

> > > > > > > > > ;

00 Tq0 re00d ¼ e0d e00d þ ðXq0 Xq00 Þiq

:

ð6:50Þ

In (6.47) we still have two basic parameters Xaf and Rf. To avoid these two parameters in the expression, we need to choose proper base values such that in per unit system we have Xaf ¼ Rf and hence Efq ¼ vf. This choice of base values is usually called ‘‘unit excitation voltage/unit stator voltage’’ per unit system. The details are as follows. As we have introduced previously, SB is decided by the choice of base values on the generator stator side. For each winding on the rotor, we have to choose a base value for either voltage or current and derive the other. In the ‘‘unit excitation voltage/unit stator voltage’’ per unit system, we first choose the base value for the voltage of the field winding VfB and then derive the base value for field current IfB from (6.23). We choose VfB such that when synchronous generator operates at steady state, is subject to no load and rotates at synchronous speed, the voltage of stator winding is equal to the base value of stator voltage. Obviously, VfB can be gained by experiment. From the above definition about VfB, in (6.14) and (6.15) we only have if 6¼ 0, we have 9 vd ¼ 0 > = vq ¼ oB maf if ¼ VB : > ; vf ¼ Rf if ¼ Vf B So we can obtain Vf B ¼

Rf VB : oB maf

Because Zf B ¼ Vf B =IfB , we have Rf oB maf oB maf Rf ¼ ¼ Rf If B ¼ Zf B Rf VB ZB

If B : IB

Comparing the above equation with Xaf in (6.28), we can see Rf* ¼ Xf*. Hence in per unit Xaf Efq ¼ vf ¼ vf : ð6:51Þ Rf Up to this point, we have established the mathematical model of synchronous generator represented by 11 machine parameters that consists of the voltage

6.2 Mathematical Model of Synchronous Generator

351

equation of armature windings (6.45), flux linkage equation of armature windings (6.49), and voltage equation of rotor windings (6.50). We should point out that this model only needs the specific choice of base value for field winding. Base value for the voltage or current of damping windings can be selected according to (6.23). Besides, voltage of field winding vf is affected by excitation control and hence Efq in (6.50) will be discussed further in Sect. 6.3.

6.2.3

Simplified Mathematical Model of Synchronous Generator

In the above discussion, we established the mathematical model of synchronous generator where four rotor windings, f, g, D, and Q, are used. From (6.50) we can see that the electromagnetic transient of rotor windings is depicted by four differential equations. In a modern power system, there could be over 1,000 generators in synchronous operation. Higher-order differential equations could result in numerical difficulty in power system analysis and calculation. Therefore, in practice the mathematical model of a synchronous generator is often simplified according to requirements of computing accuracy, and only for those generators that we are particularly concerned about are higher-order models used. The simplification can be classified according to how to ignore certain rotor windings, leading to three rotor-winding model, two rotor-winding model, nondamping-winding model and constant e0q model (classical model). All these models can be derived from the full four rotor winding model of a synchronous generator. 1. Three rotor winding model (f, D, Q). For a salient-pole generator, usually we only consider one equivalent damping winding Q on q-axis and ignore the existence of g winding. This means that in the four rotor winding model, ig ¼ fg ¼ 0. Hence in (6.41), ed1 ¼ 0 and in (6.42), e0d ¼ 0 and Xq0 ¼ Xq . The voltage equations of rotor windings are reduced to an order three model 0 Td0 re0q ¼

9 Xd Xd00 0 Xd Xd0 00 > > e þ e þ E fq > q q = Xd0 Xd00 Xd0 Xd00

00 Td0 re00q ¼ e0q e00q ðXd0 Xd00 Þid 00 Tq0 re00d ¼ e00d þ ðXq0 Xq00 Þiq

> > > ;

:

ð6:52Þ

There is no change in the voltage and flux linkage equation of the armature windings. 2. Two winding model (f, g or double-axis model). We only consider one damping winding g on q-axis and ignore D, Q damping winding. This is the same as the assumption iD ¼ iQ ¼ fD ¼ fQ ¼ 0 in four winding rotor model. Hence in (6.41), we have eq2 ¼ ed2 ¼ 0 and in (6.42), e00q ¼ e00d ¼ 0. The flux linkage equation of armature windings becomes

352

6 Mathematical Model of Synchronous Generator and Load

)

’d ¼ e0q Xd0 id

:

’q ¼ e0d Xq0 iq

ð6:53Þ

Voltage equation of rotor windings is reduced to a second-order model 0 Td0 re0q ¼ e0q ðXd Xd0 Þid þ Efq

)

0 Tq0 re0d ¼ e0d þ ðXq Xq0 Þiq

:

ð6:54Þ

There is no change in the voltage equation of armature windings. 3. Nondamping winding model (f, or variable e0q model). Ignoring damping windings, we have iD ¼ iQ ¼ ig ¼ fD ¼ fQ ¼ fg ¼ 0. Hence in (6.41), ed1 ¼ eq2 ¼ ed2 ¼ 0 and in (6.42), e0d ¼ e00q ¼ e00d ¼ 0. The flux linkage equation of armature windings becomes ’d ¼ e0q Xd0 id ’q ¼ Xq iq

) :

ð6:55Þ

Voltage equation of rotor winding is reduced to a first-order model 0 Td0 re0q ¼ e0q ðXd Xd0 Þid þ Efq :

ð6:56Þ

There is no change in the voltage equation of armature windings. 4. Constant e0q model. We neglect damping windings and transient of field winding. Also we consider the right-hand side of (6.56) to be zero due to the control function of AVR, i.e., e0q ðXd0 Xd Þid þ Efq ¼ constant: Thus the mathematical model of synchronous generator is comprised of only the voltage and flux linkage equation of armature windings of (6.45) and (6.55). There is no differential equation for the rotor windings. Constant e0q model usually is used when the rotor motion equation of the synchronous generator is described by electrical torque. 5. Classical model. This is to use Xd0 ¼ Xq0 to further simplify the expression of output electrical power of a synchronous generator. The discussion above is the simplification of depicting rotor windings to reduce the order of the mathematical model of a synchronous generator. On the other hand, in the analysis of power system steady-state operation, the voltage equation of armature windings can be simplified in the following two ways:

6.2 Mathematical Model of Synchronous Generator

353

1. Ignoring the electromagnetic transient of armature windings. This is to neglect the induced voltage due to the variations of ’d and ’q in the voltage equation of armature windings. Thus the voltage equation of armature winding becomes vd ¼ o’q Ra id vq ¼ o’d Ra iq

) :

ð6:57Þ

For power system stability studies the above simplification is very important. From the flux linkage equation of armature winding (6.49) we can see that the differential of flux linkage of armature windings with respect to time will involve that of armature current with respect to time. Because the armature windings of a synchronous generator are connected to a transmission network that is formed by a certain topology of resistance, inductance, and capacitance, the differentiation of armature current with respect to time will require the description of the network by differential equations. This will greatly increase the order of the mathematical model of whole power system. In addition, if the electromagnetic transients of armature windings and network are not ignored, the armature current of the synchronous generator will contain high-frequency components. Under this circumstance, we must take very small integration time steps to achieve the required computing accuracy in the numerical solution of power system mathematical equations. For a modern large power system, increase of the order of its mathematical model and decrease of the required integration time step would add a heavy computational burden such that normal calculation would become impossible. In fact, compared to the electromechanical process of the synchronous generator, the electromagnetic transient behavior of the power network is sufficiently fast that it can be ignored as far as its influence on power system stability analysis and computation is concerned. From (6.57) we can see that when the electromagnetic transient of armature windings of synchronous generator is neglected, its voltage equations become algebraic equations, i.e., those depicting steady-state operation of the synchronous generator. 2. In the voltage equation of armature windings, we consider the rotor speed of synchronous generator o always to be the synchronous speed. This does not mean that during the transient, the rotor speed of synchronous generator does not change. It is because the range of o is small due to the existence of various control functions in generator operation. Hence in the voltage equation of the armature winding, the numerical variation caused by the small change of o is very small and hence can be ignored. This simplification does not result in great saving in computation. However, it has been shown that taking o ¼ 1 in the voltage equation of the armature winding of synchronous generator can partly correct the computational errors caused by ignoring the

354

6 Mathematical Model of Synchronous Generator and Load

electromagnetic transient [153]. Therefore, the voltage equation of armature winding becomes vd ¼ ’q Ra id vq ¼ ’d Ra iq

6.2.4

) :

ð6:58Þ

Steady-State Equations and Phasor Diagram

Mathematically, transient analysis of power systems is to solve a group of differential equations depicting power system transient behavior. Usually the steady-state operating point is the initial condition to solve the differential equations. In the following, we will derive the formula to calculate the initial conditions from the steady-state equations of the synchronous generator. In steady-state operation, the generator rotates at synchronous speed, all electrical variables are balanced and the current of damping windings is zero. Current id, iq, if, and eq1 associated with if as well as flux linkage of all windings are constant. In the following, we shall use capital letters to denote various steady-state electric variables. 1. Steady-state equations represented by synchronous reactance From (6.43) we have ) Fd ¼ Xd Id þ Eq1 : Fq ¼ Xq Iq

ð6:59Þ

At steady state Eq1 ¼ Xaf If ¼ Xaf

Vf ¼ Efq : Rf

Substituting (6.59) into (6.58) we obtain Efq ¼ Vq þ Ra Iq þ Xd Id 0 ¼ Vd þ Ra Id Xq Iq

) :

ð6:60Þ

After load flow calculation, we have had terminal voltage V_ t and current I_t of the synchronous generator in xy coordinate. To obtain Vd, Vq, Id, and Iq in dq coordinate, we need to find the connection between these two coordinate systems, i.e., to find the angle between them. For this purpose, we multiply the first equation of (6.60) by j and add it to the second equation jEfq jðXd Xq ÞId ¼ V_ t þ ðRa þ jXq ÞI_t :

6.2 Mathematical Model of Synchronous Generator

355

We can define an imaginary voltage E_ Q according to the above equation to be E_ Q ¼ V_ t þ ðRa þ jXq ÞI_t :

ð6:61Þ

Because E_ Q and jEfq are in the same phase, from phasor diagram (Fig. 6.2a), we can see that the angle between E_ Q and x, d, is that between d–q and x–y coordinate. Hence from (6.61) we can find d and obtain the transformation between two coordinate systems as follows

¼

sin d cos d cos d sin d

Ax ; Ay

ð6:62Þ

is Ax

is y Ax

It

EQ

vd

E fq

δ

vt

q

jxq I q

Ra I d

Id

R a It

It

R a I q jx′d I d

Ix

E′q Axis q E′

d

Axis q

It

t

R a Id

Ix

jxd Id

jxqIq

v

δ

Id

R a Iq

jx

vd

vq

Iq

jx

d

Iy

vq

Iq

y

It

Ad Aq

jx ′

Ra I t

It

is Ax

x

x

is Ax

Axis d

Axis d

a

b is Ax

Iq

δ

Id Ix

E q′′

vq

vt

It

Axis q

E ′′ Ed′′

It Ra

vd

y

jx ′q′ I q

jxd′′ I d

is x Ax

Axis d

c Fig. 6.2 Phasor diagram of steady-state operation of synchronous generator (a) when synchronous reactance is used (b) when transient reactance is used (c) when subtransient reactance is used

356

6 Mathematical Model of Synchronous Generator and Load

Ax Ay

sin d cos d ¼ cos d sin d

Ad ; Aq

ð6:63Þ

where A denotes current, voltage, flux linkage, and various EMF. After Vd, Vq, Id, and Iq are found, from (6.60) we can calculate the initial value of vf, Vf ¼ Efq. 2. Steady-state equations represented by transient reactance From the first and third equation in (6.44) we have E0q ¼ ðXd Xd0 ÞId þ Eq1

) :

E0d ¼ ðXq Xq0 ÞIq

Noting Eq1 ¼ Efq at steady state and canceling Xd and Xq by substituting the first and second equation in (6.60) into the first and second above equation, we have E0q ¼ Vq þ Ra Iq þ Xd0 Id

) :

E0d ¼ Vd þ Ra Id Xq0 Iq

ð6:64Þ

3. Steady-state equations represented by subtransient reactance From the second and fourth equation of (6.44), we can have E00q ¼ ðXd Xd00 ÞId þ Eq1

) :

E00d ¼ ðXq Xq00 ÞIq

Taking the similar procedure, from (6.64) we can obtain E00q ¼ Vq þ Ra Iq þ Xd00 Id E00d ¼ Vd þ Ra Id Xq00 Iq

) :

ð6:65Þ

Equations (6.60), (6.64), and (6.65) comprise steady-state equations of a synchronous generator adopting the four rotor winding model. From those three equations we can calculate the initial values of five state variables, vf, e0q ; e0d ; e00d , and e00q . Phasor diagrams related to those three equations are shown in Fig. 6.2. When a simplified model of the synchronous generator is used, we can calculate required initial values of state variables directly from the above steady-state equations of the four rotor winding model. For example, when the damping windings are ignored, we have 9 Efq ¼ Vq þ Ra Iq þ Xd Id > =

0 ¼ Vd þ Ra Id Xq Iq : > ; 0 0 Eq ¼ Vq þ Ra Iq þ Xd Id

6.2 Mathematical Model of Synchronous Generator

6.2.5

357

Mathematical Equations Considering Effect of Saturation

In the above discussion, we have established mathematical equations of the synchronous generator under the condition that the magnetic circuit of machine is unsaturated. In practice, to save materials, the design and manufacture of synchronous generator usually makes the iron core of both stator and rotor slightly saturated when operating at rated conditions. At some particular operating conditions, with the increase of flux density, saturation would become very obvious and serious. In system planning and operation analysis, errors caused by ignoring saturation are small. However, in certain applications, such as in transient stability analysis, with detailed model of AVR and its limiters included, the effect of machine saturation can greatly affect the accuracy of analysis and calculation. Study on the effect of saturation started as early as about 60 years ago. The mathematical model of a synchronous generator will become extremely complicated if machine saturation is modeled in great detail. This is because the extent of saturation of a magnetic circuit is closely related to the total mmf in the machine air gap. It is required to combine d- and q-axis mmf to air-gap total mmf and then to find the corresponding magnetic flux and linkage from the saturation curve. Even though air-gap total mmf has a strict sinusoidal distribution in space, mmf varies in different positions. Thus saturation at various positions in space is different, which will cause distortion of the flux wave in the air gap. Therefore, in practice, considering the simplicity of model used, effectiveness of parameters and accuracy of computation, proper approximation is applied to take account of the effect of machine saturation [155–158]. In the following, we shall introduce a method commonly used in stability analysis [156]. The assumptions to apply the method are: 1. The effect of saturation is simply considered on d- and q-axis separately. The difference of magnetic reluctance in d- and q-axis magnetic circuits is only caused by that of length of air gap in the direction of d- and q-axis. 2. On a same axis, the extent of saturation depends on the Potier voltage behind Potier reactance Xp. The higher the Potier voltage, the more serious the saturation. Potier voltage on d- and q-axis is given by the following equation vdp ¼ vd þ Ra id Xp iq vqp ¼ vq þ Ra iq þ Xp id

) :

ð6:66Þ

In addition, the extent of saturation of voltage and flux linkages of armature and rotor windings is approximately considered to be same on the same axis. 3. The distortion of the distribution wave of air-gap flux does not affect the selfinductance and mutual inductance of various windings and the unsaturated values of winding reactance.

358

6 Mathematical Model of Synchronous Generator and Load

Fig. 6.3 No-load saturation characteristic of synchronous generator

Vf

The unsaturated characteristics

(Vqp)

Vf 0

The no-load saturated characteristics

Vf

0

if0 if

if

The extent of saturation is described by saturation factor. For d-axis, saturation factor Sd can be calculated from the saturation characteristic of machine in no-load operation. This is because vqp is equivalent to the voltage of q winding induced from the resultant d-axis air-gap flux. From Fig. 6.3 of no-load saturation characteristic of synchronous generator, we can find the unsaturated value of vqp0 from a certain value of vqp. Hence we can define Sd to be Sd ¼ f ðvqp Þ ¼

vqp0 1: vqp

ð6:67Þ

Obviously, the bigger the value of Sd is, the more saturated is the synchronous generator. Zero Sd indicates the case of no saturation. For q-axis, the saturation characteristic is difficult to obtain through experiment. Hence from the first assumption above, the saturation factor Sq is also determined by using the no-load saturation characteristic of synchronous generator, using the following equation Sq ¼

Xq f ðvdp Þ: Xd

ð6:68Þ

To calculate the saturation factor, one commonly used method is to approximately fit the no-load saturation characteristic curve of Fig. 6.3 by an analytical function, such as if ¼ aVt þ bVtn : Obviously when b ¼ 0, we have the characteristic curve without saturation if 0 ¼ aVt : Hence according to triangle similarity of Fig. 6.3, we have Sd ¼

vqp0 vqp0 vqp if if 0 avqp þ bvnqp avqp b n1 1¼ ¼ ¼ ¼ vqp : vqp vqp if 0 avqp a

6.2 Mathematical Model of Synchronous Generator

359

That is Sd ¼ cvn1 qp ;

ð6:69Þ

where c ¼ b/a. Similarly, from (6.68) we have Sq ¼ c

Xq n1 v : Xd dp

ð6:70Þ

In the following, we shall discuss the voltage equations of the field winding, voltage and flux linkage equations of armature windings of a synchronous generator, taking account of the saturation effect. From the derivation of voltage equations of field winding of (6.50) without considering saturation effect, we can see that on the righthand side of the equations we have the voltage drop across the equivalent resistance of the field winding caused by field current and the external voltage applied on the rotor windings (i.e., the excitation voltage Vf). Hence there should no problem of saturation about this part in the equations. Hence when we consider the saturation effect, we shall still use unsaturated values for those on the right-hand side of the equations. On the other hand, on the left side of the equations, we have the induced voltage by variations of flux linkage with time. Hence when saturation effect is taken into account, for those terms on the left-hand side of the equations we should use their saturated values associated with actual flux linkage. According to the previous assumption (2) and (6.67), we know that on d-axis, the ratio of unsaturated value to the saturated of each voltage and flux linkage is (1 þ Sd). Similarly, on qaxis, the ratio is (1 þ Sq). Therefore, when we consider the saturation effect, the voltage equations of the field winding of a synchronous generator are 9 Xd Xd00 Xd Xd0 > 0 00 > ð1 þ S Þe þ ð1 þ S Þe þ E d qs d qs fq > > > Xd0 Xd00 Xd0 Xd00 > > > > 00 00 0 00 0 00 = Td0 reqs ¼ ð1 þ Sd Þeqs ð1 þ Sd Þeqs ðXd Xd Þid 0 Td0 re0qs ¼

0 Tq0 re0ds

Xq Xq00 Xq Xq0 0 ¼ 0 ð1 þ S Þe þ ð1 þ Sq Þe00ds q ds Xq Xq00 Xq0 Xq00

00 Tq0 re00ds ¼ ð1 þ Sq Þe0ds ð1 þ Sq Þe00ds þ ðXq0 Xq00 Þiq

> > > > > > > > > ;

;

ð6:71Þ

where subscript s denotes the saturated value of each voltage. Taking saturation effects into account, we have the flux linkage equations of armature windings of (6.49) becoming ð1 þ Sq Þ’qs ¼ ð1 þ Sq Þe00ds Xq00 iq ð1 þ Sd Þ’ds ¼ ð1 þ Sd Þe00qs Xd00 id

) :

ð6:72Þ

When we do not consider the saturation effect, from the voltage equations of armature windings of synchronous generator (6.58) and the definition of Potier

360

6 Mathematical Model of Synchronous Generator and Load

voltage of (6.66), we can obtain the relationship between the Potier voltage and flux linkage of armature windings to be vdp0 ¼ ’q Xp iq vqp0 ¼ ’d þ Xp id

) :

ð6:73Þ

According to the relationship between saturated and unsaturated value, we can have ð1 þ Sq Þvdp ¼ ð1 þ Sq Þ’qs Xp iq ð1 þ Sd Þvqp ¼ ð1 þ Sd Þ’ds þ Xp id

) :

Substituting (6.72) into the above equation we can establish the relationship between the Potier voltage and the EMF with saturation considered, to be vdp vqp

9 Xq00 Xp > > ¼ þ iq > = 1 þ Sq : Xd00 Xp > > 00 > ¼ eqs id ; 1 þ Sd e00ds

ð6:74Þ

Substituting the above equation into the defining equation of the Potier voltage, we have the voltage equations of armature windings with saturation being considered, to be 9 Xq00 Xp > > vd ¼ R a id þ þ Xp iq > = 1 þ Sq : 00 > Xd Xp > 00 > vq ¼ eqs Ra iq þ X p id ; 1 þ Sd e00ds

ð6:75Þ

Equations (6.66), (6.67), (6.71), and (6.75) form the mathematical model of synchronous generator with machine saturation being taken into account. From the model it would be straightforward to derive the steady-state equations of a synchronous generator. In practice, we often assume that stator leakage flux does not saturate. Hence we can use Xs as the Potier reactance Xp.

6.2.6

Rotor Motion Equation of Synchronous Generator

6.2.6.1

Rotor Motion Equation of Stiff Rotor

If we consider the prime mover and generator rotor to be a single mass, the rotor motion equation of the whole generation unit is

6.2 Mathematical Model of Synchronous Generator

9 dd = ¼ ðo 1Þos > dt ; do > TJ ¼ Tm Te ; dt

361

ð6:76Þ

where TJ ¼ 2Wk/SB, d is the electrical angle between q-axis of generator rotor and a reference axis x that rotates at synchronous speed. This angle is a dimensionless number and can be measured in radians (rad), TJ is the moment of inertia of generation unit measured in seconds (s), Wk the rotating kinetic energy of the rotor rotating at synchronous speed and measured in Joules (J), SB the base value of generation capacity in V A; Tm and Te are the output mechanical torque of prime mover and the electromagnetic torque of the generator in per unit; their base value is SB/Os measured in radian/second (rad s1), where Os is the mechanical synchronous speed of rotor. The positive direction of Tm and Te is taken to be as same as and opposite to that of rotation of the rotor, respectively. In some references, the mechanical inertia is represented by H ¼ Wk/SB. Obviously, TJ ¼ 2H. In addition, we ought to note the following two issues: 1. Since the product of torque and speed is the power and O/Os ¼ o/os ¼ o*, in per unit we can have ) Pm ¼ Tm o ; ð6:77Þ Pe ¼ Te o where Pm is the output mechanical power from the prime mover and Pe the electromagnetic power of the synchronous generator. As pointed out before, various functions of power system stability control result in a small change of o*. Hence in order to save computational time, sometimes we can just simply take o* to be 1. Thus in per unit, torque is equal to power. 2. Rotor rotation is always subject to air resistance and friction between bearing and shaft. This results in a damping torque to rotor motion. Often we assume that this damping torque is approximately proportional to rotor speed and represent it by the product of a damping coefficient D and speed o*. Considering what has been discussed above, when time is also represented in per unit, the rotor motion equation becomes 9 dd > > ¼ o 1 = dt : ð6:78Þ do > > ; TJ ¼ Do þ Pm Pe dt We would point out that the mechanical torque and power involved in the above rotor motion equation are subject to the control of the governing system of generation unit. Hence the appearance of mechanical torque and power will lead to the establishment of equations for the governing system. This will be discussed in Sect. 6.4.

362

6 Mathematical Model of Synchronous Generator and Load

In (6.78), we consider a combined rotor of generator and prime mover to be a single lumped mass. This consideration will usually bring about no obvious errors when carrying out transient stability analysis. However, when the subsynchronous resonance of power systems is studied, we cannot ignore the existence of rotor shaft elasticity, since large steam-turbine generation units often consist of multiple stage turbines and their shafts can be as long as several tens of meters. In this case, we can consider the exciter, generator rotor, and each turbine section to be separate lumped masses. Thus elasticity of the whole shaft system can be treated as torsional springs between each mass. Therefore, with elasticity being taken into account, rotation speed of each mass could be different during a transient process, resulting in difference in relative angular position of each mass. The motion equation of each mass forms the motion equation of the whole shaft system. Detailed discussion can be found in [159].

6.2.6.2

Electromagnetic Torque and Power of Synchronous Generator

In the rotor motion equation of (6.78), mechanical torque (or power) from prime mover and electromagnetic torque (or power) of synchronous generator are introduced. The former is included in the mathematical model of the prime mover and governing system of the generation unit, which will be discussed in Sect. 6.4. Here we shall introduce the computing model of electromagnetic torque and power. Electromagnetic torque represents the function of force applied on the rotor from the mutual electric and magnetic interactions between stator and rotor of the synchronous generator. Theoretical proof has been provided that electromagnetic torque is equal to the partial differentiation of total magnetic field energy stored in various windings to rotor angle [148], i.e., Te ¼

@WF ; @y

ð6:79Þ

where y is the angle between d-axis and a-axis of armature winding (see Fig. 6.1a) and WF is the total magnetic energy stored in three-phase armature windings and rotor windings, which can be represented as 1 1 WF ¼ ð’a ia þ ’b ib þ ’c ic Þ þ ð’f if þ ’D iD þ ’g ig þ ’Q iQ Þ: 2 2

ð6:80Þ

Because the reference positive direction of armature current is opposite to that of associated flux linkage, we have a negative sign in the above equation. From the base value for torque TB ¼ SB/OB and (6.2)–(6.7), we can obtain

6.3 Mathematical Model of Generator Excitation Systems

Te ¼ ’d iq ’q id :

363

ð6:81Þ

The above equation shows that electromagnetic torque is independent of zerosequence components, because they do not couple with rotor windings. In addition, although the above equation has been established from the four winding model, it is applicable to other higher or lower winding models with only slight differences in derivation. When the four rotor winding model is used, substituting the flux linkage equation of (6.49) into the above equation, we can obtain the expression of electromagnetic torque to be Te ¼ e00d id þ e00q iq ðXd00 Xq00 Þid iq :

ð6:82Þ

From the above expression and (6.77), we can directly establish the expression of electromagnetic power where state variables o*, e00d , and e00q are included. This will bring about a heavy computing burden in the solution. Hence to solve this problem, we can substitute the voltage equation of armature windings of (6.45) into (6.81) and use (6.77) to obtain pe ¼ ud id þ uq iq þ Ra i2d þ i2q id p ’d iq p ’q ;

ð6:83Þ

where Ra i2d þ i2q is the copper loss of the armature windings. When the transient of armature windings is ignored, from the comparison of the above equation with the output power expression of the synchronous generator in (6.29), we can see that the electromagnetic power of the generator is the sum of generator output power and copper loss of generator armature windings. Finally we would like to mention here that (6.83) is also applicable to cases when other types of rotor winding model is used and/or machine saturation is considered.

6.3

Mathematical Model of Generator Excitation Systems

In (6.50) we introduced variable Efq in the per unit system as ‘‘unit excitation voltage/unit stator voltage,’’ that is equal to voltage vf applied to the field winding. Hence we need to establish the mathematical model of generator excitation systems. The basic function of a generator excitation system is to provide the generator field winding with appropriate DC current to generate a magnetic field in the distributed space of the generator armature windings. In earlier times, the excitation system regulated the excitation voltage through manual control, to maintain the required terminal voltage of the generator and reactive power supply from the generator. More recently, various types of excitation and AVR were proposed

364

6 Mathematical Model of Synchronous Generator and Load

and used. In the 1960s, the proposal and application of power system stabilizers (PSS) further enhanced the role played by excitation control systems to improve power system stability. With the advancement of control theory and computer control technology, further new types of excitation regulators have been proposed. Their control tasks have been extended from simple terminal voltage regulation of the generator to multiple excitation control functions. Feedback signals used have developed from a single deviation of generator terminal voltage to the superimposition of various signals on the voltage deviation, based on factors such as electromagnetic power, electrical angular speed, system frequency, armature current, and deviation of excitation current or voltage and their combinations. The control strategy started with simple proportional control and has been enhanced by applying proportional–integral–differential (PID) control, multivariable linear system control schemes, self-tuning control, adaptive control, fuzzy control, and nonlinear control. In recent years, digital excitation controllers based on microprocessors or microcomputers have been developed and installed. In the near future, research into, and innovative applications of, excitation control will involve the development of digital excitation control systems realized by microcomputers and using modern control theory. Relatively accurate analysis of power system dynamics must be supported by mathematical models of the excitation system. Development and design of new types of excitation controller need to establish mathematical models for simulation to check if the dynamic performance is satisfactory. In this section, we shall only introduce the mathematical models of widely used excitation systems and the design principle of excitation regulators will not be discussed. Also we shall not discuss the newer type of excitation controllers, such as linear optimal excitation controller (LOEC), nonlinear optimal excitation controller (NOEC), because they are still at the stage of further theoretical research and testing. Figure 6.4 shows the construction of a general excitation system. The exciter provides field current to the field winding of the generator. The regulator controls the field current. The measurement unit for generator terminal voltage and load compensation measures generator terminal voltage V_ t and compensates for the load current of generator I_t , respectively. The auxiliary control signals are sent through the auxiliary controller. One of the most widely used

Protection and limiter Terminal voltage measurement and load compensation Reference

Regulator

Exciter

Generator Auxiliary controller

Fig. 6.4 Excitation system of generator

To power network

6.3 Mathematical Model of Generator Excitation Systems

365

auxiliary controllers is PSS. Protection and limiter are incorporated to ensure the generator’s operation within various allowed constraints. In Sect. 6.2, we have discussed the mathematical model of a synchronous generator. In the following we shall introduce the mathematical models of excitation systems of generators for power system stability analysis, as shown in Fig. 6.4, block by block. These models are applicable to power system operation when system frequency deviation is within 5% and system oscillation frequency is below 3 Hz. Generally speaking, for the study of SSR or other problems of shaft torsional oscillations, these models would not be precise enough.

6.3.1

Mathematical Model of Exciter

According to the different means of providing excitation power sources, exciters can be classified into three types: DC exciter systems, AC exciter systems, and static excitation systems. The two former types are also called rotational excitation systems. In the following, we shall introduce each of the three types of exciter.

6.3.1.1

Mathematical Model of DC Exciter

Due to the high cost of maintenance, DC exciters have not been used in recently built large generation units. However, in some power systems, we can still see DC exciters in operation. Hence it is necessary to introduce their mathematical model. We shall introduce the establishment of a mathematical model of the general case of a DC exciter that has both self-excitation and separate excitation. Figure 6.5 shows the configuration of the DC exciter. In Fig. 6.5, E represents armature of the exciter; Ref and Lef, Rsf and Lsf is the resistance and self-inductance of the self-excited and separately excited windings, respectively; ief, isf, and icf are the currents of the self-excitation, separate excitation,

Rc

isf Vsf

Rsf

Ref

Fig. 6.5 Configuration of a DC exciter

ief E

icf Lsf

if

Lef

Vf

366

6 Mathematical Model of Synchronous Generator and Load

and compound excitation, respectively; vsf is the voltage externally applied on the separately excited winding; and Rc is a variable regulating resistor. For simplicity of analysis, we assume that self-excited and separately excited windings have the same number of turns, or number of turns and parameters of the separately excited winding have been transferred to the side of the self-excited winding. Hence we can obtain the following voltage equations and flux linkage equations (without considering magnetic saturation). vf ¼ Rc ief þ Ref ðicf þ ief Þ þ p’ef

) ;

vsf ¼ Rsf isf þ p’sf ’ef ¼ Lef ðicf þ ief Þ þ Mes isf ’sf ¼ Mes ðicf þ ief Þ þ Lsf isf

ð6:84Þ

) :

ð6:85Þ

In the above flux linkage equations, we can approximately consider that the selfexcited winding and the separately excited winding are coupled completely. Hence leakage reactance of each winding is zero and unsaturated self-inductance and all mutual inductance have the same value. From (6.85) we can have ’L0 ¼ ’ef ¼ ’sf ¼ Lif S ;

ð6:86Þ

where L ¼ Lef ¼ Lsf ¼ Mes if S ¼ icf þ ief þ isf

) :

ð6:87Þ

’L0 is the flux linkage of the self-excited winding and separately excited, winding without considering saturation, ifS is the total excitation current provided by the DC exciter. If the saturation effect is considered, the relationship between the actual flux linkage ’L and the total excitation current provided by DC exciter ifS is determined according to Fig. 6.6a, which shows the saturation characteristic curve of the DC exciter. Similarly to (6.67), we define the saturation factor of the DC exciter to be SE ¼

if S ’L0 1¼ 1: ’L if S0

ð6:88Þ

As shown in Fig. 6.6, in (6.88), ifS0 is the total excitation current required to generate ’L without considering saturation. The value of SE represents the level of saturation of the DC exciter, describing the relationship between saturated flux linkage ’L and unsaturated flux linkage ’L0. It is usually obtained from the load characteristic curve of the exciter. Figure 6.6b shows that because the load of the exciter is fixed, i.e., when the influence of excitation current of generator if on the

6.3 Mathematical Model of Generator Excitation Systems

yL

vf 0

yL0 =Lif ∑ The saturated characteristics

yL

The unsaturated characteristics vf 0 = bif ∑

vf

The unsaturated characteristics

yL0

367

The saturated characteristics

vf

if ∑

if ∑ a 0

if ∑0 if ∑

b

if ∑0 if ∑

Fig. 6.6 Saturation characteristic curve of DC exciter (a) Relationship between flux linkage and excitation current (b) Load characteristic curve

armature voltage of the exciter during transients is ignored, the output voltage of exciter is approximately proportional to its internal EMF. If the variation of speed is neglected, flux linkage ’L is proportional to voltage vf. Hence the unsaturated characteristic in Fig. 6.6b can be expressed as vf 0 ¼ bif S :

ð6:89Þ

b is the slope of the unsaturated load characteristic curve of exciter, measured in Ohms. From the equation above and (6.86) we can obtain ’L0 ¼

L vf 0 : b

Because flux linkage ’L is proportional to voltage vf, the equation above can be extended to be ’L ¼

L vf : b

ð6:90Þ

Dividing both sides of the first equation, (6.84), by Rc þ Ref, the second equation by Rsf and adding these two equations, as well as using (6.86), (6.87), and (6.90) we can obtain vf vsf Rc 1 L L þ ¼ if S icf þ þ rvf : ð6:91Þ Rc þ Ref Rsf Rc þ Ref b Rc þ Ref Rsf

368

6 Mathematical Model of Synchronous Generator and Load

From (6.90), (6.88), and (6.89) we have bif S b b ’L0 b Lif S vf ¼ ’ L ¼ ¼ ¼ : L L 1 þ SE L 1 þ SE 1 þ SE Substituting the above equation into (6.91) and canceling variable ifS, we can obtain

b SE þ 1 Rc þ Ref

þ ðTef þ Tsf Þr vf ¼

b bRc vsf þ icf ; Rsf Rc þ Ref

ð6:92Þ

where ) Tef ¼ L=ðRc þ Ref Þ ; Tsf ¼ L=Rsf

ð6:93Þ

where Tef and Tsf are the time constants of self-excited and separately excited windings, respectively (measured in seconds). Equation (6.92) gives the relationship between input vsf, icf, and output vf of the exciter using physical units. In order to combine it with the mathematical model of generator in per unit, established in Sect. 6.2, we need to convert (6.92) into per unit form. Here we should use the same base value VfB that has been chosen in Sect. 6.2 for vf. To decide the base value for vsf and icf, we divide both sides of (6.92) by VfB. Then we can see that when base voltage for the excitation current and voltage of the separately excited winding of the exciter are chosen according to the following equation, the equation in per unit is in the most simple form. If SB ¼ Vf B =b Vsf B ¼ Rsf Vf B =b

) :

ð6:94Þ

Hence (6.92) in per unit becomes ðSE þ KE þ TE rÞvf ¼ vsf þ Kcf icf ;

ð6:95Þ

9 KE ¼ 1 b=ðRc þ Ref Þ > = TE ¼ T ef þ Tsf : > ; Kcf ¼ Rc =ðRc þ Ref Þ

ð6:96Þ

where

KE, TE, and Kcf are termed self-excitation factor, time constant, and gain of compound excitation, respectively. By changing variable resistance Rc, these three parameters can be adjusted properly. Equation (6.95) is the mathematical

6.3 Mathematical Model of Generator Excitation Systems Fig. 6.7 Block diagram of DC exciter

vsf icf

+

369

∑ +

Kcf

+

∑

1 sTE

vf

KE + SE

model of the exciter shown in Fig. 6.5. Figure 6.7 is its block diagram where the per unit subscript * has been omitted. Using the same method that has been adopted to consider the saturation effect in synchronous generators, in Sect. 6.2, we can obtain the relationship between the saturation factor SE and output voltage of the DC exciter. To match the saturated load characteristic of the exciter by an approximate function, we can derive the following equation, as we have done (6.69) SE ¼ aE vnf E 1 =bE :

ð6:97Þ

1. The case without separately excited winding is equivalent to Rsf ¼ 1, vsf ¼ 0. Hence from (6.93) and (6.96) we have TE ¼ Tef. 2. The case with only a separately excited winding is equivalent to Rc ¼ 1, icf ¼ 0. Hence from (6.93) and (6.96) we have TE ¼ Tsf and KE ¼ 1.

6.3.1.2

Mathematical Model of AC Exciter

An AC exciter uses a synchronous machine (alternator), usually rotating on the shaft of the generator. AC output from the armature winding of the exciter is rectified through a three-phase noncontrollable, or controllable, bridge rectifier to supply current to the field winding of the generator. There are two types of rectifiers, stationary rectifiers and rotating rectifiers, and two methods of excitation: self-excitation and separate excitation. Hence there are different combinations of types of rectifiers and means of excitation. In the following, we shall first discuss the mathematical model of the exciter and then that of the rectifiers. The majority of AC exciters use separate excitation. In this case, we can use the mathematical model of a synchronous generator, established in Sect. 6.2, to represent the AC exciter. However, the load of an AC exciter is the field winding of the generator and its operating conditions are much simpler than the generator’s. Hence to reduce the effort in analysis and calculation, we need not describe an AC exciter in such detail as we have done for a generator. There are several methods to simplify the mathematical model of a synchronous generator to derive a mathematical model of an AC exciter. Here we shall introduce one simple and commonly used method as follows.

370

6 Mathematical Model of Synchronous Generator and Load

Because the load of the exciter is the field winding of the generator, the armature current of the exciter is almost purely inductive. Hence the q component of armature current of exciter is approximately zero. In the mathematical model of a synchronous generator without damping windings considered, ignoring (6.55) and (6.58), we can obtain the voltage equation of the armature winding of the exciter to be vd ¼ 0 vq ¼ ’d ¼ e0q Xd0 id

) :

ð6:98Þ

In the above equation, we can further ignore the influence of stator current of the exciter on stator voltage. This leads to stator voltage being equal to the transient voltage. In (6.56), due to the adoption of the base value system of ‘‘unit excitation/ unit stator voltage’’, Efq is equal to field voltage. Using the same assumption, denoting the exciter’s field voltage by vR, stator voltage by vE, stator current by iE, using subscript E to denote synchronous reactance, transient reactance, and various time constants of the exciter and following a similar procedure as for deriving (6.56), we can establish the mathematical model of the exciter without considering saturation as follows TE rvE ¼ vR eqE 0 eqE ¼ vE þ ðXdE XdE ÞiE

) :

ð6:99Þ

With saturation being considered, similarly to the procedure for deriving (6.71), we can obtain 0 ð1 þ SE ÞvE ¼ eqE ðXdE XdE ÞiE ;

ð6:100Þ

where we use the same method to gain the saturation factor of exciter SE as we have done for a DC exciter, i.e., to fit the saturation curve of exciter by the approximate function SE ¼ aE vnEE 1 =bE :

ð6:101Þ

We should note that stator voltage vE and current iE only enter the field winding of the generator after rectification. The relationship between vE and vf will be established later in the mathematical model of rectifier. Here we shall derive the connection between iE and if first. When the exciter supplies the field winding of the generator through a threephase noncontrollable bridge rectifier, output current from the rectifier if is the field current of the generator that is approximately proportional to the input current of the 0 rectifier, i.e., armature current of exciter iE. Hence replacing ðXdE XdE ÞiE in (6.100) by KDif, we can describe this relationship as

6.3 Mathematical Model of Generator Excitation Systems Fig. 6.8 Block diagram of transfer function of separately excited AC exciter

vR

+

371

vE

1 sTE

∑

eqE

+

1 + SE

∑

+

TE rvE ¼ vR eqE eqE ¼ ð1 þ SE ÞvE þ KD if

KD

if

) :

ð6:102Þ

From this, the mathematical model of a separately excited AC exciter, using a threephase noncontrollable bridge rectifier, can be expressed by the block diagram of Fig. 6.8. For self-excited exciter, replacing (1 þ SE) in (6.102) and Fig. 6.8 by (KE þ SE) we can obtain its mathematical model [160], where KE is self-excitation factor and KE < 1. Because an AC exciter is connected to the field winding of the synchronous generator through a rectifier, base values of its armature and field voltage and current must not only satisfy various rules used when the mathematical model of synchronous generator is established in Sect. 6.2, but also be related to the mathematical model of the rectifier. This will be discussed in Sect. 6.3.1.3. 6.3.1.3

Mathematical Model of Power Rectifier

An AC exciter usually supplies excitation to the generator through a three-phase noncontrollable or controllable rectifying circuit. In the following, we shall introduce a mathematical model of a noncontrollable rectifier. The input to the rectifier is the stator voltage of the AC exciter vE, the output voltage and current are the field voltage and current of the synchronous generator, respectively. It is very complicated to accurately model the transient response of a rectifier. Engineering practice also suggests that a transient rectifier model is unnecessary. Consequently, a so-called quasisteady-state mathematical model is usually adopted. That is, although during a transient vE, vf, and if satisfy the transient equations of the rectifier, for their instantaneous values in numerical solutions we approximate them as satisfying a steady-state equation. In this way, the transient process is approximated as a series of continuous steady-state processes. A rectifier has three operational modes according to the value of its commutating angle, g, being less than, equal to, or greater than 60 . When g is less than 60 and harmonics are ignored, the steady-state equation of the rectifier using actual values of variables is pﬃﬃﬃ 3 2 3Xg Vf ¼ VE If ; ð6:103Þ p p

372

6 Mathematical Model of Synchronous Generator and Load

where VE is the effective value of stator line voltage of the AC exciter, Xg is the commutating reactance of the rectifier (that is often taken to be the subtransient reactance or negative-sequence reactance of the exciter). Comparing with (4.37), we can see that the equation above in fact treats the three-phase uncontrollable bridge rectifier as for the case of a six-pulse rectifier in HVDC when its firing angle a is zero. In the above equation, 3Xg If/p reflects commutating voltage drop. To connect with the mathematical model of a generator, we need to convert (6.103) into per unit form. Hence, we divide both sides of the equation by the base value of field voltage of the generator VfB Vf ¼ FEX VE ;

ð6:104Þ

pﬃﬃﬃ 3 2VE ¼ ; pVf B

ð6:105Þ

where VE

pﬃﬃﬃ FEX ¼ 1 IN = 3 IN ¼ KC If =VE pﬃﬃﬃ KC ¼ Xg =ð 3pZf B Þ

ð6:106Þ

and KC is a constant. We should point out that commutating angle g is not included in (6.106). In fact, when g is less that 60 , IN is in the range of (0–0.433). It can be proved that when g is equal to or greater that 60 , (5.104) can still be used as the mathematical model of the rectifier. However, in this case, the relationship between FEX and IN has changed. When IN is between zero and 1, FEX is given by the following equation. 8 pﬃﬃﬃ > 1 IN = 3 0 IN < 0:443 > < pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 FEX ¼ ð6:107Þ 0:75 IN 0:443 IN 0:75 : > > : pﬃﬃﬃ 3ð1 IN Þ 0:75 < IN < 1 To use the above model, IN must be nonnegative and less that 1. If for some reason, the value of IN is greater than 1, FEX should be set to zero. From (6.104), (6.106), and (6.107), we can obtain the block diagram of the transfer function of the rectifier, and the relationship of FEX and IN as shown in Fig. 6.9, where the subscript of per unit has been omitted as before. In the following, we shall discuss the base value of various variables in the mathematical model of an AC exciter. Because we have established the mathematical model of an exciter directly by using that of a synchronous generator, choice of per unit system of the exciter should be kept consistent with that of the synchronous generator. From (6.105) and noting that VE is the effective value of stator line voltage of the exciter, obviously the base value of stator line voltage of the exciter, VLEB, should be

6.3 Mathematical Model of Generator Excitation Systems

Vf FEX

0. 8

If

g = 60°

FEX

π

VE

g < 60°

1. 0

373

IN =

KcIf VE

1−IN 3 0 ≤ IN < 0.433

IN

FEX =

0.75− IN2 0.433 ≤ IN ≤ 0.75 3(1−IN) 0.75 < IN < 1

b

0. 6

0. 4

g > 60°

VE

0. 2

Vf

π FEX

a

0

0.2

0.4

0.6

IN

0.8

1.0

c

if

FEX (VE, if , IN)

Fig. 6.9 Mathematical model of power rectifier (a) FEX–IN relationship curve (b) block diagram of transfer function (c) simplified representation of block diagram of transfer function

p VLEB ¼ pﬃﬃﬃ Vf B : 3 2 From the relationship between line and phase voltage as well as effective and maximum value, we can obtain the base value of the maximum value of stator phase voltage of the exciter to be VEB

pﬃﬃﬃ 2 p ¼ pﬃﬃﬃ VLEB ¼ pﬃﬃﬃ Vf B : 3 3 3

ð6:108aÞ

From (6.20), we can give the base value of the maximum value of stator phase current of the AC exciter to be IEB

pﬃﬃﬃ 2 3SB ¼ : pVf B

ð6:108bÞ

The base value of field voltage of the exciter VRB should be decided through experiment according to the principle of ‘‘unit excitation voltage/unit stator voltage.’’ Base value of the field current of exciter is obtained by (6.23) to be IRB ¼ SB =VRB :

ð6:109Þ

In (6.102), KDif represents the load effect of the exciter. From previous derivation of (6.102) we have known that if is approximately proportional to iE, that is, if ¼ kiE

374

6 Mathematical Model of Synchronous Generator and Load

Fig. 6.10 Mathematical model of AC exciter when generator excitation is provided through controllable rectifier

VR

VRmax−KCIf Vf VRmin

where k is a coefficient of proportionality. From the second term on the right-hand side of (6.100), we have 0 ðXdE XdE ÞiE ¼

0 0 0 if If B XdE XdE iE XdE XdE XdE XdE ¼ ¼ if : ZEB IEB ZEB kIEB kZEB IEB

Hence, KD in (6.102) is given by the following equation as KD ¼

0 XdE XdE If B : kVEB

ð6:110Þ

When an AC exciter supplies generator excitation by use of a controllable rectifier, the AC exciter itself often is excited in the form of self-excitation. The voltage regulator of the exciter controls the firing angle of its rectifiers to maintain an approximately constant output voltage. In this case, the mathematical model of the exciter is simplified. In addition, in this case, terminal voltage of the AC exciter is set at a high level. This results in relatively small commutating voltage drop for the controllable commutating bridge. At normal operating conditions, commutating voltage drop can be ignored. Only under automatic field-forcing or reduction, is commutating voltage drop represented as KCIf in the upper limit of output voltage. Hence when an AC exciter provides generator excitation through controllable rectifiers, the mathematical model of the AC exciter is a unit with bidirectional limiters, as shown in Fig. 6.10. We shall discuss input–output relationships of various limiters later.

6.3.1.4

Mathematical Model of Stationary Exciter

A stationary exciter takes terminal voltage or terminal current plus voltage as the power source of excitation for the generator. The former is called a self-excited potential-source system; the latter is a self-excited compound-source system. In a self-excited potential-source system, generator terminal voltage is reduced via an exciter transformer to supply generator excitation through a controllable rectifier. Firing angle of the controllable rectifier is set by a regulator. The block diagram of the transfer function of the self-excited potential-source system is shown in Fig. 6.11. We can see that it is very similar to that of the AC exciter of Fig. 6.10. This is because the only physical difference between the two is that of the excitation power

6.3 Mathematical Model of Generator Excitation Systems Fig. 6.11 Mathematical model of stationary exciter

375 VtVRmax−KCIf

VR

Vf

VtVRmin

V·t

I·t

Vref

Vc = V·t + (Rc + jXc)It •

Vc

+ 1 VM − Σ 1+sTR

Fig. 6.12 Voltage measurement and load compensation

source. This difference is demonstrated in the output limiters of the two excitation systems. In self-excited potential-source system, the excitation power source is the generator itself. Hence upper and lower limit of output voltage is related to the terminal voltage of generator Vt to be VtVRmax KCIf and VtVRmin, respectively. VRmax and VRmin, respectively, are the maximum and minimum value of no-load voltage of the rectifier when Vt ¼ 1. In a self-excited compound-source system, the power source of the controllable rectifier is supplied by an exciter voltage transformer and current transformer. Measured voltage and current can be accumulated before or after rectification in the form of parallel or series addition. This results in many different types and here we shall not introduce their mathematical models. Details can be found in [160].

6.3.2

Voltage Measurement and Load Compensation Unit

The function of an AVR is to maintain generator terminal voltage at an ideal level. The voltage measurement unit takes generator terminal voltage V_ t to convert it into a DC signal through stages of voltage reduction, rectification and filtering, etc. The voltage measurement and conversion unit can be described by a first-order simple lag, as shown by the block diagram transfer function of Fig. 6.12. The function of load compensation is to compensate the load current of generator I_t so as to maintain the required constant voltage at a controlled voltage point in steadystate operation when the load changes. RC þ jXC represents the impedance between the controlled voltage point and generator terminal. When RC and XC are positive, the controlled voltage point is inside the generator; otherwise, outside the generator. In addition, automatic distribution of reactive load among electrically close generators is related to the voltage droop of the generator; while the voltage droop of the generator is realized by adjusting parameter RC and XC. For simplicity, often RC is ignored and set to zero. When XC is greater than zero, we have positive droop; that is, the larger the load current, the higher the terminal voltage. On the other hand, when XC is less than zero we have negative droop. Terminal voltage decreases with the increase of load current. In the case without compensation,

376

6 Mathematical Model of Synchronous Generator and Load

parameters RC, and XC are zero. Voltage measurement and load compensation units may have different time constants. For simplicity, usually we just use a single time constant TR for their description. TR is called the time constant of the measurement unit and usually is less than 60 ms. For many systems, it is very close to zero. Hence in computation, we often take it to be zero. Its output voltage VM is compared against reference voltage Vref. After amplification, the error signal is used as the control signal of the excitation system of the generator. Although the reference voltage Vref is set artificially, it reflects the ideal value for the controlled voltage point of the generator and must satisfy the initial steady-state operating conditions of the whole power system.

6.3.3

Limiters

In the mathematical model of an excitation system, due to functional limitations, physical limits, or the existence of saturation, the output of certain units is subject to limitations, which we represent via limiters. There are two types of limiters, windup limiters and nonwindup limiters. Limiters often appear in integral units, the firstorder simple lag and lead-lag units. Figure 6.13a, b show block diagrams of those two types of limiters. In the following, we shall discuss the example of an integral unit and its input–output relationship. We leave those of windup and nonwindup first-order simple lag and lead-lag units for readers to establish. Equation of an integral unit is dv/dt ¼ u. The limiting function of the two types of limiters is different. For a windup limiter, if variable v is greater than lower limit B and lower than upper limit A, output variable y is v; If v is greater or equal to upper limit A, output variable y is constrained to be upper limit A; If v is less or equal to lower limit B, output variable y is constrained to be lower limit B. We should note that variable v is not constrained and only the next output variable y is. If v is beyond limitation, output variable y is constrained to be the value of the upper or lower limit. For a nonwindup limiter, output variable y is directly constrained between upper and lower limit. If y is within the limits, input–output relationship is dv/dt ¼ u. If y is equal to upper limit and tends to increase with time, i.e., dy/dt > 0, input–output relationship is dy/dt ¼ 0 and y takes the value of upper limit A. If y is equal to lower limit and tends to decrease, i.e., dy/dt < 0, the relationship is dy/dt ¼ 0 and y takes value of lower limit B. When the output variable y takes the value of the upper or lower limit, once input variable u changes sign, y enters within

u Fig. 6.13 Limiters (a) integral unit with windup limiter (b) integral unit with nonwindup limiter

a

1 s

v

A

y

A

u

B

1 s B

b

y

6.3 Mathematical Model of Generator Excitation Systems

377 VS max

VIS

ks 1

1 1 + sT6 2

V1

sT5 1 + sT5 3

V2

1 + sT1 1 + sT2

V3

1 + sT3 1 + sT4

4

5

VS

V4 VS min

6

Fig. 6.14 Block diagram of transfer function of power system stabilizer

the limits. However, for a windup limiter, only when variable v returns within the limits, so does the output variable.

6.3.4

Mathematical Model of Power System Stabilizer

Power system stabilizer (PSS) is a widely used auxiliary regulator in excitation control. Its function is to suppress power system low-frequency oscillations or increase system damping. Its basic principle is to provide the AVR with an auxiliary control signal to make the generator produce an electrical torque in phase with the deviation of rotor speed. Details about the PSS principle, parameter setting, and installation locations can be found in [153]. There are several forms of PSS. Here we give a commonly used block diagram of a PSS transfer function as shown in Fig. 6.14. In Fig. 6.14, block is the gain of PSS; is the measurement unit with a time constant T6 (usually very small and can be ignored); is a wash-out unit or lowfrequency filter to block steady-state input signal to disable PSS at steady-state operation. T5 usually is as large as about 5 s. ; and ; are two lead-lag networks. PSS should consist of at least one lead-lag network. ; is a limiter. Input signal to PSS, VIS, usually is generator speed, terminal voltage, power, system frequency, or combination of some of them. Output signal Vs is superimposed on the AVR input signal. For PSS to play an effective role, its installing location must be selected and parameters be set properly.

6.3.5

Mathematical Model of Excitation Systems

The function of an AVR is to treat and amplify the input control signal to generate a suitable excitation control signal. The AVR usually consists of power amplifier, excitation system stabilizer, and limiters. In the following, we shall introduce mathematical models of different excitation systems. In each block diagram shown below, basic input signal VC is the output from voltage measurement and load compensation unit of Fig. 6.12 and Vs is an auxiliary regulation signal of the AVR, such as the output signal from a PSS.

378

6 Mathematical Model of Synchronous Generator and Load

6.3.5.1

Excitation System with DC Exciter

With different types of AVR being used, there are three types of DC excitation systems: controllable phase compound regulator, compound excitation plus load compensation, and thyristor-controlled regulator. The former two DC excitation systems are usually used for small generation units (100 MW or below) and have been gradually passing out of use. The block diagram of an excitation system adopting controllable phase compound regulator is shown in Fig. 6.15, where V_ t and I_t are the terminal voltage and current of generator, respectively. Block ; represents phase compound excitation, blocks ; and ; are load compensation and measurement unit, ; is composite amplifying unit, ; is limiter with input signal being compound excitation current of exciter, and are units of the DC exciter. To improve performance of the excitation system, a soft negative feedback unit is often used to provide a series adjustment to field voltage of the generator. Control parameters that can be set are KV, KI, RC, XC, KE, KA, TA, KF, and TF. For the excitation system adopting compound excitation plus load compensation, the block diagram can still be used except that block ; needs to be replaced by a simple amplifier of It. Figure 6.16 shows the block diagram of a DC excitation system using thyristorcontrolled regulator. TB and TC are time constants of the excitation regulator itself. They are usually very small and considered to be zero. Time constant and gain of composite amplifying unit is TA and KA. Due to the saturation of the amplifier and limitation of power output, the block for the amplifier has a nonwindup limiter. VF

1

+

KVVt + jK I It Vt It

Vc = Vt + (Rc + jX c )It

1 1 + sTR

2

3

VM − + +

−

Σ

KA 1 + sTR

Vs

VRmax VR min

+

Vref

Σ

5

+

Σ

−

Vf

K E + SE

8

4

6 1 sTE

7

sK F 1 + sTF

Fig. 6.15 Block diagram of transfer function of AVR of DC exciter using controllable phase compound excitation Vs VM

−

+ Vref

+

∑

− VF

1 + sTC 1 + sTB

KA 1 + sTA VRmin

VRmax VR +

∑

−

1 sTE KE + SE

sKF 1 + sTF

Fig. 6.16 Block diagram of transfer function of a DC excitation system

Vf

6.3 Mathematical Model of Generator Excitation Systems

379

is the output of the soft negative feedback unit of excitation voltage to improve the dynamic performance of whole excitation system. VR is the excitation voltage of the DC exciter. Parameters to be set for the normal operation of excitation control are RC, XC, KE, KA, TA, KF, and TF.

6.3.5.2

Excitation System with AC Exciter

Excitation system with an AC exciter is widely used for 100 MW or above generation units. Most excitation systems with AC exciter adopt uncontrollable power rectifier. They can be classified into two groups: stationary rectifier excitation systems and rotating rectifier excitation systems. Here we introduce one type of AC excitation system as shown by the block diagram of Fig. 6.17. Introduction to other types of AC excitation systems can be found in [160]. In Fig. 6.17, parameters TB, TC, KA, TA, KF, and TF describe three blocks belonging to the excitation regulator similar to that in Fig. 6.16. The input signal to the series regulation unit is the no-load voltage eqE of the AC exciter (6.102). Another kind of arrangement is to use the field voltage of the generator Vf as the feedback input signal. Field current If is also an input signal of the excitation regulator and constant KD represents the equivalent load effect of the AC exciter. In Fig. 6.17, the exciter is separately excited. When self-excitation is used, we need to replace the block 1 þ SE by kE þ SE, where kE and SE are the self-excitation coefficient and saturation factor of the AC exciter, respectively. Because the input to rectifier requires VE not to be negative, in the block of the exciter the integral unit represented by TE has a single-directional windup limiter to prevent VE from becoming negative. Parameters to be set for the normal operation of excitation control are RC, XC, KE, KA, TA, KF, and TF. The block diagram of an AC excitation system adopting a controllable rectifier to supply generator excitation is shown in Fig. 6.18. The rectifier is controlled by an independent voltage regulator and hence its output is kept approximately constant. Therefore, the mathematical model of an AC exciter and controllable rectifier is shown in Fig. 6.10. In Fig. 6.18, this has been combined with an equivalent Vs + VM − ∑ + Vref

−

VF

1+sTC 1+sTB

VRmax KA VR − ∑ 1+sTA eqE−

1 VE sTE

+

∑

Vf

0

VRmin sKF 1+sTF

π

+

1+sE KD

FEX (VE, if, IN) If

Fig. 6.17 Block diagram of transfer function of excitation system with AC exciter adopting uncontrolled power rectifier

380

6 Mathematical Model of Synchronous Generator and Load Vs VM −

Vref

VRmax − kcIf

VIm ax

+

VI

Σ

1 + sTc

VR

KA 1 + sTA

1 + sTB

+V Im in

Vf

VRm in

Fig. 6.18 Block diagram of transfer function of excitation system with AC exciter adopting controllable power rectifier Vs VM − +

Vref

+

Σ

−

VImin

VtVRmax − KcIf

VAmax

VImax KA 1+sTA

VI 1+sTc 1+sTc1 1+sTB 1+TB1

+

If

+

Σ

− ILR

Vf −

VAmin

VF

Σ VtVRmin

KLR

0

sKF 1 + sTF

Fig. 6.19 Block diagram of transfer function of self-excited potential-source system

composite amplifying unit, where time constant TA and gain KA depict the dynamic performance of the controllable rectifier and its regulator. To improve system dynamic performance, this type of excitation system usually adopts a series regulator instead of a shunt regulator. The time constants of the series regulator are TB and TC. We should point out that the load of the controllable rectifier is limited to ensure IN between 0 and 0.433 (6.107). Load effect of the excitation system is reflected in the upper limit of the bidirectional limiter. Parameters to be set for the normal operation of excitation control are RC, XC, KA, TA, TC, and TB. Here, because an independent AC exciter is used, the values of upper and lower limits of the bidirectional nonwindup limiter are not connected to the terminal voltage of the generator.

6.3.5.3

Stationary Excitation System

Figure 6.19 shows the block diagram of a self-excited potential-source system and controllable rectifier described by a bi-directional limiter. As has been introduced before, the power into stationary excitation is from the generator terminal. Hence the value of upper and lower limit is related to the terminal voltage of the generator. This type pf excitation system can provide very high automatic field forcing

6.4 Mathematical Model of Prime Mover and Governing System

381

voltages. To avoid overloading of generator field and rectifier, the field current of generator If is constrained by KLR and ILR in the diagram. Proportional unit KLR has a windup lower limit. To avoid this unit we can simply set KLR to zero. KA and TA are the system composite equivalent gain and time constant, respectively. Both series regulation and shunt regulation are displayed in the diagram. Usually only one of them is used. Hence when series regulation is used, we can set KF to zero. Or when shunt regulation is used, we just set time constants TB and TC to zero. Time constants TB1 and TC1 are for the increase of system dynamic gain. Usually we have TC1 > TB1. To simplify the model, this unit can be ignored by setting both of these time constant to zero. Here we should point out that the block diagram of Fig. 6.19 can represent the excitation system adopting full-wave controllable rectifying bridge. When a half-wave controllable rectifying bridge is used, we can simply set the lower limit of the bidirectional limiter at the system output to zero. Parameters to be set for the normal operation of excitation control are RC, XC, KA, TA, KF, TF, TC, TB, KLR, and ILR. In [160], we can find more about the mathematical model and block diagrams of other types of stationary excitation system.

6.4

Mathematical Model of Prime Mover and Governing System

Variable Pm in the rotor movement equation of the generator (6.78) is the mechanical power output from the prime mover. Pm is related to the operating condition of the prime mover and controlled by a governing system. Excluding wind, sun, and wave power generation, there are two types of prime mover used for large-scale power generation, hydraulic turbines, and steam turbines. The hydraulic turbine (or steam turbine) converts hydraulic energy (or steam thermal energy) into rotating kinetic energy of the prime mover which is then converted into electric power by the generator. Obviously, the amount of power being converted is associated with the opening position of the wicket gate of a hydraulic turbine and steam valve of a steam turbine. Because the generator rotor is driven by the prime mover and rotates on the same shaft with the prime mover, if we assume that the generator output power is fixed, when the opening position increases, the generator will accelerate; and conversely it decelerates. Therefore, regulation of the gate or valve position will change the output power from the prime mover to control generator speed. Hence it is easy to see that the main control signal to the opening position should be generator speed. From the rotor movement equation (6.78) we can see that when a power system is subject to a disturbance at steady-state operation, electric power output from the generator changes. This change destroys the balance between electric power output from the generator and mechanical power input to the generator from the prime mover, leading to variation of the generator speed. Change of generator speed results in a response of the governing system to adjust the opening position of the wicket gate (of a hydraulic turbine) or steam valve (of a

382

6 Mathematical Model of Synchronous Generator and Load

steam turbine). The disturbance causes the system to engage in a complex transient process of mechanical, magnetic, and electrical interactions. Therefore, when the function of the governing system is considered, resulting in the variable Pm, we need to establish a mathematical model of the prime mover and the governing system in order to quantitatively analyze electromechanical transients in power systems.

6.4.1

Mathematical Model of Hydroturbine and Governing System

6.4.1.1

Mathematical Model of Hydraulic Turbine

Dynamics of hydraulic turbines are closely related to those of water flow through a penstock, whereas the characteristics of water flow through a penstock are affected by many factors, such as water inertia, water compressibility, and pipe wall elasticity in the penstock. For example, due to water inertia inside a penstock, change of water flow inside a hydraulic turbine lags the opening position change of the wicket gate. When the opening position of the wicket gate increases suddenly, water volume at the wicket gate increases. However, due to the water inertia, speed of water flow at other points inside the pipe cannot increase immediately. This results in input water pressure of the hydraulic turbine decreasing instead of increasing for a short of period of time after the change, leading to a decrease of input power of the hydraulic turbine instead of an increase. On the other hand, when the opening position of the wicket gate decreases suddenly, input water pressure and input power will increase temporarily and then decrease. This phenomenon is usually called the water hammer effect. Furthermore, for the movement of a compressible fluid inside an elastic pipe, the change of water flow volume and pressure at each point inside the pipe is a wave movement, quite similar to the wave process of transmission lines with evenly distributed parameters. A detailed derivation of the mathematical model of input water pressure on the turbine with wave effects considered requires extensive application of fluid mechanics. This is only necessary for the case with a long pressure pipe. In the following, we shall establish a mathematical model of a hydraulic turbine, useful for the analysis of power system stability, with the wave effect of water flow ignored. That is, to assume that the pressure pipe is inelastic, and water is not compressible. Additionally we shall only consider an ideal hydraulic turbine, i.e., (1) neglecting the mechanical power loss caused by the resistance against water flow from the penstock wall; (2) power output of the hydraulic turbine being proportional to the product of net water head and water flow volume; and (3) speed of water flow being proportional to the product of the opening position of the wicket gate and square root of the stationary water head. Hence we can obtain the hydraulic equations as follows: pﬃﬃﬃﬃ U ¼ KU m H ; ð6:111Þ

6.4 Mathematical Model of Prime Mover and Governing System

383

Pm ¼ KP HU;

ð6:112Þ

dU g ¼ ðH H0 Þ; dt L

ð6:113Þ

where U is the water velocity; KU the proportional constant; H the net water head of hydraulic turbine; m the opening position of wicket gate; Pm the mechanical power output of hydraulic turbine; KP the proportional constant; g the gravity acceleration constant; L the length of penstock; and H0 is the steady-state value of H. Taking the initial value of various variables as their base value, the above hydraulic equations can be converted into the following per unit form (subscript * is omitted as before) pﬃﬃﬃﬃ U ¼ m H;

ð6:114Þ

Pm ¼ HU;

ð6:115Þ

dU 1 ¼ ðH 1Þ; dt Tw s

ð6:116Þ

Tw ¼ LU0 =ðgH0 Þ

ð6:117Þ

where

Tw is the time constant of equivalent water hammer effect and physically it is the time required for water head H0 to accelerate water flow in penstock from a stationary state to the flowing speed U0. We ought to point out that this time constant is affected by U0, i.e., related to the load condition of the hydraulic turbine. The heavier the load is, the higher the time constant. Usually under full load condition, Tw is set by the manufacturer between 0.5 and 4 s. Assuming that at initial steady state, the operating point of the hydraulic turbine shifts slightly due to small disturbances from the load, the above hydraulic equations can be linearized at the initial steady-state operating point and after Laplace transformation they become DU ¼

pﬃﬃﬃﬃﬃﬃ 1 m0 ﬃ DH; H0 Dm þ pﬃﬃﬃﬃﬃ 2 H0

ð6:118Þ

DP ¼ H0 DU þ U0 DH;

ð6:119Þ

Tw sDU ¼ DH:

ð6:120Þ

384

6 Mathematical Model of Synchronous Generator and Load Δm

Fig. 6.20 Transfer function of classical model of hydraulic turbine

m

Fig. 6.21 Relationships between actual and ideal opening position

AtgFL

1−TwS 1 + 0.5TwS

ΔPm

Full load

AtgNL No-load loss 0

gNL

gFL 1.0

g

Eliminating variables DH and DU in the above three equations we can obtain (per unit value of H0 is 1) DPm ¼

1 Tw s Dm: 1 þ 0:5Tw s

ð6:121Þ

The model above is called the classical model of a hydraulic turbine. Its transfer function block diagram is shown in Fig. 6.20. In the analysis of power system stability, the above classical model of a hydraulic turbine is used. From the assumptions used to derive the model we know that the classical model is applicable to cases with relatively small variations of load. When load changes over a wide range, the model may cause a large computational error [161]. In the following, we shall establish a nonlinear model of a hydraulic turbine. Basic assumptions will be the same as those for deriving the classical model except that mechanical power loss and dead zone are taken into account. Opening position of wicket gate m in (6.111) is that of an ideal wicket gate with the dead zone of the hydraulic turbine caused by factors such as friction being ignored, i.e., it is assumed that when m changes from 0 to 1, operation of the hydraulic turbine goes from no load to full load. With mechanical power loss being considered, position change of the wicket gate from closing to opening, initially has to overcome stationary friction forces in the hydraulic turbine without causing the turbine to start rotation immediately. Hence we need to replace the ideal opening position m by the actual one g. From Fig. 6.21, we can see their relationship to be m ¼ At g;

ð6:122Þ

where At ¼

1 : gFL gNL

ð6:123Þ

6.4 Mathematical Model of Prime Mover and Governing System

385

When the actual opening position is gNL, the hydraulic turbine is still at no-load. When it is gFL, the hydraulic turbine operates at full load. With power loss being considered, hydraulic equations (6.112) become Pm ¼ KP HU PL ;

ð6:124Þ

PL ¼ KP UNL H;

ð6:125Þ

where PL is no-load loss of the hydraulic turbine; UNL critical water speed when the hydraulic turbine changes from stationary to rotating. Obviously that is when the actual opening position is gNL. Taking the rated parameters of the hydraulic turbine as the corresponding base value, we can convert (6.111), (6.113), (6.124), and (6.125) into the following per unit form pﬃﬃﬃﬃ U ¼ m H;

ð6:126Þ

Pm ¼ ðU UNL ÞH;

ð6:127Þ

dU 1 ¼ ðH H0 Þ; dt TW

ð6:128Þ

where TW ¼

LUB : gHB

ð6:129Þ

TW is the time constant of the equivalent water hammer effect at rated load. From (6.117) we can see that the relationship of the time constant between any load condition and at rated load is Tw ¼

U0 HB TW : UB H0

ð6:130Þ

In (6.127), base value of power is the rated power of the hydraulic turbine. To connect it to the mathematical model of the generator, we can convert the base value of power to the rated power of generator SB Pm ¼ Pr ðU UNL ÞH;

ð6:131Þ

Pr ¼ PB =SB :

ð6:132Þ

2 U : m

ð6:133Þ

Rewriting (6.126) H¼

386

6 Mathematical Model of Synchronous Generator and Load

From (6.122) and (6.133) we can eliminate H in (6.128) and (6.131) to obtain dU 1 ¼ dt TW

! 2 U H0 ; At g

Pm ¼ Pr ðU UNL Þ

U At g

ð6:134Þ

2 :

ð6:135Þ

The two equations above are the nonlinear model of a hydraulic turbine. From the physical meaning of UNL we know that when the actual opening position of wicket gate g is gNL, acceleration of water flow is zero. From (6.134) we have dU 1 ¼ dt U¼UNL TW UNL

UNL At gNL pﬃﬃﬃﬃﬃﬃ ¼ At gNL H0

!

2 H0

¼ 0;

:

ð6:136Þ

Normally H0 is 1 and hence UNL is a constant. From (6.122), (6.133), (6.128), and (6.131) we can show the nonlinear model of the hydraulic turbine in Fig. 6.22. 6.4.1.2

Mathematical Model of Governing System of Hydraulic Turbine

Modern generation units usually use an electrical-hydraulic governing system. However, the principle of mechanical hydraulic governing system is easier to illustrate. Hence we shall take it as representative to establish the mathematical model of a governing system of a hydraulic turbine. Figure 6.23 shows the configuration of a governing system using a centrifugal pendulum (fly-ball governor). In the following we shall present the equation of motion of each component of the governing system, where variables are in per unit and their positive direction is indicated in Fig. 6.23. Compressibility of hydraulic oil will be neglected: 1. Equation of the centrifugal pendulum. The function of the centrifugal pendulum (fly balls) is to measure generator speed. Relative ring position of the fly balls is denoted by . When generator speed increases, the fly balls move away from each other due to the increase of centrifugal force and consequently decreases. On the other hand, when generator speed decreases, the fly balls come closer because of the decrease of centrifugal force and hence increases. Ignoring the γ

At

μ

π

H +

Σ H0

1 sTw

U +

Σ UNL

Fig. 6.22 Block diagram of transfer function of hydraulic turbine

π

Pr

Pm

Tm

ω

6.4 Mathematical Model of Prime Mover and Governing System

A

I

C

B

G

η

ζ

ζ2 D

F

E

ζ1

387

VI H V

μ

σ a

III

b IV

II

Fig. 6.23 Illustration of governing system of a centrifugal pendulum

mass of the fly balls and damping of motion, is approximately proportional to the deviation of generator speed with a proportional coefficient kd, that is ¼ kd ðo0 oÞ:

ð6:137Þ

2. Equation of pilot valve. If the servomotor does not function (point D in Fig. 6.23 is fixed) and the inertia of the pilot valve is ignored, the relationship between position of pilot valve, s, and that of point B, z, is s ¼ B:

ð6:138Þ

3. Equation of relay valve. Position of relay valve, m, is the integral of the position of the pilot valve, s, with respect to time, i.e., position speed of relay valve is proportional to position of pilot valve, s. The proportional constant is called the time constant of the relay. Hence Ts rm ¼ s:

ð6:139Þ

4. Feedback equation. From Fig. 6.23, we can see that when increases, s increases accordingly and m also increases. Increase of and m results in an increasing z, leading s to decrease and m to decrease accordingly. Hence z is a position variable exhibiting feedback from m. There are two parts in z, z1, and z2. z1 is a soft feedback due to the existence of a spring and dashpot; while z2 is proportional to m and hence it is a hard feedback. We have B ¼ B1 þ B2 ¼

kb Ti s m þ ka m; 1 þ Ti s

ð6:140Þ

where ka ¼ a/d, kb ¼ b/d, d ¼ 1/kd. kb and Ti are the gain and time constant of the soft feedback, respectively; ks the gain of hard feedback; d the sensitivity of

388

6 Mathematical Model of Synchronous Generator and Load

measuring component; b coefficient of soft feedback; and a is the droop coefficient. Due to the inertia, water flow cannot follow the change of opening position of the wicket gate quickly. Hence when speed deviation of the generator changes fast, the governor needs a strong negative feedback from the opening position m to slow down the change of opening position of wicket gate such that water flow and output power from the hydroturbine can follow that of m. For slow variations of generator speed in steady-state operation, the governor needs to respond promptly. Hence the gain of the negative feedback should take a small value. From (6.140) we can see that dynamic gain of the whole negative feedback unit is high. When t ¼ 0, it is the summation of kb and ka. The time constant of the soft negative feedback is large, usually between 0.5 and 5 s. At steady state, the steady-state gain of the soft negative feedback is zero and the gain of the whole negative feedback is only that of the hard feedback ka. This provides the generator with a certain droop coefficient at steady-state operation, such that a generator speed decrease will increase generator output. The droop characteristic ensures stable load sharing among multiple generation units in parallel steady-state operation, to realize the function of primary frequency control. ka and kb are often about 0.04 and 0.4, respectively. Opening position of both pilot valve and wicket gate have certain limitations imposed. In addition, due to the existence of mechanical friction and gap, there exists a certain dead zone of the governing function. Hence in the mathematical model, there are associated limiters and a nonlinear unit representing the dead zone. From (6.137) to (6.140) we can obtain the transfer function block diagram of the governing system of a hydraulic turbine as shown in Fig. 6.24. The function of an electrical-hydraulic governing system of a hydraulic turbine is quite similar to that of the mechanical hydraulic system introduced above, but more simple and flexible as far as the regulation of basic parameters is concerned. A mathematical model of an electrical-hydraulic governing system of the hydraulic turbine adopting PID control can be found in [162, 163].

ωref ω −

+

Σ

kδ

η +

εkδ

Σ

−

σ max σ

σ

1 sTs

σ min

ζ

Σ +

+

ζ2

ζ1

+

μ ref +

Σ

μ max μ

μ

μ min

kσ skβ Ti 1 +sTi

Fig. 6.24 Block diagram of governing system of centrifugal pendulum (fly ball) of a hydraulic turbine

6.4 Mathematical Model of Prime Mover and Governing System

6.4.2

Mathematical Model of Steam Turbine and Governing System

6.4.2.1

Mathematical Model of Steam Turbine

389

Dynamics of steam turbines are mainly related to the volume effect of steam. In the following, we shall first derive the time constant for the general steam volume effect. As shown in Fig. 6.25, volume of the vessel is V (m3) and input and output steam mass flow rates are Qin and Qout (kg s1), respectively. We have dW dr ¼V ¼ Qin Qout ; dt dt

ð6:141Þ

where W is the weight of steam in the vessel (kg) and r is the density of steam (kg m3). Assuming that the output steam flow is proportional to steam pressure in the vessel, we have Qout ¼

QN P; PN

ð6:142Þ

where P is steam pressure in the vessel (kPa), PN the rated steam pressure in the vessel (kPa), and QN is the rated output of steam out of the vessel (kg s1). With the steam temperature in the vessel being constant, we have dr dP @r ¼ ; dt dt @P

ð6:143Þ

where the rate of change of steam density with pressure, at a given temperature, ∂r/∂P, can be obtained from steam parameter tables and is a constant. From (6.141) to (6.143) and after Laplace transformation, we have Qout ¼

1 Qin ; 1 þ sTV

where TV ¼

ð6:144Þ

PN @r V : QN @P

ð6:145Þ

Qout

Qin V

Fig. 6.25 Steam vessel

390

6 Mathematical Model of Synchronous Generator and Load From boiler

Q0

Steam valve

Reheater

crossover

Q2

Q1 HP

shaft

Q3 IP

shaft

LP

LP

shaft

To condenser

Fig. 6.26 Illustration of a multistage steam turbine

TV is called the time constant of steam volume effect. From (6.145) we can see that the bigger the volume of the vessel, the higher is the time constant of volume effect. From (6.144) we can see that when the input steam flow increases (or decreases) suddenly, the output steam flow will not increases (or decreases) immediately because the pressure inside the vessel cannot increase (or decrease) instantly. Change of output steam flow lags that of input steam flow. This is the steam volume effect phenomena. There are many types of configuration of steam turbines. Modern steam turbine units consist of multiple-stage steam turbines to drive a single generator. According to the difference in rated operating steam pressure, multiple-stage turbines can be classified as high pressure (HP), intermediate pressure (IP), and low pressure (LP) turbines. Medium and small steam turbine units may have only a one-stage turbine. To increase thermal efficiency, modern steam turbine units usually have an intermediate reheater (RH). Figure 6.26 shows the configuration of a steam turbine with reheater. From Fig. 6.26 we can see that the high-pressure high-temperature steam from the boiler enters the HP stage through a main valve and steam chest. We should note the existence of a certain volume of steam in the pipe and chest from the main valve to the nozzle of the HP stage. Exhaust steam from the HP stage is sent into the reheater section to raise temperature before entering the IP stage. Similarly we ought to note that there exists a certain volume of steam between the output point of the HP stage and input point of IP stage. Exhaust steam from IP stage enters the LP stage through crossover that also has a certain volume. Volume effects of the three volumes mentioned above can be described by time constant TCH, TRH, and TCO, respectively. Usually TCH is between 0.2 and 0.3 s, time constant of reheater TRH is large, between 5 and 10 s, and TCO is about 0.5 s. Output mechanical torque of the steam turbine is proportional to the steam flow at the nozzle. In addition, we assume that input steam flow to the HP stage is approximately proportional to the opening position of the main steam valve m. We denote the proportionality coefficient of mechanical power of HP, IP, and LP stages to be FHP, FIP, FLP. Usually FHP, FIP, FLP is 0.3, 0.3, 0.4 and their summation is one.

6.4 Mathematical Model of Prime Mover and Governing System + +

TmH

μ Q0

1 1 + sTcH

+

Σ TmI

FHP

FIP

Q1

1 + sTRH Q 2

Tm

Σ +

TmL

FLP

1

391

π

Pm

ω

1

1 + sTco

Q3

Fig. 6.27 Block diagram of transfer function of a multi-stage steam turbine

From the analysis above and taking proper base values for per unit expressions, we can obtain the mathematical model of the steam turbine in per unit to be 9 1 > Q1 ¼ Q0 > > 1 þ TCH s > > > = 1 ð6:146Þ Q2 ¼ Q1 ; 1 þ TRH s > > > > > 1 ; Q3 ¼ Q2 > 1 þ TCO s 9 > > > > > > > > > = TmI ¼ FIP Q2 ; > TmL ¼ FLP Q3 > > > > Tm ¼ TmH þ TmI þ TmL > > > > ; F þF þF ¼1

m ¼ Q0 TmH ¼ FHP Q1

HP

IP

ð6:147Þ

LP

where TmH, TmI, TmL is the output mechanical torque of HP, IP, and LP turbines, respectively, flows Q0–Q3 are shown in Fig. 6.26. Block diagram of the transfer function of the mathematical model above is shown in Fig. 6.27. Other types of mathematical models of steam turbines can be found in [164, 165]. 6.4.2.2

Mathematical Model of Governing System of Steam Turbine

Basic functions of the governing system of a steam turbine include normal primary frequency control, secondary frequency control, over-speed control, over-speed generation shedding and generation starting and stopping control in normal operation, as well as auxiliary steam pressure control. Normal primary frequency control and secondary frequency control of the steam turbine is quite similar to those of a hydroturbine. Primary frequency control provides a droop around 4–5% to ensure stable load sharing among parallel generation units. Secondary frequency control is achieved through adjusting the load reference. In modern steam turbine units, usually there are more control valves in addition to the main valve shown in Fig. 6.26. For example, in a steam turbine unit equipped with a reheater there is a stop

392

6 Mathematical Model of Synchronous Generator and Load

valve behind the RH stage. When over-speeding generation requires an emergency reduction of output power from the steam turbine, it would not be enough to just turn off the main valve, because the steam volume of the RH stage is very large. Under this circumstance, usually the main valve and stop valve must be turned off simultaneously. Primary frequency control and secondary frequency control function only by adjusting the main valve shown in Fig. 6.26. Usually in the study of power system stability, only the control of the main valve is considered and that of other valves is ignored. However, if emergency stop and generation shedding are used as the method for stability control, the control function of other valves needs to be taken into account. In this book, we shall only introduce the control model of the main valve. The control model of other valves can be found in [164, 165]. Governing systems of steam turbines can be classified into three types, mechanical hydraulic, electrohydraulic, and power-frequency electrohydraulic. The principle of mechanical hydraulic governing system is the same as that of a centrifugal pendulum governor introduced previously, except that the governing system of a steam turbine does not need the soft feedback unit and only uses hard feedback. The coefficient of hard feedback is 1. Hence a mechanical hydraulic governing system of a steam turbine can be shown by the transfer function block diagram of Fig. 6.28, where the simple lag with time constant T1 represents the pilot valve in the governor. The value of T1 usually is small and hence this unit can be ignored. In an electrohydraulic governor, the low power output unit in the mechanical hydraulic governor, i.e., the part from speed measurement to servomotor, is realized by an electronic circuit. Compared to the mechanical hydraulic governor, an electrohydraulic governor is of better applicability and flexibility, with quicker responds speed. In order to obtain better performance and linear response, the feedback channel from steam flow (or steam pressure at the first stage in the HP turbine) and valve position of the servomotor is introduced in the electrical-hydraulic governor. The transfer function block diagram is shown in Fig. 6.29. The transfer function block diagram of the power-frequency electrohydraulic governor is shown in Fig. 6.30. By comparing frequency and power signals with the given reference, an error signal is obtained and then amplified. A PID controller conditions the amplified signal. Its output electrical signal is converted into a hydraulic signal by an electrical-hydraulic converter to drive a relay and servomotor to regulate the main valve of the steam turbine. In Fig. 6.30, kP, kI and kD are the gains of proportional, integral, and differential units, respectively; TEL the time constant of the electrical-hydraulic converter; and Ts is the time constant of the relay.

ω−

ω0 +

Σ

kδ

η +

Σ ζ

−

ε kδ

μ ref

ρ max

1 1 + sT1

ρ ρ min

1 sTs

+

+

μ max μ

Σ μ min

Fig. 6.28 Block diagram of mechanical hydraulic governing system of steam turbine

6.5 Mathematical Model of Load

393

ω ref ω−

μmax

μ open •

+

kδ

Σ

kp

+

+

Σ

−

Σ

−

μ close •

1 sTs

kp−1

μ

μ min

qHP

Fig. 6.29 Block diagram of transfer function of electrohydraulic governing system

ω−

ω0 +

Σ P0

P−

Σ

kI s kδ +

Σ

+

Σ

+

correcting signal

kp skD 1 + sTD

+

+

Σ

+

μ max

1 + 1 + sTEL

Σ

−

1 sTs

μ μ min

Fig. 6.30 Transfer function of power-frequency electro hydraulic governor

In the recent 20 years, digital governing systems for steam turbines have been developed, in which the operating unit of the main valve of the steam turbine is connected to a digital controller via a digital-analogue hybrid unit. The control function is realized by software. A digital governor provides more flexible and universal functions than an electrical-hydraulic governor. Response speed is enhanced greatly with the time constant being about 0.03 s. More details about digital governors can be found in [167]. We ought to point out that (6.147) is equivalent to ignoring the transients of the thermodynamic system. If thermodynamics is considered, obviously Q0 will be determined by a mathematical model describing the thermodynamic system. As far as the time scale for the computation of power system stability (usually for 5 s following a disturbance) is concerned, the time constant of the thermodynamic system is very large. Hence the thermodynamic system can be considered as operating in steady state. However, for long-term power system stability analysis, involving system dynamics for several minutes after a disturbance, the dynamics of the thermodynamic systems, such as the boiler, will play an important role. Mathematical modeling of the thermodynamic system is still a research subject at the moment.

6.5

Mathematical Model of Load

Load is an important part in a power system. To study power system behavior in various operational states, we need to establish a mathematical model of system load. It is not difficult to establish a mathematical model of certain power-consuming

394

6 Mathematical Model of Synchronous Generator and Load

equipment in the power system. However, it is neither necessary nor possible mathematically to describe each of hundreds and thousands of loads in detail. Hence in this section, power system load refers to all electrical equipment connected at a common node in the power system. It includes not only various end-users of power-consuming equipment but also under-load tap changing transformer, distribution network, various kinds of reactive power compensators, voltage regulation units, and even some small generators, etc. The relationship between active and reactive power absorbed, by all those mentioned above, at the node and the node voltage and system frequency constitutes the mathematical model of nodal load. Obviously, for different types of node, such as residential, commercial, industrial, and rural, the composition of load is quite different. Besides, for the same node, during different time periods, such as different seasons in a year, different days in a week, and different hours in a day, the composition of nodal load can vary. Due to the variety, randomness, and time variance of load, it is an extremely difficult problem to establish an accurate load model. A large number of studies have demonstrated that the conclusions from power system analysis are greatly affected by whether the mathematical model of the load has been established properly or not. From the point of view of system operation analysis and control, improper mathematical modeling of the load will result in analytical conclusions being poorly matched with practical results, either being too conservative – leading to inefficient utilization of the system, or too optimistic – causing hidden risks to system operation. An even more difficult problem is that at the moment there is no way to know if a certain load model is always conservative or optimistic under any disturbance. The importance and complexity of establishing mathematical models of the load has become a special research field, resulting in a large number of studies over many years [168–170]. There are many methods for the establishment of mathematical models of load, but these can be classified into two groups: ‘‘method of theoretical aggregation’’ [170] and ‘‘method of identification aggregation’’ [171]. In theoretical aggregation, nodal load is considered to be the combination of various individual users. Firstly those users are electrically categorized and average characteristics of each category are determined. Then a statistical percentage of each category of users is worked out and finally the total load model is aggregated. The method of identification aggregation uses collected field data. After a proper structure for the load model is chosen, the model parameters are identified by using field data. The two methods have their own merits and disadvantages. The former is simple and easy to use, but its accuracy is not satisfactory. The latter can produce more accurate mathematical models by treating and analyzing field data using modern identification theory. However, it is still difficult to obtain an accurate dynamic model of load because voltage and frequency of the real power system cannot vary over a large range. Therefore, power system load modeling remains a research topic to be pursued and no fully matured method is available. There are quite a few methods to classify power system load models. With regard to whether the load model can describe load dynamics, a model is categorized as either static or dynamic. Obviously, a static load model is a set of algebraic

6.5 Mathematical Model of Load

395

equations, while a dynamic model includes differential equations. Other classifications include: linear load model or nonlinear load model and voltage-related model or frequency-related model. Conventionally, we consider load models related to both voltage and frequency to be frequency-related models. According to the way that the model is established, we have derived-model or input–output model. A derived model has clear physical meaning and can easily be understood. It is usually adopted when few types of load are considered. Nonderived models only concern the mathematical relationship between load input and output. Due to the limitation of space, in this section, we shall only introduce several commonly used types of load. The simplest load model is to use a constant impedance to represent the load. That is, to assume that the equivalent impedance of the load does not change during system transients and its value is determined by the node voltage and power absorbed by the load at steady state before the occurrence of a disturbance. This load model is rather rough. However, due to its simplicity, it is still widely used when requirements on computational accuracy are not high.

6.5.1

Static Load Model

The static characteristic of load is the relationship between node voltage or frequency and power absorbed by the load, when voltage or frequency varies slowly. The usual forms of static load model are as follows. 1. Static load voltage or frequency characteristic described by a polynomial. Without considering variations of frequency, the relationship between node voltage and power absorbed by load is taken to be 9 " #

> VL 2 VL > 2 > PL ¼ PL0 aP þ bP þ cP ¼ PL0 aP VL þ bP VL þ cP > > = VL0 VL0 " #

> > VL 2 VL > QL ¼ QL0 aQ þ bQ þ cQ ¼ QL0 aQ VL2 þ bQ VL þ cQ > > ; VL0 VL0 ð6:148Þ where PL0, QL0, and VL0 are the active, reactive power absorbed by the load and node voltage before the occurrence of a disturbance. Parameters, aP, bP, cP, aQ, bQ, and cQ have different values for different nodes and satisfy ) aP þ bP þ c P ¼ 1 : aQ þ bQ þ cQ ¼ 1

ð6:149Þ

From (6.148) we can see that this model is in fact equivalent to representing the load in three parts. Coefficient a, b, and c represent the percentage of constant

396

6 Mathematical Model of Synchronous Generator and Load

impedance (Z), constant current (I), and constant power (P) in the total nodal load, respectively. Hence this type of load model is also called a ZIP model. Because system frequency does not vary much during transients, static frequency characteristics of load can be represented linearly. Without considering variation of node voltage, the relationship between node power and system frequency is 9 f f0 > > > PL ¼ PL0 1 þ kP = f0 ; ð6:150Þ f f0 > > > QL ¼ QL0 1 þ kQ ; f0 where PL0, QL0, and f0 are the active, reactive power absorbed by load and system frequency before the occurrence of a disturbance, respectively. Parameters kP and kQ have different values at different nodes and their physical meaning is the differential of node power to variation of system frequency at steady state, that is 9 f0 dPL dPL > > kP ¼ ¼ > PL0 df f ¼f0 df f ¼f0 = : ð6:151Þ > f0 dQL dQL > > kQ ¼ ¼ ; QL0 df f ¼f0 df f ¼f0 With variation of voltage and frequency being taken into account, the mathematical model of load is the product of the two per unit model expressions above, that is )

PL ¼ aP VL2 þ bP VL þ cP ð1 þ kP Df Þ ð6:152Þ

: QL ¼ aQ VL2 þ bQ VL þ cQ 1 þ kQ Df We would like to point out here that in statistical computation, various base values must be converted to maintain consistency with system base values. 2. Static load voltage characteristics expressed by exponentials. Without considering variation of frequency, static load voltage characteristics can be described by the following exponential form a 9 VL > > PL ¼ PL0 > = VL0 ð6:153Þ b : > VL > > ; QL ¼ QL0 VL0 For composite load, power a usually is between 0.5 and 1.8, b changes significantly between different nodes, typically between 1.5 and 6. With the effect of frequency change being taken into account, we have

6.5 Mathematical Model of Load

397

9 VL a f f0 > > 1 þ kP > = VL0 f0 b : QL VL f f0 > > > ¼ 1 þ kQ ; QL0 VL0 f0 PL ¼ PL0

ð6:154Þ

Although static load models are widely used in routine computation of power system stability due to their simplicity, computational errors could be very high when the magnitude of node voltage involved in the computation varies over a wide range. For example, discharge lighting load takes over 20% of commercial load. When the per unit voltage value reaches as low as 0.7 p.u., the light goes off and the load consumes zero power. When the voltage recovers, the light goes on after a short delay. Some induction motors are equipped with low voltage protection. When the voltage decreases below a certain level, the motor will be disconnected from the network. Also, due to transformer saturation at higher voltages, reactive power absorbed is very sensitive to changes in the magnitude of nodal voltage. All these factors make static load models inapplicable when nodal voltage varies over a large range. A common method to cope with this problem is to use different model parameters in different voltage ranges or to use a simple constant impedance load when the node voltage is below 0.3–0.7 p.u. Other algebraic forms of static load model can be found in [170].

6.5.2

Dynamic Load Model

When node voltage changes quickly over a large range, adoption of purely static load models will bring about excessive computational errors; especially in the study of voltage stability (or load stability) where high accuracy is required in load modeling. Many studies using different types of load model have shown that at sensitive nodes, dynamic load models should be used [172–175]. In computational practice, those nodal loads are considered to consist of two parts: static and dynamic. Although there are many different types of industrial load, induction motors takes the largest share. Hence, load dynamics are mainly determined by the transient behavior of an induction motor. In the following, we shall introduce mathematical models of induction motors of two types: a model considering only mechanical transients and a more detailed model including both electromechanical transients and mechanical transients. Induction motors of large and small capacity have obviously different dynamics. For small capacity motors, only mechanical transients need to be considered [168]. 6.5.2.1

Dynamic Load Model Considering Mechanical Transients of an Induction Motor

In this type of model, electromechanical transients of an induction motor are ignored, with only the mechanical transient being taken into consideration. From

398

6 Mathematical Model of Synchronous Generator and Load

machine theory we know that the dynamics of an induction motor can be described by the equivalent circuit of an induction motor as in Fig. 6.31, where X1 and X2 are leakage reactance of armature and field windings, respectively; Xm the mutual impedance between armature and field windings; R2/s the equivalent rotor resistance. If system frequency and motor speed are denoted by o and om, respectively, machine slip speed s ¼ (o om)/o ¼ 1 om* should satisfy the following equation of motion of the rotor TJM

ds ¼ TmM TeM ; dt

ð6:155Þ

where TJM is the equivalent moment of inertia of the machine rotor and mechanical load and TmM and TeM are the mechanical torque of load and machine electrical torque, respectively. Derivation of above equation is the same as that used to derive the rotor motion equation for a synchronous generator, noting its reference positive direction of torque is just opposite to that for the synchronous generator. From the above equation we can see that when load torque is greater than electrical torque, slip speed of the induction motor increases, i.e., motor speed decreases. Ignoring electromechanical transients, electrical torque of an induction motor can be expressed to be TeM

2TeM max VL 2 ¼ s ; scr V LN þ scr s

ð6:156Þ

where TeM max is the maximal electrical torque of the induction motor at rated voltage and scr is the critical slip speed for steady-state stability of the induction motor. For a certain induction motor, TeM max and scr are constant when change of frequency is not considered. VL and VLN are the terminal voltage and rated voltage of the induction motor, respectively. Mechanical torque of an induction motor is related to the characteristics of the mechanical load and often a function of motor speed. Traditionally it is given as TmM ¼ k½a þ ð1 aÞð1 sÞpm ;

ð6:157Þ

where a is the portion of mechanical load torque that is independent of motor speed, pm the exponent associated with the characteristic of the mechanical load, and k is R1 + jX1 VL

Fig. 6.31 Equivalent circuit of induction motor

jX2 R2 / s

Rμ + jX μ

6.5 Mathematical Model of Load

399

the percentage of motor load. To achieve better flexibility and wider applicability of computation, currently mechanical torque is expressed as the summation of polynomial and exponential forms [168] TmM om 2 om om g ¼ am þ bm þ c m þ dm ; ð6:158Þ TmM0 om0 o0 om0 where TmM0 and om0 are mechanical torque and motor speed before occurrence of disturbance. am, bm, cm, dm, and g are the characteristic parameters of mechanical torque. Parameter cm is calculated from the following equation cm ¼ 1 ðam þ bm þ dm Þ:

ð6:159Þ

From Fig. 6.31, we can obtain the equivalent impedance of an induction motor to be ZM ¼ R1 þ jX1 þ

ðRm þ jXm ÞðR2 =s þ jX2 Þ ; ðRm þ R2 =sÞ þ jðXm þ X2 Þ

ð6:160Þ

where ZM is a function of motor slip speed. Rotor motion equation of an induction motor (6.155), electrical torque ignoring electromechanical transients (6.156), load mechanical torque (6.157), (6.158), or (6.159), and equivalent impedance (6.160) form the mathematical model of an induction motor load with electromechanical transients neglected. Input variables to the model are node voltage and system frequency. Output variable is the equivalent impedance. Hence when VL and o, as functions of time, are known, s can be found by solving the above equations to obtain the equivalent impedance ZM at any time. As pointed out previously, nodal load includes all electrical equipment connected at the node. Because so many types of electrical equipment may be connected, the dynamics of nodal load are very complicated. In the following, we shall introduce a method of simplifying nodal load by use of the classical model of an induction motor. The key issue in the simplification is to obtain the equivalent impedance of nodal load at any time. Step 1. We separate the total power PL(0) and QL(0) absorbed by the nodal load, in steady-state operation, into two parts. One part is expressed by a static load model with power PLS(0) and QLS(0). The corresponding equivalent impedance is denoted 2 as ZLSð0Þ ¼ VLð0Þ ½PLSð0Þ jQLSð0Þ . Another part is modeled by an induction motor with only mechanical transients considered. The power of this part is denoted as 2 PLM(0) and QLM(0) with corresponding equivalent impedance ZLMð0Þ ¼ VLð0Þ ½PLMð0Þ jQLMð0Þ . Equivalent impedance of nodal load is ZL(0) ¼ ZLS(0) þ ZLM(0). Step 2. It is approximated that all equipment connected at the node, with their dynamics being considered, is a certain typical induction motor. Model parameters of the typical motor are s(0), TJM, TeM max, scr, R1, X1, R2, X2, Rm, Xm and k, a, pm or am, bm, dm, g. From (6.160) we can find the steady-state equivalent impedance of the typical motor ZM(0). Obviously, steady-state equivalent impedance of the typical

400

6 Mathematical Model of Synchronous Generator and Load

motor does not have to be equal to the steady-state equivalent impedance of an equivalent motor. Step 3. In a system transient, node voltage and system frequency vary with time. By using some numerical methods to solve system equations and rotor motion equation of the typical motor (details about the numerical method are introduced in Chaps. 7 and 8), we can obtain the slip speed s(t) of the typical motor, nodal voltage magnitude VL(t) and system frequency o(t) at time t. From (6.160) we then can calculate the equivalent impedance of the typical motor ZM(t) at time t. From the static load model we can find the equivalent impedance of static load ZLS(t). Step 4. We suppose that at any time, the ratio between the equivalent impedance of equivalent motor and equivalent impedance of typical motor is a constant. Hence at any time t, the equivalent impedance of the equivalent motor is ZLMðtÞ ¼ ðcr þ jci ÞZMðtÞ ;

ð6:161Þ

where the proportionality constant can be found from steady-state conditions cr þ jci ¼ ZLMð0Þ =ZMð0Þ :

ð6:162Þ

Finally we obtain the equivalent impedance of nodal load at time t to be ZLðtÞ ¼ ZLSðtÞ þ ZLMðtÞ :

6.5.2.2

ð6:163Þ

Load Dynamic Model with Electromechanical Transients of Induction Motors Considered

Compared to the model introduced above, this model considers electromechanical transients of the field winding of induction motors. Similar to the case of a synchronous generator, because the transient of the armature winding is very fast, we do not consider the electromechanical transient of armature windings for an induction motor either. Details about deriving the mathematical model of an induction motor with electromechanical transients of the field winding being taken into account can be found in [153]. In the following, we shall give a simple derivation method by use of the mathematical model of a synchronous generator established in Sect. 6.2. In fact, as far as the transient equation of the machine is concerned, an induction motor can be considered to be a synchronous generator being completely symmetrical in the two directions of d- and q-axes. Hence in some algorithms of power system transient stability analysis, modeling of induction motors and synchronous generators is treated in the same way. When an induction motor is considered individually; for simplicity, the subtransient process of a synchronous generator is ignored. In the mathematical model, the f winding has the same structure as that of

6.5 Mathematical Model of Load

401

the g winding but is short-circuited. Under these conditions, in equations of the synchronous generator ((6.43)–(6.46)), letting Xd ¼ Xq ¼ X, Xd0 ¼ Xq0 ¼ X, 0 0 eq2 ¼ ed2 ¼ e00q ¼ e00d ¼ 0, p’d ¼ p’q ¼ 0, Td0 ¼ Tq0 , o ¼ 1 s, Ra ¼ R1, we have per unit equations of an induction motor to be 9 vq ¼ ð1 sÞðe0q X0 id Þ R1 iq > > > > vd ¼ ð1 sÞðe0d þ X0 iq Þ R1 id = ; 0 > Td0 pe0q ¼ e0q ðX X0 Þid > > > ; 0 Td0 pe0d ¼ e0d þ ðX X0 Þiq

ð6:164Þ

0 where machine parameters, X, X0 , and Td0 can be derived from Fig. 6.31. Because dand q-axis are completely symmetrical and the structure of f and g windings is the same, in (6.32) and (6.33) we have

Xaf ¼ Xag ¼ Xm :

ð6:165Þ

Hence according to the definition of synchronous reactance, we have the following equation for the stator side, X ¼ Xd ¼ Xq ¼ X1 þ Xm :

ð6:166Þ

Similarly on the rotor side, we have Xf ¼ Xg ¼ X2 þ Xm :

ð6:167Þ

Substituting (6.166) and (6.167) into (6.40b), we can obtain X0 ¼ Xd0 ¼ Xq0 ¼ X1 þ

X2 Xm : X2 þ Xm

ð6:168Þ

We denote the resistance in (6.30) and (6.31) Rf ¼ Rg as R2. Substituting (6.167) into (6.40b) we have 0 0 Td0 ¼ Tq0 ¼ ðX2 þ Xm Þ=R2 :

ð6:169Þ

Equation (6.164) can be simplified by converting it in dq coordinates from (6.62) to system unified xy coordinates. Differentiation of (6.62) to per unit time can result in p

Ad Aq

¼

A sin d cos d cos d sin d Ax p x þ pd: Ay cos d sin d sin d cos d Ay

ð6:170Þ

402

6 Mathematical Model of Synchronous Generator and Load

From the geometrical meaning of a and (6.78) we know pd ¼ s. Hence in xy coordinates (6.164) becomes vx ¼ ð1 sÞe0x þ ð1 sÞX0 iy R1 ix vy ¼ ð1 sÞe0y ð1 sÞX0 ix R1 iy

)

0 0 Td0 pe0x ¼ Td0 se0y e0x þ ðX X0 Þiy 0 0 Td0 pe0y ¼ Td0 se0x e0y ðX X0 Þix

;

ð6:171Þ

) :

ð6:172Þ

At quasisteady state, multiplying the second equation in (6.171) and (6.172) by j and adding to the first equation, we have V_L ¼ ð1 sÞE_ 0M ½R1 þ jð1 sÞX0 I_M ;

ð6:173Þ

0 0 Td0 pE_ 0M ¼ ð1 þ j sTd0 ÞE_ 0M jðX X0 ÞI_M ;

ð6:174Þ

where V_L ¼ Vx þ jVy , I_M ¼ Ix þ jIy , E_ 0M ¼ E0x þ jE0y . However, with subtransient process ignored, the mathematical model of a synchronous generator cannot be converted into the form of (6.173) and (6.174) if d- and q-axis are not symmetrical. Treating an induction motor as a synchronous generator and from (6.81), (6.43), and (6.44), we can obtain the electrical torque of an induction motor to be TeM ¼ ðe0q iq þ e0d id Þ ¼ ðe0x ix þ e0y iy Þ;

ð6:175Þ

where the negative sign is because the positive reference direction of electrical torque of an induction motor is just opposite to that of a synchronous generator. Because a generator model is used, the reference direction of current is going out of, instead of into the induction motor. Therefore, the mathematical model of an induction motor considering electromagnetic transients consists of (6.155), (6.173)–(6.175) and the load mechanical torque of (6.157) or (6.158). For nodal composite load, we can use the same method adopted previously with mechanical transients being considered. For the typical motor, pE_ 0M ¼ 0 in steady-state operation, from (6.173) and (6.174) we can find I_Mð0Þ ; E_ 0Mð0Þ . Hence the equivalent impedance of the typical motor at steady state is ZMð0Þ ¼ V_ Lð0Þ I_Mð0Þ . Equivalent impedance of the equivalent motor can be calculated from nodal voltage and load power at steady state. Thus the ratio between equivalent impedance of typical and equivalent motors can be computed from (6.162). During transients, solving the combined equation describing the typical motor and system we can obtain I_MðtÞ , V_ LðtÞ , and ZM(t). Hence the equivalent impedance of equivalent motor and composite load can be calculated from (6.161) and (6.163). During transients, variation of slip speed has little effect on armature voltage,

Thinking and Problem Solving

403

numerically. It can be ignored in a simple computation and hence in the armature voltage equation of the motor, (6.173), s is taken to be a constant 0. Typical parameters of induction motors can be found in [168, 176]. There are other forms of load dynamic model. For some special loads with large capacity, such as large rolling machines, electric-arc furnaces, electric trains, large units of temperature control, and synchronous motors in pumping or energy storage power plants, etc., the model needs to be established individually. For long-term stability analysis, transformer saturation, adjustment of under-load tap changing transformers, voltage regulation arising from reactive compensators, and the action of low-frequency low-voltage load-shedding equipment, etc., ought to be represented within the load models. Overall, load modeling is still a developing subject.

Thinking and Problem Solving 1. How is the relationship between the electrical quantities in stator and rotor of a synchronous generator set up? 2. Does the mutual inductance between stator winding and rotor winding vary with time, according to whether the generator is round-rotor or salient-pole? 3. What is the function of Park conversion? 4. In the state equation of (6.1), each winding flux linkage is a state variable. Considering the motion equation of the rotor, the electrical rotational speed o of generator is also a state variable. Discuss the nonlinearity of the generator model according to this formula. 5. Discuss the physical significance of the right-hand three items in (6.14), and thereby explain the electrical mechanism of power output of a generator. 6. What are the usual applications of round-rotor generators and salient-pole generators in electrical power systems, and why? 7. The form of synchronous generator model will be influenced by such factors as the choosing of positive direction of magnetic axis, the suppositions taken during converting original parameters into rotor parameters, selection method of base values, and so on. By consulting other books, compare the common and different points of synchronous machine models with the forms that are introduced in this book. 8. By consulting books on synchronous generator experiments, find out about and describe the methods that can be used to empirically determine the parameters of a synchronous generator. 9. During a transient in an electrical power system, the electromagnetic transient process in the electrical network is much faster than the rotor flux dynamics of the generator, so in the synchronous generator model that is used to analyze the electromechanical transient process, the time derivative of stator winding flux linkage is taken to be zero. Analyze the effect of this approximation on calculation quantity.

404

6 Mathematical Model of Synchronous Generator and Load

10. There are three kinds of coordinates used to describe the electrical quantities of a generator. These are the electrical quantities in three-phase static coordinates a, b, c; three-axis rotating coordinates d, q, 0; and complex plane x – y. Discuss the relationship among these three kinds of description. 11. Given one salient-pole synchronous generator, its terminal voltage U_ t ¼ 1:0, and unit power output P þ jQ ¼ 1.0 þ j0.1. The parameters of the generator unit are Xd ¼ 1.0, Xq ¼ 0.6, Xd0 ¼ 0:3; Xq0 ¼ 0:2; Xd00 ¼ 0:15; Xq00 ¼ 0:1. If the stator resistance is neglected, calculate the emfs E0q ; E0d ; E00q , and E00q of this generator. 12. During the formulation of a synchronous generator model, in which formulae can the electrical rotational speed o be considered approximately as invariable, and in which formulae can the electrical rotational speed o not be considered approximately invariable? Why? 13. Discuss the effect of excitation current on the operating state of a synchronous generator, according to (6.50) and (6.51). 14. Discuss the working mechanism of the Washout link in PSS (in Fig. 6.14). 15. Discuss the necessity and difficulty of building steady and dynamical synthetic load models.

Chapter 7

Power System Transient Stability Analysis

7.1

Introduction

The mechanical–electrical transient of a power system that has experienced a large disturbance can evolve into two different situations. In the first situation, the relative rotor angles among generators exhibit swing (or oscillatory) behavior, but the magnitude of oscillation decays asymptotically; the relative motions among generators eventually disappear, thus the system migrates into a new stable state, and generators remain in synchronous operation. The power system is said to be transiently stable. In another situation, the relative motions of some generator rotors continue to grow during the mechanical–electrical transient, and the relative rotor angles increase, resulting in the loss of synchronism of these generators. The system is said to be transiently unstable. When a generator loses synchronization with the remaining generators in the system, its rotor speed will be above or below what is required to produce a voltage at system frequency, and the slip motion between the rotating stator magnetic field (relative to system frequency) and rotor magnetic field causes generator power output, current and voltage to oscillate with very high magnitudes, making some generators and loads trip and, in the worst case, causing the system to split or collapse. A necessary condition that a power system maintains normal operation is the synchronous operation of all generators. Therefore, analyzing the stability of a power system after a large disturbance is equivalent to analyzing the ability of generators to maintain synchronous operation after the system experiences a large disturbance, this is called power system transient stability analysis. The aforementioned power system transient stability analysis typically involves the short-term (within some 10 s) dynamic behavior of a system, nevertheless, sometimes we have to study system midterm (10 s to several minutes) and longterm (several minutes to tens of minutes) dynamic behavior, this would be termed power system midterm and long-term stability analysis. Midterm and long-term stability mainly concerns the dynamic response of a power system that experiences a severe disturbance. A severe disturbance can cause system voltages, frequency, and power flows to undergo drastic changes; therefore, it is meaningful to look into certain slow dynamics, control, and protection X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853‐7, # Springer Science þ Business Media, LLC 2008

405

406

7 Power System Transient Stability Analysis

performance that are not addressed in a short-term transient stability analysis. The response time of devices that affect voltage and frequency can be from a few seconds (such as the response time of generator control and protection devices) to several minutes (such as the response time of a prime mover system and on-load tap changing regulators, etc.) A long-term stability analysis focuses on the slow phenomenon with long duration that occur after a large disturbance has happened, and the significant mismatch between active/reactive power generation and consumption. The phenomena of concern include: boiler dynamics, water gate and water-pipe dynamics of hydraulic turbines, automatic generation control (AGC), control and protection of power plant and transmission system, transformer saturation, abnormal frequency effects of load and network, and so forth. When performing long-term stability analysis, one is often concerned about the responses of a system under extremely severe disturbances that are not taken into consideration in system design. After the occurrence of an extremely severe disturbance, a power system can undergo cascading faults and can be split into several isolated parts. The question a stability study has to answer is whether or not each isolated part can reach acceptable stable operation after any load-shedding occurs. Midterm response refers to response whose timeframe is between that of shortterm response and long-term response. Midterm stability study investigates the synchronous power oscillations among generators, including some slow phenomena and possibly large voltage and frequency deviations [177]. Large disturbances are severe threatens to power system operation, but in reality they cannot be avoided. The consequence of losing stability after a power system experiences a disturbance is in general very serious, it can even be a disaster. In fact the various large disturbances, such as short circuit, tripping or committing of large capacity generator, load, or important transmission facility, appear as probabilistic events, therefore when designing and scheduling a power system, one always ensures that the system can maintain stable operation under a set of reasonably specified credible contingencies, rather than requiring that the system can sustain the impact of any disturbance. Because every country has their own stability requirements, the selection of credible contingencies can be based on different standards. To check if a power system can maintain stable operation under credible contingencies, one needs to perform transient stability analysis. When the system under study is not stable, efficient measures that can improve system stability need to be sought. When a system experiences extremely severe stability problems, fault analysis is required to find the weak points in the system, and develop corresponding strategies. In power system stability analysis, the mathematical models of system components not only directly relate to the analysis results, but also have a significant effect on the complexity of the analysis. Therefore, if appropriate mathematical models for each system component are developed, stability analysis can be made simple and accurate. This is a crucial step in stability analysis.

7.2 Numerical Methods for Transient Stability Analysis

Differential Eqs

Rotor Circuit Eqs

Network Eqs

Excitation System Eqs

Coordinate Transformation Eqs

PSS Eqs

Rotormotion Eqs

Stator Voltage Eqs

Primer-mover & Governer Eqs

407

Other generotots Loads DC System Other dynamic devices Such as SVC, TCSC, etc

Algebraic Eqs

Fig. 7.1 Conceptual framework of mathematical models for stability studies

Figure 7.1 conceptualizes the mathematical model of all system components for power system stability studies. From the figure one can see that the mathematical model consists of the models of synchronous machines and the associated excitation systems, prime mover and speed-governing system, electrical load, and other dynamic devices and electrical network. Apparently, all the dynamic components of the system are independent; it is the electrical network that connects them with each other. Mathematically, the complete system model can be described as a set of differential-algebraic equations as follows: dx ¼ fðx; yÞ; dt

ð7:1Þ

0 ¼ gðx; yÞ:

ð7:2Þ

This chapter first introduces the composition of the component models for transient stability analysis and the numerical solution algorithms for differentialalgebraic equations, then describes the mathematical relationship between dynamic components and electrical network, followed by an exposition of how to model network switches and faults. The chapter also presents in detail the solution algorithms for simple transient stability analysis and for analysis of systems with FACTS devices represented by full mathematical models.

7.2

Numerical Methods for Transient Stability Analysis

Power system transient stability analysis can be viewed as an initial value problem of differential-algebraic equations. In this section we first introduce the numerical methods for ordinary differential equations, and then discuss the numerical methods for differential-algebraic equations. We provide a general procedure for transient stability analysis at the end of the section.

408

7 Power System Transient Stability Analysis

7.2.1

Numerical Methods for Ordinary Differential Equations

7.2.1.1

Fundamental Concept

Consider the following first-order differential equation: dx ¼ f ðt; xÞ; dt

xðt0 Þ ¼ x0 :

ð7:3Þ

In general, the function f in the above equation is a nonlinear function of x and t. In many real world situations, f is not an explicit function of time t, therefore the above equation reduces to dx ¼ f ðxÞ; dt

xðt0 Þ ¼ x0 :

ð7:4Þ

In power system stability analysis, the right-hand side of all the differential equations does not contain explicitly the time variable t. When f in (7.4) is a linear function of x, one can easily obtain the closed-form solution of the differential equation. For example, consider the following differential equation: dx ¼ x: dt

ð7:5Þ

The closed-form solution is given as x ¼ A et ;

ð7:6Þ

where A is a constant. Equation (7.6) represents a family of integral curves. Given an initial condition in the form of x(t0) ¼ x0, one can determine a solution curve. For instance, if x(0) ¼ 1, then from (7.6) the integral constant can be found as A ¼ 1, thus the solution curve is as follows: x ¼ et :

ð7:7Þ

The differential equations of real world engineering problems appear to be more complex, the right-hand sides of the equations are typically not integrable, therefore closed-form solutions, like (7.6), of such differential equations cannot be obtained. To solve these differential equations, one must rely on numerical methods. The idea of numerical methods is to employ a certain integral formula to solve for the approximate value of xn at each instant in the time series tn ¼ t0 þ nh, n ¼ 1,2, . . . (here h is the step size) in a step-by-step fashion, starting from the initial state (t ¼ t0, x ¼ x0). This method of solving differential equation is called step-by-step integration.

7.2 Numerical Methods for Transient Stability Analysis Fig. 7.2 Illustration of Euler’s method for solving differential equations

409

x

x3

true solution x = x(t)

x2 x0

x1

0

t1

t2

t3

t

In the following, we illustrate the basic idea of step-by-step integration using Euler’s method as an example. Suppose that the exact solution of the first-order differential (7.3) at t0 ¼ 0, x(t0) ¼ x0 is as follows: x ¼ xðtÞ:

ð7:8Þ

The graph of the function, that is, the integral curve of the differential (7.3) passing through the point (0, x0) is depicted in Fig. 7.2 . Euler’s method is also called the Euler’s tangent method or Euler’s polygon method. The idea of the method is to approximate the integral curve by an Euler’s polygon, the slope of each Euler’s polygon is obtained by solving for (7.3) with the initial value of the Euler’s polygon as input. Specifically, the computational steps are as follows: For the first segment, the slope of the integral curve at point (0, x0) is dx ¼ f ðx0 ; 0Þ: dt 0 Replacing the first segment with a straight line which has a slope of dx dt 0 , one can find the incremental of x at t1 ¼ h (h is the step size) as follows: dx Dx1 ¼ h: dt 0 Therefore the approximation of x at t1 ¼ h should be dx x1 ¼ x0 þ Dx1 ¼ x0 þ h: dt 0

410

7 Power System Transient Stability Analysis

For the second segment, the integral curve will be approximated by another straight line segment, the slope of which can be obtained by substituting the initial value of the segment (that is, the starting point of the segment (t1, x1)) into (7.3): dx ¼ f ðx1 ; t1 Þ: dt 1 An approximate value of x at t2 ¼ 2h can be found based on x2 ¼ x1 þ

dx h dt 1

as illustrated in Fig. 7.2. The above procedure can be repeated to find an approximate value of x3 at t3 and so forth. In general, the recursive formula for computing an approximate value of the n þ 1 point is as follows: dx xnþ1 ¼ xn þ h; dt n

n ¼ 0; 1; 2; . . . :

ð7:9Þ

Now we turn to analyzing the error introduced by this recursive formula which is used to compute (tnþ1, xnþ1) from (tn, xn). To do so, expand the integral function (7.8) at (tn, xn) using Taylor’s formula as follows: xnþ1 ¼ xn þ x0n h þ x00n

r h2 ðrÞ h þ þ x xn ; 2! r!

ð7:10Þ

where x0n ; x00n ; . . . are the first-order, second-order,. . . derivatives of the integral function with regard to variable t. The symbol xn represents a number in the interval ðrÞ [tn, tnþ1], and xxn is the residual of the Taylor’s series. When r ¼ 2, (7.10) becomes h2 n 2!

xnþ1 ¼ xn þ x0n h þ x00x0

ð7:11Þ

or in an alternative form xnþ1

dx d2 x h2 ¼ xn þ h þ 2 : dt n dt x0n 2!

ð7:12Þ

Here the symbol x0n still represents a number in interval [tn, tnþ1] and in general 6¼ xn . Obviously, Euler’s recursive (7.9) can be obtained after neglecting the residual 2 2 term ddt2x 0 h2! in (7.12). x0n

xn

7.2 Numerical Methods for Transient Stability Analysis

411

Therefore when computing the function value at point n þ 1 from that at n, the error introduced by the approximation is Enþ1

d2 x h2 ¼ 2 : dt x0n 2!

ð7:13Þ

Suppose that within the computing interval [0,tm], the maximum value 2 of ddt2x ¼ f 0 ðx; tÞ is M, then the error Enþ1 should satisfy Enþ1

M 2 h ; 2

ð7:14Þ

where M is independent of the choice of step size h. The errors in (7.13) and (7.14) are due to the approximation made when computing the function value at point n þ 1 from that at n, it is called local truncation error. The truncation error of Euler’s formula is in proportion to h2, and often expressed as of order h2 or 0(h2). It should be noted that before obtaining xnþ1, xn is solved using the same recursive formula, therefore xn itself also contains error. As a matter of fact, when computing xnþ1 based on (7.9), one should take into account the impact of the error of xn, in addition to the impact of the local truncation error associated with neglecting residual term. This error is called global truncation error or simply put truncation error. Consequently the error introduced by the inaccuracy of Euler’s formula is larger than the local truncation error expressed in (7.13) and (7.14). It can be proved that the global truncation error of Euler’s formula is in proportion to h, in other words, it is 0(h). Based on the above discussion, a smaller step size h should be selected to reduce the computational error of the Euler’s formula. But it is false to assert that the smaller the step size h is, the smaller the error would be. In the aforementioned discussion, we did not take into consideration the roundoff error of the computer. When a small step size h is used, the computational effort adversely increases; thus, the impact of rounding errors increases, as illustrated in Fig. 7.3. In the figure, hmin is the step size associated with the minimum error, therefore we cannot merely rely on reducing the step size to reduce computational error. If higher computational precision is desired, a better computational algorithm has to be used. In the above calculations, when computing the function value at tnþ1, only the function value xn at the previous point tn is required, this algorithm is called a single-step algorithm. The algorithms to be presented in this section belong to this category. There are multistep or multivalue algorithms which are more accurate. These algorithms require the information of previous steps (tn, xn), (tnþ1, xnþ1), . . . , (tnkþ1, xnkþ1) when solving for the value xnþ1 corresponding to time tnþ1.

412

7 Power System Transient Stability Analysis

error

minimum error truncation error round-off error

h

hmin

Fig. 7.3 Relationship between error and step size

7.2.1.2

Modified Euler’s Method

The large error of Euler’s method comes from the fact that the derivative dx dt n ¼ f ðxn ; tn Þ of the starting point of an Euler’s polygon is used for the entire segment [tn, tnþ1]. In other words, the slope of each Euler’s polygon is entirely determined by the starting point of the polygon. If the slope of an Euler’s segment is replaced with the average of slopes of starting point and end point, we should expect improved solution precision. This is the basic idea of the modified Euler’s method. For first-order differential equation (7.3), let the initial value is given as t0 ¼ 0, x(t0) ¼ x0, in what follows we introduce the computational steps of the modified Euler’s method. To find out the function value x1 at t1 ¼ h, first compute an approximate value of x1 using Euler’s method: ð0Þ x1

dx ¼ x0 þ h; dt 0

where dx dt 0 ¼ f ðx0 ; t0 Þ. ð0Þ

ð7:15Þ

ð0Þ

When x1 is obtained based on (7.15), substitute t1, x1 into (7.3) to solve for the derivative at the end point of the segment: dxð0Þ ð0Þ ¼ f ðx1 ; t1 Þ: dt 1

7.2 Numerical Methods for Transient Stability Analysis

413

x

Fig. 7.4 Geometrical explanation of the modified Euler’s method

x = x(t) dx dt

(0)

dx dt

1

dx dx + dt dt

x1

0

(1)

x1

2

1 (0) 1

(0)

x1 x0

dx dt

h t2

t1

0

t

dxð0Þ Now the average of dx dt 0 and dt 1 can be used to calculate an improved solution of x1 as follows:

ð1Þ

x1

dx dxð0Þ þ dt 0 dt 1 ¼ x0 þ h: 2

ð7:16Þ

ð1Þ

The solution x1 computed this way better approximates the true solution x1 than ð0Þ

does x1 which is computed using a standard Euler’s method. Figure 7.4 provides a geometrical explanation. To compute (tnþ1, xnþ1) from (tn, xn), the following general formula can be used 9 dx > > ¼ f ðx ; t Þ n n > > dt n > > > > > dx > ð0Þ > > xnþ1 ¼ xn þ h > > dt n > = ð0Þ : ð7:17Þ dx ð0Þ > ¼ f ðxnþ1 ; tnþ1 Þ > > dt nþ1 > > > ð0Þ > > > > dx dx > > þ > > dt dt ð1Þ n nþ1 > xnþ1 ¼ xnþ1 ¼ xn þ h; 2 Eliminate xn in (7.17), the fourth formula of (7.17) can be modified as xnþ1 ¼

ð0Þ xnþ1

a dx þ h; dt nþ1

ð7:18Þ

414

7 Power System Transient Stability Analysis

where a ð0Þ ! dx 1 dx dx ¼ : dt nþ1 2 dt nþ1 dt n As such, the general formula of the modified Euler’s method can be summarized as follows: 9 dx > > ¼ f ðxn ; tn Þ > > > dt n > > > > > dx ð0Þ > > xnþ1 ¼ xn þ h > = dt n a : ð7:19Þ dx 1 dx > ð0Þ > > ¼ f ðx ; t Þ > nþ1 nþ1 dt nþ1 2 dt > > > a n > > > > dx ð1Þ ð0Þ > xnþ1 ¼ xnþ1 ¼ xnþ1 þ h > ; dt nþ1 When solving for xnþ1 based on (7.19), because it takes the same form as that of ð0Þ xnþ1 , the computer code can be simplified. In addition, xn need not be saved after ð0Þ

xnþ1 is obtained, thus computer memory can be saved. In what follows we discuss the local truncation error of modified Euler’s method. To do so, recall the Taylor’s expansion formula of equation (7.10): xnþ1 ¼ xn þ x0n h þ x00n

h2 h3 þ x000 ; x00n 2! 3!

ð7:20Þ

h3 is the residual term of the Taylor’s expansion. n 3! The fourth equation of the modified Euler’s method (7.17) can be re-cast as

where x000 x00

h h ð0Þ ð1Þ xnþ1 ¼ xn þ x0n þ f ðxnþ1 ; tnþ1 Þ: 2 2 Substituting the first formula in (7.17) into the above equation, one obtains h h ð1Þ xnþ1 ¼ xn þ x0n þ f ðxn þ x0n h; tn þ hÞ: 2 2

ð7:21Þ

Expand the third term in the right-hand side of the above equation using Taylor’s formula, h h h2 @f 0 h2 @f f ðxn þ x0n h; tn þ hÞ ¼ f ðxn ; tn Þ þ x þ þ 0ðh3 Þ: 2 2 2 @xn n 2 @t n

7.2 Numerical Methods for Transient Stability Analysis

Since x00n

415

@f 0 @f ¼ xn þ ; @x n @t n

therefore h h h2 f ðxn þ x0n h; tn þ hÞ ¼ x0n þ x00n þ 0ðh3 Þ; 2 2 2 substituting the above formula into (7.21), it follows ð1Þ

xnþ1 ¼ xn þ x0n h þ x00n

h2 þ 0ðh3 Þ; 2

ð7:22Þ

subtracting the above formula from (7.20), we have ð1Þ

h3 0ðh3 Þ: n 3!

Enþ1 ¼ xnþ1 xnþ1 ¼ x000 x00

The above equation shows that the local truncation error of the modified Euler’s method is 0(h3). By the same token, it can be proved that the global truncation error of the modified Euler’s method is 0(h2). [Example 7.1] Solve the following differential equation by the modified Euler’s method dx 2t ¼ x ; dt x where the initial values are t0 ¼ 0 and x0 ¼ 1. [Solution] Taking 0.2 as step length, the computational results are summarized in the following table:

n 0 1 2 3 4 5

tn 0 0.2 0.4 0.6 0.8 1.0

xn 1 1.18667 1.34832 1.49837 2.62790 1.75430

dx dt n

xnþ1

1 0.84959 0.75499 0.69036 0.64500

1.2 1.35658 1.49932 1.63179 1.75690

ð0Þ

tnþ1 0.2 0.4 0.6 0.8 1.0

The true solution of this differential equation is x¼

pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2t þ 1:

dx0 dt nþ1 0.8667 0.7669 0.6990 0.6513 0.6185

ð0Þ dx dx þ dt n dt nþ1 2 0.9333 0.8083 0.7270 0.6708 0.6318

xn 1.18667 1.34832 1.49372 1.62788 1.75430

416

7 Power System Transient Stability Analysis

When t ¼ 1, x ¼ 1.73205, therefore the error is equal to j1:73205 1:7543j ¼ 0:02225: The modified Euler’s method can also be employed to solve first-order differential equations. For instance, for the following differential equations: 9 dx ¼ f1 ðx; y; tÞ > = dt : ð7:23Þ dy > ¼ f2 ðx; y; tÞ ; dt Let the initial values be t0, x0, y0, when step length h is determined, for the first segment, one can compute the approximate value of the true solution as follows: dx ð0Þ x1 ¼ x0 þ h dt 0 ; dy ð0Þ y1 ¼ y0 þ h dt 0 where dx ¼ f1 ðx0 ; y0 ; t0 Þ dt 0 : dy ¼ f ðx ; y ; t Þ 2 0 0 0 dt 0 ð0Þ

ð0Þ

From t1 ¼ h; x1 ; y1 , we have ð0Þ dx ð0Þ ð0Þ ¼ f1 ðx1 ; y1 ; t1 Þ dt 1 ; ð0Þ dy ð0Þ ð0Þ ¼ f2 ðx1 ; y1 ; t1 Þ dt 1 thus the true solution of the differential equation at t should be dx dxð0Þ a þ dt 0 dt 1 dx ð0Þ x1 ¼ x0 þ h ¼ x1 þ h; 2 dt 1 ð0Þ dy dy þ dt 0 dt 1 dya ð0Þ y1 ¼ y0 þ h ¼ y1 þ h; 2 dt 1

7.2 Numerical Methods for Transient Stability Analysis

417

where a ð0Þ dx 1 dx ¼ dt 1 2 dt 1 dya 1 dyð0Þ ¼ dt 1 2 dt 1

! dx dt 0 ! dy dt 0

and so forth. From (7.17), it can be concluded that the modified Euler’s method applied to one segment requires computational effort, that is, two times of that of the Euler’s method. On the other hand, if the same step length is used, the modified Euler’s method provides more accurate calculation results than the Euler’s method. As discussed before, the truncation error of the modified Euler’s method is 0(h2), while the Euler’s method is 0(h). Figure 7.5 illustrates that, when the tolerance is equal to e1, the difference between the required step length of the modified Euler’s method h01 and the Euler’s step length h1 is small. Under such circumstance, the computational effort required by the modified Euler’s method is larger than that of the Euler’s method. When the tolerance is equal to e2, the required step length of the modified Euler’s method h02 is significantly larger than the Euler’s step length h2. Obviously, if h02 > 2h2 , then the total computational effort of the modified Euler’s method is smaller than that of the Euler’s method.

7.2.1.3

Runge–Kutta Method

The modified Euler’s method is based on the observation that xnþ1 can be estimated using the derivatives or slopes of two points in the interval [tn, tnþ1], and since the Taylor’s series of the integral function is approximated by the first three terms, the local truncation error is 0(h3). This has motivated the following question: is it possible to estimate xnþ1 using the derivatives of more points in the interval [tn, tnþ1],

e

Euler Method Modified Euler Method

e1 e2 Fig. 7.5 Comparison between the modified Euler’s method and Euler’s method

h2

h2′

h1

h1′

h

418

7 Power System Transient Stability Analysis

such that more terms of the Taylor’s series can be included? The answer to this question is positive. The well-known Runge–Kutta method is built upon this idea. The most popular Runge–Kutta method is the fourth-order method. In this method, xnþ1 is estimated using the derivatives of four points in the interval [tn, tnþ1], thus the first five terms of Taylor’s series are included in the approximation: xnþ1 ¼ xn þ x0n h þ x00n

4 h2 h3 ð4Þ h þ xð3Þ þ x þ 0ðh5 Þ: n n 2! 3! 4!

The local truncation error of the method is 0(h5), and the global truncation error is 0(h4). For differential (7.3), the following Runge–Kutta formula should be used: 9 1 > xnþ1 ¼ xn þ ðk1 þ 2k2 þ 2k3 þ k4 Þ > > > 6 > > > > k1 ¼ hf ðxn ; tn Þ > > > > = k1 h k2 ¼ hf xn þ ; tn þ 2 2 > > > > > k2 h > > k3 ¼ hf xn þ ; tn þ > > > 2 2 > > ; k4 ¼ hf ðxn þ k3 ; tn þ hÞ

ð7:24Þ

to solve for x1, x2, x3, . . .. [Example 7.2] Solve the first-order differential equation in Example 7.1 using the Runge–Kutta method [Solution] Let the step length h ¼ 0.2, the computational steps are described in the following table: tn 0 0.2 0.4 0.6 0.8 1

xn 1 1.1832292 1.3416668 1.483281 1.612513 1.732141

k1 0.2 0.1698342 0.1490788 0.1348528 0.1240546

tn þ 0.1 0.3 0.5 0.7 0.9

h 2

xn þ

k1 2

1.1 1.267746 1.416026 1.550707 1.674541

k2 0.1836364 0.1588930 0.1420188 0.1295786 0.1199240

tn þ 0.1 0.3 0.5 0.7 0.9

h 2

k2 2 1.0918182 1.262676 1.412676 1.548070 1.672475 xn þ

k3 0.1817274 0.1574990 0.1409600 0.1287436 0.1192452

tn þ h

xn þ k3

0.2 0.4 0.6 0.8 1.0

1.181727 1.340728 1.482627 1.612025 1.731759

k4 0.1686478 0.1488074 0.1346506 0.1238970 0.1153728

The above table shows that, based on the Runge–Kutta method, the value of x at t ¼ 1 is x ¼ 1.732141. Comparing this result with the true solution, the error is equal to j1:73205 1:732141j ¼ 0:00009; which is a much better result in comparison with the result obtained in Example 7.1.

7.2 Numerical Methods for Transient Stability Analysis

419

The Runge–Kutta method can also be used to solve first-order differential equations. As an example, the differential equation (7.23) can be solved using the following recursive formula: 1 xnþ1 ¼ xn þ ðk1 þ 2k2 þ 2k3 þ k4 Þ; 6 1 ynþ1 ¼ yn þ ðl1 þ 2l2 þ 2l3 þ l4 Þ; 6 where 9 k1 ¼ hf1 ðxn ; yn ; tn Þ > > > > k1 l1 h > > > > k2 ¼ hf1 xn þ ; yn þ ; tn þ 2 2 2 = ; k2 l2 h > > > k3 ¼ hf1 xn þ ; yn þ ; tn þ > 2 2 2 > > > > ; k4 ¼ hf1 ðxn þ k3 ; yn þ l3 ; tn þ hÞ 9 l1 ¼ hf2 ðxn ; yn ; tn Þ > > > > k1 l1 h > > > > l2 ¼ hf2 xn þ ; yn þ ; tn þ 2 2 2 = : k2 l2 h > > > l3 ¼ hf2 xn þ ; yn þ ; tn þ > 2 2 2 > > > > ; l4 ¼ hf2 ðxn þ k3 ; yn þ l3 ; tn þ hÞ Although the Runge–Kutta method has the advantage of higher precision, it requires larger computational effort which is four times that required by the Euler’s method. The trend is that multiple-step methods, which require less computational effort, are replacing Runge–Kutta methods when higher computational accuracy is required. Runge–Kutta methods are typically used as auxiliary methods only to initiate multiple-step methods in the first few steps. 7.2.1.4

Implicit Integration Methods

Explicit and implicit methods are the major categories of solution methods for differential equations. The methods described in the previous sections belong to the category of explicit methods. From (7.9), (7.17), and (7.24), one can see that the right-hand sides of the formulas are known quantities; therefore, the value of the end point xnþ1 can be directly computed using those recursive formulas. In contrast, an implicit method does not work with recursive equations, it first converts differential equations into difference equations, then solves for the value xnþ1 using the methods of difference equations. Let us first introduce the method of the trapezoidal rule.

420

7 Power System Transient Stability Analysis

Fig. 7.6 Geometrical illustration of trapezoidal rule

dx dt

f (xn, tn)

f (xn+1, tn)

D

C

A

0

tn

B tn+1

t

When xn at tn is known, the function value xnþ1 at time tnþ1 ¼ tn þ h of the differential equation (7.3) can be solved using the following formula: Z tnþ1 xnþ1 ¼ xn þ f ðx; tÞdt: ð7:25Þ tn

The solution of the definite integral of the above equation is equal to the area of the shaded region in Fig. 7.6. Observe that if the step size h is sufficiently small, the graph of the function f(x, t) between tn and tnþ1 can be approximated by a straight line as illustrated in the figure. Apparently the area of the shaded region is equal to the area of the trapezoid ABCD. Equation (7.25) can thus be reformulated as h xnþ1 ¼ xn þ ½f ðxn ; tn Þ þ f ðxnþ1 ; tnþ1 Þ: 2

ð7:26Þ

This is the difference equation of the trapezoidal rule. Obviously, one cannot rely on certain recursive formula to compute xnþ1 because the right-hand side of (7.26) also includes unknown xnþ1. Equation (7.26) has to solve as an algebraic equation to find xnþ1. Generally speaking, the idea of implicit methods is to transform a numerical initial value problem of differential equations into a sequence of algebraic equation problems. For example, given starting point t0 and x0, according to (7.26) the difference equation for the first step should be h x1 ¼ x0 þ ½f ðx0 ; t0 Þ þ f ðx1 ; t0 þ hÞ; 2 where the only unknown variable is x1, which can be solved for using the methods for solving algebraic equation. Given t1 and x1, based on (7.26), the difference formula for the next step should be h x2 ¼ x1 þ ½f ðx1 ; t1 Þ þ f ðx2 ; t1 þ hÞ 2 from which x2 can be computed, and so forth.

7.2 Numerical Methods for Transient Stability Analysis

421

If f(xn, tn) and f(xnþ1, tnþ1) are viewed as the slopes of the integral curve at the starting point and terminating point of the interval [tn, tnþ1], then it is reasonable to term the implicit trapezoidal rule as an implicit modified Euler’s method. In other words, difference equation (7.26) can be viewed as the solution formula of the implicit modified Euler’s method. In fact, the idea of implicit methods are applicable not only to the modified Euler’s method, but also to the previously mentioned Euler’s method, Runge–Kutta method, and multistep methods. For example, the recursive formula of the Euler’s method (7.9) can be rewritten as xnþ1 ¼ xn þ x0nþ1 h ¼ xn þ f ðxnþ1 ; tnþ1 Þh:

ð7:27Þ

Changing the derivate value x0n of the starting point of the interval [tn, tnþ1] to one obtains the implicit Euler’s method. Equation (7.27) is the difference formula of the implicit Euler’s method. The difference equations (7.26) and (7.27) can be nonlinear as a result of the nonlinearity of the function f(x, t) in (7.3). Therefore the algorithms for implicit methods are more complex than those of explicit methods. It is not difficult to find out that the truncation error of implicit trapezoidal rule is introduced by the approximation of replacing the trapezoid with the shaded area (see Fig. 7.6). Using the same arguments as before, one can prove that the local truncation error of difference equation (7.26) is 0(h3). The advantage of implicit methods over explicit methods is that a larger step size can be used. This issue involves the numerical stability of numerical initial value problems; readers are referred to relevant references. Here we illustrate using a simple example. Suppose we have the following differential equation: x0nþ1 ,

dx ¼ 100x: dt

ð7:28Þ

The initial values are t ¼ 0 and x0 ¼ 1. For the above differential equation, the true solution is x ¼ e100t. This is an exponential function, as depicted in Fig. 7.7. When the step length is equal to h ¼ 0.025, the numerical solution using the Euler’s method is as follows: Steps 0 1 2 3

tn 0.000 0.025 0.050 0.075

xn 1 1.5 2.25 3.375

x0n

x0n h

100 150 225

2.5 3.75 5.625

Observe that the function value oscillates as time increases, and the magnitude of the oscillation increases, as illustrated in the dotted line in Fig. 7.7. Mathematically,

422

7 Power System Transient Stability Analysis

Fig. 7.7 Illustration of solutions obtained using different methods

x

Implicit

Euler method

Euler Method

0

0.025

0.05

0.075

t

this indicates that the numerical solution obtained using the Euler’s method is not stable. This situation can be avoided if the implicit Euler’s method is used. Let us first transform (7.28) into a difference equation as follows: xnþ1 ¼ xn þ x0nþ1 h ¼ xn 100xnþ1 h: Thus xnþ1 ¼

xn : 1 þ 100h

When h ¼ 0.025, xnþ1 ¼

xn : 3:5

One obtains the following calculation results:

Steps 0 1 2 3

tn 0.000 0.025 0.050 0.075

xn 1 1/3.5 (1/3.5)2 (1/3.5)3

The function value in the above table decays as time increases, see Fig. 7.7.

7.2 Numerical Methods for Transient Stability Analysis

423

To explain the relationship between step size and numerical solutions, we rewrite differential equation (7.28) into more general form: dx x ¼ ; dt T

ð7:29Þ

where the constant T has the unit of time, which is termed the time constant. Substituting (7.29) into the Euler’s equation (7.9), we have h xnþ1 ¼ xn 1 : T Therefore h nþ1 xnþ1 ¼ x0 1 : T

ð7:30Þ

Obviously in order for x to be a monotonically decaying function, the right-hand side of (7.30) has to meet the following condition: 0 > ¼ bx ¼ > > R2a þ Xd Xq R2a þ Xd Xq > > > > Rag sin d Xqg cos d Ra cos d Xd sin d > > > ¼ g ¼ y = 2 2 Ra þ Xd Xq Ra þ Xd Xq : 2 2 Ra ðXd Xq Þ sin d cos d Xd cos d þ Xq sin d > > > ¼ B ¼ x > > R2a þ Xd Xq R2a þ Xd Xq > > > 2 2 Xd sin d Xq cos d Ra þ ðXd Xq Þ sin d cos d > > > ; ¼ G ¼ y R2 þ Xd Xq R2 þ Xd Xq a

ð7:40Þ

a

Substituting the current injection formula derived from (7.39) into network equation (7.36), and applying some simple manipulations, one can conclude that the interconnection of a generator is equivalent to a current injection at the corresponding node: 0 Ix gx bx Ed ¼ : Iy0 by gy Eq

7.3 Network Mathematical Model for Transient Stability Analysis

433

This current is termed generator pseudocurrent. Furthermore, the corresponding block of the admittance matrix of the network should be added to by a matrix as follows: Gx Bx : By Gy It is not difficult to see that, after connecting a generator into the system, the network equations during the stability study period are still linear, however, the generator pseudocurrents and the corresponding admittance matrix are functions of the generator variables Ed , Eq , and d. Thus these linear equations are time varying. If simpler synchronous machine models are used in the study, the network equations can be simplified too. These simplified equations appear as n-order equations in the complex plane. Unless there is a fault or switch change, the network equations remain unaltered. Thus during the study period, the coefficient matrix of the network equations needs to be refactorized using triangular factorization only when there is a fault or switch change. In what follows we discuss the network model associated with two simplified machine models. If the effect of damper windings is not considered, the varying E0 q or E0 q ¼ C model for synchronous machines in Table 7.1 should be applied. In this case, (7.39) can be reformulated as 2 3 0 0 Xd0 Xq X þ X X X q q d d Ra sin 2d þ cos 2d 7 6 2 2 2 7 6 0 0 2 2 6 7 Ix Ra þ Xd Xq Ra þ Xd Xq 6 7 ¼6 0 0 0 7 Iy Xd Xq 6 Xd þ Xq þ Xd Xq cos 2d Ra þ sin 2d 7 4 5 2 2 2 2 0 2 0 Ra þ Xd Xq Ra þ Xd Xq 0 E cos d Vx q0 : ð7:41Þ Eq sin d Vy From the above, one obtains the formula of generator current into node i represented in the complex domain: I_i ¼ I_i0 Yi0 V_ i ;

ð7:42Þ

0 Rai j 12 ðXdi þ Xqi Þ ; 0 X R2ai þ Xdi qi

ð7:43Þ

where Yi0 ¼

9 1 0 Rai jXqi _ 0 2 ðXdi Xqi Þ j2di ^_ > = I_i0 ¼ 2 E j e V i 0 0 Rai þ Xdi Xqi qi R2ai þ Xdi Xqi : > ; 0 0 jd i E_ qi ¼ Eqi e

ð7:44Þ

434

7 Power System Transient Stability Analysis

Fig. 7.9 Generator equivalent circuit when damper winding is not considered

· Vi

o

I·i I·i′

i

Yi′

o

The concept underlying (7.42) can be explained using the circuit model illustrated in Fig. 7.9, where Y0 i is called generator pseudoadmittance and is dependent only on generator parameters. The generator pseudoadmittance can be incorporated into the network admittance matrix; I_i0 is the generator pseudocurrent injection which is related to generator terminal voltage. The network equations are now nonlinear, thus can only be solved using an iterative procedure. As one example, assume an initial value of voltage V_ i , compute I_i0 based on (7.44), then solve the network equations for an improved solution of V_ i , taking I_i0 as current injection. This procedure is repeated until convergence is reached. In normal computational steps, the iteration converges within 2–3 steps; while under fault or switching conditions, it may take a few more steps to obtain a converged solution [196]. If synchronous machines are represented by classical models, the effects of damper windings and salient poles are neglected; in addition, the transient voltages E0 of generators behind X0 d are assumed to be constants during the stability study period. This situation is shown in Table 7.1, where E0 ¼ Eq0 ¼ C and Xq ¼ Xd0 . Correspondingly, from (7.42)–(7.44), it follows that Yi0 ¼

1 0 ; Rai þ jXdi

9 1 = _0 > E i 0 Rai þ jXdi : > ; 0 0 jdi _ Ei ¼ Ei e I_i0 ¼

ð7:45Þ

ð7:46Þ

Obviously generator pseudocurrent I_i0 is independent of generator terminal voltage V_i ; thus, once pseudoadmittance Y0 i is incorporated into the network admittance matrix, the network equations can be solved by direct Gauss elimination since I_i0 is a known quantity. 7.3.1.2

Relationship Between Loads and Network

Depending upon the characteristics of loads, the ways loads are treated are different: 1. If loads are represented by constant impedance models, the constant impedances can be incorporated into the network admittance matrix.

7.3 Network Mathematical Model for Transient Stability Analysis

435

2. If loads are modeled as dynamic devices and only the mechanical–electrical interactions of induction motors in synthesized loads are taken into consideration, loads are still modeled as impedances. However, these impedances are not constant during the stability study period, but vary as the slip-speeds of the induction motors vary. Therefore the impedances representing induction motor loads must be updated, given the slips of the induction motors in each step of the transient stability computation. This means that the diagonal elements of the network admittance matrix are varying in the calculation. The network admittance matrix has to be refactorized in each step of the transient stability calculation when solving the network equations. 3. Again, if loads are modeled as dynamic devices and only the mechanical– electrical interactions of induction motors in synthesized loads are taken into consideration, they can be represented using the Norton equivalent circuit described in Sect. 5.5.2, as illustrated in Fig. 7.10. That is, the load impedances R þ jX and KM (r1 þ jx0 ) are incorporated into the network; thus, the loads become simple current sources. This treatment is similar to the way generators connected to the network are treated. In the above load representations, the underlying networks are linear. 4. If loads are modeled based on steady-state voltage characteristics, the corresponding node current injections are nonlinear functions of node voltage; as a result, the network equations are nonlinear. According to (6.148) and (6.153), the steady-state voltage characteristics of loads have two formulations, these are the second-order polynomial formulation and exponential formulation: " # 9 9 > Vi 2 Vi > Vi m > > > Pi ¼ Pið0Þ aP þ bP þ cP > > > = Pi ¼ Pið0Þ V = Við0Þ Við0Þ ið0Þ " # : ð7:47Þ n 2 > > Vi > > Vi Vi > > Qi ¼ Qið0Þ ; Qi ¼ Qið0Þ aQ þ bQ þ cQ > > ; Við0Þ Við0Þ Við0Þ Note that the active and reactive powers in the above equations are the loads absorbed from the network.

VM

Network

eM ′

R+jX

KM (r1 + jx′) KM (r1 + jx′)

Fig. 7.10 Load representation

436

7 Power System Transient Stability Analysis

Node voltage, current injection, and power injection are connected by the following relationship: Pi jQi ¼ V_ i^I_i ¼ ðVxi þ jVyi ÞðIxi jIyi Þ from which it is easy to find the relationships between load current injections and node voltages. If loads are represented by second-order polynomial forms, the load current injections are found to be 9 Pið0Þ aP Vxi þ Qið0Þ aQ Vyi Pið0Þ bP Vxi þ Qið0Þ bQ Vyi Pið0Þ cP Vxi þ Qið0Þ cQ Vyi > > q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ > 2 > Við0Þ Vxi2 þ Vyi2 > = Við0Þ Vxi2 þ Vyi2 ; Qið0Þ aQ Vxi Pið0Þ aP Vyi Qið0Þ bQ Vxi Pið0Þ bP Vyi Qið0Þ cQ Vxi Pið0Þ cP Vyi > > > q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ Iyi ¼ þ þ > 2 > Við0Þ Vxi2 þ Vyi2 ; Við0Þ Vxi2 þ Vyi2 Ixi ¼

ð7:48Þ where the terms proportional to the square of voltages can be incorporated into the network admittance matrix as constant admittances, thus (7.48) is reduced to the last two terms only. If loads are represented by exponential functions, the load current injections are found to be 9 Pið0Þ Vim2 Vxi Qið0Þ Vin2 Vyi > > > Ixi ¼ > m n = Vð0Þ Vð0Þ : Qið0Þ Vin2 Vxi Pið0Þ Vim2 Vyi > > > Iyi ¼ > ; n m Vð0Þ Vð0Þ

7.3.1.3

ð7:49Þ

The Relationship Between FACTS Devices and the Network

Here we will only describe the relationship between SVC/TCSC and the network; also the relationship between the other FACTS devices and the network can be derived following the same concept. 1. SVC: In general an SVC is connected to a high-voltage bus of the network through a transformer (let the index of this bus be i). Thus the shunt susceptance of the device is equal to j

BSVC : 1 XT BSVC

7.3 Network Mathematical Model for Transient Stability Analysis

437

From the relationship between nodal voltage V_i and current injection I_i it is not difficult to find the real and imaginary parts of the current injection as follows: Ixi ¼

9 > > = ; > > Vxi ;

BSVC Vyi 1 XT BSVC

Iyi ¼

BSVC 1 XT BSVC

ð7:50Þ

where XT is the impedance of the transformer, BSVC is the equivalent susceptance of the SVC, Vxi and Vyi are the real and imaginary parts of the voltage of the high-voltage bus. 2. TCSC: Regardless of the place where the TCSC is connected in series in a line, it is always possible to put two nodes around the TCSC, let the nodes be i and j. As a matter of the fact, the role a TCSC plays is equivalent to two current sources having the same magnitude but opposite directions at node i and j, the current injections are easily derived as Ixi ¼ Ixj ¼ BTCSC ðVyi Vyj Þ Iyj ¼ Iyi ¼ BTCSC ðVxi Vxj Þ

) ;

ð7:51Þ

where BTCSC is the equivalent susceptance of the TCSC, Vxi, Vyi, Vxj, and Vyj are the real and imaginary parts of the voltages of the two nodes.

7.3.1.4

The Relationship Between Two-Terminal HVDC and the Network

Let variables with subscript ‘‘d’’ denote quantities on the DC side, and subscripts ‘‘R’’ and ‘‘I’’ denote rectifier and inverter sides (they have the same meaning in subsequent text), respectively. From (5.52)–(5.54) and (5.57) (where kg 1), the steady-state equations of the rectifier are as follows: VdR ¼ kR VR cos a XcR IdR VdR ¼ kR VR cos ’R IR ¼ kR IdR PR ¼ VdR IdR

9 > > > > > > > > > > > =

> > > pﬃﬃﬃ > > ¼ 3VR IR cos ’R > > > > > > ;

QR ¼ PR tg ’R

ð7:52Þ

438

7 Power System Transient Stability Analysis

and the inverter side steady-state equations: 9 > > > > > > ¼ kI VI cos ’I > > = ¼ kI IdI : > > pﬃﬃﬃ > > ¼ VdI IdI ¼ 3VI II cos ’I > > > > ; ¼ PI tg ’I

VdI ¼ kI VI cos b þ XcI IdI VdI II PI QI

ð7:53Þ

Based on (7.52) and (7.53), the power injections into the AC system by the DC system can denoted by functions of Id, a, b, VxR, VyR, VxI, and VyI. The power injection into the AC bus from the rectifier is given by 9 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 > 2 þ V 2 cos a > PR ¼ PR ¼ VdR IdR ¼ XcR IdR kR IdR VxR > yR > > > pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 2 = kR VR VdR 2 2 2 QR ¼ QR ¼ PR ¼ IdR kR VR VdR VdR > > rﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > 2 2 2 2 2 2 2 2 > ; ¼ IdR kR ðVxR þ VyR Þ sin a þ 2kR XcR IdR VxR þ VyR cos a XcR IdR > ð7:54Þ and the power injection from the inverter is 9 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > 2 2 > PI ¼ PI ¼ VdI IdI ¼ þ kI IdI VxI þ VyI cos b > > > > pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > q ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ 2 = kI2 VI2 VdI 2 2 2 QI ¼ QI ¼ PI ¼ IdI kI VI VdI : VdI > rﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ> > > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ > 2 2 > ; ¼ IdI kI2 ðVxI2 þ VyI2 Þ sin2 b 2kI XcI IdI VxI2 þ VyI2 cos b XcI IdI > 2 XcI IdI

ð7:55Þ

Thus the current injections into the AC system from the rectifier and inverter are obtained as IxR

PR VxR þ QR VyR ¼ ; 2 þ V2 VxR yR

PI VxI þ QI VyI IxI ¼ ; 2 VxI2 þ VyI

9 PR VyR QR VxR > > > IyR ¼ > 2 þ V2 = VxR yR : > PI VyI QI VxI > > > IyI ¼ ; 2 VxI2 þ VyI

ð7:56Þ

Substituting (7.54) and (7.55) into (7.56), and eliminating variables PR , QR , PI , and QI , the current injections IxR and IyR become functions of IdR, a, VxR and VyR, and IxI and IyI are functions of variables IdI, , bVxI, and VyI.

7.3 Network Mathematical Model for Transient Stability Analysis

7.3.2

439

Modeling Network Switching and Faults

When a fault or switch change is applied to a network, the network admittance matrix needs to be correctly modified. If the fault or switch is three-phase symmetrical, for example, a three-phase short circuit, the removal of three phases of a device, the forced connection of a series capacitor, the introduction or removal of a braking resistor, etc., the modification to the admittance matrix is straightforward because such a fault or switching operation results in a parameter change in a shunt branch or series branch of the network. Most of short circuits and device removals are unsymmetrical, and thus have to be analyzed using a symmetrical components method. In addition to dealing with the positive sequence network of the power system under study, one has to consider the negative sequence and zero sequence networks. On the other hand, in stability studies we are mostly interested in the quantities of the positive sequence network, paying little if any attention to the quantities of the negative and zero sequence networks. The effects of the negative and zero sequence networks can be modeled using an equivalent impedance viewed from the positive sequence network. When analyzing unsymmetrical problem using the concept of symmetrical components, phase A is often taken as the reference, the boundary conditions of various types of short circuit or open-conductor are expressed in terms of the sequence quantities of phase A. When a short circuit or an open-conductor occurs, the phase that exhibits different behavior compared with the other two phases is called the special phase. For instances, the special phase in a single-line-to-ground fault is the phase connected to ground; the special phase in a double-line-to-ground or line-to-line fault is the phase that is not faulted. The special phase in a singleline-open-conductor is the phase that is open, and the special phase in a double-lineopen-conductor is the phase that is intact. When the special phase of a short circuit or open-conductor is phase A, the three sequence networks can be directly connected to form the so-called composite sequence network according to certain boundary conditions. This is equivalent to connecting supplementary impedance to the faulted terminals of the positive sequence network. The size of the supplementary impedance depends on the type of fault, as illustrated in Tables 7.2 and 7.3. Here the ‘‘faulted terminals’’ mean, in a short circuit the terminals between faulted bus and ground and in an open-conductor fault the two nodes resulting from the open-conductor. The network admittance matrices under these circumstances can be easily formed. Table 7.2 Supplemental impedances of short circuits Type of short circuit Supplemental impedance ð2Þ ð0Þ Single-line-to-ground Z þZ Double-line-to-ground Line-to-line

S S ð2Þ ð0Þ

ð2Þ

ð0Þ

ZS ZS =ðZS þ ZS Þ ð2Þ ZS

ð2Þ

ZS is the self-impedance of the short circuit in negative sequence network, ð0Þ

ZS is the self-impedance of the short circuit in zero sequence network

440

7 Power System Transient Stability Analysis Table 7.3 Supplemental impedances of open-conductor Type of open-conductor Supplemental impedance Single-line-open-conductor Z(2)Z(0)/(Z(2) þ Z(0)) Double-line-open-conductor Z(2) þ Z(0) (2) Z is the equivalent impedance of the open-conductor terminals in the negative sequence network, Z(0) is the equivalent impedance of the openconductor terminals in the zero sequence network Table 7.4 The ratios of ideal transformers Special phase Sequence Zero Positive Negative A 1 1 1 B 1 a2 a C 1 a a2

When the special phase is not phase A, there is a complex operator a ¼ ej120 in the boundary conditions, therefore the three sequence networks cannot be directly connected to form a combined sequence network. However, we can connect the three sequence networks via three ideal transformers with ratios 1:n0 (0), 1:n0 (1), and 1:n0 (2) in the zero, positive, and negative sequence networks. The two sides of these transformers have the same voltage/current ratios thus the transformers introduce no losses. For different special phases, these transformers have different ratios in different sequence networks, as described in Table 7.4. After introducing ideal transformers, the various types of unsymmetrical short circuit and open-conductor can be classified into two categories: series and shunt (or parallel) faults based on the topology of the three sequence networks. The faults belonging to the series category include single-line-to-ground, double-line-openconductor, and single-line-to-ground of a series capacitor. The boundary conditions of these faults are as follows: the sum of three sequence voltages is zero, and the sequence currents are identical in the nonstandard ratio side of the transformer. The faults belonging to the shunt category include double-line-to-ground, single-lineopen-conductor, and double-line short circuit of capacitors. The boundary conditions of this class of faults are as follows: in the nonstandard ratio side of the transformer, the sum of sequence currents is equal to zero, and the sequence voltages are equal. When simultaneous short circuits or open-conductors occur, and they occur in different phases, the method for handling single faults can still be applied to modify the admittance matrix of the positive sequence network, but now the concept of supplementary impedance is generalized to that of synthesized impedance matrix. In what follows, we introduce the basic concept of synthesized impedance matrix using single-line-to-ground and single-line-open-conductor faults as examples. Suppose a single-line-to-ground fault occurs at bus k (let this be fault 1), and a single-line-open-conductor occurs between buses i and j (let this be fault 2), and the two faults occur in different phases. By the boundary conditions of the three sequence components at the place where a fault occurs, the combined sequence

7.3 Network Mathematical Model for Transient Stability Analysis

441

•

n 1′ (1 ) : 1

V1(1 )

k

i

o

j

1 : n ′2 (1)

I 2(1 ) •

I 1(1 ) •

k

i

o

j

Positive sequence

•

V2(1 )

Positive sequence

n 1′ ( 2 ) : 1

k

i

o

j

k

i

o

j

Negative sequence

zero sequence

a

I 2( 2 ) •

•

I1( 2 )

1 : n 2′ ( 2 )

•

V1 ( 2 )

k

i

o

j

Negative sequence

n1( 2 ) : 1

•

V2( 2 ) 1 : n 2( 2 )

•

•

I1( 0 )

I 2( 0 )

•

V1( 0 )

i

o

j

0ð1Þ

0ð1Þ

Zero sequence

n1( 0 ) : 1

b

k

•

V2( 0 ) 1 : n 2( 0 )

Fig. 7.11 Combined sequences of two simultaneous faults 0ð2Þ

0ð2Þ

network can be obtained as in Fig. 7.11a. In the figure, n1 , n2 , n1 , and n2 are the ratios of ideal transformers, the specific values of them depending on the special phases. For ease of mathematical manipulation, let us recast the combined sequence network in Fig. 7.11a as that in Fig. 7.11b. It is not difficult to see that the ratios in the two figures obey the following relationships: ð2Þ

0ð2Þ

0ð1Þ

n1 ¼ n1 =n1 ;

ð2Þ

0ð2Þ

0ð1Þ

n2 ¼ n2 =n2 ;

ð0Þ

0ð1Þ

n1 ¼ 1=n1 ;

ð0Þ

0ð1Þ

n2 ¼ 1=n2 :

In the following, we derive the impedance matrix Zf viewed from the fault buses of the positive sequence network into the negative and zero sequence network based on the combined sequence network. We call Zf the synthesized impedance matrix of simultaneous faults. In Fig. 7.11b, the single-line-to-ground part on the left forms a loop circuit, let ð1Þ the loop current be I_ , and the single-line-open-conductor part on the right forms 1

ð1Þ ð0Þ two independent loop circuits, let the currents in these circuits be I_2 and I_2 . ð2Þ ð2Þ ð0Þ ð0Þ Therefore the currents I_1 , I_2 , I_1 , and I_2 of the faulted buses in the negative and zero sequence networks can be obtained in terms of these loop currents as follows:

IS ¼ CIL ;

ð7:57Þ

where C the coincidence matrix is dependent on fault conditions. The definitions of the symbols are 2

3 ð2Þ I_1 6 _ð2Þ 7 6I 7 IS ¼ 6 2ð0Þ 7; 4 I_1 5 ð0Þ I_ 2

2

3 ð1Þ I_1 6 7 IL ¼ 4 I_2ð1Þ 5; ð0Þ I_ 2

2

3 1 0 0 6 0 1 1 7 7: C¼6 41 0 0 5 0 0 1

ð7:58Þ

442

7 Power System Transient Stability Analysis

Based on loop voltage equations, the relationship among the voltages of the faulted buses in each sequence can be obtained as VL ¼ CT VS ;

ð7:59Þ

where CT is the transpose of matrix C and 2

3 ð1Þ V_ ok 6 7 VL ¼ 4 V_ ð1Þ 5; ji 0

2

ð2Þ 3 V_1 6 _ ð2Þ 7 6V 7 VS ¼ 6 2ð0Þ 7: 4 V_1 5 ð0Þ V_

ð7:60Þ

2

From the transformer nonstandard ratio side point of view, the relationship among the currents and voltages of negative and zero sequence networks is expressed as "

"

ð2Þ V_ 1 ð2Þ V_ 2 ð0Þ V_ 1 ð0Þ V_ 2

#

" ¼

#

" ¼

Z11 ð2Þ Z21

ð2Þ

Z12 ð2Þ Z22

ð0Þ

Z12 ð0Þ Z22

Z11 ð0Þ Z21

ð2Þ

ð0Þ

#"

#"

# ð2Þ I_1 ð2Þ ; I_2

ð7:61Þ

# ð0Þ I_k ð0Þ : I_i

ð7:62Þ

Because of the existence of ideal transformers in negative and zero sequence networks, the impedance matrices in (7.61) and (7.62) are unsymmetrical, in general. The computation of the elements of these matrices will be introduced in subsequent sections. Let us incorporate (7.61) and (7.62) into a single equation: 2

3 2 ð2Þ ð2Þ V_ 1 Z11 6 _ ð2Þ 7 6 ð2Þ V 6 2 7 6 Z21 6 ð0Þ 7 ¼ 6 4 V_ 1 5 4 0 ð0Þ V_ 0 2

ð2Þ

Z12 ð2Þ Z22 0 0

0 0 ð0Þ Z11 ð0Þ Z21

32 ð2Þ 3 0 I_1 76 _ð2Þ 7 0 76 I2 7 ð0Þ 76 ð0Þ 7 Z12 54 I_1 5 ð0Þ ð0Þ Z22 I_2

ð7:63Þ

or in compact form: VS ¼ ZIS :

ð7:64Þ

Making use of the matrix Z and coincidence matrix C, the relationship between positive sequence voltage and current can be derived. To this end, substituting (7.64) and (7.57) into (7.59), the relationship between faulted loop voltage and current is found to be VL ¼ ZL IL ;

ð7:65Þ

7.3 Network Mathematical Model for Transient Stability Analysis

443

where ZL is termed the loop impedance matrix, defined by ZL ¼ CT ZC:

ð7:66Þ

In this example, 2 ð2Þ 3 Z11 1 0 1 0 6 ð2Þ 6Z ZL ¼ 4 0 1 0 0 56 21 4 0 0 1 0 1 0 2 0 3 0 0 Z11 Z12 Z13 0 0 0 5 4 ¼ Z21 Z22 Z23 : 0 0 0 Z31 Z32 Z33

ð2Þ

Z12 ð2Þ Z22 0 0

2

0 0 ð0Þ Z11 ð0Þ Z21

32 0 1 76 0 76 0 ð0Þ 76 Z12 54 1 ð0Þ 0 Z22

0 1 0 0

3 0 7 1 7 7 05 1 ð7:67Þ

ð0Þ

Eliminating current I_2 in (7.65), it follows: "

ð1Þ V_ ok ð1Þ V_ ji

#

" ¼

Z11

Z12

Z21

Z22

#"

# ð1Þ I_1 ð1Þ ; I_

ð7:68Þ

2

where the elements Zmn (m and n can be equal to 1 or 2) of the impedance matrix are computed based on: 0 Zmn ¼ Zmn

0 0 Zm3 Z3n ; 0 Z33

ð7:69Þ

(7.68) is rewritten in compact form as Vf ¼ Z f I f :

ð7:70Þ

Finally, the impedance matrix Zf, viewed from the faulted buses of the positive sequence network into the negative and zero sequence networks, is obtained. Equation (7.70) can also be expressed in the form of synthesized admittance matrix as follows: I f ¼ Yf Vf ;

ð7:71Þ

where Yf ¼ Z1 f . Once Yf is determined, the elements of the matrix can be incorporated into the correct position of the admittance matrix of the positive sequence network. In this example, notice that ð1Þ ð1Þ V_ok ¼ V_k ;

ð1Þ ð1Þ ð1Þ V_ji ¼ V_ j V_i ;

ð1Þ ð1Þ I_k ¼ I_1 ;

ð1Þ ð1Þ I_i ¼ I_2 ;

ð1Þ ð1Þ I_j ¼ I_2 :

ð7:72Þ

444

7 Power System Transient Stability Analysis

The above relationships together with (7.71) give us the relationship among the voltages and currents at node k, i and j in the positive sequence network: 2

3 2 ð1Þ Y11 I_k 6 _ð1Þ 7 6 Y 4 Ii 5 ¼ 4 21 ð1Þ Y21 I_

Y12 Y22 Y22

j

32 ð1Þ 3 Y12 V_ k 7 6 Y22 54 V_ ið1Þ 7 5: ð1Þ _ Y22 Vj

ð7:73Þ

In summary, the calculation of synthesized impedance matrix includes the following steps: 1. Form the impedance matrix of the faulted buses of the negative and zero sequence network (refer to (7.63)) 2. By use of the coincidence matrix that represents the boundary conditions of simultaneous faults, form the loop impedance matrix ZL (refer to (7.66) and (7.67)) 3. Eliminate the closed circuit from the synthesized impedance matrix Zf (refer to (7.68) and (7.69)) In what follows we describe the above steps in detail. (1) Forming the impedance matrices of the faulted buses of the negative and zero sequence networks: In a transient stability study, the admittance matrices of each sequence network should be formed first, followed by calculation of the triangular factors for these matrices. In this way the impedance matrices of the faulted buses of each sequence network can be easily obtained given the fault information. For the negative sequence network, observing Fig. 7.11b it is not difficult to see that, if one injects unity current into the nonstandard ratio node k of the ideal ð2Þ transformer with zero current injections to the other nodes, that is, I_k ¼ 1 and ð2Þ I_m ¼ 0 (m is a node other than node k), then solve the equation of the negative ð2Þ sequence network including the ideal transformer for voltages V_ and k

ð2Þ ð2Þ ð2Þ V_ ij ¼ V_ i V_ j . These quantities are the desired quantities for the first ð2Þ

ð2Þ

column Z11 and Z21 of the impedance matrix in (7.71). More specifically, injecting unity current into node k of the nonstandard transformer is equivalent to injecting into node k of negative sequence network 0ð2Þ ð2Þ a current I_k ¼ n^1 , thus after performing sparse forward substitution and backward substitution on the admittance matrix of the negative sequence 0ð2Þ 0ð2Þ 0ð2Þ 0ð2Þ network, voltages V_ k and V_ ij ¼ V_ i V_ j are obtained; in addition, we ð2Þ ð2Þ 0ð2Þ ð2Þ ð2Þ 0ð2Þ have V_ ¼ n V_ and V_ ¼ n V_ . k

1

k

ij

2

ij

By the same token, injecting into nodes i and j of the negative sequence ð2Þ ð2Þ network currents þ n^2 and n^2 , and performing sparse forward substitution

7.3 Network Mathematical Model for Transient Stability Analysis

445

0ð2Þ 0ð2Þ 0ð2Þ 0ð2Þ and backward substitution, one obtains voltages V_ k and V_ij z ¼ V_ i V_j , ð2Þ ð2Þ 0ð2Þ ð2Þ ð2Þ 0ð2Þ furthermore, V_ ¼ n V_ , V_ij ¼ n V_ ij , the quantities which we seek for k ð2Þ

1

k ð2Þ

2

the elements Z12 and Z22 of the second column in the impedance matrix in (7.61). The same principle applies to compute the elements of the impedance matrix in (7.62). (2) Forming loop impedance matrix from the coincidence matrix: As discussed before, the combined sequence network of a series fault is formed by putting the three sequence networks together in series, therefore there is only one independent loop circuit. The combined sequence network of a shunt fault is formed by putting together the three sequence networks in parallel, resulting in two independent loop circuits. Besides, a line-to-line fault is viewed as a special shunt fault. From (7.57) and (7.59), the coincidence matrix C expresses the relationship between the loop current of the boundary circuit of the combined sequence network and the current of the faulted buses of the negative and zero sequence networks. Thus the number of rows of the coincidence matrix equals the dimension of IS, that is, two times the number of simultaneous faults (when a line-to-line fault occurs, an empty faulted bus in the zero sequence network is designated). The number of columns of the coincidence matrix equals the dimension of IL. A series fault occupies one column in the coincidence matrix as illustrated below: ½ 0 0 1 0 0 0 0 1 0 . . . 0 T ; |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} corresponding to the current of the faulted bus in negative sequence network

corresponding to the current of the faulted bus in zero sequence network

where the column number of the nonzero is equal to the index number of the fault among all faults. A shunt fault occupies two columns in the coincidence matrix as illustrated below:

T 0 0 1 0 0 0 0 ; 0 0 1 0 0 0 0 1 0 0 |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} corresponding to the current of the faulted bus in negative sequence network

corresponding to the current of the faulted bus in zero sequence network

where the first column contains the information on how the negative sequence network is connected with the positive sequence network, while the second column describe the connectivity between the zero sequence and negative sequence networks. The column number of the nonzero corresponds to the index number of the fault among all faults. For a line-to-line fault, the coincidence matrix has only the first column because there is no circuit connection between negative and zero sequence networks.

446

7 Power System Transient Stability Analysis

Based on the above principle, we can easily find the coincidence matrix that represents the boundary conditions of arbitrarily complex simultaneous faults given the types of the faults. For example, if three faults simultaneously occur, and the faults are, in order, single-line-to-ground, single-line-open-conductor, and line-to-line, then the coincidence matrix is as follows: 3 9 1 0 0 0 > = 7 6 0 1 1 0 6 7 negative sequence part 6 0 7 > 0 0 1 6 7 ; 6 1 7 9 0 0 0 C¼6 : 7 6 0 0 1 0 7 = 6 7 > 6 0 7 0 0 0 5 zero sequence part 4 |{z} |ﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄ} |{z} > ; singleline singleline linetoline 2

toground openconductor

With the coincidence matrix that describes the boundary conditions of complex simultaneous faults, the loop impedance matrix of the combined sequence network can be obtained using (7.66) and (7.67). The manipulations on these matrices can be accomplished by simple addition and subtraction operations. (3) Eliminating the closed circuit of shunt faults to form synthesized impedance matrix: The order of the loop impedance matrix equals the number of independent loop currents in the combined sequence network. To seek the synthesized impedance matrix, the currents of zero and negative sequence network must be eliminated (refer to (7.68) and (7.69)).

7.4

Transient Stability Analysis with Simplified Model

For a regional power system, the duration of losing synchronous stability is very short, typically a simulation study of the first swing (1–1.5 s) after a disturbance is applied suffices to judge whether or not the system can maintain synchronous operation. In stability studies like this, the effects of speed-governing systems can be neglected, thus the output of prime movers can be assumed to be constant, the reason is that the inertias of the prime movers are sufficient to keep the outputs of the prime movers constant; besides, because the time constants of the excitor windings are relatively large, their flux linkages do not change drastically in a short range of time, as a result the effect of the excitation system can be modeled as keeping generator transient voltages Eq0 or E0 constant. In other words, the free current components of excitor windings are compensated by the regulation of excitation systems, thus the flux linkages cf of excitor windings remain constant. Correspondingly, the effects of damper windings are also ignored. The simplified models for transient stability analysis are widely used in power system operation and planning. Specific applications include feasibility studies on

7.4 Transient Stability Analysis with Simplified Model

447

system topologies and operating schedules, computation of maximum transfer capabilities, calculation of critical clearing times, and investigations into the effects of stability controls, etc. Using different models for generators, loads, and network, one can build codes for various simplified stability analyses. Which portfolio of models to use depends on the fundamental characteristics of the problem under study. To explain the principles and procedures of simplified transient stability analysis, the subsequent sections assume the following mathematical models and solution algorithms have been applied to the transient stability analysis procedure: Generators: Generator transient voltage Eq0 remains constant Loads: Small loads are modeled as constant impedances, while larger loads are modeled as motors with mechanical–electrical interactions Network: Modeled with admittance matrix The differential equations are solved by the modified Euler’s method while the network equations are solved by Gauss elimination method. The overall procedure for a transient stability calculation is still as described in Fig. 7.8. The computer implementation of the calculation is provided below.

7.4.1

Computing Initial Values

Before starting the numerical integration, the initial values of the differential equations should be calculated based on the prefault operating state obtained by performing a load flow study. In a simplified transient stability study, the calculation of initial values include prefault generator transient voltages, rotor angles, the output of prime movers, and the slips and equivalent admittances of motors representing loads, etc. These parameters do not change discontinuously at the instant immediately after the fault is applied. In what follows the initial value variables are marked with subscripts (0). First we describe how to calculate the initial values of generators. From a load flow study the generator terminal voltages before the disturbance and the generator powers are given by V_ ð0Þ ¼ Vxð0Þ þ jVyð0Þ and S(0) ¼ P(0) + jQ(0). Furthermore, the generator currents injected into the network are computed by I_ð0Þ ¼ Ixð0Þ þ jIyð0Þ ¼

S^ð0Þ : ^_ V

ð7:74Þ

ð0Þ

Thus by (6.61), one can find the pseudovoltage E_ Qð0Þ as E_ Qð0Þ ¼ EQxð0Þ þ jEQyð0Þ ¼ V_ ð0Þ þ ðRa þ jXq ÞI_ð0Þ :

ð7:75Þ

Subsequently, the generator rotor angles are calculated by dð0Þ ¼ arctgðEQyð0Þ =EQxð0Þ Þ:

ð7:76Þ

448

7 Power System Transient Stability Analysis

Under steady-state operation, generators rotate at synchronous speed, therefore: oð0Þ ¼ 1:

ð7:77Þ

Using coordinate transformation formula (6.62), the d, q components of generator stator voltages and currents are given by Vdð0Þ sin dð0Þ cos dð0Þ Vxð0Þ Idð0Þ ¼ Vqð0Þ cos dð0Þ sin dð0Þ Vyð0Þ Iqð0Þ sin dð0Þ cos dð0Þ Ixð0Þ ¼ : ð7:78Þ cos dð0Þ sin dð0Þ Iyð0Þ Now based on (6.64), the values of transient voltages are obtained as E0qð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd0 Idð0Þ :

ð7:79Þ

In addition, the electrical powers Pe(0) of generators under steady-state operation are equal to the mechanical powers of the prime movers Pm(0), that is, 2 2 Pmð0Þ ¼ Peð0Þ ¼ Pð0Þ þ ðIxð0Þ þ Iyð0Þ ÞRa :

ð7:80Þ

The calculation of the initial values of loads is simple. The prefault node voltages V_ ð0Þ and powers S(0) consumed by loads are obtained from a load flow study, therefore the equivalent admittances of loads are computed by Yð0Þ ¼

S^ð0Þ : 2 Vð0Þ

ð7:81Þ

When loads are modeled as constant impedances, the corresponding equivalent admittances remain constant in the study period, and thus can be incorporated into the network admittance matrix as discussed earlier. For loads representing motors with mechanical–electrical interactions, since the slips of motors do not jump at the instant of disturbance, the equivalent admittances of loads do not change. In other words, the equivalent admittances of loads after the disturbance are identical to those of loads under normal steady-state operation.

7.4.2

Solving Network Equations with Gauss Elimination Method

In this solution method, the network equations are represented in the domain of real numbers, as in (7.36). Before starting the simulation, the loads represented by constant impedances should be incorporated into the network to obtain the network with constant impedance loads, this set of network equations remains constant during the simulation period.

7.4 Transient Stability Analysis with Simplified Model

449

Suppose a motor load is connected at node j. In the transient period the motor slip sj is time varying, and given the sj at an instant, the actual impedance of the motor load can be calculated based on (6.160):

ZMjð0Þ ðRm þ jXm ÞðR2 =sj þ jX2 Þ ZMj ¼ R1 þ jX1 þ ; ð7:82Þ ðRm þ jXm Þ þ ðR2 =sj þ jX2 Þ ZMð0Þ where ZMj(0) and ZM(0) are the equivalent impedance of all the motors under normal operation and the equivalent impedance of a typical motor. The admittance associated with the actual impedance can be rewritten as YMj ¼

1 ¼ GMj þ jBMj : ZMj

ð7:83Þ

Now suppose a generator is located at node i of the network. When the generator is represented by a varying E0 q model, with reference to Table 7.1, in (7.39) let 0 Edi ¼ 0, Eqi ¼ E0qi , Xdi ¼ Xdi , and Xqi ¼ Xqi , the formula for generator current is as follows:

Ixi Iyi

bxi 0 Gxi ¼ E gyi qi Byi

Bxi Gyi

Vxi ; Vyi

ð7:84Þ

where the elements can be rewritten, based (7.40), as follows: Rai cos di þ Xqi sin di ; 0 R2ai þ Xdi Xqi 0 Rai ðXdi Xqi Þ sin di cos di Gxi ¼ ; 0 R2ai þ Xdi Xqi bxi ¼

Byi ¼

0 Xdi sin2 di Xqi cos2 di ; 0 X R2ai þ Xdi qi

9 > > > > > > > > = 2 0 2 Xdi cos di þ Xqi sin di : Bxi ¼ 0 X > R2ai þ Xdi qi > > > 0 > Rai þ ðXdi Xqi Þ sin di cos di > > > Gyi ¼ ; 0 X R2ai þ Xdi qi gyi ¼

Rai sin di Xqi cos di 0 R2ai þ Xdi Xqi

ð7:85Þ Substitute the generator current representations (7.84) into the network equations with constant impedance loads, and do the same for the equivalent admittance [(7.83)] of motors, we obtain the new set of network equations. Obviously, the new network equations are just modifications of the original network equations: the diagonal elements of the admittance matrix are modified, and there are nonzero pseudocurrents in elements of the current vector associated with generators, the current injections of other nodes are zero, that is: The ith diagonal block of the admittance matrix is changed to

Gxi þ Gii Byi þ Bii

Bxi Bii Gyi þ Gii

ð7:86Þ

450

7 Power System Transient Stability Analysis

and the jth diagonal block changes to

GMj þ Gjj BMj þ Bjj

BMj Bjj : GMj þ Gjj

ð7:87Þ

The pseudocurrent injections at generator nodes are given by

Ixi0 Iyi0

¼

bxi 0 E : gyi qi

ð7:88Þ

Now the linear equations obtained in each integration step can be solved by Gauss elimination or the triangular factorization method. This gives us the real and imaginary part Vx and Vy of the network voltages for this step. Finally, based on (7.84), the generator currents Ix and Iy can be found.

7.4.3

Solving Differential Equations by Modified Euler’s Method

In a transient stability analysis using simplified models, the differential equations comprise the motion (6.76) of generator rotors and the motion (6.155) of motor rotors representing loads: 9 ddi > > ¼ os ðoi 1Þ > > dt > > = doi 1 ¼ ðPmi Pei Þ : dt TJi > > > > > dsj 1 ; ¼ ðMmMj MeMj Þ > dt TJMi

ð7:89Þ

Suppose the simulation of mechanical–electrical interactions has been completed up to time t, now let us discuss how to calculate the system states for time t + Dt. Before calculating system states for the next step, whether or not there is a fault or switch operation at time t should be checked first. If the answer is no, then one proceeds to compute the states of the next step, given the states of time t; otherwise, one has to calculate the postswitch or postfault network operating parameters first, and then continue the calculation for the next step. The computational procedure for solving differential equations based on the modified Euler’s method is as follows: (1) Given generator di(t) and motor sj(t) at time t, compute system voltages Vx(t) and Vy(t), and generator currents Ixi(t) and Iyi(t) based on the method described in Sect. 7.4.2.

7.4 Transient Stability Analysis with Simplified Model

451

(2) Based on (7.89), compute the derivatives for time t: 9 ddi > > ¼ o ðo 1Þ s iðtÞ > > dt t > > > = doi 1 ¼ ðP P Þ ; mi eiðtÞ > dt t TJi > > > > > dsj 1 > ; ¼ ðM M Þ mMjðtÞ eMjðtÞ dt t TJMi

ð7:90Þ

where generator power Pei(t) is calculated by 2 2 PeiðtÞ ¼ ðVixðtÞ IixðtÞ þ ViyðtÞ IiyðtÞ Þ þ ðIixðtÞ þ IiyðtÞ ÞRai :

ð7:91Þ

The mechanical torque TmMj(t) of generators and electrical torque TeMj(t) of motors are computed based on (6.157) and (6.156) as follows: 9 MmMjðtÞ ¼ k½a þ ð1 aÞð1 sjðtÞ Þ2 > > > = 2 2 2MeM max VjxðtÞ þ VjyðtÞ ; MeMjðtÞ ¼ s > scrj V 2 þ V 2 jðtÞ > > jxð0Þ jyð0Þ ; þ scrj sjðtÞ

ð7:92Þ

in which Vjx(0) and Vjy(0) denote the real and imaginary parts of prefault node voltage at node j. (3) Compute an initial estimate of state variables for time t + Dt: ½0

9 ddi > > Dt > > dt t > > > doi = ¼ oiðtÞ þ Dt : dt t > > > > > > dsj ; ¼ sjðtÞ þ Dt > dt t

diðtþDtÞ ¼ diðtÞ þ ½0

oiðtþDtÞ ½0

sjðtþDtÞ

½0

ð7:93Þ

½0

(4) Similar to step (1), given generator diðtþDtÞ and motor sjðtþDtÞ , compute system ½0

½0

½0

½0

node voltages VxðtþDtÞ and VyðtþDtÞ , generator currents IxiðtþDtÞ and IyiðtþDtÞ based on the method of Sect. 7.4.2. (5) Similar to step compute the estimated derivatives (2), ½0 ½0 ½0 dsj ddi doi ; dt ; and dt for step t + Dt. To this end, one should replace dt tþDt

tþDt

tþDt

½0

½0

½0

oi(t), Pei(t), MmMj(t), and MeMj(t) in (7.92) with oiðtþDtÞ , PeiðtþDtÞ , MmMjðtþDtÞ , ½0

and MeMjðtþDtÞ . To compute them, one should also replace Vix(t), Viy(t), Iix(t),

452

7 Power System Transient Stability Analysis ½0

½0

½0

½0

Iiy(t), sj(t), Vjx(t), and Vjy(t) with VixðtþDtÞ , ViyðtþDtÞ , IixðtþDtÞ , Iiy(t), sj(t), VjxðtþDtÞ , ½0

and VjyðtþDtÞ . (6) Finally, compute the variable values for step t + Dt, that is: # 9 " ½0 > Dt ddi ddi > > diðtþDtÞ ¼ diðtÞ þ þ > > 2 dt t dt > > tþDt > #> " > > ½0 = Dt doi doi oiðtþDtÞ ¼ oiðtÞ þ þ : 2 dt t dt > tþDt > > > # " > > ½0 > > Dt dsj dsj > > sjðtþDtÞ ¼ sjðtÞ þ þ > ; 2 dt t dt

ð7:94Þ

tþDt

[Example 7.3] Consider the 9-bus system in Fig. 7.12 [178]. This system consists of three generators, three loads, and nine branches. The generator and branch parameters are listed in Tables 7.5 and 7.6, respectively. The system load flow under normal operation is illustrated in Table 7.7, and the system frequency is 60 Hz. [Solution] A stability analysis based on the simplified system model will be described below. The disturbances are as follows: at time zero a three-line-toground fault occurs in line 5–7 at the node 7 side, the fault is cleared five cycles (about 0.08333 s) later by the removal of line 5–7. Generators are modeled as constant Eq0 , loads are modeled as impedances, the network is modeled by admittance matrix, the differential equations are solved by the modified Euler’s method, and the network equations are solved by a direct method.

18kV

230kV

13.8kV

2

3 2

8

3

7

9 6

5

4

16.5kV

1 1

Fig. 7.12 Single-line diagram of 9-bus system

7.4 Transient Stability Analysis with Simplified Model Table 7.5 Branch data From-end To-end bus bus 4 4 5 6 7 8 1 2 3

5 6 7 9 8 9 4 7 9

Resistance (in per unit) 0.010 0.017 0.032 0.039 0.0085 0.0119 0.0 0.0 0.0

Reactance (in per unit) 0.085 0.092 0.161 0.170 0.072 0.1008 0.0576 0.0625 0.0586

453

Half of the admittance (in per unit) 0.088 0.079 0.153 0.179 0.0745 0.1045

Non-standard ratio of transformer

1.0 1.0 1.0

Table 7.6 Generator data 0 0 Generator Bus TJ Ra Xd Xd0 Xq Xq0 Td0 Tq0 D 1 1 47.28 0.0 0.1460 0.0608 0.0969 0.0969 8.96 0.0 2 2 12.80 0.0 0.8958 0.1198 0.8645 0.1969 6.00 0.535 0.0 3 3 6.02 0.0 1.3125 0.1813 1.2578 0.2500 5.89 0.600 0.0 The units for all time constants are ‘‘seconds,’’ the units of all damping coefficients D, resistances and impedances are in ‘‘per unit’’ Table 7.7 Load flow under normal system operation Bus Voltage Generator Magnitude Phase angle Active Reactive (degree) power power 1 1.040 0.0000 0.7164 0.2705 2 1.0250 9.2800 1.6300 0.0665 3 1.0250 4.6648 0.8500 0.1086 4 1.0258 2.2168 5 0.9956 3.9888 6 1.0127 3.6874 7 1.0258 3.7197 8 1.0159 0.7275 9 1.0324 1.9667

Load Active power

Reactive power

1.2500 0.9000

0.5000 0.3000

1.0000

0.3500

Based on the general procedure described in Fig. 7.8 and the method described in the previous section, the transient stability analysis can be summarized below: 1. Initial value computation: Compute the equivalent shunt admittances of loads according to (7.81), and the results are as follows: Load (node 5): 1.26099 j0.50440 Load (node 6): 0.87765 j0.29255 Load (node 8): 0.96898 j0.33914 Then compute, based on (7.74)–(7.80), generator transient voltage Eq0 , initial rotor angle d(0), and mechanical power Pm(0). The results are in Table 7.8. The

454

7 Power System Transient Stability Analysis Table 7.8 E0 q, d(0), and Pm(0) of generators Generator Neglecting the effect of With the effect of sasalient poles lient poles Eq0 d(0) Eq0 d(0) 1 1.05664 2.27165 1.05636 3.58572 2 1.05020 19.73159 0.78817 61.09844 3 1.01697 13.16641 0.76786 54.13662

Pm(0)

0.71641 1.63000 0.85000

initial values of generator rotor angles are set to o1(0) ¼ o2(0) ¼ o3(0) ¼ 1. In the calculations to be described below, the effect of generator salient poles is neglected, which is to say Eq0 ¼ C and Xq ¼ Xd0 . This is the classical model of generators. 2. The fault-on system and post-fault system model: In the fault-on system, a shunt branch with zero impedance is connected at node 7, to model this shunt branch, the diagonal element Y77 of the admittance matrix Y is set to a very high value (say 1020). The admittance matrix of the fault-on system is YF. In the postfault network, branch 5–7 is removed. Because the contribution of line 5–7 to admittance matrix is equal to: 2

Ylð57Þ

5

.. 6 . 6 1 6 þ jb 6 r þ jx 56 6 . .. ¼ 6 6 .. . 6 1 76 6 6 r þ jx 4 .. .

7 .. . .. . .. .

.. .

1 r þ jx .. .

1 þ jb r þ jx .. .

.. .

3 7 7 7 7 7 7 7; 7 7 7 7 7 5

where r ¼ 0.032, x ¼ 0.161, and b ¼ 0.153. Thus the postfault admittance matrix is YP ¼ Y Yl(5–7). 3. Integrating the differential-algebraic equations: We will only compute the transient duration from the instant the fault occurs to time equals 2 s. Thus the system for the duration 0–2 s is divided into two autonomous systems: that is, the fault-on system for duration 0–0.08333 s, and the postfault system for duration 0.08333–2 s. The step size for numerical integration is 0.001 s. Table 7.9 lists the rotor angles d(t) and relative maximum rotor angles, with and without consideration of the effect of salient pole. The later is also depicted in Fig. 7.13. From Fig. 7.13, observe that the system is stable, whether or not the salient pole effect is taken into consideration. When the salient pole effect is considered, the maximum relative rotor angle is d21 ¼ 151.48396 (t ¼ 0.80133s). When the

0.00000 0.04200 0.08333 0.13333 0.18333 0.23333 0.28333 0.33333 0.38333 0.43333 0.48333 0.53333 0.58333 0.63333 0.68333 0.73333 0.78333 0.83333 0.88333 0.93333 0.98333 1.03333 1.08333 1.13333 1.18333 1.23333

2.27165 2.28779 2.34848 2.40803 2.58251 3.19401 4.52397 6.79447 10.16415 14.73304 20.54980 27.61528 35.87927 45.23062 55.48563 66.38413 77.60323 88.79177 99.61769 109.81335 119.20630 127.73265 135.43743 142.46732 149.05619 155.50082

19.73159 22.15764 29.28237 41.21540 53.48395 65.30378 76.12258 85.62591 93.68682 100.29635 105.50168 109.36539 111.95248 113.34805 113.70536 113.31385 112.66232 112.45974 113.57983 116.92253 123.22495 132.88223 145.83696 161.57457 179.23461 197.81063

13.16641 14.63856 18.86248 25.92757 33.68320 41.94182 50.42907 58.80933 66.72892 73.85628 79.91656 84.73139 88.27281 90.72587 92.52996 94.35019 96.94304 100.94170 106.66335 114.05272 122.79557 132.52906 143.03225 154.30791 166.52345 179.84443

Table 7.9 Generator d(t) and relative maximum rotor angles Without consideration of salient pole effects d1 d2 d3 3.58572 3.69016 4.00270 4.48409 5.09318 6.11234 7.80288 10.37892 13.99684 18.75788 24.71833 31.90242 40.31440 49.94781 60.79180 72.83490 86.06717 100.48127 116.07313 132.84229 150.79173 169.92694 190.25360 211.77290 234.47339 258.31838

61.09844 63.52449 70.64922 82.69359 95.37520 108.03487 120.26523 131.92171 143.06625 153.88636 164.62086 175.50595 186.74309 198.48469 210.83172 223.83807 237.51766 251.85143 266.79269 282.27022 298.18912 314.43044 330.85111 347.28707 363.56384 379.52044

54.13662 55.31617 58.74084 64.79619 72.05804 80.45556 89.78211 99.74929 110.07065 120.53143 131.02170 141.53491 152.14627 162.98542 174.21138 185.99198 198.48689 211.83293 226.13087 241.43372 257.73701 274.97226 293.00522 311.64166 330.64456 349.76706

With consideration of salient pole effects ~d1 ~d2 ~ d3 17.45994 19.86985 26.93389 38.80737 50.90144 62.10977 71.59862 78.83143 83.52267 85.56332 84.95188 81.75011 76.07321 68.11743 58.21973 46.92972 35.05909 23.66797 13.96214 7.10918 4.01865 5.14958 10.39953 19.10725 30.17842 42.30981

d2 d1 57.51272 59.83433 66.64652 78.20951 90.28202 101.92253 112.46235 121.54280 129.06941 135.12847 139.90254 143.60353 146.42869 148.53688 150.03992 151.00317 151.45049 151.37016 150.71956 149.42793 147.39740 144.50351 140.59752 135.51417 129.09045 121.20206 (continued)

~d2 ~d1

7.4 Transient Stability Analysis with Simplified Model 455

1.28333 1.33333 1.38333 1.43333 1.48333 1.53333 1.58333 1.63333 1.68333 1.73333 1.78333 1.83333 1.88333 1.93333 1.98333

162.12683 169.25129 177.15096 186.04298 196.07909 207.34945 219.88920 233.68131 248.65233 264.66308 281.50116 298.88496 316.48551 333.96386 351.01413

216.36810 234.19924 250.87411 266.20363 280.15828 292.78226 304.12763 314.22084 323.06766 330.69730 337.23926 343.01311 348.59530 354.81755 362.66346

194.25664 209.48217 225.02567 240.30096 254.76080 267.98613 279.73658 289.98411 298.94296 307.08305 315.08928 323.73003 333.64692 345.15535 358.17917

Table 7.9 (continued) Without consideration of salient pole effects d1 d2 d3 283.22935 309.06874 335.62983 362.64368 389.80754 416.82762 443.45984 469.53546 494.97070 519.76777 544.01356 567.87632 591.59440 615.45111 639.73684

395.05209 410.17107 425.07106 440.16296 456.04799 473.42157 492.94327 515.12417 540.25876 568.39552 599.32930 632.61227 667.60021 703.55468 739.78895

368.80340 387.65294 406.38036 425.24462 444.67239 465.17298 487.22074 511.14408 537.05069 564.80111 594.03622 624.26498 655.00155 685.90126 716.82750

With consideration of salient pole effects ~d1 ~d2 ~ d3 54.24126 64.94795 73.72315 80.16065 84.07919 85.43281 84.23842 80.53953 74.41533 66.03422 55.73810 44.12814 32.10979 20.85369 11.64932

d2 d1 111.82274 101.10233 89.44123 77.51928 66.24045 56.59395 49.48343 45.58871 45.28806 48.62775 55.31574 64.73595 76.00581 88.10357 100.05211

~d2 ~d1

456 7 Power System Transient Stability Analysis

Relative Swing Angle(Degrees)

7.4 Transient Stability Analysis with Simplified Model 160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0

457

d21

d21

0 0.10.20.30.40.50.60.70.80.9 1 1.11.21.31.41.51.6 1.71.81.9 2 t(s)

Fig. 7.13 Relative rotor angles as functions of time

salient pole effect is not considered, the maximum relative rotor angle is d21 ¼ 85.65788 (t ¼ 0.44633 s), and the angle of the second swing d21 ¼ 85.43378 (t ¼ 1.53433 s), which is smaller than that of the first swing. Lastly, the critical clearing times, with and without consideration of salient pole effects, are calculated. It turns out that the critical clearing time under the former circumstance is between 0.162 and 0.163 s, and it is between 0.085 and 0.086 s under the second condition. The swing curves under these circumstances are provided in Figs. 7.14 and 7.15, respectively.

7.4.4

Numerical Integration Methods for Transient Stability Analysis Under Classical Model

In a modern energy management system (EMS), to assess system security, transient stability under various prespecified contingencies is predicted online within a limited amount of time. Because the number of contingencies is quite large, to meet the requirement of online assessment, each transient stability analysis must be completed rapidly. Obviously the traditional integration methods for transient stability analysis are no longer suitable because of the limitation of speed, fast methods customized for online applications need to be developed. Dynamic security assessment is one of the ‘‘hot’’ areas in the power system stability field. As early as 1983, IEEE established a transient stability analysis working group, the responsibility of which was to lead and review the research in

458

7 Power System Transient Stability Analysis 300 280 260

d 21 ( ct = 0.163s )

Relative Swing Angle(Degrees)

240 220 200 180 160 140 120

d 21 ( ct = 0.162 s )

100 80 60 40 20 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.81.9 2 t(s)

Fig. 7.14 Relative rotor angle near critical clearing time (salient pole effects not considered)

300 280

Relative Swing Angle(Degrees)

260

~

d21 (ct = 0.086s)

240 220 200 180 160 140 120

~

d21 (ct = 0.085s)

100 80 60 40 20 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 t/s

Fig. 7.15 Relative rotor angle near critical clearing time (salient pole effects considered)

7.4 Transient Stability Analysis with Simplified Model

459

this area. A dynamic security assessment method must be extra-rapid, especially in an online environment. Though the requirement on computational precision can be relaxed to some extent, reliability and robustness are still required. Currently, the methods for improving the speed of online dynamic security assessment include no more than: first, simplifying the mathematical model for stability analysis; and second, developing rapid algorithms for stability analysis. In what follows we introduce a rapid algorithm for transient stability analysis based on a classical model. 1. The classical model of power system stability: The ‘‘classical model’’ places the following assumptions on the mathematical model: (1) Assume that the generator mechanical power remains constant during the transient stability period, and neglect the effects of damper windings (2) Assume that generator transient voltage E0 does not change during the transient stability study period, and furthermore the phase angle of this voltage is equal to the rotor angle (3) Loads are modeled as constant impedances With the above assumptions, the motion equation of the ith generator is obtained as 9 ddi > > ¼ os ðo 1Þ = dt i ¼ 1; 2; . . . ; m: doi 1 > ; ¼ ðPmi Pei Þ > dt TJi

ð7:95Þ

From a load flow study one finds the transient voltage: 0 E_ 0i ¼ E0i ﬀdið0Þ ¼ V_ ið0Þ þ ðRai þ jXdi Þ

Pið0Þ jQið0Þ : ^_ V

ð7:96Þ

ið0Þ

And the normal operation conditions give oið0Þ ¼ 1:

ð7:97Þ

Based on (7.42), (7.45), and (7.46) in Sect. 7.3.1, which describe the generator–network relationship, we can incorporate the pseudoadmittances of generators (see (7.45)) and load equivalent admittances (see (7.81)) into the network. The diagonal elements of the admittance matrix of the network (7.35) should have the generator pseudoadmittances or load equivalent

460

7 Power System Transient Stability Analysis

admittances added. The right-hand side vector, as given by (7.46), contains nonzeros in rows corresponding to generator nodes only, the other rows are zero. The expression for electromagnetic power of generators is easily obtained as ! ^_ 0 V ^_ E i Pei ¼ Re E_ 0i i : 0 Rai jXdi

ð7:98Þ

2. Solving for the network equations: First perform triangular factorization on the admittance matrix Y (a symmetrical matrix): Y ¼ UT DU;

ð7:99Þ

where U is an upper triangular matrix, D is a matrix with nonzeros on the diagonal only. After performing the following forward substitution and backward substitution, one obtains the voltage: F ¼ D1 UT I;

ð7:100Þ

V ¼ U1 F:

ð7:101Þ

Vector I in the above equations is a sparse vector. To compute the electromagnetic powers of generators, it is only necessary to know the generator voltages. Thus the unknown vector V is also sparse. Therefore, the network equations can be solved using rapid forward and backward substitutions. The experience of medium size computations demonstrates that, some 1/3 computational effort can be saved if a sparse vector method is used to perform rapid forward and back substitution. When solving the network equations with a sparse vector method, the majority of the time is spent in factorizing the admittance matrix. In dynamic security assessment, the admittance matrices for fault-on and postfault networks under different contingency scenarios are different. If these admittance matrices are factored every time, it would take a large amount of computer time. However, in general, the fault-on and postfault network admittance matrices differ from the prefault network admittance matrix only in a few places. This allows the utilization of compensation methods for solving network equations. The idea of a compensation method is to avoid matrix refactorizations, thus computational burden can be greatly relieved. Consider network equation: ðY þ DYÞV ¼ I;

ð7:102Þ

7.4 Transient Stability Analysis with Simplified Model

461

where Y is the prefault network admittance matrix, and DY is the adjustment to Y due to network switching or a fault and can be represented as DY ¼ MdyMT ;

ð7:103Þ

where dy is a (q q) matrix, including the information for the adjustment to Y, q is in general of order 1 or 2, and M is a (n q) coincidence matrix related to the specific fault or switch. By the matrix inversion lemma, (7.102) and (7.103) become V ¼ ðY1 Y1 MCMT Y1 ÞI;

ð7:104Þ

where the (q q) matrix C is equal to C ¼ ½ðdyÞ1 þ Z1 ;

ð7:105Þ

while the (q q) matrix Z is Z ¼ MT Y1 M:

ð7:106Þ

Thus according to (7.104), taking into account (7.99), the computational steps for solving the network (7.102) using compensation are as follows: Preparatory calculation: ð1Þ

9 > > > > > > =

W ¼ UT M

~ ¼ D1 W ð2Þ W ð3Þ

~ TW Z ¼W

ð4Þ

C ¼ ½ðdyÞ1 þ Z1

> > > > > > ;

:

ð7:107Þ

Solving the network equations:

ð2Þ

9 > > > > > > T > ~ ~ DF ¼ WCF F =

ð3Þ

F ¼ F~ þ DF

ð4Þ

V ¼ U1 D1 F

ð1Þ F~ ¼ UT I

> > > > > > > ;

:

ð7:108Þ

The forward and backward substitutions in (7.107) and (7.108) are all completed using a sparse vector method.

462

7 Power System Transient Stability Analysis

3. A numerical integration algorithm for second-order conservative systems: The differential equations in (7.95) can be rewritten in the following compact form: d2 d ¼ fðdÞ; dt2

ð7:109Þ

9 > > > = T fðdÞ ¼ ½f1 ðdÞ; . . . ; fm ðdÞ : > os > > ; fi ðdÞ ¼ ðPmi Pei Þ TJi

ð7:110Þ

where d ¼ ½d1 ; . . . ; dm T

The right-hand side functions of the differential equations in (7.109) do not contain arguments with first-order derivatives, the equations are thus termed a second-order conservative system. Compared with solving two first-order equations, the equations can be solved by direct differencing which results in an efficiency one level higher. Consider the Stormer and Numerov integration formula [186]: dkþ2 ¼ 2dkþ1 dk þ h2 fðdkþ1 Þ; dkþ2 ¼ 2dkþ1 dk þ

ð7:111Þ

h2 ½fðdkþ2 Þ þ 10fðdkþ1 Þ þ fðdk Þ: 12

ð7:112Þ

Equation (7.111) is an explicit second-order method, while (7.112) is an implicit forth-order method. To solve the differential (7.109) based on (7.111) requires a smaller step size because of the poor numerical stability. To solve (7.109) based on (7.112) allows a larger step size because the method has higher order and has a larger region of absolute stability. However this method still takes a large amount of computational effort because it involves solving a set of nonlinear simultaneous ½0 equations. On the other hand, if good initial estimates dkþ2 are provided when solving (7.112), the convergence can be speeded up. This suggests the application of a predictor–corrector method for solving (7.109); specifically, the explicit method (7.111) is adopted for the predictor, while the implicit method (7.112) is adopted for the corrector. Let P and C represent the application of one predictor and one corrector, E represent computing function f(d) once, the pair of predictor–corrector is formed as ½0 PECE. More concretely, one computes dkþ2 based on the predictor, and calculates ½0

½0

½1

f kþ2 ¼ fðdkþ2 Þ, substitute the result into the corrector to obtain dkþ2 , and finally ½1

½1

compute f kþ2 ¼ fðdkþ2 Þ.

7.5 Transient Stability Analysis with FACTS Devices

463

The above method falls into the category of multistep methods, the procedure can be started using the following special fourth-order Runge–Kutta formula [185]: 9 h2 > dkþ1 ¼ dk þ þ ðk1 þ 2k2 Þ > > > > 6 > > > h > 0 0 > dkþ1 ¼ dk þ ðk1 þ 4k2 þ k3 Þ > > > 6 > = k1 ¼ fðdk Þ : > > > h 0 h2 > > k2 ¼ f d k þ dk þ k 1 > > 2 8 > > > > 2 > h > 0 > ; k3 ¼ f dk þ hdk þ k2 2 hd0k

ð7:113Þ

The classical model for transient stability analysis applies to ‘‘first swing’’ (about 1.5 s after the disturbance). This model is free from the stiffness problem and therefore permits the use of larger step size (0.1–0.2 s).

7.5

Transient Stability Analysis with FACTS Devices

To study in detail the transient stability of a large scale interconnected power system experiencing various large disturbances and to analyze the effects of control devices on system stability, often for the purpose of seeking mechanisms for improving stability, a detailed component model for transient stability analysis is required. As the technology of HVDC develops, HVDC systems are widely used in long‐ distance transmission, under‐sea cable transmission and system interconnection. The technology of flexible AC transmission (FACTS), matured only in recent years, is also receiving much acceptance from the industry. FACTS devices not only help to improve system steady‐state performance, they also improve the dynamic performance of power systems to a certain degree, as a result system transfer capabilities are enlarged considerably. The dynamic performance of a power system is also affected by generator prime movers and speed‐governing systems, excitation systems, PSSs, and other control devices. A power system with increasing scale, and increasing installations of dynamic devices, exhibits complex behavior after it experiences a disturbance. The mechanical–electrical interaction of such a system lasts longer, and the duration of oscillation of the system before loss of stability occurs can be as long as several seconds to a dozen seconds. This section introduces the basic transient stability analysis method for large‐ scale interconnected power systems with many dynamic devices which are modeled in detail. It should be noted that the material presented does not address the detailed implementation of a commercial code, rather it concerns the basic principles.

464

7 Power System Transient Stability Analysis

The mathematical models for the dynamic devices are as follows: synchronous machines which are modeled by a sixth‐order model with varying Eq0 , E00q, Ed0 , E00d, and rotor variables; hydroprime mover and their speed‐governing system; excitation systems with thyristor‐based DC excitors, PSSs with generator speed deviation as input; two‐terminal HVDC; SVC and TCSC of the FACTS family; constant impedance loads or loads with second‐order voltage characteristics. If a different model other than those described above is used for a component, the same principle applies. The above large scale dynamic system is a typical stiff system because of the existence of dynamic devices with drastically different time constants. To solve such systems with an explicit numerical method, a very small step size has to be assumed because the stability region is relatively small. The implicit trapezoidal rule is a second‐order algorithm with the left half plane being the stability region, therefore it allows for the use of a larger step size. In early commercial codes explicit methods such as forth‐order Runge–Kutta method were quite popular. Because of their better numerical properties, adaptability to stiffness, and the introduction of fast control schemes with small time constants, the second‐order trapezoidal rule has become almost an industry standard since the 1970s. Many commercial grade codes, for example, the transient stability analysis package developed by Bonneville power administration (BPA), the power system analysis software package, are based on this method. In a typical transient stability analysis, the trapezoidal rule with constant step size, between 0.01 and 0.02 s (or even longer), is assumed. The difference and algebraic equations are solved by a simultaneous method or a sequential method. In the large‐scale transient stability analysis procedure to be presented below, the implicit trapezoidal rule is used to solve the differential equations, while a Newton method is used to solve the simultaneous difference‐algebraic equations of the detailed system model.

7.5.1 7.5.1.1.

Initial Values and Difference Equations of Generators Generators

The mathematical model of a synchronous machine comprises rotor motion equations, rotor electromagnetic equations, etc., together with stator voltage equations and the expressions for electromagnetic powers. Based on (6.1)–(6.4), these equations can be rewritten as follows: Rotor motion equations: 9 dd > > ¼ os ðo 1Þ = dt : do 1 > ; ¼ ðPm Pe DoÞ > dt TJ

ð7:114Þ

7.5 Transient Stability Analysis with FACTS Devices

465

Rotor electromagnetic equations: 9 dE0q 1 0 00 > ¼ 0 ½Efq kd Eq þ ðkd 1ÞEq > > > > dt Td0 > > > > 00 > dEq 1 0 > 00 0 00 > ¼ 00 ½Eq Eq ðXd Xd ÞId > = dt Td0 ; > dE0d 1 > > ¼ 0 ½kq E0d þ ðkq 1ÞE00d > > dt Tq0 > > > > > 00 > dEd 1 0 00 0 00 > ¼ 00 ½Ed Ed þ ðXq Xq ÞIq > ; dt Tq0

ð7:115Þ

Xq Xq00 Xd Xd00 and k ¼ q Xd0 Xd00 Xq0 Xq00 Stator voltage equations:

where kd ¼

Vd ¼ E00d Ra Id þ Xq00 Iq Vq ¼ E00q Xd00 Id Ra Iq

) :

ð7:116Þ

The electrical power is equal to the output power plus stator copper loss: 2 Pe ¼ Pout þ I_ Ra ¼ Vx Ix þ Vy Iy þ ðIx2 þ Iy2 ÞRa :

ð7:117Þ

Given a load flow solution, some of the initial values of generators can be computed based on (7.74)–(7.78). Note that the current flows in damper windings under steady‐state operation are equal to zero, based on (6.60), (6.64), and (6.65), the initial values of generator no‐load synchronous voltages, transient voltages, and sub‐transient voltages can be easily obtained as Efqð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd Idð0Þ ; E0qð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd0 Idð0Þ

) ;

E0dð0Þ ¼ Vdð0Þ þ Ra Idð0Þ Xq0 Iqð0Þ E00qð0Þ ¼ Vqð0Þ þ Ra Iqð0Þ þ Xd00 Idð0Þ

ð7:118Þ ð7:119Þ

)

E00dð0Þ ¼ Vdð0Þ þ Ra Idð0Þ Xq00 Iqð0Þ

:

ð7:120Þ

Besides, the electrical power Pe(0) of generators under steady‐state operation can be computed directly from (7.117): 2 2 Peð0Þ ¼ Pð0Þ þ ðIxð0Þ þ Iyð0Þ ÞRa :

ð7:121Þ

466

7 Power System Transient Stability Analysis

Set

do dt

¼ 0 in (7.114), the prime mover outputs Pm(0) are equal to Pmð0Þ ¼ Peð0Þ þ D:

ð7:122Þ

To solve the difference equations we first apply the trapezoidal rule for the rotor motion (7.114), dðtþDtÞ ¼ dðtÞ þ

os Dt ðoðtþDtÞ þ oðtÞ 2Þ; 2

oðtþDtÞ ¼ oðtÞ þ

ð7:123Þ

Dt ðPmðtþDtÞ PeðtþDtÞ DoðtþDtÞ þ PmðtÞ PeðtÞ DoðtÞ Þ: 2TJ ð7:124Þ

From (7.124) one obtains the expression for o(tþDt), substituting this into (7.123), it follows that: dðtþDtÞ ¼ aJ ðPmðtþDtÞ PeðtþDtÞ Þ þ d0 ;

ð7:125Þ

where aJ ¼

os ðDtÞ2 ; 4TJ þ 2DDt

ð7:126Þ

4TJ d0 ¼ dðtÞ þ aJ PmðtÞ PeðtÞ þ oðtÞ os Dt: Dt

ð7:127Þ

In (7.126), aJ is a function of step size Dt and some other constants. If a fixed step size is assumed, it becomes a constant. As for d0 in (7.127), it is a constant only in difference equation (7.125), it takes different values in each computational step. After d(tþDt) is found, o(tþDt) is calculated based on (7.123): oðtþDtÞ ¼

2 ðdðtþDtÞ dðtÞ Þ oðtÞ þ 2: os Dt

ð7:128Þ

Now applying the trapezoidal rule to the electromagnetic equation (7.115), it follows: 9 Dt 0 00 > ½E k E þ ðk 1ÞE d qðtþDtÞ d fqðtþDtÞ > qðtþDtÞ > 0 > 2Td0 > > > > 0 00 = þ EfqðtÞ kd E þ ðkd 1ÞE

E0qðtþDtÞ ¼ E0qðtÞ þ

qðtÞ

E00qðtþDtÞ ¼ E00qðtÞ þ

qðtÞ

Dt 0 00 0 00 00 ½EqðtþDtÞ EqðtþDtÞ ðXd Xd ÞIdðtþDtÞ 2Td0

þ E0qðtÞ E00qðtÞ ðXd0 Xd00 ÞIdðtÞ

> > > > > > > > ;

;

ð7:129Þ

7.5 Transient Stability Analysis with FACTS Devices

E0dðtþDtÞ

E00dðtþDtÞ

9 > > ¼ > > > > > > > 0 00 = kq EdðtÞ þ ðkq 1ÞEdðtÞ : > Dt > ¼ E00dðtÞ þ 00 ½E0dðtþDtÞ E00dðtþDtÞ þ ðXq0 Xq00 ÞIqðtþDtÞ > > > 2Tq0 > > > > ; 0 00 0 00 þ E E þ ðX X ÞI E0dðtÞ

467

Dt þ 0 ½kq E0dðtþDtÞ þ ðkq 1ÞE00dðtþDtÞ 2Tq0

dðtÞ

dðtÞ

q

q

ð7:130Þ

qðtÞ

Eliminating variables E0 q(tþDt) and E0 d(tþDt) in (7.129) and (7.130), we have E00qðtþDtÞ ¼ a00d ðXd0 Xd00 ÞIdðtþDtÞ þ a00d ad1 EfqðtþDtÞ þ E00q0 ;

ð7:131Þ

E00dðtþDtÞ ¼ a00q ðXq0 Xq00 ÞIqðtþDtÞ þ E00d0 ;

ð7:132Þ

where n 9 E00q0 ¼ a00d ad1 EfqðtÞ ðXd0 Xd00 ÞIdðtÞ þ 2ð1 kd ad1 ÞE0qðtÞ > > > > >

> > 1 > 00 > > þ ad1 ðkd 1Þ þ 2 EqðtÞ = ad2 n ; > > E00d0 ¼ a00q ðXq0 Xq00 ÞIqðtÞ þ 2ð1 kq aq1 ÞE0dðtÞ > > > >

> > 1 > 00 > þ aq1 ðkq 1Þ þ 2 EdðtÞ ; aq2

ð7:133Þ

9 > > > > > > = Dt Dt : ad2 ¼ 00 ; ad2 ¼ 00 > > 2Td0 þ Dt 2Td0 þ Dt > > > > a00d ¼ ½ad1 ð1 kd Þ þ 1=ad2 1 ; a00q ¼ ½aq1 ð1 kq Þ þ 1=aq2 1 ;

ð7:134Þ

ad1 ¼

Dt ; 0 2Td0 þ kd Dt

aq1 ¼

Dt 0 2Tq0 þ kq Dt

The coefficients ad1, ad2, a00d, aq1, aq2, and a00q in (7.134) are all constants if a fixed step size Dt is assumed, while in (7.133), E00q0 and E00d0 are known quantities at step t, although they take different values in each step. After E00 q(tþDt) and E00 d(tþDt) are calculated, now based on (7.129) and (7.130), 0 E q(tþDt) and E0 d(tþDt) can be obtained by

E0dðtþDtÞ

0 2Tdo kd Dt 0 EqðtÞ þ EfqðtþDtÞ þ EfqðtÞ þ ðkd 1ÞðE00qðtþDtÞ þ E00qðtÞ Þ Dt 0 2Tqo kq Dt 0 00 00 ¼ aq1 EdðtÞ þ ðkq 1ÞðEdðtþDtÞ þ EdðtÞ Þ Dt

E0qðtþDtÞ ¼ ad1

9 > > > = : > > > ;

ð7:135Þ

468

7 Power System Transient Stability Analysis

7.5.1.2

Exitation System and PSS

Taking the DC exitor with thyristor‐based regulator, illustrated in Fig. 6.16, as an example, let us derive the differential‐algebraic equations based on the transfer function diagram. We will neglect the effects of RC, and equivalent time constants TB and TC of the analog regulator. Under the ‘‘one per unit exitation voltage/one per stator voltage’’ system, by (6.51) it follows that Vf ¼ Efq. The measurement and filter system dVM 1 ¼ ðVC VM Þ; dt TR

VC ¼ V_ þ jXC I_

ð7:136Þ

The transient droop feedback: dðKF Efq TF VF Þ ¼ VF dt

ð7:137Þ

The amplifier: 9 1 > > f ¼ ½KA ðVREF þ VS VM VF Þ VR > > TA > > > > dVR > = if VR ¼ VRMAX and f > 0; ¼ 0; VR ¼ VRMAX > dt > dVR > > if VRMIN < VR < VRMAX ; ¼f > > dt > > > > dVR > if VR ¼ VRMIN and f < 0; ¼ 0; VR ¼ VRMIN ; dt

ð7:138Þ

The exiter dEfq 1 ¼ ½VR ðKE þ SE ÞEfq ; dt TE

ð7:139Þ

where the saturation coefficient SE is modeled as an exponential function according to (6.101). Under one per unit excitation voltage/one per unit stator voltage system, (6.101) is simplified to SE ¼ CE ENfqE 1 :

ð7:140Þ

The saturation function can also be piece‐wise linearized as follows: SE Efq ¼ K1 Efq K2 :

ð7:141Þ

7.5 Transient Stability Analysis with FACTS Devices

469

From Fig. 6.14 we have the PSS equations:

9 dV1 1 > > ¼ ðKS VIS V1 Þ > > > dt T6 > > > > dðV1 V2 Þ 1 > > ¼ V2 = dt T5 : > dðT1 V2 T2 V3 Þ > > ¼ V3 V2 > > > dt > > > > dðT3 V3 T4 V4 Þ ; ¼ V4 V3 > dt

ð7:142Þ

The limits of PSS output are If V4 VSmax ;

9 > = VS ¼ V4 : > ;

VS ¼ VSmax

If VSmin < V4 < VSmax ; If V4 VSmin ; VS ¼ VSmin

ð7:143Þ

The initial values of excitation system variables can be found by setting, in the transfer function diagram, s ¼ 0, or alternatively setting the left‐hand side of the differential equations of the excitation system to zero. The effects of limiters can in general be ignored since the variables with limiters under normal operation do not in general exceed their corresponding limits. In the following, we describe how to compute the initial values of the excitation system mentioned above, the other excitation systems can be dealt with likewise. Setting the left‐hand side of (7.139) to zero, one obtains the initial value for the amplifier VRð0Þ ¼ ðSEð0Þ þ KE ÞEfqð0Þ ;

ð7:144Þ

where the saturation coefficient is calculated based on (7.140), that is, E 1 SEð0Þ ¼ CE ENfqð0Þ :

Setting the left‐hand side of (7.136), (7.137), and (7.138) to zero, it follows: 9 VFð0Þ ¼ 0; VMð0Þ ¼ V_ð0Þ þ jXC I_ð0Þ = : ð7:145Þ VRð0Þ ; VREF ¼ VMð0Þ þ KA Setting the left‐hand side of (7.142) to zero, and taking into account the relationship expressed in (7.142), we have the initial value of PSS: ) VSð0Þ ¼ V4ð0Þ ¼ V3ð0Þ ¼ V2ð0Þ ¼ 0 ; ð7:146Þ V1ð0Þ ¼ KS VISð0Þ ¼ 0 where VIS is equal to zero since it often takes the form of speed, or change of active power.

470

7 Power System Transient Stability Analysis

Applying the trapezoidal rule to (7.136), we have the difference equations of measurement and filter systems: VMðtþDtÞ ¼ aR VCðtþDtÞ þ VM0 ;

ð7:147Þ

in which aR ¼

Dt ; 2TR þ Dt

ð7:148Þ

2TR Dt VMðtÞ ; 2TR þ Dt ) ¼ V_ ðtþDtÞ þ jXC I_ðtþDtÞ : ¼ V_ ðtÞ þ jXC I_ðtÞ

VM0 ¼ aR VCðtÞ þ VCðtþDtÞ VCðtÞ

ð7:149Þ ð7:150Þ

Applying the trapezoidal rule to (7.137), we have VFðtþDtÞ ¼ aF EfqðtþDtÞ þ VF0 ;

ð7:151Þ

where aF ¼ VF0 ¼

2KF ; 2TF þ Dt

ð7:152Þ

2TF Dt VFðtÞ aF EfqðtÞ : 2TF þ Dt

ð7:153Þ

When limiters are not taken into consideration, applying the trapezoidal rule to (7.138), we have the difference equation: VRðtþDtÞ ¼ aA ðVSðtþDtÞ VMðtþDtÞ VFðtþDtÞ Þ þ VR0 ;

ð7:154Þ

where aA ¼

KA Dt ; 2TA þ Dt

VR0 ¼ aA ð2VREF þ VSðtÞ VMðtÞ VFðtÞ Þ þ

ð7:155Þ 2TA Dt VRðtÞ : 2TA þ Dt

ð7:156Þ

Substituting (6.141) into (6.139), and applying trapezoidal rule, we have the difference equations of the excitor: EfqðtþDtÞ ¼ aE VRðtþDtÞ þ VE0 ;

ð7:157Þ

where aE ¼

Dt ; 2TE þ ðKE þ K1 ÞDt

ð7:158Þ

7.5 Transient Stability Analysis with FACTS Devices

471

VE0 ¼ aE ½VRðtÞ 2ðKE þ K1 ÞEfqðtÞ þ 2K2 þ EfqðtÞ : Applying the trapezoidal rule to (7.142), it follows: 9 V1ðtþDtÞ ¼ a1 VISðtþDtÞ þ V10 > > > > V2ðtþDtÞ ¼ a2 V1ðtþDtÞ þ V20 = ; V3ðtþDtÞ ¼ a3 V2ðtþDtÞ þ V30 > > > > ; V4ðtþDtÞ ¼ a4 V3ðtþDtÞ þ V40

ð7:159Þ

ð7:160Þ

in the above formula a1 ¼

KS Dt ; 2T6 þ Dt

a2 ¼

a4 ¼

2T3 þ Dt ; ð7:161Þ 2T4 þ Dt

9 > > > > > > > > > 2T5 Dt > > ¼ V2ðtÞ a2 V1ðtÞ = 2T5 þ Dt : > 2T2 Dt 2T1 Dt > ¼ V3ðtÞ V2ðtÞ > > > 2T2 þ Dt 2T2 þ Dt > > > > > 2T4 Dt 2T3 Dt ; ¼ V4ðtÞ V3ðtÞ > 2T4 þ Dt 2T4 þ Dt

ð7:162Þ

2T5 ; 2T5 þ Dt

V10 ¼ a1 VISðtÞ þ V20 V30 V40

a3 ¼

2T1 þ Dt ; 2T2 þ Dt

2T6 Dt V1ðtÞ 2T6 þ Dt

Eliminating the intermediate variables V1(tþDt), V2(tþDt), and V3(tþDt) in (7.160), it follows: V4ðtþDtÞ ¼ a4 a3 a2 a1 VISðtþDtÞ þ V40 þ a4 ½V30 þ a3 ðV20 þ a2 V10 Þ:

ð7:163Þ

If the input of the PSS is set to VIS = o os, apparently VIS(t) = o(t) os. Substituting VIS(tþDt)¼o(t¼Dt) os into (7.163), and making use of (7.128) to eliminate variable o(tþDt), we have V4ðtþDtÞ ¼ aS dðtþDtÞ þ VS0 ;

ð7:164Þ

where aS ¼

2a4 a3 a2 a1 ; os Dt

ð7:165Þ

VS0 ¼ V40 þ a4 ½V30 þ a3 ðV20 þ a2 V10 Þ aS dðtÞ þ a4 a3 a2 a1 ð2 os oðtÞ Þ: ð7:166Þ If the limits of PSS outputs are not considered, obviously we get VSðtþDtÞ ¼ V4ðtþDtÞ :

ð7:167Þ

If PSS takes other forms of input signals, following the same derivations, we should be able to find the corresponding expressions.

472

7 Power System Transient Stability Analysis

Eliminating the intermediate variables V4(tþDt), VS(tþDt), VM(tþDt), VF(tþDt), and VR(tþDt) in (7.164), (7.167), (7.147), (7.151), (7.154), and (7.157), the difference equations of the excitation system without taking into account the affects of limiters are obtained as EfqðtþDtÞ ¼ b1 dðtþDtÞ b2 V_ðtþDtÞ þ jXC I_ðtþDtÞ þ Efq0 ; ð7:168Þ where b1 ¼ Efq0 ¼ 7.5.1.3

aE aA aS ; 1 þ aE aA aF

b2 ¼

aE aA aR ; 1 þ aE aA aF

VE0 þ aE ½VR0 þ aA ðVS0 VM0 VF0 Þ : 1 þ aE aA aF

ð7:169Þ ð7:170Þ

The Prime Movers and Their Speed-Governing Systems

Taking the hydrogenerator and its speed-governing system illustrated in Fig. 6.24 as an example, based on the transfer function we have the differential-algebraic equations: The acentric flyball ¼ Kd ðoREF oÞ

ð7:171Þ

9 eKd eKd > > > 2 2 > = eKd eKd : If x

; s¼x > 2 2 > > > eKd eKd > ; If x ; s¼xþ 2 2

ð7:172Þ

The valve The dead zones are If

The limits of value position are If sMIN < s < sMAX ; If s sMAX ; If s sMAX ;

9 s ¼ s> =

s ¼ sMAX s ¼ sMIN

: > ;

ð7:173Þ

The servo system dm s ¼ dt TS

ð7:174Þ

7.5 Transient Stability Analysis with FACTS Devices

473

The limits of the valve 9 If mMIN < m < mMAX ; m ¼ m > = If m mMAX ; m ¼ mMAX > ; If m mMAX ; m ¼ mMIN

ð7:175Þ

d½x ðKb þ Ki Þm 1 ¼ ðKi m xÞ dt Ti

ð7:176Þ

dðPm þ 2KmH mÞ 2 ¼ ðKmH m Pm Þ; dt To

ð7:177Þ

The feedback system

The hydrogenerator

where parameter KmH is defined as follows: KmH ¼

PH ðMWÞ : SB ðMVAÞ

ð7:178Þ

In general, the parameters of a prime mover and its speed-governing system are provided in the per unit system with the nominal capability of the generator being the base. With the introduction of parameters KmH, Pm, and Pe all expressed in per unit system with system base SB. Similar to the calculation of excitation system initial values, the initial values of prime mover and speed-governing systems can be found by setting, in the transfer functions, s = 0, or alternatively by setting the left-hand side of the differential equations to zero. Again the dead zones of measurement systems and various limiters need not be considered in general. Setting the left-hand side of (7.177), (7.176), and (7.174) to zero, and making use of the linear relationships in (7.171), (7.172), (7.173), and (7.175), together with (7.77), we have the initial values of each state variable: 9 Pmð0Þ > mð0Þ ¼ mð0Þ ¼ ; ð0Þ ¼ xð0Þ ¼ Ki mð0Þ ; sð0Þ ¼ sð0Þ ¼ 0; > = KmH ð7:179Þ xð0Þ xð0Þ > > ; oREF ¼ oð0Þ þ ¼1þ : Kd Kd Based on (7.171), the equation corresponding to instant t + Dt for the acentric flyball is as follows: ðtþDtÞ ¼ Kd ðoREF oðtþDtÞ Þ:

ð7:180Þ

474

7 Power System Transient Stability Analysis

Neglecting the measurement dead zone, based on (7.172), it follows: sðtþDtÞ ¼ ðtþDtÞ xðtþDtÞ :

ð7:181Þ

Also neglecting the limits on valve position, and based on (7.173), obviously we have: sðtþDtÞ ¼ sðtþDtÞ :

ð7:182Þ

Applying the trapezoidal rule to (7.174), we obtain the following difference equation: mðtþDtÞ ¼ aS sðtþDtÞ þ m0 ;

ð7:183Þ

where aS ¼

Dt ; 2TS

m0 ¼ aS sðtÞ þ mðtÞ :

ð7:184Þ ð7:185Þ

Neglect again the limit on valve position, based on (7.175), we have mðtþDtÞ ¼ mðtþDtÞ :

ð7:186Þ

Applying the trapezoidal rule to (7.176), we have the difference equation of the feedback block: xðtþDtÞ ¼ ai mðtþDtÞ þ x0 ;

ð7:187Þ

where ai ¼ K i þ x0 ¼

2Ti Kb ; 2Ti þ Dt

2T i Dt 2T i Kb ½xðtÞ Ki mðtÞ m : 2T i þ Dt 2T i þ Dt ðtÞ

ð7:188Þ ð7:189Þ

Applying the trapezoidal rule to (7.177), the difference equations of hydrogenerators are obtained, as follows: PmðtþDtÞ ¼ aH mðtþDtÞ þ P0 ;

ð7:190Þ

where aH ¼

KmH ð2To DtÞ ; To þ Dt

ð7:191Þ

7.5 Transient Stability Analysis with FACTS Devices

P0 ¼

To Dt KmH ð2To þ DtÞ PmðtÞ þ mðtÞ : To þ Dt To þ Dt

475

ð7:192Þ

Eliminating the intermediate variables Z(tþDt), sðtþDtÞ , s(tþDt), mðtþDtÞ , m(tþDt), and x(tþDt) in (7.180), (7.182), (7.183), (7.186), (7.187), and (7.190), and eliminating variable o(tþDt) based on (7.128), we find the difference equations for step tþDt of hydrogenerators and their speed-governing systems, without consideration of limiters: PmðtþDtÞ ¼ b3 dðtþDtÞ þ Pm0 ;

ð7:193Þ

in which b3 ¼ Pm0 ¼ P0 b3 dðtÞ þ

2aH aS Kd ; ð1 þ aS ai Þos Dt

aH ½aS Kd ð2 oREF oðtÞ Þ þ aS x0 m0 : 1 þ aS ai

ð7:194Þ ð7:195Þ

Finally, substituting (7.117) of Pe(t + Dt) and (7.193) of Pm(t + Dt) into (7.125), and substituting difference equation (7.168) of Efq(t + Dt) into (7.131), together with (7.132) we obtain the difference equations of generators for step t + Dt. Let us transform the state currents under d–q coordinates into those under x–y coordinates, for notational simplicity, neglecting the subscripts (t + Dt), it follows: 9 ð1 aJ b3 Þd þ aJ ½Vx Ix þ Vy Iy þ Ra ðIx2 þ Iy2 Þ aJ Pm0 d0 ¼ 0 > > > > 00 00 0 00 00 > = Eq þ ad ðXd Xd ÞðIx sin d Iy cos dÞ ad ad1 b1 d qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ : > þ a00d ad1 b2 ðVx XC Iy Þ2 þ ðVy þ XC Ix Þ2 a00d ad1 Efq0 E00q0 ¼ 0 > > > > ; E00d a00q ðXq0 Xq00 ÞðIx cos d þ Iy sin dÞ E00d0 ¼ 0 ð7:196Þ The set of simultaneous (7.196) consists of three equations, the first reflects the mechanical motion of the generators, while the other two reflect the electromagnetic interactions. Based on (7.39), generator currents Ix and Iy are functions of Vx, Vy, d, E00q, and E00d (refer to (7.258) for details), and therefore can be eliminated. The above set of simultaneous equations thus has three state variables d, E00q , and E00d , plus two operating parameters Vx and Vy.

7.5.2

Initial Values and Difference Equations of FACTS and HVDC

7.5.2.1

SVC

Here we will focus on an SVC model comprising a fixed capacitor (FC) and a thyristor-controlled reactor (TCR). For ease of exposition, we will take a proportional regulator-based SVC as an example; its transfer function is illustrated in (7.16).

476

7 Power System Transient Stability Analysis VRef V

−

Σ

+

KS

BS1

1+sTS2

1+sTS

BC

BS2

1+sTS1

BSVC BC − BL

Fig. 7.16 A simple model of SVC

An SVC is generally connected to a high-voltage system via a transformer. The equivalent admittance of TCR is controlled by the firing angle a of a thyristor, thus the equivalent admittance BSVC of the SVC is manipulated. This mechanism facilitates the control of voltage V given the input VREF. The mathematical model of the SVC is obtained easily from Fig. 7.16 as 9 dBS1 1 > ¼ ½KS ðVREF VÞ BS1 > = dt TS : > dðTS2 BS2 TS1 BS1 Þ > ; ¼ BS1 BS2 dt

ð7:197Þ

The limit on SVC output is 9 BSVC ¼ BS2 > > = ¼ BC ; > > ; BSVC ¼ BC BL

If BC BL < BS2 < BC ; If BS2 BC ;

BSVC

If BS2 BC BL ;

ð7:198Þ

where BC = oC is the susceptance of the fixed capacitor, BL = 1/oL is the susceptance of the reactor, the output BSVC is the equivalent susceptance of the SVC. The upper limit of the SVC corresponds to the point at which the thyristor is completely shut off, while the lower limit corresponds to the point at which the thyristor is like a lossless conductor. The position between the limits corresponds to a point at which the thyristor is partially closed. Although an SVC is connected at the low-voltage side of a transformer, it can still be viewed as a reactive power source at the high-voltage side, intended to control the voltage at the high-voltage side bus of the transformer. Therefore, the high-voltage bus can be effectively set as a PV node in load flow studies (P = 0, V = VSP). From the result of a load flow study, one obtains V_ ð0Þ ¼ V SP ﬀyð0Þ and the power injection from the SVC S(0) = jQ(0). Let the reactance of the transformer be XT, the power injected into the network from the SVC is given by Qð0Þ ¼

2 Vð0Þ

1 BSVCð0Þ

XT

:

ð7:199Þ

7.5 Transient Stability Analysis with FACTS Devices

477

Setting both sides of (7.197) to zero, and noticing the relationship in (7.198) and (7.199), we find the initial values of the SVC as BSVCð0Þ ¼ BS2ð0Þ ¼ BS1ð0Þ VREF

BSVCð0Þ ¼ V SP þ KS

9 > > ¼ 2 > > Vð0Þ = XT þ Qð0Þ : > > > > ; 1

ð7:200Þ

Applying the trapezoidal rule to the first (7.197), it follows: BS1ðtþDtÞ ¼ n1 VðtþDtÞ þ BS10 ;

ð7:201Þ

where n1 ¼

KS Dt ; 2TS þ Dt

BS10 ¼ n1 ð2VREF VðtÞ Þ þ

2TS Dt BS1ðtÞ : 2TS þ Dt

ð7:202Þ ð7:203Þ

And applying the trapezoidal rule to the second equation in (7.197), and eliminating BS1(t + Dt) from (7.201), it follows: BS2ðtþDtÞ ¼ BSVC0 nS

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxðtþDtÞ þ VyðtþDtÞ ;

ð7:204Þ

where nS ¼ n1 BSVC0 ¼

2TS1 þ Dt ; 2TS2 þ Dt

2TS1 þ Dt 2TS2 Dt 2TS1 Dt BS10 þ BS2ðtÞ BS1ðtÞ : 2TS2 þ Dt 2TS2 þ Dt 2TS2 þ Dt

ð7:205Þ ð7:206Þ

If the limit of the SVC is ignored, then BS(tþDt) ¼ BS2(tþDt), thus: BSVCðtþDtÞ ¼ BSVC0 nS 7.5.2.2

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxðtþDtÞ þ VyðtþDtÞ :

ð7:207Þ

TCSC

A thyristor-controlled series compensator (TCSC) is connected into a transmission line in series, it changes its equivalent admittance thus achieving the goal of controlling the equivalent admittance of the transmission line. Here we will only

478

7 Power System Transient Stability Analysis

give the mathematical model of a TCSC composed of FC and TCR connected in parallel (similar to an SVC): 9 dBT1 1 > ¼ ½KT ðPREF PT Þ BT1 > = dt TT ; > dðTT2 BT2 TT1 BT1 Þ > ; ¼ BT1 BT2 dt

ð7:208Þ

where the input signal PT is the power flowing through the line in which the TCSC is connected, the output BTCSC is the equivalent susceptance of the TCSC. The limits of the TCSC are MAX If BMIN TCSC < BT2 < BTCSC ;

If BT2 BMAX TCSC ;

BTCSC

If BT2 BMIN TCSC ;

BTCSC

9 BTCSC ¼ BT2 > > = MAX ; ¼ BTCSC > > ; ¼ BMIN TCSC

ð7:209Þ

MIN where the specific values of BMAX TCSC and BTCSC depend on the sizes of L and C. They can be computed based on (5.153)–(5.155). As usual a load flow study provides BTCSC(0) and PT(0) = PSP, similar to computing the initial values of an SVC, we have

9 BTCSCð0Þ ¼ BT2ð0Þ ¼ BT1ð0Þ = : BTCSCð0Þ ; PREF ¼ PTð0Þ þ KT

ð7:210Þ

If the measured value of PT of TCSC flows from bus i to bus j, the expression of PT is easily obtained as PT ¼ BTCSC ðVxi Vyj Vyi Vxj Þ:

ð7:211Þ

Apply the trapezoidal rule to the first equation in (7.208), one obtains BT1ðtþDtÞ ¼ z1 PTðtþDtÞ þ BT10 ;

ð7:212Þ

in which z1 ¼

KT Dt ; 2TT þ Dt

BT10 ¼ z1 ð2PREF PTðtÞ Þ þ

2TT Dt BT1ðtÞ : 2TT þ Dt

ð7:213Þ ð7:214Þ

7.5 Transient Stability Analysis with FACTS Devices

479

Now applying the trapezoidal rule to the second equation of (7.208), eliminating BT1(tþDt) and PT(tþDt) based on (7.212) and (7.211), we have ½1 þ zT ðVxiðtþDtÞ VyjðtþDtÞ VyiðtþDtÞ VxjðtþDtÞ ÞBT2ðtþDtÞ BTCSC0 ¼ 0;

ð7:215Þ

in which zT ¼ z1 BTCSC0 ¼

2TT1 þ Dt ; 2TT2 þ Dt

2TT1 þ Dt 2TT2 Dt 2TT1 Dt BT10 þ BT2ðtÞ BT1ðtÞ : 2TT2 þ Dt 2TT2 þ Dt 2TT2 þ Dt

ð7:216Þ ð7:217Þ

If the output limits of the TCSC are neglected, obviously we have BTCSC(t + Dt) = BT2(t + Dt), then ½1 þ zT ðVxiðtþDtÞ VyjðtþDtÞ VyiðtþDtÞ VxjðtþDtÞ ÞBTCSCðtþDtÞ BTCSC0 ¼ 0: 7.5.2.3

ð7:218Þ

HVDC Systems

In stability studies, the network equations of the AC system appear in terms of positive sequence quantities, this places a fundament limitation on the model of an HVDC system. In particular, commutation failure in the HVDC system cannot be predicted. A commutation failure may be the result of a severe three-line-to-ground fault occurring close to the rectifier, an unsymmetrical fault on the AC side of the rectifier, or saturation of HVDC transformer operating during a transient period. Earlier HVDC models included the dynamic characteristics of transmission lines and converter dynamics. In recent years, there is a trend toward adopting simpler models. Two models for an HVDC system are popular, these are, a simplified model and a steady-state model.

1. The simplified model An HVDC system some distance away from the study area has little impact on the results of a stability study, and thus can be modeled using a simple model: the system is viewed as a pair of active and reactive power sources connected at the converter AC substation. A more realistic model is termed the steady-state model. Based on (5.2), the DC line is modeled by the algebraic equation of a resistor: VdR ¼ VdI þ Rdc Id ; where Rdc denotes the resistance of the line.

ð7:219Þ

480

7 Power System Transient Stability Analysis

Noticing that IdR ¼ IdI ¼ Id, from (7.52) and (7.53), eliminating VdR and VdI in (7.219), it follows: RId ¼ kR VR cos a kI VI cos b;

ð7:220Þ

R ¼ Rdc þ XcR þ XcI :

ð7:221Þ

where

The pole control action is assumed to be instantaneous; many of the control functions are represented in terms of their net effects, rather than actual characteristics of the hardware. This model appears in the form of an algebraic equation, the interaction between AC and DC system is similar to that in a load flow model. 2. Quasi-steady-state model If the short circuit currents in any of the converters are relatively low, then the dynamics of DC system elements has a non-negligible impact on the AC system. As a result, a more detailed DC model is necessary for conducting a transient stability analysis. In a quasi-steady-state model, the converter characteristics are still modeled by the equation governing the relationship between average DC values and the nominal values of fundamental frequency components. In this setting, the DC transmission line can adopt different models given different requirements on precision. The simplest DC line model is just that of a steady-state model, as in (7.220). A more detailed model is based on an R–L circuit: L

dId þ RId ¼ kR VR cos a kI VI cos b; dt

ð7:222Þ

where R is defined in (7.221), besides, L ¼ Ldc þ LR þ LI ;

ð7:223Þ

where Ldc, LR, and LI are the reactance of DC line, and the smoothing reactors. For the control system, taking the control mode of constant current and constant voltage as an example, from the transfer function given in Fig. 5.18, we have the differential equations: 9 > > = : > dðKc1 x1 aÞ Kc2 > ; ¼ ðIdREF x1 Þ dt Tc2 dx1 1 ¼ ðId x1 Þ dt Tc3

ð7:224Þ

7.5 Transient Stability Analysis with FACTS Devices

481

The limits on delayed ignition angle include 9 < aMAX ; a ¼ a > If aMIN < a = aMAX ; a ¼ aMAX If a ; > ; aMIN ; a ¼ aMIN If a

ð7:225Þ

9 dx4 1 > > ¼ ðVdI x4 Þ = dt Tv3 : > dðKv1 x4 bÞ Kv2 ; ¼ ðVdREF x4 Þ > dt Tv2

ð7:226Þ

The limits on ignition advance angle include MAX ; = b : If b ; b ¼ b MAX MAX > ; If b bMIN ; b ¼ bMIN

ð7:227Þ

When the rectifier is under constant current control, and the inverter is under SP constant voltage control, we have Idð0Þ ¼ IdSP and VdIð0Þ ¼ VdI . From a load flow study we have VR(0) and VI(0). Based on (7.224)–(7.227), noticing the relationships in (7.219) and/or (7.222), and (7.52) and (7.53), it follows: 9 IdREF ¼ x1ð0Þ ¼ Idð0Þ > > > > > V þ ðR þ X ÞI dc cR dIð0Þ dð0Þ > 1 > að0Þ ¼ að0Þ ¼ cos > = kR VRð0Þ : > VdREF ¼ x4ð0Þ ¼ VdIð0Þ > > > > > > 1 VdIð0Þ XcI Idð0Þ > ; bð0Þ ¼ bð0Þ ¼ cos kI VIð0Þ

ð7:228Þ

Applying trapezoidal rule to (7.224), we find x1ðtþDtÞ ¼ g1 IdðtþDtÞ þ x10 ;

ð7:229Þ

where g1 ¼

Dt ; 2Tc3 þ Dt

x10 ¼ g1 IdðtÞ þ

2Tc3 Dt x1ðtÞ : 2Tc3 þ Dt

ð7:230Þ ð7:231Þ

482

7 Power System Transient Stability Analysis

Using the second formula in (7.224), and eliminating x1(tþDt) by making use of (7.229), we have aðtþDtÞ ¼ g2 IdðtþDtÞ þ a0 ;

ð7:232Þ

where g2 ¼ g1

Kc2 Dt Kc1 þ ; 2Tc2

Kc2 Dt Kc2 Dt Kc2 Dt Kc1 þ x10 þ aðtÞ IdREF Kc1 x1ðtÞ : 2Tc2 Tc2 2Tc2

a0 ¼

ð7:233Þ ð7:234Þ

Neglecting the limits on ignition angle a, it is obvious that: aðtþDtÞ ¼ aðtþDtÞ :

ð7:235Þ

Applying the trapezoidal rule to the first formula of (7.226), it follows: x4ðtþDtÞ ¼ g3 VdIðtþDtÞ þ x40 ;

ð7:236Þ

where g3 ¼

Dt ; 2Tv3 þ Dt

x40 ¼ g3 VdIðtÞ þ

2Tv3 Dt x4ðtÞ : 2Tv3 þ Dt

ð7:237Þ ð7:238Þ

Applying the trapezoidal to the second formula in (7.226), making use of (7.236) to eliminate x4(tþDt), and noticing the relationship in (7.53) allows VdI(tþDt) to be eliminated, after simple manipulations, we have b ðtþDtÞ ¼ g4 VIðtþDtÞ cos bðtþDtÞ þ g5 IdðtþDtÞ þ b0 ;

ð7:239Þ

in which g3 Kv2 Dt g4 ¼ Kv1 þ ; nI 2Tv2 b0 ¼

g5 ¼ g4 nI RcI ;

Kv2 Dt Kv2 Dt VdREF Kv1 Kv2 Dt x4ðtÞ : Kv1 þ x40 þ b ðtÞ 2Tv2 Tv2 2Tv2

ð7:240Þ ð7:241Þ

7.5 Transient Stability Analysis with FACTS Devices

483

Neglecting the limits on ignition angle b, it follows: bðtþDtÞ ¼ b ðtþDtÞ :

ð7:242Þ

Under a quasi-steady-state model, different difference equations can be developed, with or without consideration for the transient duration of the DC transmission line. If the transient duration of the DC line is not considered, the DC line is modeled based on (7.220), where Id can be expressed as a function of a, b, VxR, VyR, VxI, and VyI: Id ¼

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ kR qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 þ V 2 cos a kI VxR VxI2 þ VyI2 cos b: yR R R

ð7:243Þ

ðtþDtÞ and Id(tþDt) in (7.232), (7.235), and (7.243), we then Let us eliminate a obtain the difference equation of the rectifier under constant current control, when the limits on a are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 þ r2 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ a0 ¼ 0;

aðtþDtÞ r1

ð7:244Þ

where r1 ¼

kR g; R 2

r2 ¼

kI g : R 2

ð7:245Þ

Similarly, eliminating variables b ðtþDtÞ and Id(tþDt) in (7.239), (7.242), and (7.243), we get the difference equation of the inverter under constant voltage control, when the limits on b are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 r4 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ b0 ¼ 0;

bðtþDtÞ r3

ð7:246Þ

where r3 ¼

kR g ; R 5

r 4 ¼ g4

kI g: R 5

ð7:247Þ

If the transient response of the DC line is considered, the DC line is modeled by (7.222), applying the trapezoidal rule to this equation, it follows: IdðtþDtÞ ¼ g6 VRðtþDtÞ cos aðtþDtÞ g7 VIðtþDtÞ cos bðtþDtÞ þ Id0 ;

ð7:248Þ

484

7 Power System Transient Stability Analysis

where g6 ¼

kR Dt ; 2L þ RDt

g7 ¼ g6

Id0 ¼ g6 VRðtÞ cos aðtÞ g7 VIðtÞ cos bðtÞ þ

kI ; kR

2L RDt IdðtÞ : 2L þ RDt

ð7:249Þ ð7:250Þ

Now eliminating aðtþDtÞ and Id(tþDt) in (7.232), (7.235), and (7.248), we find the difference equation of the rectifier, under constant current control, when the limits on a are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 þ r6 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ u0 ¼ 0;

aðtþDtÞ r5

ð7:251Þ

where r5 ¼ g 2 g 6 ;

r6 ¼ g 2 g 7 ;

u0 ¼ a0 þ g2 Id0 :

ð7:252Þ ð7:253Þ

By the same token, eliminating b ðtþDtÞ and Id(tþDt) in (7.239), (7.242), and (7.248), we find the difference equation of the inverter, under constant current control, when the limits on b are not considered: qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 VxRðtþDtÞ þ VyRðtþDtÞ cos aðtþDtÞ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 2 þ r8 VxIðtþDtÞ þ VyIðtþDtÞ cos bðtþDtÞ n0 ¼ 0;

aðtþDtÞ r7

ð7:254Þ

where r7 ¼ g 5 g 6 ;

r8 ¼ g 4 g 5 g 7 ;

v0 ¼ b0 þ g5 Id0 :

7.5.3

ð7:255Þ ð7:256Þ

Forming Network Equations

The network equations expressed in the domain of real numbers are provided in (7.36). In transient stability studies, the nodes in the network are divided into three classes: nodes connected in parallel with dynamic devices (including generator nodes, SVC nodes, and load nodes); nodes connected in series with dynamic

7.5 Transient Stability Analysis with FACTS Devices

485

devices (including the AC buses of an HVDC system, TCSC nodes, etc.); and the faulted nodes or nodes not connected with any device. Substitute the network current expressions of each dynamic device, illustrated in Sect. 7.3.1, into the network equations, and properly dealing with a fault or switch, as described in Sect. 7.3.2, we obtain the network equations ready for subsequent simulation.

7.5.3.1

Nodes Connected in Parallel with Dynamic Devices

If a dynamic device is connected at node i, then the network equation for this node is DIxi ¼ Ixi

X k2i

DIyi ¼ Iyi

X k2i

9 ðGik Vxk Bik Vyk Þ ¼ 0 > > = ðGik Vyk þ Bik Vxk Þ ¼ 0 > > ;

:

ð7:257Þ

The expressions for currents Ixi and Iyi at node i depend on what device is connected. (1) Connected with a generator: Note that a generators is represented using a varying Eq0 , E00q, Ed0 , and E00d model, therefore assigning the corresponding values to the elements in (7.40), based on Table 7.1, it turns out that the expression for the generator current (7.39) can be rewritten as 9 n 1 00 00 00 00 > > ðR cos d þ X sin d ÞE þ ðR sin d X cos d ÞE ai i i ai i i > qi qi di di 00 00 > R2ai þ Xdi Xqi > > o > > > 00 00 00 2 00 2 > = ½Rai ðXdi Xqi Þ sin di cos di Vxi ðXdi cos di þ Xqi sin di ÞVyi : n 1 00 00 00 00 > > > Iyi ¼ 2 ðR sin d X cos d ÞE ðR cos d þ X sin d ÞE ai i i qi ai i i di > qi di 00 X00 > Rai þ Xdi > qi > o > > > 00 00 00 00 ; þ ðXdi sin2 di þ Xqi cos2 di ÞVxi ½Rai þ ðXdi Xqi Þ sin di cos di Vyi

Ixi ¼

ð7:258Þ (2) Connected with a load: As described in Sect. 7.3.1, if the load is a constant impedance, it can be incorporated in the network. If the load is modeled as a second-order polynomial function, it is modeled as a current injection in (7.48), note that the constant impedance part of the load can also be incorporated into the network. The last two terms in (7.48) are treated as current injections. If the load is modeled as an exponential function of voltage, it can be viewed as a current injection (7.49).

486

7 Power System Transient Stability Analysis

(3) Connected with an SVC: The expression of the current injection of an SVC is given in (7.50). 7.5.3.2

Nodes Connected with Series Devices

If a dynamic device is connected in series between node i and node j, the network equations between node i and j are X

9 ðGik Vxk Bik Vyk Þ ¼ 0 > > > > k2i > > X > > DIyi ¼ Iyi ðGik Vyk þ Bik Vxk Þ ¼ 0 > > > = k2i X : DIxj ¼ Ixj ðGjk Vxk Bjk Vyk Þ ¼ 0 > > > > > k2j > > X > > DIyj ¼ Iyj ðGjk Vyk þ Bjk Vxk Þ ¼ 0 > > ; DIxi ¼ Ixi

ð7:259Þ

k2j

The currents Ixi, Iyi, Ixj, and Iyj take different forms, depending upon what device is connected between node i and j. 1. Series TCSC The expressions for currents Ixi, Iyi, Ixj, and Iyj are given by (7.51). 2. Series HVDC system If the AC–DC system is solved by a simultaneous approach, the expressions for DC currents Ixi, Iyi, Ixj, and Iyj injected into the AC nodes are given in (7.56). In the formula of (7.54) and (7.55), the DC current Id can be replaced by (7.243) or (7.248), thus the current injections are functions of a, b, VxR, VyR, VxI, and VyI only. 7.5.3.3

A Connection Node or Faulted Node

A connection node has zero current injection. As discussed before, any type of fault can be modeled by adjusting the admittance matrix of the positive sequence network based on the concept of the synthesized impedance matrix. Therefore, in the faulted node of the extended positive sequence network, there is no current injection. The faulted node is therefore a connection node. The network equation of a connection node or faulted node is X

9 ðGfk Vxk Bfk Vyk Þ ¼ 0 > > = k2f X : DIyf ¼ 0 ðGfk Vyk þ Bfk Vxk Þ ¼ 0 > > ; DIxf ¼ 0

k2f

ð7:260Þ

7.5 Transient Stability Analysis with FACTS Devices

7.5.4

487

Simultaneous Solution of Difference and Network Equations

All the equations of the system for step t þ Dt have been given, they include the network equations and the difference equations of each dynamic device. In this system of equations, the unknowns include the operating variables of the power system under study Vx and Vy; all the state variables of dynamic devices, for example, d, E00q, and E00d of each generator, the BSVC of each SVC, the BTCSC of each TCSC, and the a and b of the HVDC system. Assume that there are n nodes, nG generators, nS SVCs, nT TCSCs, nD HVDC systems, then the number of unknowns is equal to 2n þ 3nG þ nS þ nT þ 2nD, which is just equal to the number of equations. The system of equations is well defined. Network equations (7.257), (7.259), (7.260), generator difference (7.196), SVC (7.207), TCSC (7.218), together with HVDC difference equations (7.244), (7.246) or (7.251), (7.254) comprise a set of nonlinear equations. The current injections and the difference equations are time varying, while the network equations assume the same structure, except for the steps at which a disturbance (either a fault or a switch operation) occurs. To compute the system states immediately after a disturbance, only the network equations need to be resolved. The state variables d, E00q, E00d, BSVC, BTCSC, a, and b of the dynamic devices should take the values obtained before the disturbance. The set of nonlinear equations comprising the difference and network equations is solved in a recursive manner to provide the states of the study system at each integration step. The above set of nonlinear equations is typically solved using a Newton method. Since the Newton method is already fairly familiar, the computational procedure of the method will only be briefly described below: 1. Set, for step t + Dt, the initial values of generator state variables d, E00q, E00d, the initial values of SVC state variables BSVC, the initial values of TCSC state variables BTCSC, the initial values of HVDC state variables a, b, and the initial values of network voltage Vx and Vy. These initial values either can be set to the values at step t, or may be extrapolated from the values of the previous steps. 2. For the set of nonlinear equations comprising generator difference equations, SVC difference equations, TCSC difference equations, HVDC difference equations, and the network equations, compute the Jacobian matrix and mismatches given the initial values obtained in step (1), then solve the linear equations for the updates to the variables. 3. Check if the iteration has converged. If yes, stop; otherwise, return to step (2). The iteration continues until convergence is reached. 4. After the quantities of the state variables for t + Dt are obtained, proceed to compute the values of the other dynamic variables according to the difference/ algebraic equations derived in Sects. 7.5.1 and 7.5.2. These values will be useful for the computation of the next step. It should be noted that, the effects of limiters should be considered in this step.

488

7 Power System Transient Stability Analysis

Thinking and Problem Solving 1. What is meant by the transient stability of electrical power systems? What methods can be adopted to analyze it? How can we judge the transient stability of electrical power systems? 2. What are the consequences of loss of transient stability in a power system? 3. What suppositions are made within the transient stability analysis model of electrical power systems and what is the underlying theory? 4. What main measures exist to improve the transient stability of electrical power systems? What is the principle of each measure? 5. Give the method used to modify an admittance matrix when short-circuit failures at different locations and of different fault types occur on one transmission line in an electrical network, and list essential calculation formulas. 6. What aspects should be considered in choosing appropriate integration methods when numerical integration is used to analyze the transient stability of electrical power systems? 7. What kinds of initial value problems of differential equations belong to the class of ‘‘stiff’’ problems? What requirements are there for numerical integration methods to solve stiff problems? 8. What are the advantages and disadvantages of the alternating solution method and the simultaneous solution method in solving initial value problems of differential-algebraic equations? 9. How can we deal with limiters when a numerical integration method is used to find the time-domain solution of each state variable in an electrical power system? 10. Although there are many numerical integration methods, the implicit trapezoidal integration method obtains wide application in transient stability analysis of electrical power systems, why? 11. During dynamic security evaluation, it is required to carry out rapid transient stability analysis of electrical power systems under each contingency. What aspects can be considered to improve the speed of the transient stability analysis? 12. When analyzing the transient stability of electrical power systems, each generator can be represented by one of the following models: E0 = C; Eq0 = C; Eq0 vary; Eq0, Ed0 vary; Eq0, E00q, E00d vary; Eq0, E00q, Ed0, E00d vary. Explain the applicability of each model. 13. It can be seen from the numerical solution of the transient stability analysis of a real electrical power system that the current on an inductance and the voltage across the two terminals of a capacitance will change significantly at the second that failure occurs, which seems to not satisfy the law of electromagnetic induction. Why? 14. During transient stability calculation using the improved Euler’s method, when considering the transient process of the excitation winding and the influence of excitation system, select a type of excitation system, and list the relevant formulas of the transient process calculation for one step.

Chapter 8

Small-Signal Stability Analysis of Power Systems

8.1

Introduction

Small-signal stability analysis is about power system stability when subject to small disturbances. If power system oscillations caused by small disturbances can be suppressed, such that the deviations of system state variables remain small for a long time, the power system is stable. On the contrary, if the magnitude of oscillations continues to increase or sustain indefinitely, the power system is unstable. Power system small-signal stability is affected by many factors, including initial operation conditions, strength of electrical connections among components in the power system, characteristics of various control devices, etc. Since it is inevitable that power system operation is subject to small disturbances, any power system that is unstable in terms of small-signal stability cannot operate in practice. In other words, a power system that is able to operate normally must first be stable in terms of small-signal stability. Hence, one of the principal tasks in power system analysis is to carry out small-signal stability analysis to assess the power system under the specified operating conditions. The dynamic response of a power system subject to small disturbances can be studied by using the method introduced in Chap. 7 to determine system stability. However, when we use the method for power system small-signal stability analysis, in addition to slow computational speed, the weakness is that after a conclusion of instability is drawn, we cannot carry out any deeper investigation into the phenomenon and cause of system instability. The Lyapunov linearized method has provided a very useful tool for power system small-signal stability analysis. Based on the fruitful results of eigensolution analysis of linear systems, the Lyapunov linearized method has been widely used in power system small-signal stability analysis. In the following, we shall first introduce the basic mathematics of power system small-signal stability analysis. The Lyapunov linearized method is closely related to the local stability of nonlinear systems. Intuitively speaking, movement of a nonlinear system over a small range should have similar properties to its linearized approximation.

X.‐F. Wang et al., Modern Power Systems Analysis. doi: 10.1007/978-0-387-72853-7, # Springer Science þ Business Media, LLC 2008

489

490

8 Small-Signal Stability Analysis of Power Systems

Assume the nonlinear system described by dx ¼ f ðxÞ: dt Its Taylor expansion at the origin is dDx ¼ ADx þ hðDxÞ; dt

ð8:1Þ

where @fðxe þ DxÞ @fðxÞ A¼ ¼ : @Dx @x x¼xe Dx¼0 If in the neighborhood of Dx = 0, h(Dx) is a high-order function of Dx, we can use the stability of the following linear system dDx ¼ ADx: dt

ð8:2Þ

To study the stability of the nonlinear system at point xe (1) If the linearized system is asymptotically stable, i.e., all eigenvalues of A have negative real parts, the actual nonlinear system is asymptotically stable at the equilibrium point. (2) If the linearized system is unstable, i.e., at least one of eigenvalues of A has a positive real part, the actual nonlinear system is unstable at the equilibrium point. (3) If the linearized system is critically stable, i.e., real parts of all eigenvalues of A are nonpositive but the real part of at least one of them is zero, no conclusion can be drawn about the stability of the nonlinear system from its linearized approximation. The basic principle of the Lyapunov linearized method is to draw conclusions about the local stability of the nonlinear system around the equilibrium point from the stability of its linear approximation. When carrying out small-signal stability analysis of a power system, we always assume that the system at normal operation at equilibrium point x ¼ xe or Dx ¼ 0 is disturbed instantly at the moment t ¼ t0 when system state moves from 0 to Dx(t0). Dx(t0) is the initial state of system free movement after disappearance of the disturbance. Because the disturbance is sufficiently small, Dx(t0) is within a sufficiently small neighborhood of Dx ¼ 0. Thus in the neighborhood of Dx ¼ 0, h(Dx) is a high-order indefinitely small variable. Hence according to the Lyapunov linearized method, we can study the stability of the linearized system to investigate that of the actual nonlinear power system. Linearizing the differential-algebraic dynamic description of a power system of (8.1) and (8.2) at steady-state operating point (x(0), y(0)), we can obtain

8.1 Introduction

491

~ ~ dDx=dt A B Dx ¼ ~ ~ ; 0 C D Dy

ð8:3Þ

where 2

@f1 6 @x1 6 ~¼6 . A 6 .. 6 4 @f n @x1 2 @g1 6 @x1 6 ~ 6 C ¼ 6 .. 6 . 4 @g m @x1

3 @f1 @xn 7 7 .. 7 ; . 7 7 5 @fn @xn x¼xð0Þ y¼yð0Þ 3 @g1 @xn 7 7 .. 7 ; . 7 7 5 @gm @xn x¼xð0Þ y¼yð0Þ

2 @f

1

6 @y1 6 6 ~ B ¼ 6 .. 6 . 4 @f n

@y1 2 @g

1

6 @y1 6 ~ 6 . D¼6 . 6 . 4 @g m @y1

@f1 3 @ym 7 7 .. 7 ; . 7 7 5 @fn @ym x¼xð0Þ y¼yð0Þ : @g 3 1

@ym 7 7 .. 7 . 7 7 @gm 5 @ym x¼xð0Þ y¼yð0Þ

R denotes the set of real numbers, Rn is the n-dimensional space of real vectors, Rmn is the set of m-row n-column real matrices. We define Rn to be Rn1 , i.e., elements in Rn are column vectors. On the other hand, elements in R1n are row ~ ~ ~ vectors. Obviously in the above equation, A 2 Rnn , B 2 Rnm , C 2 Rmn , ~ mm D2R . Deleting Dy in (8.3), we have dDx ¼ ADx; dt

ð8:4Þ

~ ~ ~ 1 ~ A ¼ A BD C:

ð8:5Þ

where

Matrix A 2 Rnn is often referred to as the state matrix or coefficient matrix. Therefore, small-signal stability studies local characteristics of the power system, i.e., asymptotic stability of an equilibrium point before the system is disturbed. Obviously, the theoretical basis to study power system small-signal stability by using the Lyapunov linearized method is that the disturbance must be sufficiently small. When the power system is subject to any such disturbance, state variables of the transient system model vary over a very small range. Hence asymptotic stability of the linearized system can guarantee a certain type of asymptotic stability of the actual nonlinear system.

492

8 Small-Signal Stability Analysis of Power Systems

We know that when the power system is subject to a sufficiently small disturbance at steady-state operation, there can be two consequences. One is that the disturbance approaches zero with time (i.e., disturbed movement approaches the undisturbed movement and all eigenvalues of corresponding matrix A have negative real parts) in this case the system is asymptotically stable at steady-state operation. The disturbed system will eventually return to the steady-state operation before the occurrence of disturbance. Another possible consequence is that disturbance Dx increases indefinitely with time, no matter how small the disturbance is (i.e., the real part of at least one of the eigenvalues of A is positive). Obviously the system is then unstable at this steady-state operating point. For the operation of a real power system, study of the critically stable situation is not so important, except that we can see it as the limiting case of small-signal stability. Here we need to point out that in our previous discussion about system stability, we assumed that the disturbance was instantaneous. That is, the system state moves instantly from Dx ¼ 0 to Dx(t0), and the disturbance disappears when the movement happens. However, the same theory is applicable to the study of stability when the system is subject to a permanent disturbance, because we can consider this as a case of stability subject to an instantaneous disturbance but operating at a new equilibrium point. Furthermore, for certain operating conditions in which the system is unstable in terms of small-signal stability or lacks damping, we can determine relationships between some controller parameters and the system eigenvalues (representing system stability) by using eigensolution analysis. In doing so, we can find certain ways to improve the power system small-signal stability. Hence small-signal stability analysis is a very important aspect of power system analysis and control. Therefore, power system analysis for operation at a steady state and subject to small disturbances includes: (1) Computation of steady-state values of various variables of the power system at a given steady-state operating condition, (2) Linearization of the differential-algebraic description of power system nonlinear dynamics to obtain the linearized differential-algebraic equations, (3) Formation of system state matrix A from the system linearized differentialalgebraic equations to determine system stability by calculation of the eigenvalues of A. In our above discussion, only the electromechanical oscillations between generators are considered in small-signal stability analysis. That is, we consider generators to be lumped rigid masses. However, the mechanical structure of real large-scale steam-turbine generation units is very complicated. It consists of several major lumped masses, such as turbine rotor, generator rotor, exciter rotor, etc. These lumped masses are connected by a rigid shaft of limited length. When generation units are disturbed, rotational speeds of the lumped masses are different during the system transient, due to elasticity between the lumped mass. This leads to torsional oscillations between each lumped mass. Because the inertia of each lumped mass is smaller than the total inertia of generation units, and taking into account the

8.2 Linearized Equations of Power System Dynamic Components

493

relevant stiffness, the frequency of torsional oscillations between each lumped mass is higher than that of electromechanical oscillations between generation units. Frequency of torsional oscillation is between about 10 and 50 Hz. This oscillation is often referred to as sub-synchronous oscillation (SSO). When SSO occurs, there is an oscillating torsional torque between lumped masses connected by the common shaft of a generation unit. Fatigue accumulation due to repeated episodes of torsional oscillation on the shaft will reduce shaft operating life. If the torsional torque exceeds a certain limit, shaft cracking, even breaking, can happen. Occurrence of SSO is mainly related to excitation control, governing control, HVDC control, and interaction between transmission line and series compensation of the line. When carrying out torsional oscillation analysis, we need to first establish a mathematical model of the shaft system of the steam turbine and generator. In addition, because frequency of torsional oscillation is high, a quasi-steady-state model of various components cannot be used. Instead, the electromagnetic transients in the power system need to be considered. Detailed analysis on torsional oscillation is outside the scope of this book. In this chapter, we first derive linearized models of various dynamic components in power systems, to establish the linearized equations of the whole system, in order to demonstrate the basic steps for computation of small-signal stability in power systems. Then, we will discuss the eigensolution problem in power system smallsignal stability analysis and the analytical methods required to study power system oscillations.

8.2

Linearized Equations of Power System Dynamic Components

In power system small-signal stability analysis, we need to linearize various dynamic components in the power system. In linearization, limiters in control devices often need not be considered. This is because in normal steady-state operation, the values of state variables associated with control devices are within the range determined by the limiters. If disturbances are sufficiently small, variations of state variables will not go beyond these limitations. As far as dead zones associated with certain control devices are concerned, we generally consider the dead zone to be small and hence ignored. If the dead zone is large, we can simply consider that in this case the control device does not function.

8.2.1

Linearized Equation of Synchronous Generator

8.2.1.1

Linearized Equation of Each Part of a Synchronous Generator

(1) Synchronous generator: For a synchronous generator described by (7.114)– (7.116) at a given steady-state operating condition, steady-state values of

494

8 Small-Signal Stability Analysis of Power Systems

various variables are d(0), o(0), E0 q(0), E00 q(0), E0 d(0), E00 d(0), Id(0), Iq(0), Vd(0), Vq(0), Pm(0), Pe(0), Efq(0) which can be calculated from (7.74–7.78) and (7.118–7.122). Linearizing each equation at these steady-state values, we obtain the linearized equation of a synchronous generator 9 dDd > > ¼ os Do > > > dt > > n h i > > dDo 1 00 00 00 00 00 > ¼ DDo Iqð0Þ DEq Idð0Þ DEd þ DPm Edð0Þ Xd Xq Iqð0Þ > > > dt TJ > > h i o > > > 00 00 00 > DId Eqð0Þ Xd Xq Idð0Þ DIq > > > > > 0 h i > dDEq > 1 = 0 00 ¼ 0 kd DEq þ ðkd 1ÞDEq þ DEfq dt Td0 ; > > 00 h i > dDEq > 1 > > ¼ 00 DE0q DE00q Xd0 Xd00 DId > > dt Td0 > > > > 0 >

dDEd 1 > 0 00 > > ¼ 0 kq DEd þ ðkq 1ÞDEd > > dt Tq0 > > > > h i 00 > dDEd 1 > 0 00 0 00 > > ¼ 00 DEd DEd þ Xq Xq DIq ; dt T q0

ð8:6Þ DVd ¼ DE00d Ra DId þ Xq00 DIq DVq ¼ DE00q Xd00 DId Ra DIq

) :

ð8:7Þ

(2) Excitation system: Taking an excitation system consisting of a DC exciter with thyristor-controlled regulator as an example, we can derive the linearized equation of (7.136–7.140) as follows. For measurement unit with VC ¼ V_ þ jXC I_, from coordinate transformation of (6.63), d, q components of voltage and current at generator terminals can be represented as V_ ¼ ðVd þ jVq Þejðdp=2Þ ;

I_ ¼ ðId þ jIq Þejðdp=2Þ :

ð8:8Þ

Obviously we have VC ¼ ½ðVd þ jVq Þ þ jXC ðId þ jIq Þejðdp=2Þ ¼ ðVd þ jVq Þ þ jXC ðId þ jIq Þ : qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ¼ ðVd XC Iq Þ2 þ ðVq þ XC Id Þ2

ð8:9Þ

Linearizing the above equations at steady-state values, we can obtain DVC ¼ Kcd ðDVd XC DIq Þ þ Kcq ðDVq þ XC DId Þ;

ð8:10Þ

8.2 Linearized Equations of Power System Dynamic Components

495

where

9 Kcd ¼ ðVdð0Þ XC Iqð0Þ Þ=VCð0Þ ; Kcq ¼ ðVqð0Þ þ XC Idð0Þ Þ=VCð0Þ = qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ : ; VCð0Þ ¼ ðVdð0Þ XC Iqð0Þ Þ2 þ ðVqð0Þ þ XC Idð0Þ Þ2

ð8:11Þ

Linearizing (7.136), substituting (8.10) into it to cancel DVC, we can obtain the linearized equation of measuring and filtering unit to be dDVM 1 ¼ ðDVM þ Kcq XC DId Kcd XC DIq þ Kcd DVd þ Kcq DVq Þ: dt TR

ð8:12Þ

Representing the saturation characteristic of the exciter by (7.140) and linearizing (7.139) at steady-state operating point, we can have the linearized equation of the exciter to be i dDEfq 1 h E 1 ¼ KE þ nE cE Enfqð0Þ DEfq þ DVR : ð8:13Þ dt TE Finally, linearizing (7.137, 7.138) and rearranging them together with (8.12) and (8.13), we can obtain the linearized equation of the whole DC excitation system to be 9 E 1 KE þ nE cE Enfqð0Þ > dDEfq 1 > > ¼ DEfq þ DVR > > > dt TE TE > > > > dDVR 1 KA KA KA > > ¼ DVR DVF DVM þ DVS > = dt TA TA TA TA : ð8:14Þ E 1 > KF KE þ nE cE Enfqð0Þ > > dDVF KF 1 > > ¼ DEfq þ DVR DVF > > dt TE TF TE TF TF > > > > > dDVM 1 Kcq XC Kcd XC Kcd Kcq ; ¼ DVM þ DId DIq þ DVd þ DVq > dt TR TR TR TR TR (3) Power system stabilizer: For a Power system stabilizer (PSS) of Fig. 6.14, from (7.142) and (7.143) we can establish the following linearized equations when input to PSS is the deviation of rotor speed, VIS = o os 9 dDV1 KS 1 > > ¼ Do DV1 > > > dt T6 T6 > > > > dðDV1 DV2 Þ 1 > > ¼ DV2 = dt T5 : > dðT1 DV2 T2 DV3 Þ > > ¼ DV3 DV2 > > > dt > > > > dðT3 DV3 T4 DVS Þ ; ¼ DVS DV3 > dt

ð8:15Þ

After rearrangement, we can obtain linearized state equations of the PSS as follows

496

8 Small-Signal Stability Analysis of Power Systems

dDV1 dt dDV2 dt dDV3 dt dDVS dt

9 KS 1 > > Do DV1 > > T6 T6 > > > > > KS 1 1 > > ¼ Do DV1 DV2 = T6 T6 T5 : K S T1 T1 T1 T5 1 > > > ¼ Do DV1 DV2 DV3 > > T2 T6 T 2 T6 T2 T5 T2 > > > > > K S T1 T 3 T1 T3 T3 ðT1 T5 Þ T3 T2 1 ; ¼ Do DV1 DV2 DV3 DVS > T2 T4 T6 T2 T4 T6 T2 T4 T5 T2 T4 T4 ð8:16Þ ¼

(4) Prime mover and governing system: For the hydraulic turbine and its governing system, of Fig. 6.24, we can obtain its linearized equation from (7.171)–(7.177) to be 9 dDm Kd 1 > > ¼ Do Dx > > > dt TS TS >

> = dDx Kd ðKi þ Kb Þ Ki 1 Ki þ Kb ¼ Do þ Dm þ Dx : ð8:17Þ > dt TS Ti Ti TS > > > > dDPm 2KmH Kd 2KmH 2KmH 2 > ; ¼ Do þ Dm þ Dx DPm > dt TS To TS To 8.2.1.2

Matrix Description of Linearized Equation of Synchronous Generator Unit and Coordinate Transformation

(1) Matrix description of generation unit: For a generation unit described by (8.6), (8.7), (8.9), (8.15) and (8.17), its state variables can be arranged to form the following vector: Dxg ¼ ½Dd; Do; DE0q ; DE00q ; DE0d ; DE00d ; DEfq ; DVR ; DVF ; DVM ; DV1 ; DV2 ; DV3 ; DVS ; Dm; Dx; DPm T

:

ð8:18Þ

We define DVdqg ¼ ½DVd ; DVq T ;

DIdqg ¼ ½DId ; DIq T :

ð8:19Þ

Linearized differential equations of each generation unit can be written in the following matrix form dDxg Ig DIdqg þ B Vg DVdqg : ¼ Ag Dxg þ B dt

ð8:20Þ

Linearized equations of armature voltage equations can be arranged as g Dxg þ Z g DIdqg : DVdqg ¼ P

ð8:21Þ

g; B Ig ; In the two equations above, elements in the coefficient matrices A Vg ; P g; Z g , can be obtained easily by comparing (8.20) and (8.6), (8.9), B (8.15), (8.17) and comparing (8.21) and (8.7) as follows:

⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ Ag = ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢⎣

−

1 Td′′0

1 Td′′0

1 Tq′′0

1 Tq′′0

1 TA

−

1 T5

TS

2 K mH Tω

1 T4

2 K mH Kδ TS

−

Ki Ti

Kδ TS

1 T2 T2 − T3 T2T4

−

KA TA

Kδ ( K i + K β )

−

−

1 T6

T3 (T5 − T1 ) T2T4T5

1 T6

T1T3 T2T4T6

1 TR

KA TA

K ST1T3 T2T4T6

−

−

T5 − T1 T2T5

1 TF

KA TA

T1 T2T6

−

−

−

KF TETF

−

1 TE

K S T1 T2T6

K F K E′ TETF

K E′ TE

−

−

−

1 Td′0

KS T6

−

kq − 1 Tq′0

kq

TJ

Tq′0

−

−

I d (0)

−

−

kd − 1 Td′0

TJ

k − d Td′0

−

I q (0)

KS T6

D − TJ

ωs ⎤ ⎥ 1 ⎥ TJ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥, ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ 1 ⎥ − ⎥ TS ⎥ ⎥ ⎛ 1 K + Kβ ⎞ −⎜ + i ⎥ ⎟ T T ⎥ S ⎝ i ⎠ 2 K mH 2⎥ − ⎥ TS Tω ⎥ ⎦

8.2 Linearized Equations of Power System Dynamic Components 497

498

8 Small-Signal Stability Analysis of Power Systems

⎡ ⎢ ( X ′′ − X ′′) I − E ′′ q q (0) d (0) ⎢ d ⎢ TJ ⎢ ⎢ ⎢ X d′′ − X d′ ⎢ Td′′0 ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ B Ig = ⎢ ⎢ ⎢ K cq X C ⎢ ⎢ TR ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣⎢ ⎡ Pg = ⎢ ⎣

⎤ ( X d′′ − X q′′) I d (0) − Eq′′(0) ⎥⎥ ⎡ ⎥ TJ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ X q′ − X q′′ ⎥ ⎢ ⎥ ⎢ Tq′′0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ , B = ⎥ Vg ⎢ K cd ⎥ ⎢ ⎥ ⎢ TR ⎥ K cd X C ⎢ − ⎥ ⎢ TR ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢⎣ ⎥ ⎥ ⎦⎥ 1

1

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥, K cq ⎥ ⎥ TR ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦⎥

⎤ ⎥, ⎦

X q′′ ⎤ ⎡ −R Zg = ⎢ a ⎥ ⎣ − X d′′ − Ra ⎦

If different models of synchronous generator, excitation system and governing system are adopted, using the same procedure we can always first derive the linearized equation of each unit and then represent them in the form of (8.20) and (8.21). Of course, in (8.18), the sequential arrangement of state variables can be different, which would lead to different coefficient matrices.

8.2 Linearized Equations of Power System Dynamic Components

499

(2) Coordinate transformation: In (8.20) and (8.21), DVdqg and DIdqg are the deviation of d, q voltage and current components of each generator, respectively. Hence we must convert them into a representation in a unified x–y coordinate rotating at the same speed, so that they can then be connected to a common power network. Coordinate transformation for the terminal voltages of the generator is given by (6.62)

Vd Vq

sin d cos d ¼ cos d sin d

Vx : Vy

ð8:22Þ

Steady-state values Vd(0), Vq(0), Vx(0), Vy(0), and d(0) should also satisfy (8.22). That is Vdð0Þ sin dð0Þ cos dð0Þ Vxð0Þ ¼ : ð8:23Þ Vqð0Þ cos dð0Þ sin dð0Þ Vyð0Þ Linearizing (8.22) at steady-state values we have sin dð0Þ cos dð0Þ DVx cos dð0Þ DVd ¼ þ cos dð0Þ sin dð0Þ sin dð0Þ DVy DVq V xð0Þ Dd: Vyð0Þ

sin dð0Þ cos dð0Þ

ð8:24Þ

From (8.23) we can rewrite (8.24) as

DVd DVq

sin dð0Þ ¼ cos dð0Þ

cos dð0Þ sin dð0Þ

Vqð0Þ DVx þ Dd Vdð0Þ DVy

ð8:25Þ

that can be written simply as DVdqg ¼ Tgð0Þ DVg þ RVg Dxg ;

ð8:26Þ

where DVg ¼ Tgð0Þ

Vqð0Þ DVx ; RVg ¼ Vdð0Þ DVy sin dð0Þ cos dð0Þ ¼ : cos dð0Þ sin dð0Þ

0 0

0 ; 0

Note that Tg(0) is an orthogonal matrix, satisfying T T1 gð0Þ ¼ Tgð0Þ :

ð8:27Þ

500

8 Small-Signal Stability Analysis of Power Systems

Similarly, for generator current we can obtain DIdqg ¼ Tgð0Þ DIg þ RIg Dxg ; where

DIg ¼

DIx ; DIy

RIg ¼

Iqð0Þ Idð0Þ

ð8:28Þ 0 : 0

0 0

Substituting (8.26) and (8.28) into (8.21) to cancel DVdqg and DIdqg, we have DIg ¼ Cg Dxg þ Dg DVg ; where

ð8:29Þ

9 1 ðRVg P g Þ RIg = Cg ¼ TTgð0Þ ½Z g ;

1

Tgð0Þ Dg ¼ TTgð0Þ Z g

:

ð8:30Þ

Substituting (8.26) and (8.28) into (8.20) to cancel DVdqg and DIdqg and using (8.29), (8.30) to cancel DIg, we can obtain

where

dDxg ¼ Ag Dxg þ Bg DVg ; dt

ð8:31Þ

9 g þ B Ig Z 1 ðRVg P gÞ þ B Vg RVg = Ag ¼ A g : ; Ig Z 1 þ B Vg ÞTgð0Þ Bg ¼ ðB g

ð8:32Þ

Equations (8.31) and (8.29) consist of linearized equations of every generator, in the form of the state equation and output equation for a general time-invariant linear system.

8.2.2

Linearized Equation of Load

In small-signal stability analysis, a static load model is usually adopted. If a certain amount of induction motor load needs to be considered, we can use procedures similar to those used to derive the linearized equations of a synchronous generator, to establish the linearized equations of an induction motor. No matter which form is adopted to model the static voltage characteristics of load, deviation of injected current into the load has the following relationship to nodal voltage: DIl ¼ Yl DVl ;

ð8:33Þ

where

DIx DIl ¼ ; DIy

Gxx Yl ¼ Byx

Bxy ; Gyy

DVx DVl ¼ : DVy

ð8:34Þ

8.2 Linearized Equations of Power System Dynamic Components

501

The coefficients can be calculated from the following relationship between injected current and nodal voltage at the load node @Ix @Ix Gxx ¼ ; Bxy ¼ ; @Vx Vx ¼Vxð0Þ @Vy Vx ¼Vxð0Þ Vy ¼Vyð0Þ Vy ¼Vyð0Þ @Iy @Iy Byx ¼ ; Gyy ¼ : @Vx Vx ¼Vxð0Þ @Vy Vx ¼Vxð0Þ Vy ¼Vyð0Þ

ð8:35Þ

Vy ¼Vyð0Þ

If the static voltage characteristic of the load is modeled by a quadratic polynomial, we can use the relationship of (8.48) between injected current and node voltage and (8.35) to calculate relevant coefficients in (8.34) directly 9 > > > Gxx ¼ > 4 > > Vð0Þ > > > > 2 > Qð0Þ Vyð0Þ ðbQ þ 2cQ Þ þ Pð0Þ Vxð0Þ Vyð0Þ ðbP þ 2cP Þ Qð0Þ > > > Bxy ¼ 2 > 4 = V V > 2 Pð0Þ Vxð0Þ ðbP þ 2cP Þ þ Qð0Þ Vxð0Þ Vyð0Þ ðbQ þ 2cQ Þ

Pð0Þ 2 Vð0Þ

ð0Þ

ð0Þ

> Qð0Þ > > Byx ¼ 2 > > 4 Vð0Þ Vð0Þ > > > > > > 2 > Pð0Þ Vyð0Þ ðbP þ 2cP Þ Qð0Þ Vxð0Þ Vyð0Þ ðbQ þ 2cQ Þ Pð0Þ > > Gyy ¼ 2 > > 4 V V ; 2 Qð0Þ Vxð0Þ ðbQ þ 2cQ Þ Pð0Þ Vxð0Þ Vyð0Þ ðbP þ 2cP Þ

ð0Þ

:

ð8:36Þ

ð0Þ

When an exponential function is used to model static voltage characteristics of the load, the relationship between load injected current and node voltage, of (7.49), can be used jointly with (8.35) to derive relevant coefficients in (8.34) directly as Pð0Þ Gxx ¼ 2 Vð0Þ Qð0Þ Bxy ¼ 2 Vð0Þ Qð0Þ Byx ¼ 2 Vð0Þ Pð0Þ Gyy ¼ 2 Vð0Þ

ð2 mÞ ð2 nÞ ð2 nÞ ð2 mÞ

2 Vxð0Þ 2 Vð0Þ 2 Vyð0Þ 2 Vð0Þ 2 Vxð0Þ 2 Vð0Þ 2 Vyð0Þ 2 Vð0Þ

! 1 ! 1

Qð0Þ þ 2 Vð0Þ þ

! 1 ! 1

Pð0Þ 2 Vð0Þ

Pð0Þ 2 Vð0Þ

Qð0Þ 2 Vð0Þ

!9 > > > > > > > > !> > > Vxð0Þ Vyð0Þ > > > > ð2 mÞ > 2 = Vð0Þ ! : > Vxð0Þ Vyð0Þ > > > ð2 mÞ > 2 > Vð0Þ > > > !> > > Vxð0Þ Vyð0Þ > > > > ð2 nÞ ; 2 V Vxð0Þ Vyð0Þ ð2 nÞ 2 Vð0Þ

ð8:37Þ

ð0Þ

Especially, when there is not enough information about the static voltage characteristics of the load, a normally acceptable load model is to represent load active power by a constant current (i.e., taking m ¼ 1) and load reactive power by a constant impedance (i.e., taking n ¼ 2).

502

8 Small-Signal Stability Analysis of Power Systems

8.2.3

Linearized Equation of FACTS Components

1. SVC From (7.197) and (7.198) we can obtain the following linearized equation directly 9 dDBS1 KS 1 > > ¼ DV DBS1 = dt TS TS : > dðTS2 DBSVC TS1 DBS1 Þ > ; ¼ DBS1 DBSVC dt

ð8:38Þ

Because V 2 ¼ Vx2 þ Vy2 , after linearization we have DV ¼

Vxð0Þ Vyð0Þ DVx þ DVy : Vð0Þ Vð0Þ

ð8:39Þ

Substituting the above equation into (8.38) and after rearrangement we obtain dDxs ¼ As Dxs þ Bs DVs ; dt

ð8:40Þ

where Dxs ¼ " As ¼

DBS1 DBSVC

;

DVx

DVs ¼ DVy #

T1S

0

TS TS1 TS TS2

T1S1

;

KS Bs ¼ TS Vð0Þ

"

Vxð0Þ TS1 TS2

Vxð0Þ

Vyð0Þ TS1 TS2

Vyð0Þ

#

9 > > > > = > > > > ;

:

ð8:41Þ

In addition, from (7.50) we can obtain the relationship of deviation between SVC injected current and nodal voltage to be DIs ¼ Cs Dxs þ Ds DVs ; where

DIx DIs ¼ ; DIy

DVx DVs ¼ DVy " # 0 Vyð0Þ 1 Cs ¼ ; ð1 XT BSVCð0Þ Þ2 0 Vxð0Þ

BSVCð0Þ 0 Ds ¼ 1 XT BSVCð0Þ 1

ð8:42Þ 9 > > > > = : 1 > > > > 0 ;

ð8:43Þ Hence (8.40) and (8.42) form the linearized equation of the SVC. 2. TCSC From (7.208) and (7.209) we can obtain the following linearized equation directly

8.2 Linearized Equations of Power System Dynamic Components

503

9 dDBT1 KT 1 > > ¼ DPT DBT1 = dt TT TT : > dðTT2 DBTCSC TT1 DBT1 Þ ; ¼ DBT1 DBTCSC > dt

ð8:44Þ

From (7.211) we have DPT ¼ ðVxið0Þ Vyjð0Þ Vyið0Þ Vxjð0Þ ÞDBTCSC þ BTCSCð0Þ Vyjð0Þ DVxi BTCSCð0Þ Vxjð0Þ DVyi BTCSCð0Þ Vyið0Þ DVxj þ BTCSCð0Þ Vxið0Þ DVyj

:

ð8:45Þ

Substituting the above equation into (8.44) and after rearrangement we have dDxt ¼ At Dxt þ Bt DVt ; dt where

Dxt ¼

DBT1

ð8:46Þ

9 > > > > > > > > > > > > > > =

T

; DVt ¼ DVxi DVyi DVxj DVyj DBTCSC 3 1 KT ðV V V V Þ yið0Þ xjð0Þ xið0Þ yjð0Þ 6 7 TT TT 7 At ¼ 6 4 1 : TT1 1 TT1 KT 1 5 > ðVyið0Þ Vxjð0Þ Vxið0Þ Vyjð0Þ Þ > > TT2 TT2 TT TT2 TT TT2 > 2 3> > > Vyjð0Þ Vxjð0Þ Vyið0Þ Vxið0Þ > > > KT BTCSCð0Þ > 4 5 > Bt ¼ TT1 TT1 TT1 TT1 > TT ; Vyjð0Þ Vxjð0Þ Vyið0Þ Vxið0Þ > TT2 TT2 TT2 TT2 ð8:47Þ 2

In addition, from (7.51) we can directly obtain the relationship of deviation in TCSC injected current and nodal voltage to be DIt ¼ Ct Dxt þ Dt DVt ; where DIt ¼ DIxi 2 0 60 6 Ct ¼ 6 40 0

DIyi

DIxj

Vyið0Þ Vyjð0Þ

DIyj 3

Vxjð0Þ Vxið0Þ 7 7 7; Vyjð0Þ Vyið0Þ 5 Vxið0Þ Vxjð0Þ

ð8:48Þ

T 2

0 6 1 6 Dt ¼ BTCSCð0Þ 6 4 0 1

1

0

0 1 1 0 0 1

9 > > > > > 3> = 1 > 7 0 7 >: ð8:49Þ 7> > > 1 5> > > ; 0

Thus (8.46) and (8.48) form the linearized equations of a TCSC.

8.2.4

Linearized Equation of HVDC Transmission System

When transient behavior of an HVDC transmission line is considered, the control equations of HVDC transmission line, rectifier, and inverter are given by (7.222),

504

8 Small-Signal Stability Analysis of Power Systems

and (7.224)–(7.227). Canceling VdI in (7.226) by using the first equation in (7.53) and ignoring the limitation on a and b, we can obtain the following linearized equation around steady state 9 kI VIð0Þ sin bð0Þ kR VRð0Þ sin að0Þ dDId R > > ¼ DId Da þ Db > > dt L L L > > > > kI cos bð0Þ kR cos að0Þ > > > þ DVR DVI > > L L > > > > dDx1 1 > > > ¼ ðDId Dx1 Þ = dt Tc3 : dðKc1 Dx1 DaÞ Kc2 > > > ¼ Dx1 > > dt Tc2 > > > > kI VIð0Þ sin bð0Þ kI cos bð0Þ > dDx4 XcI 1 > ¼ DId Dx4 Db þ DVI > > > dt Tv3 Tv3 Tv3 Tv3 > > > > > dðKv1 Dx4 DbÞ Kv2 > ; ¼ Dx4 dt Tv2 ð8:50Þ

Relationships between the magnitude of AC bus voltages of rectifier and inverter 2 2 and their x, y components are VR2 ¼ VxR þ VyR , VI2 ¼ VxI2 þ VyI2 . Linearizing the above equations, we have 9 VxRð0Þ VyRð0Þ > DVR ¼ DVxR þ DVyR > > = VRð0Þ VRð0Þ : VxIð0Þ VyIð0Þ > > DVI ¼ DVxI þ DVyI > ; VIð0Þ VIð0Þ

ð8:51Þ

Substituting (8.51) into (8.50) to cancel DVR and DVI and after rearrangement we obtain dDxd ¼ Ad Dxd þ Bd DVd ; dt

ð8:52Þ

where Dxd ¼ ½DId Dx1 Dx4 Da DbT

T DVd ¼ DVxR DVyR DVxI DVyI

) ;

ð8:53Þ

where coefficient matrices Ad and Bd can be easily obtained by comparing (8.52) and the original equation.

8.2 Linearized Equations of Power System Dynamic Components

505

Algebraic equations of a two-terminal HVDC transmission system can be derived from relationships of power and current on the AC and DC sides of the converter. For the rectifier, the power relationship is VxR IxR þ VyR IyR ¼ XcR Id2 kR Id VR cos a:

ð8:54Þ

Linearizing the above equation we have VxRð0Þ DIxR þ VyRð0Þ DIyR ¼ IxRð0Þ DVxR IyRð0Þ DVyR þ 2XcR Idð0Þ DId kR VRð0Þ cos að0Þ DId kR Idð0Þ cos að0Þ DVR : ð8:55Þ þ kR Idð0Þ VRð0Þ sin að0Þ Da In addition, from the third equation in (8.52) we have 2 2 IR2 ¼ IxR þ IyR ¼ kR2 Id2 :

ð8:56Þ

The linearized form of the above equation is IxRð0Þ DIxR þ IyRð0Þ DIyR ¼ kR2 Idð0Þ DId :

ð8:57Þ

Substituting (8.51) into (8.55) to cancel DVR and noting the reactive power injection into the AC system from the rectifier, QR(0) = VyR(0) IxR(0) VxR(0) IyR(0), is always nonzero, we can derive the deviation of node injected current from (8.55) and (8.57) and have the following matrix form DIR ¼ CR Dxd þ DR DVR ; " DIR ¼

DIxR DIyR

#

" ;

DVR ¼

DVxR DVyR "

#

C11 1 CR ¼ VxRð0Þ IyRð0Þ VyRð0Þ IxRð0Þ C 21

0 0

C14

0

0 0

C24

0

#

C11 ¼ 2XcR Idð0Þ IyRð0Þ kR VRð0Þ IyRð0Þ cos að0Þ kR2 Idð0Þ VyRð0Þ C14 ¼ kR VRð0Þ IyRð0Þ Idð0Þ sin að0Þ C21 ¼ 2XcR Idð0Þ IxRð0Þ þ kR VRð0Þ IxRð0Þ cos að0Þ þ kR2 Idð0Þ VxRð0Þ

ð8:58Þ 9 > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > =

> > > > > C24 ¼ kR VRð0Þ IxRð0Þ Idð0Þ sin að0Þ > > > > > " # > > D D > 11 12 1 > > > DR ¼ > > VxRð0Þ IyRð0Þ VyRð0Þ IxRð0Þ D > D22 > 21 > > > > > kR VxRð0Þ Idð0Þ cos að0Þ kR VyRð0Þ Idð0Þ cos að0Þ > > > > D11 ¼ IyRð0Þ IxRð0Þ þ ; D12 ¼ IyRð0Þ IyRð0Þ þ > > VRð0Þ VRð0Þ > > > > > > kR VxRð0Þ Idð0Þ cos að0Þ kR VyRð0Þ Idð0Þ cos að0Þ > > > D21 ¼ IxRð0Þ IxRð0Þ þ ; D22 ¼ IxRð0Þ IyRð0Þ þ > ; VRð0Þ VRð0Þ

:

ð8:59Þ

506

8 Small-Signal Stability Analysis of Power Systems

Power relationship of the inverter is VxI IxI þ VyI IyI ¼ XcI Id2 þ kI Id VI cos b:

ð8:60Þ

After linearization we have VxIð0Þ DIxI þ VyIð0Þ DIyI ¼ IxIð0Þ DVxI IyIð0Þ DVyI þ 2XcI Idð0Þ DId þ kI VIð0Þ cos bð0Þ DId þ kI Idð0Þ cos bð0Þ DVI

ð8:61Þ

kI Idð0Þ VIð0Þ sin bð0Þ Db: Similarly from the third equation in (8.53) and after linearization we have IxIð0Þ DIxI þ IyIð0Þ DIyI ¼ kI2 Idð0Þ DId :

ð8:62Þ

Again, substituting (8.51) into (8.61) to cancel DVI, from (8.61) and (8.62) we obtain the following matrix form for current and voltage deviation DII ¼ CI Dxd þ DI DVI ;

ð8:63Þ

where

DIxI DII ¼ ; DIyI

DVxI DVI ¼ : DVyI

ð8:64Þ

Equation (8.58) and (8.63) form the algebraic equation of the DC system DId ¼ Cd Dxd þ Dd DVd ;

ð8:65Þ

where

DIR DId ¼ ; DII

DVR DVd ¼ ; DVI

CR Cd ¼ ; CI

DR Dd ¼ 0

0 : ð8:66Þ DI

When different mathematical models of the DC system are used, we can follow similar procedures to derive linearized equations like (8.52) and (8.65).

8.3 8.3.1

Steps in Small-Signal Stability Analysis Network Equation

For convenience of expression, we write the network equation of (8.36) in the form of block matrices. Noting that the network equation is itself linear, we can write the linear equation for the relationship between deviation of node injection current and node voltage in x–y coordinates directly, to be

8.3 Steps in Small-Signal Stability Analysis

2

3 2 DI1 Y11 6 .. 7 6 .. 6 . 7 6 . 6 7 6 6 DIi 7 ¼ 6 Yi1 6 . 7 6 . 4 . 5 4 . . . DIn Yn1

507

Y1i .. . Yii .. . Yni

32 3 Y1n DV1 .. 76 .. 7 . 76 . 7 76 7 Yin 76 DVi 7; 7 6 .. 54 .. 7 5 . . DVn Ynn

ð8:67Þ

where DIi ¼

DIxi ; DIyi

DVi ¼

DVxi ; DVyi

Yij ¼

Gij Bij

Bij ; Gij

ð8:68Þ

i; j ¼ 1; 2; . . . ; n: For load nodes, we can substitute the relationship between deviation of injected current and node voltage into the above equation to cancel the current deviation at the load node. Assuming load is connected at node i, then the network equation after canceling this node is just a simple correction to the original network equation of (8.67): current deviation at node i becomes zero, the ith diagonal block in the network admittance matrix changes to Yii Yli and nothing more. Without loss of generality, we assume the sequence of nodes in the network is: firstly each generator node, then each SVC node followed by two-terminal nodes of TCSC, then AC bus nodes of each HVDC transmission line (the node on the rectifier side first and inverter side second), and finally the remaining nodes. Canceling current deviation of all load nodes, we have the following block-matrix form of the network equation 2

3 2 DIG YGG 6 DIS 7 6 YSG 6 7 6 6 DIT 7 ¼ 6 YTG 6 7 6 4 DID 5 4 YDG 0 YLG

YGS YSS YTS YDS YLS

YGT YST YTT YDT YLT

YGD YSD YTD YDD YLD

32 3 YGL DVG 6 7 YSL 7 76 DVS 7 7 6 YTL 76 DVT 7 7; YDL 54 DVD 5 YLL DVL

ð8:69Þ

where DIG and DVG are vectors consisting of deviation of injected current and node voltage of all generators, respectively; DIS and DVS vectors of deviation of node injection current and node voltage of all SVC nodes; DIT and DVT all TCSC nodes; DID and DVD those at AC busbars of all converters; DVL associated with voltage of remaining nodes. All those vectors can be written as T T 9 DIG ¼ ½ DIg1 DIg2 ; DVG ¼ ½ DVg1 DVg2 > > > > DIS ¼ ½ DIs1 DIs2 T ; DVS ¼ ½ DVs1 DVs2 T > > = T T : ð8:70Þ DIT ¼ ½ DIt1 DIt2 ; DVT ¼ ½ DVt1 DVt2 > T T> > DID ¼ ½ DId1 DId2 ; DVD ¼ ½ DVd1 DVd2 > > > ; DVL ¼ ½ DV1 DV2 T

508

8.3.2

8 Small-Signal Stability Analysis of Power Systems

Linearized Differential Equations of Whole Power System

Equations of all generation units are formed from (8.31) and (8.29) to be dDxG ¼ AG DxG þ BG DVG ; dt

ð8:71Þ

DIG ¼ CG DxG þ DG DVG ;

ð8:72Þ

where )

AG ¼ diagfAg1

Ag2

g;

BG ¼ diagfBg1

Bg2

g

CG ¼ diagfCg1

Cg2

g;

DG ¼ diagfDg1

Dg2

g

:

ð8:73Þ

Equations (8.40) and (8.42) of each SVC can form equations of all SVCs to be dDxS ¼ AS DxS þ BS DVH ; dt

ð8:74Þ

DIS ¼ CS DxS þ DS DVS ;

ð8:75Þ

where AS ¼ diagfAs1

As2

g;

BS ¼ diagfBs1

Bs2

g

CS ¼ diagfCs1

Cs2

g;

DS ¼ diagfDs1

Ds2

g

) :

ð8:76Þ

TCSC equations are formed from (8.46) and (8.48) ðA s1 IÞvA ¼ ðlA s1 ÞvA ðA s2 IÞvA ¼ ðlA s2 ÞvA

) ;

ð8:77Þ

DIT ¼ CT DxT þ DT DVT ;

ð8:78Þ

where AT ¼ diagfAt1

At2

g;

BT ¼ diagfBt1

Bt2

g

CT ¼ diagfCt1

Ct2

g;

DT ¼ diagfDt1

Dt2

g

) :

ð8:79Þ

All two-terminal HVDC transmission lines have the following equations dDxD ¼ AD DxD þ BD DVD ; dt

ð8:80Þ

DID ¼ CD DxD þ DD DVD ;

ð8:81Þ

8.3 Steps in Small-Signal Stability Analysis

509

where AD ¼ diagfAd1 CD ¼ diagfCd1

Ad2 Cd2

g; g;

BD ¼ diagfBd1 Bd2 DD ¼ diagfDd1 Dd2

) g : g

ð8:82Þ

Substituting (8.72), (8.75), (8.78), and (8.81) into (8.69) to cancel DIG, DIS, DIT, and DID, together with (8.71), (8.74), (8.77), and (8.80), we can obtain the matrix formulations, as required by (8.3): 9 9 = Dx ¼ ½DxG DxS DxT DxD T > > > > > > T; > Dy ¼ ½DVG DVS DVT DVD DVL > > > > 2 3 > > AG 0 0 0 > > > > 6 0 A 7 > 0 0 7 S > ~ 6 > > A¼6 7 > > 4 0 5 0 AT 0 > > > > > 0 0 0 AD > > 2 3> > > CG 0 0 0 > 2 3 > BG 0 0 0 0 6 7= 0 C 0 0 7 6 S 6 7 >: ð8:83Þ 0 07 7 ~ 6 ~ 6 0 BS 0 6 7> B¼6 ; C ¼ 0 0 C 0 7 T > 6 7> 4 0 0 BT 0 0 5 > 6 7> 0 0 CD 5 > > 4 0 > > 0 0 0 BD 0 > > > 0 0 0 0 > > 2 3 > > > YGG DG YGS YGT YGD YGL > > > 6 7 > > Y Y D Y Y Y 6 SG SS S ST SD SL 7 > > 6 7 > ~ 6 > 7 > D ¼ 6 YTG YTS YTT DT YTD YTL 7 > > > 6 7 > > YDS YDT YDD DD YDL 5 4 YDG > > ; YLG YLS YLT YLD YLL ~ ~ ~ ~ Obviously, A, B, and C are sparse block matrices, as is D which is also an ~ ~ ~ ~ admittance matrix. Using matrices A, B, C and D, and from (8.5) we can obtain the system state matrix A. By now, we have obtained the linearized equations of a power system at a steady-state operating point. Finally, we would like to point out: 1. If this linearized system is asymptotically stable, i.e., the real part of all eigenvalues of matrix A are negative, the actual nonlinear system is asymptotically stable at this equilibrium point. 2. The method used to form matrix A is different in various commercial software packages. In the above, we only give one way to form it to introduce the principles and techniques used in forming matrix A [189, 190]. There are various ~ ~ ~ ~ alternative formats of matrices A, B, C, and D in (8.83) that are related with the sequence order of state variables, format of network equations, algebraic equations of various dynamic components and ways to treat the network equations.

510

8 Small-Signal Stability Analysis of Power Systems

Different methods determine the complexity of, and flexibility in developing, the program, but do not change the resulting eigensolution. 3. In the formation of the above linearized equations, we have considered generation units, SVC, TCSC, two-terminal HVDC transmission lines. We can also treat other dynamic components in power systems in similar ways. For example, for dynamic components (such as induction motor loads) we can derive their linearized equation in the same way as treating generators; for multiterminal HVDC transmission lines, we can obtain the linearized equation as we have done in treating two-terminal HVDC transmission lines. We can then arrange the linearized equations into the equation of the whole power system. 4. Matrix A, as formed, must have a zero eigenvalue. A zero eigenvalue exists because the absolute angle of the generator rotors is not unique. In other words, there is a redundant rotor angle in a power system model. In fact, power distribution among generators is determined by the relative rotor angle of generators. If the absolute rotor angle of all generators is added to by a fixed value, the power distribution does not change at all. Hence this does not affect system stability. To eliminate the zero eigenvalue, we only need to choose the rotor angle of any particular generator as a reference and then use the relative rotor angle of other generators as the new state variable. In doing so, the dimension of state matrix A and the corresponding state variable vector is reduced by one. 5. In the case that all generator torques are not directly related to rotor speed, i.e., when there is no damping term in the swing equation and the governing effect is ignored; matrix A will have another zero eigenvalue. Similarly, to remove this zero eigenvalue, we only need to choose the rotor speed of any generator as a reference and use the relative rotor speed of the other generators as new state variables. Again, in doing so, the order of matrix A and the corresponding state variable vector is reduced by one. Knowing the origin of the zero eigenvalues, we do not have to apply the treatment of (4) and (5) above, but simply eliminate the corresponding zero eigenvalues in our computational results. However, due to errors in load flow calculation and in the computation of eigenvalues, we should note that the theoretically zero eigenvalues will be computed as eigenvalues with very small magnitude.

8.3.3

Program Package for Small-Signal Stability Analysis

From what we have discussed previously, we can develop a program package for small-signal stability analysis of an AC/DC power system with FACTS devices such as SVC and TCSC installed. Basic steps in developing the stability analysis program are: 1. Load flow calculation at a given steady-state operating condition of the power system. This includes finding the voltage, current, and power at each node in the system. 2. Formation of the admittance matrix in (8.67). 3. Treatment of load. Load power and voltage at steady-state operation are known to be P(0), Q(0), Vx(0), and Vy(0). From parameters of the static voltage characteristics of load, we calculate matrix elements Gxx, Bxy, Byx, and Gyy in (8.34) from

8.3 Steps in Small-Signal Stability Analysis

511

(8.36) or (8.37). These will be used to adjust diagonal blocks related to loads in the admittance matrix. 4. Establishment of linearized equations of dynamic components in the system. Firstly we calculate initial values of all variables of generators from (7.74)– (7.78) and (7.118)–(7.122). Then we can form matrices Ag , BIg , BVg , Pg , and Zg in (8.20) and (8.21) as well as matrices Tg(0), RVg, and RIg in (8.26) and (8.28). Finally we calculate matrices from (8.30) and (8.32) to establish the linearized equation of generators. In a similar way we can obtain linearized equations of all dynamic components in the power system. 5. Formation of system state matrix A from (8.5). This is obtained by forming ~ ~ ~ ~ matrices A, B, C, and D from (8.71)–(8.83). 6. Calculation of all eigenvalues of the state matrix A by using the QR method [187, 188]. The result of this calculation is used to determine system smallsignal stability. The QR method to calculate all eigenvalues of matrix A will be introduced in Sect. 8.4. [Example 8.1] Single-line diagram of the 9-node power system, line data, generator parameters, and load flow at a steady-state operating condition is given in Fig. 7.7, Tables 7.5–7.7, respectively. System frequency is 60 Hz. All loads in the system are modeled by constant impedance. Generator 1 uses the classical model, generators 2 and 3 the double-axis model with self-excited potential-source excitation system. Parameters of the exciter are: XC ¼ 0;

KA ¼ 200;

TR ¼ 0:03 s;

TA ¼ 0:02 s;

TB ¼ 10:0 s;

TC ¼ 1:0 s:

In addition, the damping coefficient of each generator Di is 1.0. [Solution] In the following, we will demonstrate the process of small-signal stability analysis for the example power system. For simplicity of expression, a blank in the matrix will represent either zero or a zero matrix: (1) From load flow calculation, (7.74)–(7.78) and (7.118)–(7.122) we calculate initial values of all variables of generators shown in Table 8.1. Equivalent admittance of loads has been calculated in example 8.1 which is included in the power network model. (2) Establishment of linearized equations of generators using the method introduced in Sect. 8.2.1. Generator 1 We can calculate coefficient matrices in (8.20), (8.21), (8.26), and (8.28) as follows: Table 8.1 Initial values of generator variables

1 2 3

d(0)

Vq(0)

Vd(0)

Iq(0)

Id(0)

Efq(0)

E0 q(0)

E0 d(0)

2.27165 61.09844 54.13662

1.03918 0.63361 0.66607

0.04122 0.80571 0.77909

0.67801 0.93199 0.61941

0.28716 1.29015 0.56147

1.78932 1.40299

1.05664 0.78817 0.76786

0.62220 0.62424

512

8 Small-Signal Stability Analysis of Power Systems

Dd1 0:0 376:99112 0:0 0:0 0:0 0:0 ; BI1 ¼ ; BV1 ¼ ; Do1 0:0 0:02115 0:0 0:02235 0:0 0:0 0:0 0:0 0:0 0:0608 P1 ¼ ; Z1 ¼ ; 0:0 0:0 0:0608 0:0 0:03964 0:99921 1:03918 0:0 T1ð0Þ ¼ ; RV1 ¼ ; 0:99921 0:03964 0:04122 0:0 0:67801 0:0 RI1 ¼ : 0:28716 0:0 A1 ¼

From (8.32), (8.30), and the matrices above, we can obtain matrices in the linearized equation of the generator of (8.29) and (8.31) to be A1 ¼

0:0

376:99112

;

0:0

0:0

B1 ¼ ; 0:38198 0:02115 0:01457 0:36729 17:36532 0:0 0:0 16:44737 C1 ¼ ; D1 ¼ : 0:68886 0:0 16:44737 0:0 Generator 2 In a similar way, we can obtain coefficient matrices of the linearized equation of generator 2 to be ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ΔE fq 2 ⎢ ⎢ ΔVR 2 ⎢ ΔVM 2 ⎢⎣

Δδ 2 Δω2 ΔEq′ 2 A 2 = ΔEd′ 2

376.99112 −0.07813

⎤ ⎥ ⎥ ⎥ −0.16667 0.16667 ⎥ −1.86916 ⎥, −0.10000 −4.90000 −1000.0 ⎥ ⎥ −50.0000 −10000.0 ⎥ −33.33333⎥⎦ −0.07281 −0.10079

⎡ ⎤ ⎡ ⎤ ⎢ −0.05422 −0.06935⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ −0.12933 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 1.24785 ⎥ , BV 2 = ⎢ BI 2 = ⎢ ⎥, ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 26.20186 20.60518⎥ ⎣ ⎦ ⎣ ⎦

8.3 Steps in Small-Signal Stability Analysis

⎡ P2 = ⎢ ⎣

1.0 1.0

513

⎤ ⎡ 0.0000 0.1969⎤ ⎥ , Z 2 = ⎢ −0.1198 0.0000 ⎥ , ⎦ ⎣ ⎦

⎡0.87545 −0.48331⎤ T2(0) = ⎢ ⎥, ⎣ 0.48331 0.87545 ⎦

⎡ 0.63361 RV 2 = ⎢ ⎣ −0.80571

⎤ ⎡ 0.93199 ⎥ , RI 2 = ⎢ −1.29015 ⎦ ⎣

⎤ ⎥, ⎦

376.99112 ⎡ ⎤ ⎢ −0.58783 −0.07813 −0.52542 0.25140 ⎥ ⎢ ⎥ ⎢ −0.86982 ⎥ −1.24624 0.16667 ⎢ ⎥ −8.20664 A 2 = ⎢ 4.01549 ⎥, ⎢ −0.10000 −4.90000 −1000.0 ⎥ ⎢ ⎥ −50.0000 −10000.0 ⎥ ⎢ ⎢ −33.33333⎥⎦ ⎣

⎡ ⎢ −0.08958 ⎢ ⎢ 0.52177 ⎢ B 2 = ⎢ 5.54816 ⎢ ⎢ ⎢ ⎢ 32.89707 ⎣

⎡ 7.25066 C2 = ⎢ ⎣1.14659

⎤ 0.56646 ⎥⎥ 0.94512 ⎥ ⎥ −3.06294 ⎥ , ⎥ ⎥ ⎥ 5.37531 ⎥⎦

7.30761 −2.45458 −4.03428 −4.44617

⎡ −1.38295 −7.58377 ⎤ D2 = ⎢ ⎥. ⎣ 5.84220 1.38295 ⎦

⎤ ⎥, ⎦

514

8 Small-Signal Stability Analysis of Power Systems

Generator 3 Similarly we have Δδ 3 ⎡ ⎢ Δω3 ⎢ ΔEq′ 3 ⎢ ⎢ A 3 = ΔEd′ 3 ⎢ ΔE fq 3 ⎢ ⎢ ΔVR 3 ⎢ ΔVM 3 ⎢⎣

376.99112 −0.16611 −0.10289 −0.09327 −0.16978

⎤ ⎥ ⎥ ⎥ 0.16978 ⎥ −1.66667 ⎥, −0.10000 −4.90000 −1000.0 ⎥ ⎥ −50.0000 −10000.0 ⎥ −33.33333⎥⎦

⎡ ⎤ ⎡ ⎤ ⎢ −0.11076 −0.13396 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ −0.19205 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 1.67967 ⎥ , BV 3 = ⎢ BI 3 = ⎢ ⎥, ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 25.33623 21.66071⎥ ⎣ ⎦ ⎣ ⎦ ⎡ P3 = ⎢ ⎣

1.0 1.0

⎤ ⎡ 0.0000 0.2500 ⎤ ⎥ , Z3 = ⎢ −0.1813 0.0000 ⎥ , ⎦ ⎣ ⎦

⎡0.81042 −0.58586 ⎤ T3(0) = ⎢ ⎥, ⎣0.58586 0.81042 ⎦ ⎡ 0.66607 RV 3 = ⎢ ⎣ −0.77907

⎤ ⎡ 0.61941 ⎥ , RI 3 = ⎢ −0.56147 ⎦ ⎣

376.99112 ⎡ ⎢ −0.83288 −0.16611 −0.71383 0.44257 ⎢ ⎢ −0.82530 −1.22910 0.16978 ⎢ −8.38533 A 3 = ⎢ 4.47508 ⎢ −0.10000 −4.90000 ⎢ −50.0000 ⎢ ⎢ ⎣

⎤ ⎥, ⎦ ⎤ ⎥ ⎥ ⎥ ⎥ ⎥, −1000.0 ⎥ ⎥ −10000.0 ⎥ − 33.33333⎥⎦

8.3 Steps in Small-Signal Stability Analysis

515

⎡ ⎤ ⎢ −0.07633 0.80903 ⎥ ⎢ ⎥ ⎢ 0.62061 0.85849 ⎥ ⎢ ⎥ B3 = ⎢ 5.44492 −3.93616 ⎥ , ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 33.22292 2.71085 ⎥ ⎣ ⎦ ⎡ 4.87039 C3 = ⎢ ⎣ 0.45951

4.47003

−2.34342

⎤ ⎥, ⎦

−3.23141 −3.24166

⎡ −0.71964 −4.99548⎤ D3 = ⎢ ⎥. ⎣ 4.52023 0.71964 ⎦

(3) Linearized equations of the system Obviously, matrices in (8.3) (see 8.83) are " A~ ¼ AG ;

B~ ¼ ½ BG

0 ;

C~ ¼

CG

# ;

" D~ ¼

0 2

A1

0

6 6 AG ¼ 6 0 4

A2

0

0

2

D1 6 6 DG ¼ 6 0 4 0

0 D2 0

0

3

7 7 0 7; 5

A3 0

2

B1

0

0

3

6 6 BG ¼ 6 0 4

B2

7 7 0 7; 5

0

0

B3

YGG DG

YGL

YLG 2

YLL C1

0

6 6 CG ¼ 6 0 4

C2

0

0

# ; 0

3

7 7 0 7; 5

C3

3

7 7 0 7; 5

D3

33.80848 ⎡ ⎤ ⎢ −33.80848 ⎥ ⎢ ⎥ ⎢ ⎥ 1.38295 23.58377 YGG − DG = ⎢ ⎥, − 21.84220 − 1.38295 ⎢ ⎥ ⎢ 0.71964 22.06033 ⎥ ⎢ ⎥ −21.58508 −0.71964 ⎥⎦ ⎣⎢

39.30889 −1.36519 −11.60410 −1.94219 −10.51068 ⎡ 3.30738 ⎢ −39.30889 3.30738 11.60410 −1.36519 10.51068 −1.94219 ⎢ ⎢ −1.36519 −11.60410 3.81379 17.84263 −1.18760 −5.97513 ⎢ 17.84263 3.81379 11.60410 − 1.36519 − 5.97513 −1.18760 ⎢ ⎢ −1.94219 −10.51068 4.10185 16.13348 −1.28201 ⎢ 10.51068 − 1.94219 − 16.13348 4.10185 5.58824 ⎢ YLL = ⎢ −1.18760 −5.97513 2.80473 35.44561 1.61712 −13.69798 ⎢ 5.97513 −1.18760 −35.44561 2.80473 13.69798 −1.61712 ⎢ ⎢ −1.61712 −13.69798 3.74119 23.64239 −1.15509 ⎢ ⎢ 13.69798 −1.61712 −23.64239 3.74119 9.78427 ⎢ 1.28201 5.58824 1.15509 9.78427 2.43710 − − − − ⎢ ⎢ 5.58824 1.28201 9.78427 1.15509 32.15386 − − − ⎣

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ −5.58824 ⎥ ⎥ −1.28201⎥ ⎥, ⎥ ⎥ −9.78427 ⎥ ⎥ −1.15509 ⎥ ⎥ 32.15386 ⎥ 2.43710 ⎦⎥

516 8 Small-Signal Stability Analysis of Power Systems

376.99112 ⎡ ⎢−0.06244 −0.02115 0.03518 0.02485 ⎢ ⎢ 376.99112 ⎢ −0.19905 −0.07813 −0.24571 ⎢ 0.11924 ⎢ 0.20108 −0.32411 −0.52072 ⎢ 1.13669 0.44018 ⎢−0.62862 ΔE fq 2 ⎢ ⎢ ΔVR 2 ⎢ A= ⎢ −1.48647 15.66751 ΔVM 2 ⎢ 1.23883 Δδ 3 ⎢ ⎢ 0.17777 0.10937 Δω3 ⎢ 0.20654 0.16072 0.21160 ΔEq′ 3 ⎢ 0.22459 ⎢ 0.15877 −1.06680 ΔEd′ 3 ⎢ −0.96568 ΔE fq 3 ⎢ ⎢ ΔVR 3 ⎢ 4.92556 −0.73652 ΔVM 3 ⎢⎣ 0.95082

Δδ1 Δω1 Δδ 2 Δω2 ΔEq′ 2 ΔEd′ 2

4.68110

−0.11476 −0.01561 1.32599

13.92522

−0.10000 −4.90000 −50.0

−1000.0 −10000.0 −33.33333

0.03553 0.13428 0.37374

0.01775

−0.06960 −0.04154 0.93904

−0.01913

−0.21429

13.98229

11.82683

4.10804 3.08961 376.99112 −0.38432 −0.16611 −0.40080 0.14517 −0.38531 −0.62589 −0.04685 2.03248 0.43705 −4.99508

0.24764

0.07982 0.12303 −0.50807

0.03212 −0.04750 −4.61925 0.16667

0.02726

−0.02020

⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥, ⎥ ⎥ ⎥ ⎥ ⎥ 0.16978 ⎥ ⎥ −0.10000 −4.90000 −1000.0 ⎥ ⎥ −50.0 −10000.0 ⎥ −33.3333 ⎥⎦

8.3 Steps in Small-Signal Stability Analysis 517

518

8 Small-Signal Stability Analysis of Power Systems

YLG

−17.36111 ⎡ ⎤ ⎢17.36111 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ =⎢ ⎥, −16.00000 ⎢ ⎥ 16.00000 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ − 17.06485 ⎢ ⎥ ⎢ ⎥ 17.06485 ⎣ ⎦

−17.36111 ⎡ ⎢17.36111 ⎢ ⎢ YGL = ⎢ ⎢ ⎢ ⎢ ⎣⎢

⎤ ⎥ ⎥ ⎥ ⎥. ⎥ −17.06485⎥ ⎥ 17.06485 ⎦⎥

−16.00000 16.00000

From (8.5) we can obtain the state matrix A ¼ AG ½ BG

0

YGG DG YLG

YGL YLL

1

CG 0

1 ¼ AG þ BG ½YGG DG YGL Y1 LL YLG CG :

(4) Eigenvalues and eigenvectors of state matrix A All eigenvalues of A are obtained by using the QR method as l1 ¼ 53:05299;

l2 ¼ 51:80217;

l5;6 ¼ 0:75497 j12:86370; l9 ¼ 5:58205;

l4 ¼ 28:21401;

l7;8 ¼ 0:15154 j8:67125;

l10 ¼ 3:72276;

l11;12 ¼ 1:13701 j0:91540; l15 ¼ 0:04571;

l3 ¼ 30:41762;

l13;14 ¼ 0:48432 j0:657417;

l16 ¼ 0:00000:

Obviously, except the zero eigenvalue we expect, the real part of the remaining eigenvalues is negative. Hence the power system is stable in terms of small-signal stability.

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

8.4

519

Eigenvalue Problem in Small-Signal Stability Analysis

Nonlinear system stability, when the system is subject to small disturbances, can be analyzed from the stability of its linearized system as determined by the eigenvalues of state matrix A. Hence, in the following, we shall introduce the method of eigensolution analysis for a state matrix A. From the discussion above we can see that state matrix A is a real asymmetric matrix. Hence, in the following, all our discussion will be under the condition that A 2 Rnn . We denote the set of complex numbers by C, n-dimensional complex vector space (column vector) by Cn , and set of all m-row n-column complex matrices by Cmn . Operations of scalar multiplication, addition and multiplication of complex matrices are similar to those for real matrices. However, transposition of a complex matrix is taken as conjugate transposition (denoted by superscript H), i.e., C ¼ AH ) cij ¼ a^ji . Dot product of n-dimensional vector x and y is n P s ¼ xH y ¼ x^i yi . In addition, unit vector (normalized vector) under norm p is a i¼1

vector x satisfying kxkp = 1. For example, unit vectors x under 1-norm, 2-norm, and infinite norm, respectively, are 9 jxj1 ¼ jx1 j þ þ jxn j ¼ 1 > > > qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ pﬃﬃﬃﬃﬃﬃﬃﬃ = 2 2 H jxj2 ¼ jx1 j þ þ jxn j ¼ x x ¼ 1 : ð8:84Þ > > > ; jxj ¼ max jx j ¼ 1 1

1in

i

The process to convert a vector to a unit vector is called normalization.

8.4.1

Characteristics of State Matrix Given by Its Eigensolution

8.4.1.1

Eigenvalue

For a scalar l 2 C and vector v 2 Cn , if equation Av ¼ lv

ð8:85Þ

has a nonsingular solution (i.e., v 6¼ 0), l is an eigenvalue of matrix A. To calculate eigenvalues, (8.85) can be written as ðA lIÞv ¼ 0:

ð8:86Þ

A sufficient and necessary condition for existence of a nonsingular solution of the equation is detðA lIÞ ¼ 0:

ð8:87Þ

520

8 Small-Signal Stability Analysis of Power Systems

Expansion of the determinant in the above equation gives the following polynomial equation a0 þ a1 l þ þ an1 ln1 þ ð1Þn ln ¼ 0:

ð8:88Þ

It is called the characteristic equation of matrix A. The polynomial on the left side of the above equation is called the characteristic polynomial. Because the coefficient of ln is nonzero, there are a total of n roots. The set of all roots is called the spectrum and is denoted by l(A). If lðAÞ ¼ fl1 ; ; ln g, we have detðAÞ ¼ l1 l2 ln : In addition, if we define the trace of A to be trðAÞ ¼

n X

aii :

i¼1

Then trðAÞ ¼ l1 þ l2 þ þ ln , can be proved. Eigenvalues of a real asymmetric matrix can be real or complex numbers. Complex eigenvalues always appear in the form of conjugate pairs. Moreover, similar matrices have the same eigenvalues and transposition of a matrix does not change its eigenvalues.

8.4.1.2

Eigenvectors

For any eigenvalue li, any nonzero vector vi 2 Cn satisfying equation Avi ¼ li vi

i ¼ 1; 2; . . . ; n

ð8:89Þ

is called a right eigenvector of matrix A corresponding to eigenvalue li. Since it is a homogenous equation, kvi (k is a scalar) is also the solution of the equation to be a right eigenvector of matrix A corresponding to eigenvalue li. In the following (unless explicitly stated otherwise) ‘‘eigenvector’’ refers to ‘‘right eigenvector.’’ An eigenvector defines a one-dimensional subspace that remains invariable under the operation of left multiplication by matrix A. Similarly, any nonzero vector ui 2 Cn satisfying equation AT ui ¼ li ui

i ¼ 1; 2; . . . ; n

ð8:90Þ

is called a right eigenvector of matrix AT corresponding to eigenvalue li. Taking transposition on both sides of equation, we have uTi A ¼ li uTi ;

i ¼ 1; 2; . . . ; n:

ð8:91Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

521

We call row vector uTi the left eigenvector of matrix A corresponding to eigenvalue li. To express the eigensolution of matrix A clearly, we form a diagonal matrix L consisting of all eigenvalues of matrix A, a matrix XR of all right eigenvectors arranged in columns, a matrix XL of all left eigenvectors in rows. That is 9 l2 ln g > = vn : > ; T un

L ¼ diagfl1 XR ¼ ½v1 v2 X L ¼ ½ u1

u2

ð8:92Þ

These three n-dimensional square matrices are called modal matrices. Using (8.92), (8.89), and (8.91) can be expressed in the following matrix form: ) AXR ¼ XR L : XL A ¼ LXL

ð8:93Þ

Premultiplying the first equation above by XL, and postmultiplying the second by XR, we have ðXL XR ÞL ¼ LðXL XR Þ

ð8:94Þ

or lj uTi vj ¼ li uTi vj ;

i; j ¼ 1; 2; . . . ; n:

Obviously, left and right eigenvectors corresponding to different eigenvalues are orthogonal; for the same eigenvalue their product is a nonzero number that can be converted to 1 after normalization of left and right eigenvectors. That is uTi vj ¼

0 1

i 6¼ j : i¼j

ð8:95Þ

Please note that uTi vj is not the normal inner product of two vectors. The matrix form of above equation is XL XR ¼ I;

X1 L ¼ XR :

ð8:96Þ

From (8.93) and (8.96) we have X1 R AXR ¼ L:

ð8:97Þ

522

8 Small-Signal Stability Analysis of Power Systems

8.4.1.3

Free Movement of Dynamic System

From the state equation, (8.4), we can see that the rate of change of every state variable is a linear combination of all state variables. Hence due to the coupling among state variables, it is difficult to clearly see the system movement. To cancel the coupling among state variables, we introduce a new state variable vector z. Its relationship with the original state variable vector Dx is defined to be Dx ¼ XR z:

ð8:98Þ

Substituting the above equation into (8.4) and using (8.14), the state equation can be written as dz ¼ Lz: dt

ð8:99Þ

The difference from the original state equation is that L is a diagonal matrix, while A usually is not. Equation (8.99) can be expressed as n decoupled first-order differential equations dzi ¼ li z i ; dt

i ¼ 1; 2; . . . ; n:

ð8:100Þ

Its solution in the time domain is zi ðtÞ ¼ zi ð0Þeli t ;

ð8:101Þ

where initial values of zi, zi(0) can be expressed from (8.98) by uTi and Dx(0) zi ð0Þ ¼ uTi Dxð0Þ:

ð8:102Þ

Substituting (8.101) and (8.102) into the transformation of (8.98), we have the solution of the original state vector in the time domain to be Dx ¼

n X

vi zi ð0Þeli t ;

i¼1

where solution of the ith state variable in the time domain is Dxi ðtÞ ¼ vi1 z1 ð0Þel1 t þ vi2 z2 ð0Þel2 t þ þ vin zn ð0Þeln t ; ¼ 1; 2; . . . ; n;

ð8:103Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

523

where vik is the ith element of vector vk. The above equation is the time response of system free movement expressed by eigenvalues, left and right eigenvectors. Eigenvalue li represents the ith mode of the system, with corresponding time characteristic eli t . Hence, time response of system free movement is the linear combination of n system modes. Therefore, system stability is determined by the eigenvalues: (1) A real eigenvalue represents a nonoscillatory mode. A negative real eigenvalue is a decaying mode and the bigger its absolute value, the faster it decays. A positive real eigenvalue indicates nonperiodic instability. Eigenvectors, and z (0), corresponding to real eigenvalues are real valued. (2) Complex eigenvalues always appear in conjugate pairs, i.e., l ¼ s jo:

ð8:104Þ

Each pair of complex eigenvalues represents an oscillation mode. Eigenvectors, and z(0), corresponding to complex eigenvalues are complex valued. Hence ða þ jbÞeðsjoÞt þ ða jbÞeðsþjoÞt ¼ est ð2a cos ot þ 2b sin otÞ should exhibit as est sin(ot + y). Obviously, the real part of the eigenvalue describes system oscillation damping and the imaginary part gives the frequency of oscillation. A negative real part is a decaying oscillation mode and positive an increasing oscillation mode. Oscillation frequency (Hz) is f ¼

o : 2p

ð8:105Þ

Damping ratio is defined to be s z ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ : 2 s þ o2

ð8:106Þ

This determines the decay rate property of the oscillation magnitude.

8.4.2

Modal Analysis of Linear Systems

8.4.2.1

Mode and Eigenvector

From the discussion above we know that the relationship among system time response, vectors Dx and z are 9 n X > DxðtÞ ¼ XR zðtÞ ¼ ½v1 v2 vn zðtÞ ¼ vi zi ðtÞ = : ð8:107Þ i¼1 > ; T zðtÞ ¼ XL DxðtÞ ¼ ½u1 u2 un DxðtÞ

524

8 Small-Signal Stability Analysis of Power Systems

Variables Dx1, Dx2, . . ., Dxn are the original state variables depicting system dynamics. Variables z1 ; z2 ; ; zn are state variables after transformation, each of which represents a mode of the system. From the first equation of (8.107) we can see that right eigenvectors decide the form of exhibition of each mode, i.e., when a specific mode is excited, the relative activity of each state variable is described by the right eigenvector. For example, when the ith mode is excited, the kth element vki of right eigenvector vi gives the level of influence of this mode on state variable xk. Magnitude of each element in vi represents the level of activity of each of the n state variables resulting from the ith mode; while the angle of each element represents the effect of the mode on the phase shift of each state variable. From the second equation of (8.107) we can see that the left eigenvector uTi represents the way the original state variables combine to effect the ith mode. Therefore, the kth element in the right eigenvector vi measures the level of activity of state variable xk in the ith mode; while the kth element of the left eigenvector uTi weights the contribution of the exhibited activity to the ith mode.

8.4.2.2

Eigenvalue Sensitivity

Firstly we consider the sensitivity of an eigenvalue to each element akj in matrix A (the k-row, j-column element in A). Taking partial derivatives to akj on both sides of (8.89), we have @A @vi @li @vi vi þ A ¼ vi þ li : @akj @akj @akj @akj

ð8:108Þ

Premultiplying both sides of the above equation by uTi and from (8.91) and (8.95) we can obtain @li @A ¼ uTi vi : @akj @akj

ð8:109Þ

Obviously, in @A=@akj the kth-row, jth-column element is 1 and remaining elements are zero. Hence @li ¼ uki vji @akj

ð8:110Þ

where, nji is the jth element in vi and uki is the kth element in ui. Assuming a is a scalar, A(a) is an n-order square matrix with elements being akj(a) and for all k and j, akj (a) is a differentiable function, we have dAðaÞ ¼ da

dakj ðaÞ : da

ð8:111Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

525

Therefore, similarly we can find the eigenvalue sensitivity to scalar a to be @li @A ¼ uTi vi : @a @a 8.4.2.3

ð8:112Þ

Participation Factor

To determine the relationship between state variables and system modes, we establish a so-called participation matrix P by combining right and left eigenvectors to measure the level of coupling between state variables and system modes. l1 u11 n11 Dx1 6 .. 6 . 6 6 P ¼ Dxk 6 uk1 nk1 6 6 .. 4 . Dxn un1 nn1 2

.. .

li u1i n1i .. .

.. .

.. .

uki nki .. . uni nni

.. .

ln 3 u1n n1n .. 7 . 7 7 7 ukn nkn 7: 7 .. 7 . 5

ð8:113Þ

unn nnn

Element pki = uki nki in matrix P is called a participation factor [193] that measures the level of participation of the ith mode and the kth state variable Dxk with each other. The ith row of matrix P, pi, is the participation vector of the ith mode. Since nki measures the level of activity of Dxk in the ith mode and uki weights the contribution of the activity to the mode, their product pki can measure the pure participation. The product of corresponding elements in left and right eigenvectors is a dimensionless result, independent of the dimensions selected for the eigenvectors. Assuming Dx(0) = ek, i.e., Dxk (0) = 1 and Dxj 6¼ k (0) = 0, from (8.102) we have zi (0) = uki. From (8.103) we can obtain Dxk ðtÞ ¼

n X

vki uki eli t ¼

i¼1

n X

pki eli t :

ð8:114Þ

i¼1

This equation shows that the ith mode excited by initial value Dxk (0) = 1 participates in response Dxk (t) with a participation coefficient pki. That is why it is called a participation factor. For all modes or all state variables, it is easy to prove that n X i¼1

pki ¼

n X

pki ¼ 1:

ð8:115Þ

k¼1

To set t = 0 in (8.114), we can easily obtain the summation of the kth-row elements of P to be 1. Summation of the ith-column elements of matrix P is equal to uTi vi

526

8 Small-Signal Stability Analysis of Power Systems

which is 1 according to (8.95). In addition, from (8.110) we can see that participation factor pki in fact is the sensitivity of eigenvalue li to diagonal element akk of matrix A, i.e., pki ¼

@li : @akk

8.4.3

Computation of Eigenvalues

8.4.3.1

QR Method

ð8:116Þ

Among numerical methods to compute all the eigenvalues of a general matrix, the QR method is usually the first choice. It was proposed by J. G. F. Francis in 1962, and has advantages such as strong robustness and fast speed of convergence. It has been found to be the most effective method of eigensolution so far. For a given A 2 Rnn and orthogonal matrix Q0 2 Rnn , we have the following iteration: A0 ¼ QT0 AQ0 ; k ¼ 1; 2; . . . ; Ak1 ¼ Qk Rk

ðQR decompositionÞ;

ð8:117Þ

A k ¼ R k Qk ; where each Qk 2 Rnn is an orthogonal matrix and Rk 2 Rnn upper triangular matrix. By an inductive approach we have Ak ¼ ðQ0 Q1 Qk ÞT AðQ0 Q1 Qk Þ:

ð8:118Þ

Hence each Ak is similar to A. Because matrix A has complex eigenvalues, Ak will not converge to a strict ‘‘eigenvalue exposed’’ upper triangular matrix, but satisfy a computational real Schur decomposition. An upper triangular matrix with diagonal elements being 1 1 blocks or 2 2 blocks is called an upper quasi-triangular matrix. Real Schur decomposition is a real operation to reduce a matrix to an upper quasi-triangular matrix. If A 2 Rnn , there exists an orthogonal matrix Q 2 Rnn to lead to 2

R11 6 0 6 QT AQ ¼ 6 .. 4 . 0

R12 R22 .. . 0

.. .

3 R1m R2m 7 7 .. 7; . 5 Rmm

ð8:119Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

527

where Rii is either a 1 1 matrix or 2 2 matrix. If it is 1 1, the element is an eigenvalue of matrix A; if 2 2, the eigenvalues of Rii are a pair of conjugate complex eigenvalue of A. To effectively complete a Schur decomposition, we select the initial orthogonal matrix of a similarity transformation as in (8.117) Q0 to make A0 become an upper Hessenberg matrix. Doing so, the computational complexity of one iteration is reduced from O(n3) to O(n2). In an upper Hessenberg matrix, except for the sub-diagonal elements, those below the diagonal are zero. For example, in a 6 6 upper Hessenberg matrix, nonzero elements are distributed as shown below: 2

6 6 6 6 6 6 4

3 7 7 7 7: 7 7 5

This form of matrix can be obtained by performing a series of Householder transformations. Since Householder transformation is a symmetrical orthogonal similarity transformation, the upper Hessenberg matrix obtained has the same eigenvalues as the original matrix. Finally, if values of elements of A have large differences, implementation of the iterative method could result in large computational errors in eigenvalues. The level of sensitivity of eigenvalue computation to round off can be reduced by a balancing operation. Since usually errors of eigensolution from numerical computation are proportional to a Euclidean norm, the idea of the balancing operation is to make the norm of corresponding rows and columns as close as possible through similarity transformation. Thus, the total norm of the matrix is reduced without changing the eigenvalues of the matrix. Implementation of the balancing operation is to determine the diagonal matrix D through O(n2) computation such that ~ A ¼ D1 AD ¼ ½c1 ; c2 ; . . . ; cn ¼ ½r1 ; r2 ; . . . ; rn T

ð8:120Þ

with k rik1 k cik1, i = 1,2, . . ., n. Diagonal matrix D is selected to have the form D ¼ diagfbi1 ; bi2 ; . . . ; bin g, where b is the floating-point base. Thus round off in ~ computing A is avoided. After A goes through the balancing operation, computation of eigenvalues will become more accurate. 8.4.3.2

The Power Method

In practical applications, often we do not need to compute all eigenvalues of matrix A, but only that with largest modulus (often called the dominant eigenvalue). The

528

8 Small-Signal Stability Analysis of Power Systems

power method is a very effective iterative method to calculate the dominant eigenvalue. Assuming that A 2 Cnn can be diagonalized and X1 A X = diag (l1, l2, . . ., ln), where X = [x1, x2, . . ., xn], jl1 j > jl2 j jln j. For a given initial unit vector under the 2-norm vð0Þ 2 Cn , the power method generates the following series of vectors v(k) 9 > zðkÞ ¼ Avðk1Þ > = ðkÞ ðkÞ ðkÞ ; v ¼z = z 2> > ðkÞ ðkÞ H ðkÞ ; l ¼ ½v Av

k ¼ 1; 2; . . . :

ð8:121Þ

Obviously, the series of vectors in the above iteration v(k) are unit vectors under the 2-norm. Because k ! l2 distðspanfvðkÞ g; spanfx1 gÞ ¼ O l1 and

l1 lðkÞ ¼ O

k ! l2 : l 1

Obviously, only l2/l1 < 1, when k ! 1, we have lðkÞ ! l1 ;

vðkÞ ! x1 :

ð8:122Þ

The power method is of linear convergence and its applicability depends on the ratio |l2| / |l|1, which reflects the rate of convergence. After the dominant eigenvalue of A is obtained by using the power method, we can compute the remaining eigenvalues through a deflation technique. There are many deflation methods but only a few of them are numerically stable. In the following, we shall introduce a deflation method based on similarity transformation. Assuming l1 and v1 are known, we can find a Householder matrix H1 to satisfy H1 v1 = k1 e1 and k1 6¼ 0. From A1 v1 = l1 v1, we have H1 A1 ðH1 1 H1 Þv1 ¼ l1 H1 v1 . 1 Obviously, H1 A1 H1 e ¼ l e , that is, the first column of H 1 1 1 A1 H1 is l1 e1. 1 1 Denoting l1 bT1 1 A2 ¼ H1 A1 H1 ¼ ; ð8:123Þ 0 B2 where B2 is an (n 1)th-order square matrix that obviously has eigenvalues to be l2, . . ., ln. Under the condition that |l2| > |l3|, we can use power method to compute the dominant eigenvalue of B2, l2, and corresponding eigenvector, y2, where

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

529

B2 y2 = l2 y2. Assuming A2 z2 = l2 z2 and to calculate z2, assuming a a constant to be found and y an (n 1)th dimensional vector, we have ( l1 a þ bT1 y ¼ l2 a a z2 ¼ ; : y B 2 y ¼ l2 y bT y

1 Because l1 6¼ l2, we can choose y = y2, a ¼ l2 l , thus we can find z2. v2 ¼ H1 1 z2 1 is the eigenvector of A corresponding to l2. With application of the above method and Householder matrix, we have 9 k1 ¼ sgnðeT1 v1 Þjjv1 jj2 > > > T 1 > = b ¼ jjv1 jj2 jjv1 jj2 þe1 v1 : ð8:124Þ > u ¼ v1 k1 e1 > > > T T ; A2 ¼ H1 A1 H1 1 ¼ ðI buu ÞA1 ðI buu Þ

After l2 and v2 are computed, we can continue to deflate B2 to calculate the rest of the eigenvalues and eigenvectors. In theory, if eigenvalues of A are arranged according to their modulus and those with higher values can be separated, we can use the above method to compute those eigenvalues. A drawback of the deflation method is that it changes elements of the original matrix, so that any sparsity in the matrix cannot be maintained during deflation. Finally, we would like to point out that it is not so straightforward to compute the dominant eigenvalue and corresponding eigenvector by using the power method as introduced above, because we only discussed the case of a single dominant eigenvalue. In fact, l1 could be one of a set of multiple real eigenvalues or l1 and l2 could have the same modulus but are real eigenvalues with opposite sign, or l1 and l2 are a pair of conjugate complex eigenvalues. For those different cases, the power method will be slightly different. Details can be found in [187].

8.4.3.3

The Inverse Power Method

Eigenvalues of inverse matrix A1 of a nonsingular matrix A are reciprocal values of the eigenvalues of A. Hence the reciprocal of the dominant eigenvalue of A1 is the eigenvalue of A with smallest modulus. Applying the power method on A1 is called the inverse power method (or inverse iterative method) to compute the eigenvalue of the nonsingular matrix A with smallest modulus and corresponding eigenvector. For a given initial unit vector under the 2-norm, vð0Þ 2 Cn , the inverse power method generates the following iterative series 9 > AzðkÞ ¼ vðk1Þ > = ðkÞ ðkÞ ðkÞ ; k ¼ 1; 2; . . . : ð8:125Þ v ¼z = z 2> > ; lðkÞ ¼ ½vðkÞ H AvðkÞ

530

8 Small-Signal Stability Analysis of Power Systems

When k ! 1, lðkÞ !

1 ðkÞ v ! xn ln

ð8:126Þ

Another more useful form of inverse power method is to apply the power method to matrix (A tI)1, where t is a real or complex constant. For a given initial unit vector under the 2-norm, vð0Þ 2 Cn , the iterative process is as the following: 9 ðA tIÞzðkÞ ¼ vðk1Þ > > = ðkÞ ðkÞ ðkÞ ; k ¼ 1; 2; . . . : ð8:127Þ v ¼z = z 2 > > ; ðkÞ ðkÞ H ðkÞ l ¼ ½v Av When k ! 1,

9 1 1 > t þ ðkÞ ! lp = lp t l ; > ; !x

lðkÞ ! v

ðkÞ

ð8:128Þ

p

where lp is that closest to t among all eigenvalues of A and xp is the corresponding eigenvector. We need to explain (8.128) further as follows. Because eigenvalues of nonsingular matrix A tI are lj t (j = 1, 2, . . ., n), those corresponding to matrix (A t I)1 are lj 1t ðj ¼ 1; 2; . . . ; nÞ. Applying the power method to matrix (A tI)1, we obtain eigenvalue

1 lp t

with largest

modulus that means lp t with smallest modulus. Hence lp is the closest to t. Hence if we need to compute the eigenvalue of matrix A with a value closest to number t and corresponding eigenvector, we can use the inverse power method given by (8.127). Another application of the inverse power method is that with a known approximation t of an eigenvalue of matrix A, we can use the inverse power method to compute the corresponding eigenvector and improve the accuracy of computation of the eigenvalue. Using (8.127), we can apply triangular decomposition on matrix A tI, A tI ¼ LU; where L is a unit lower triangular matrix and U upper triangular. Then equation of solution becomes LUzðkÞ ¼ vðk1Þ :

8.4.4

Eigensolution of Sparse Matrix

In small-signal stability analysis, the dynamics of a power system are described by ~ ~ ~ differential-algebraic equations of (8.3). From (8.83) we can see that A, B, C, and ~ D are all sparse matrices. When we obtain matrix A from (8.5) to compute its

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

531

eigenvalues, we can find that matrix A has lost its sparsity almost completely. Since the implementation of the QR method cannot take advantage of matrix sparsity, it is not important whether A is sparse or not when we compute its eigenvalues. However, when other iterative methods, such as power method, inverse power method, and sub-space method (to be introduced later), are used to compute part of the eigenvalues of matrix A; if we can take full advantage of the sparsity of the original matrices to compute those eigenvalues directly from (8.3), computational efficiency will be greatly enhanced. For an eigenvalue of A, l, a nonzero vector v 2 Cn satisfying the following equation " # ~ ~ v v A B ¼ l ð8:129Þ ~ ~ 0 C D w is the right eigenvector of A, corresponding to this eigenvalue. The matrix on the left-hand side of above equation is called the augmented state matrix. It is not difficult to prove the above conclusion. In fact, from (8.129) we have ~ ~ w ¼ D1 Cv. Canceling w and from (8.5), we can obtain ~ ~ ~ 1 ~ ð A B D CÞv ¼ Av ¼ lv:

ð8:130Þ

Hence we can compute eigenvalues and eigenvectors of matrix A from the eigensolution of the augmented state matrix of (8.129) without destroying system sparsity. In the following, we shall introduce the sparse realization of the power method and the inverse power method. In addition, the sparse expression of eigenvalue sensitivity to scalar a will be presented. 8.4.4.1

Sparse Realization of Power Method of (8.121)

Since the relationship between z(k) and v(k1) given by the equation "

zðkÞ 0

#

" ¼

~ A ~ C

~ B ~ D

#"

vðk1Þ

#

wðk1Þ

is equivalent to z(k) = Av(k 1), computation of the first equation in (8.121) can be replaced by the following equations: 9 ~ ðk1Þ = ~ ðk1Þ Dw ¼ Cv : ~ ðk1Þ ~ðk1Þ ; zðkÞ ¼ Av þB

ð8:131Þ

Before the iteration of (8.121) is implemented, we only apply sparse triangular ~ ~ decomposition on D once, i.e., D. Hence computation of (8.131) in each iteration is

532

8 Small-Signal Stability Analysis of Power Systems

only the multiplication of some sparse matrices and vectors and solution of two triangular equations.

8.4.4.2

Sparse Realization of Inverse Power Method of (8.127)

Since the relationship between z(k) and v(k 1) given by the equation "

~ A tI ~ C

~ B ~ D

#"

zðkÞ

#

wðkÞ

" ¼

vðk1Þ

#

0

ð8:132Þ

is equivalent to (A tI)z(k) = v(k 1), solution of the first equation in (8.127) can be replaced by that of (8.132) to obtain vector z(k). ~ ~ ~ ~ ~ For a given number t, we first calculate D ¼ D C ðA tIÞ1 B and apply ~ ~ sparse triangular decomposition D ¼ LU. Noting ðDÞ is a diagonal block matrix (a diagonal block is from a dynamic component in power system), we can obtain ~ ðA tIÞ1 by calculating the inverse of diagonal block matrices directly. In ~ ~ addition, D and D have the same sparse structure (2 2 block sparse matrix). Hence solution of (8.132) can be summarized in the following steps: (1) Calculate w(k) from solution of the equation ~ ~ ~ D wðkÞ ¼ CðA tIÞ1 vðk1Þ ~ ~ (2) Calculate zðkÞ ¼ ðA tIÞ1 ðvðk1Þ BwðkÞ Þ

8.4.4.3

Eigenvalue Sensitivity to Scalar a

Similar to (8.129), for a left eigenvector we have

uT

yT

~

A ~ C

~ B ¼ l T u ~ D

0 :

ð8:133Þ

Hence using a similar derivation, we can obtain

@li T ¼ ui @a

2 ~ ~3 @A @B 7

6 6 @a @a 7 vi yTi 6 ~ 7 ~ 4 @ C @ D 5 wi @a @a

ð8:134Þ

8.4 Eigenvalue Problem in Small-Signal Stability Analysis

8.4.5

533

Application of Eigenvalue Sensitivity Analysis

In analysis of power system operation and design of power system controllers, we often need to investigate the influence of certain parameters, such as the gain and time constant of a controller, on power system stability. This will help in the selection or setting of those parameters to stabilize the power system or to improve system stability. Since system state matrix A is a function of a system parameter a, i.e., A(a), any eigenvalue of matrix A, li, is also a function of parameter a, i.e., li (a), i = 1, 2, . . ., n. When parameter a varies, li (a) will accordingly change. Variation of li (a) represents the influence of variation of parameter a on power system stability. Assuming that parameter a changes from a(0) to a(0) + D a, the corresponding change of system eigenvalue is from li (a(0)) to li (a(0) = D a). Taylor expansion of li (a(0) + D a) at a(0) is li ðað0Þ þ DaÞ ¼ li ðað0Þ Þ þ

@li ðaÞ @ 2 li ðaÞ Da þ ðDaÞ2 þ : @a a¼að0Þ @a2 a¼að0Þ

When D a is very small, change of li can be approximately expressed as Dli ðað0Þ ; DaÞ ¼ li ðað0Þ þ DaÞ li ðað0Þ Þ ¼

@li ðaÞ Da; @a a¼að0Þ

ð8:135Þ

where partial derivative ∂ li/∂ a is the first-order sensitivity of eigenvalue li to parameter a, referred to simply as eigenvalue sensitivity. Hence if we can calculate ∂ li/∂a, D a can be approximately determined from the required change of eigenvalue D li. Calculation of the first-order sensitivity of eigenvalue li to parameter a can be summarized as follows: (1) Set a = a(0) to form state matrix A(a(0)) (2) Calculate eigenvalue of A(a(0)), li, and corresponding left and right eigenvector H uH i and vi such that ui vi ¼ 1 (3) Calculate @AðaÞ @a a¼a ð0Þ (4) @li ðaÞ ¼ uH @AðaÞ vi i @a @a a¼að0Þ

In the following, we shall give an example taking the gain KS, time constant T1, T2, T3, and T4 of lead-lag network of PSS as parameter a to demonstrate the calculation of @AðaÞ @a . In the equations of generation unit g of (8.20) and (8.21), except Ag ; BIg ; BVg ; Pg , and Zg are independent of a. In addition, RIg, RVg, and Tg(0) are also independent of a. Hence from (8.30) and (8.32) we have

534

8 Small-Signal Stability Analysis of Power Systems

@Bg @Cg @Dg ¼ ¼ ¼ 0; @a @a @a @A

where matrix @ag can be calculate from matrix Ag in (8.20). Obviously in the equation of the whole system, from (8.83) we can obtain 2 @A G @a ~ @A 6 0 ¼6 @a 4 0 0

0 0 0 0

0 0 0 0

3 0 07 7; 05 0

~ @B ¼ 0; @a

~ @C ¼ 0; @a

~ @D ¼ 0; @a

where ∂AG/∂a can be calculated from (8.73). Moreover, from (8.5) and the equation above we have ~ @A @ A ¼ : @a @a The partial derivative of matrix A to other parameters can be calculated similarly. In the analysis of eigenvalue sensitivity, in addition to the eigenvalue sensitivity to parameters introduced above, eigenvalue sensitivity to power system operating conditions has been proposed. To enhance computational accuracy, second-order eigenvalue sensitivity has also been suggested, with some effective computational methods proposed. Details can be found in references [215–218].

8.5

Oscillation Analysis of Power Systems

A power system cannot operate without proper control. System operators can satisfy the predicted load demand through automatic generation control, and also through switching on, or off, various other controllable devices. Certain automatic control devices, such as the governor and AVR of a generator, HVDC control and FACTS control, etc., carry out the task of fast automatic regulation to maintain system frequency and voltage within required limits, when the power system is subject to disturbances. Since the middle of the twentieth century, the power industry has found that interconnection of regional power systems can lead to more reliable and economical operation of power systems. This has resulted in the increasing scale of modern power systems. In the 1960s, the interconnection of two Northern American power systems suffered from increasing oscillations. Power system oscillations have subsequently been reported in many countries. Investigation into power system oscillations has revealed that when regional power networks are connected through long-distance transmission lines, the resulting weak coupling of large power generation centers implies weak damping of interarea power oscillations. Another cause of reduced, or even negative, damping of power system oscillation is the application

8.5 Oscillation Analysis of Power Systems

535

of high-gain, fast-acting excitation systems. Electrical engineers have found that through the introduction of a supplementary control signal from PSS, system damping can be increased. Experience of Northern American power system interconnection has shown that application of PSS is very effective in damping power system oscillations. Increasing oscillations prevent power networks from exploiting interconnection. In some interconnected power systems, power exchange between interconnected networks has to be kept below a certain limit to avoid the occurrence of oscillations. This greatly reduces the value of interconnecting regional power networks. In some interconnected power systems, low-gain AVR have to be adopted to avoid the oscillation problem. Hence before the scheme of asynchronous interconnection via HVDC was proposed, further interconnection was abandoned in some power systems. Since the 1940s, it has been known that excitation control can enhance the stability limit of a synchronous generator. Since in some cases, excitation control can successfully improve power system dynamic performance, in addition to the control being fast and efficient, electrical engineers have held high expectations of the function of excitation control. However, effectiveness of excitation control is not unlimited. Fast-acting excitation systems can improve synchronous torque to enhance system first swing stability. However, fact-acting excitation is often a negative feedback system with high gain that has little influence on oscillation damping after the first swing. Sometimes it could provide negative damping. When a power system exhibits negative oscillation damping, fast-acting excitation control (usually with high gain) often increases the negative damping to the detriment of system operating conditions. In an m machine interconnected power system, there are a total of m 1 electromechanical oscillation modes. From field records of real power system oscillation [229] and extensive experience from power system simulation, these oscillation modes can be classified, according to the area of coverage, into two types [189], local modes and interarea modes: (1) Local modes only involve power swings of generation units in a power plant to the rest of the power system. Oscillation frequency usually is between 1 and 2 Hz. (2) Interarea modes are power swings of a group of generators in an area to another group of generators in another area. This interarea oscillation often occurs between two or more generators in a weakly connected power system. Because the moment of inertia of the equivalent generator in each area is very large, the oscillation frequency of an interarea oscillation is lower than that of local-mode oscillation, being in the range of 0.1–0.7 Hz. When the oscillation is exhibited between two groups of generators, oscillation frequency is between about 0.1 and 0.3 Hz. When it is an oscillation among multiple groups of generators, oscillation frequency is about 0.4–0.7 Hz. Since the frequency of those two types of oscillation is low, they are often called power system low-frequency oscillations. In addition to electromechanical oscillation modes, control modes and torsional oscillation modes may exist in a power

536

8 Small-Signal Stability Analysis of Power Systems

system. Torsional modes have been previously introduced. Control modes are related to various control devices installed in the power system. Since regulation of control devices is fast and controllers have small time constant, frequency of control modes is usually high. Here we are only concerned about electromechanical oscillation modes. Analysis regarding control modes and torsional modes is out of the scope of this book. Small disturbances can lead to power system low-frequency oscillations. If the oscillations of all modes are decaying, the power system is stable in terms of smallsignal stability. However, in real power system operation, usually only where the damping ratio of electromechanical oscillation modes is greater than 0.05, is the power system operation acceptable. Of course this value is not fixed. With variations of system operating conditions and small changes of oscillation modes, lower damping ratios (such as 0.03) could also be acceptable. It is apparent that small-signal instability of real power systems is mainly due to system oscillations caused by lack of damping. In 1969, Demello and Concordia [218] obtained conditions of power system small-signal stability with regard to the operation of a thyristor-controlled excitation system for a single-machine infinitebus power system. These are certain requirements on the setting of AVR gain and the introduction of an auxiliary control signal of generator rotor deviation. Their work clearly revealed the cause of power system oscillation in the single-machine infinite-bus model and laid down a solid theoretical foundation for the design of PSSs. Based on their idea and principles proposed, researchers have attempted extensions into multimachine power systems for the analysis of local-mode oscillations, and further to interarea oscillations in interconnected power systems. However, we have to point out that some of the simple extensions are often found to be inappropriate. A large-scale multimachine power system is a typical nonlinear dynamic system. Increasing oscillations caused by disturbances are dependant on many factors. Network topology and parameters, characteristics of dynamic components, system operating conditions, control strategies and parameters of various controllers all play an important role in system oscillations. It is a challenging task to clearly analyze the cause of power system electromechanical oscillations and to propose effective measure to overcome the problem. With the increasing demand of economics in a modern society, especially with the trend toward electricity markets, more and more load is required to be carried over existing power networks. However, economics and security of the power system are two conflicting requirements. When a power system operates under a light load condition before it is disturbed, damping windings of generators can provide adequate torque proportional to rotor speed. This damping can usually absorb the energy involved in system oscillations and thus the magnitude of oscillations decays continuously. The power system is stable in terms of small-signal stability. If the power system operates at heavy load conditions before it is disturbed, damping windings of generators cannot completely dissipate the energy involved in the system oscillations, so that the oscillations can grow continuously. The power system is unstable in terms of smallsignal stability. Moreover, to increase the capability of power transmission and to

8.5 Oscillation Analysis of Power Systems

537

improve system transient stability or other system performance aspects, large numbers of various types of controller are installed in the power system. Some of these may clash with the damping of system oscillations, due to improper control strategies or parameters, or mismatch among controller functions. This may again lead to unstable system oscillations. The purpose of oscillation analysis of power systems is to study key factors affecting oscillation modes, so that useful measures can be worked out to suppress oscillations effectively. [Example 8.2] In Example 8.1, all the eigenvalues of the system state matrix have been calculated. In the following, we shall study system oscillation modes. [Solution] Table 8.2 gives oscillation frequency and damping ratio of several oscillation modes; corresponding left and right eigenvectors and participation vectors are given in Tables 8.3 and 8.4. In the following we will carry out modal analysis from the results in Tables 8.3 and 8.4, where all vectors have been normalized to unit vectors under the infinite norm. Firstly, we identify electromechanical oscillation modes from the participation vector of specified modes: if the component with largest modulus in a participation vector is related to generator speed, we identify that the mode is an electromechanical oscillation mode. Then we can observe the exhibition of modes from right eigenvectors: for those components in right eigenvectors related to generator speed, a group of components with similar modulus and directional phase identifies a group of coherent generators. Incoherent generators are associated with those components with opposite phase. The right eigenvector of a local mode is dominated by variables related to one or a group of closely located generators. Components of the right eigenvector of an interarea mode evenly distribute in all regions in a power system. For oscillation mode l5,6, the element with largest modulus in its participation vector is related to Do3. Hence it is an electromechanical oscillation mode. Besides, in its right eigenvector, components associated with Do1, Do2 have small modulus (being 0.00018 and 0.00121, respectively) and similar phase (being 170.62 and 166.98 , respectively); the component associated with Do3 has large modulus (0.00411) and opposite directional phase to those above (being 10.52 ). Hence this mode will exhibit as an electromechanical oscillation between generator 1, 2 and generator 3. It is a local oscillation mode with oscillation frequency being 2.04732 Hz.

Table 8.2 Oscillation frequency and damping ratio of several oscillation modes

f x

l5,6

l7,8

l11,12

l13,14

2.04732 0.05859

1.38007 0.01747

0.14569 0.77893

0.10463 0.59313

Dd1 Do1 Dd2 Do2 DE0 q2 DE0 d2 DEfq2 DVR2 DVM2 Dd3 Do3 DE0 q3 DE0 d3 DEfq3 DVR3 DVM3

0.01137 0.33268 0.02281 0.66769 0.02124 0.00888 0.00028 0.00003 0.00198 0.03416 1.00000 0.03534 0.01302 0.00047 0.00005 0.00336

Modulus

u6

90.60 176.13 85.77 178.79 85.82 110.43 178.73 13.37 53.99 92.62 0.00 94.07 78.12 1.15 166.51 126.13

Phase

0.00532 0.00018 0.03527 0.00121 0.00156 0.00688 0.03610 0.36141 0.00184 0.12030 0.00411 0.00411 0.02094 0.09989 1.0000 0.00509

Modulus

v6

l6

96.02 170.62 99.66 166.98 162.26 36.04 72.77 76.77 88.59 82.84 10.52 12.05 152.38 4.01 0.00 165.36

Phase 0.01470 0.01470 0.19568 0.19574 0.00808 0.01485 0.00241 0.00233 0.00089 0.99933 1.0000 0.03536 0.06632 0.01132 0.01091 0.00416

Modulus

p6 5.10 5.00 3.36 3.71 101.41 135.95 116.48 52.88 45.12 0.74 0.00 71.50 118.98 13.38 155.99 57.99

Phase

Table 8.3 Left and right eigenvectors and participation vectors of oscillation modes

0.02300 1.00000 0.01871 0.81346 0.02567 0.00413 0.00049 0.00005 0.00248 0.00430 0.18692 0.00870 0.00257 0.00017 0.00002 0.00086

Modulus

u8

90.86 0.00 89.81 179.71 77.72 42.44 168.06 2.07 71.83 86.22 176.13 68.23 12.05 158.57 11.56 81.33

Phase 0.04052 0.00093 0.11332 0.00261 0.00265 0.01286 0.10048 1.00000 0.00506 0.06144 0.00141 0.00054 0.00513 0.07102 0.70681 0.00358

Modulus

v8

l8

155.40 113.60 26.92 64.08 48.13 91.43 5.93 0.00 170.13 23.80 67.20 19.43 94.98 6.82 0.89 171.02

Phase

0.43955 0.43957 0.99988 1.0000 0.03212 0.02506 0.02337 0.02253 0.00591 0.12456 0.12458 0.00222 0.00621 0.00570 0.00550 0.00144

Modulus

p8

2.47 2.61 0.52 0.00 86.62 67.22 57.78 118.28 17.92 6.19 7.29 67.42 9.19 49.18 126.88 26.52

Phase

538 8 Small-Signal Stability Analysis of Power Systems

Dd1 Do1 Dd2 Do2 DE0 q2 DE0 d2 DEfq2 DVR2 DVM2 Dd3 Do3 DE0 q3 DE0 d3 DEfq3 DVR3 DVM3

0.00383 1.00000 0.00240 0.64734 0.36458 0.15380 0.04393 0.00440 0.02583 0.00149 0.41946 0.16500 0.10380 0.02025 0.00203 0.01191

Modulus

u12

140.64 0.00 31.48 170.64 34.69 156.74 103.88 75.05 171.93 52.19 171.13 42.95 156.78 95.61 83.32 179.80

Phase

0.05071 0.00020 0.05334 0.00021 0.00812 0.00066 0.06692 1.00000 0.00489 0.05286 0.00021 0.00685 0.00099 0.05590 0.83532 0.00408

Modulus

v12

l12

85.09 133.74 73.44 145.39 178.71 13.57 40.05 0.00 178.93 75.02 143.82 177.95 11.69 39.27 0.78 178.14

Phase 0.04408 0.04458 0.02911 0.03036 0.67247 0.02317 0.66740 1.00000 0.02866 0.01782 0.01949 0.25677 0.02343 0.25702 0.38511 0.01104

Modulus

p12

150.68 151.21 33.09 31.08 140.93 68.12 141.02 0.00 82.04 52.22 47.74 149.95 70.04 150.07 9.05 72.99

Phase 0.00049 0.23071 0.00194 0.94544 0.43208 0.08490 0.09457 0.00936 0.04856 0.00194 1.00000 0.50272 0.14005 0.11208 0.01109 0.05756

Modulus

Table 8.4 Left and right eigenvectors and participation vectors of oscillation modes

u14

146.95 87.88 49.58 171.29 38.37 153.90 81.94 97.30 148.04 115.83 0.00 141.75 30.18 97.94 82.82 32.08

Phase 0.04062 0.00009 0.04627 0.00010 0.01202 0.00528 0.05319 0.48473 0.00240 0.04028 0.00009 0.02494 0.01276 0.10972 1.00000 0.00495

Modulus

v14

l14

106.90 126.72 122.47 111.15 0.10 171.90 115.53 176.05 4.71 63.91 169.72 176.31 12.20 68.42 0.00 179.24

Phase

0.00159 0.00162 0.00715 0.00756 0.41422 0.03577 0.40108 0.36171 0.00930 0.00622 0.00696 1.00000 0.14246 0.98069 0.88442 0.02273

Modulus

p14

74.61 73.39 38.33 43.00 3.91 68.77 0.97 113.31 118.19 145.17 155.72 0.00 76.95 5.04 117.38 114.12

Phase

540

8 Small-Signal Stability Analysis of Power Systems

Similarly, for l7,8, the element with largest modulus in its participation vector is related to Do2. Hence it is an electromechanical oscillation mode. In addition, in its right eigenvector, components associated with Do2, Do3 have relatively large modulus (being 0.00261 and 0.00141, respectively) and the same direction (phase being 64.08 and 67.20 , respectively); component associated with Do1 has small modulus (0.00093) and opposite direction. Hence this mode will exhibit as electromechanical oscillation between generator 1, 2 and generator 3. It is also a local oscillation mode with oscillation frequency being 1.38007 Hz. Though this mode is stable, the damping ratio (0.01747) is not sufficient, exhibiting poor dynamic performance as far as oscillation decay is concerned. For mode l11,12, the element with largest modulus in its participation vector is related to D VR2; for l13,14, element with largest modulus in its participation vector is related to D E0 q3. Hence those modes are not electromechanical oscillation modes but control modes. [Example 8.3] We take the 39-node 10-machine simplified New England system as an example to demonstrate the procedure of power system oscillation analysis [221]. In the power system, ten machines are at nodes 30–39 and the machine at node 39 is an equivalent generator. Generators at nodes 30–38 have fast static excitation systems installed. [Solution] We obtain the system linearized equation by using the methods introduced above and then compute all eigenvalues of the system state matrix. Damping of nine modes associated with electromechanical oscillations is not sufficient and some of eigenvalues have positive real parts. For two eigenvalues 0.1022 j7.215 (mode 1) and 0.037 4.301 (mode 9), components associated with generator speed in their right eigenvectors are given in Table 8.5. From Table 8.5 we can see that in the eigenvector of the first mode, there are three components with large modulus (highlighted by bold figures), among these the direction of the first component (with phase being 0 ) is opposite to that of the Table 8.5 Components associated with generator speed in right eigenvectors of mode 1 and 2 Generator number

30 31 32 33 34 35 36 37 38 39

Mode 1

Mode 9

Modulus

Phase (degree)

Modulus

Phase (degree)

1.0 0.1408 0.0797 0.1851 0.4777 0.7935 0.7797 0.3468 0.1664 0.0170

0.0 44.5 241.9 152.3 32.1 170.2 170.5 10.1 111.4 191.3

0.5574 0.4757 0.5208 0.7601 1.0 0.7961 0.7977 0.5084 0.6694 0.4052

9.9 3.4 5.5 5.3 0.0 5.7 6.8 12.4 3.3 179.6

8.5 Oscillation Analysis of Power Systems

541

other two (with phase being about 170 ). This indicates that the mode mainly exhibits as an electromechanical oscillation between generator 30 and generators 35, 36; with oscillation frequency being 7.215/2p ¼ 1.148 Hz. This is a local oscillation mode. In the eigenvector of the second mode, except for the component associated with generator 39 with relatively small modulus, other components have similar values of modulus. Moreover, the first nine components have opposite direction (with phase being about 0 ) to that of the last one (with phase being about 180 ). This indicates that this mode exhibits mainly as an electromechanical oscillation between generators 30–38 and generator 39 (the equivalent generator that can be seen as a regional network). Hence this is an interarea oscillation with oscillation frequency being 4.301/2p ¼ 0.685 Hz. In addition to identifying electromechanical oscillation modes, participation vectors can also be used to estimate the relative effects of generator controls on specified oscillation modes. For example, a component associated with rotor speed in a participation vector gives eigenvalue sensitivity to the variation of damping applied on the associated generator. If it is zero, this indicates that installation of PSS on the generator will have no impact in improving oscillation damping. If it is a large positive number, this shows that the associated generator is a good candidate place to install PSS, to effectively increase damping of the relevant oscillation mode. In Table 8.6, components associated with generator rotor speed are given. From Table 8.6 we can see that for the local mode, a component in the participation vector associated with generator 30 has the largest value, about equal to the sum of components associated with generators 35 and 36. Hence applying damping control at generator 30 should almost be equivalent to similar applications at both generators 35 and 36 simultaneously. For the interarea oscillation mode, a component associated with generator 39 has the largest modulus. However, generator 39 is an equivalent machine and damping control cannot be applied there. The sum of components associated with generators 30–38 is about equal to the component related to generator 39. This means that damping control applied at generators 30– 38 will achieve a similar effect as that applied at generator 39. Moreover, we should note that although some generators have large participation factors, there will be little effect in applying damping control on those generators if their capacity is small. Applying damping control on generators with large capacity will be more effective than applying it on those with small capacity, as far as increased oscillation damping is concerned.

Table 8.6 Components in participation vector Generator number Mode 1 Mode 2

30

31

32

33

34

35

1.0 0.17

0.01 0.09

0.005 0.12

0.02 0.22

0.13 0.33

0.42 0.26

Values in the table are estimated from graphs presented in [221]

36

37

38

39

0.43 0.21

0.07 0.07

0.02 0.18

0.001 1.0

542

8 Small-Signal Stability Analysis of Power Systems

Thinking and Problem Solving 1. What are the purpose and significance of small-signal stability analysis for electrical power systems? 2. What are the basic principle and basic procedures of small-signal stability analysis? 3. What are the main methods to solve the eigenvalues of a linearized electrical power system? What are their advantages and disadvantages? 4. Why is the QR method not suitable for the eigenvalue analysis of large-scale electrical power systems? 5. What is the critical eigenvalue? What methods are there for calculating critical eigenvalues of large-scale electrical power systems? What advantages and disadvantages are there for each method? 6. How can we apply sparse matrix techniques to critical eigenvalue calculations for large-scale electrical power systems? Can the sparse matrix technique be used in the QR method? 7. How are the eigenvalue and corresponding left and right eigen vectors used to represent the modes of a linear system? 8. What is the participation factor? Why can the participation factor be used to represent both the observability and controllability of a system? 9. What are the main causes of increasing amplitude, low-frequency oscillation? 10. What are the major manifestations of low-frequency oscillation? Why is the oscillating frequency among local generators lower than that among generators in a plant, and the oscillating frequency among regional generators is lower than that among local generators? 11. What are the main measures to control low-frequency oscillation?

References

1. W.F. Tinney, I.W. Waiker, ‘‘Direct solutions of sparse network equation by optimal ordered triangular factorization,’’ Proceedings of IEEE 55(11), 1801–1809, 1967 2. W.F. Tinney, Some examples of sparse matrix methods for power system problems, Proceedings of Power Systems Computation Conference (PSCC), Rome, June 23–27, 1969 3. W.F. Tinney, V. Brandwajn, S.M. Chen, ‘‘Sparse vector methods,’’ IEEE Transactions on Power Apparatus and Systems, 104, 295–301, 1985 4. G.W. Stagg, A.H. El-Abiad, Computer Methods in Power Systems, McGraw Hill, New York, 1968 5. R.G. Andreich, H.E. Brown, H.H. Happ, C.E. Person, ‘‘The piecewise solution of the impedance matrix load flow,’’ IEEE Transactions on Power Apparatus and Systems, 87(10), 1877– 1882, 1968 6. W.F. Tinney, C.E. Hart, ‘‘Power flow solution by Newton’s method,’’ IEEE Transactions on Power Apparatus and Systems 86(4), 1449–1460, 1967 7. W.F. Tinney, ‘‘Compensation methods for network solutions by optimal ordered triangular factorization,’’ IEEE Transactions on Power Apparatus and Systems, 91(1), 123–127, 1972 8. B. Scott, O. Alsac, ‘‘Fast decoupled load flow,’’ IEEE Transactions on Power Apparatus and Systems, 93(3), 859–869, 1974 9. R. Van Amerongen, ‘‘A general-purpose version of the fast decoupled load flow,’’ IEEE Transactions on Power Systems, 4(2), 760–770, 1989 10. A. Monticeli, O.R. Savendra, ‘‘Fast decoupled load flow: Hypothesis, derivations, and testing,’’ IEEE Transactions on Power Systems, 5(4), 1425–1431, 1990 11. L. Wang, X. Rong Li, ‘‘Robust fast decoupled power flow,’’ IEEE Transactions on Power Systems, 15(1), 208–215, 2000 12. V.M. da Costa, N. Martins, J.L. Pereira, ‘‘Developments in the Newton Raphson power flow formulation based on current injections,’’ IEEE Transactions on Power Systems, 14(4), 1320–1326, 1999 13. A. Semlyen, F. de Leon, ‘‘Quasi-Newton power flow using partial Jacobian updates,’’ IEEE Transactions on Power Systems, 16(3), 332–339, 2001 14. V.H. Quintana, N. Muller, ‘‘Studies of load flow method in polar and rectangular coordinates,’’ Electric Power System Research, 20(1), 225–235, 1991 15. R.P. Klump, T. J. Overbye, ‘‘Techniques for improving power flow convergence,’’ Proceedings of PES Summer Meeting, Seattle, USA, vol. 1, July 2000 16. K.L. Lo, Y.J. Lin, W.H. Siew, ‘‘Fuzzy-logic method for adjustment of variable parameters in load flow calculation,’’ IEE Proceedings of Generation Transmission Distribution, 146(3), 276–282, 1999 17. W.L. Chan, A.T.P. So, L.L. Lai, ‘‘Initial applications of complex artificial neural networks to load-flow analysis,’’ IEE Proceedings of Generation Transmission Distribution, 147(6), 361–366, 2000

543

544

References

18. T. Nguyen, ‘‘Neural network load-flow,’’ IEE Proceedings of Generation Transmission Distribution, 142(1), 51–58, 1995 19. K.P. Wong, A. Li, T.M.Y. Law, ‘‘Advanced constrained genetic algorithm load flow method,’’ Proceedings of Generation Transmission Distribution, 146(6), 609–616, 1999 20. P.K. Mannava, L. Teeslink, A.R. Hasan, ‘‘Evaluation of efficiency of parallelization of power flow algorithms,’’ Proceedings of the 40th Midwest Symposium on Circuit and Systems, Sacramento, California, USA, August 1997, pp. 127–130 21. N. balu, T. Bertram, A. Bose, et al, ‘‘On-Line power system security analysis,’’ Proceedings of the IEEE, 80, 262–282, 1992 22. J. Carpentier, ‘‘Static security assessment and control: A short survey,’’ IEEE/NTUA Athens Power Tech Conference on Planning, Operation and Control of Today’s Electric Power Systems’’, Athens, Greece, September 5–8, 1993, pp. 1–9 23. B. Stott, O. Alsac, F.L. Alvarado, ‘‘Analytical and computational improvement in performance index raking algorithm for networks,’’ International Journal of Electrical Power and Energy Systems, 7(3), 154–160, 1985 24. T.A. Mikolinas, B.F. Wollenberg, ‘‘An advanced contingency selection algorithm,’’ IEEE Transactions on Power Apparatus and Systems, 100(2), 608–617, 1981 25. V. Brandwajn, ‘‘Efficient bounding method for linear contingency analysis,’’ IEEE Transactions on Power Systems, 3(2), 726–733, 1988 26. R. Bacher, W.F. Tinney, ‘‘Faster local power flow solutions: The zero mismatch approach,’’ IEEE Transactions on Power Systems, 4(4), 1345–1354, 1989 27. W.P. Luan, K.L. Lo, Y.X. Yu, ‘‘NN-based pattern recognition technique for power system security assessment,’’ Proceedings of the International Conference on Electric Utility Deregulation and Restructuring and Power Technologies, London, April 2000 28. Task Force on Probabilistic Aspects of Reliability Criteria of the IEEE PES Reliability, Risk and Probability Applications Subcommittee, ‘‘Probabilistic security assessment for power system operations’’, IEEE PES General Meeting, Denver, USA, June 6–10, 2004 29. A.M. Leite da Silva, L.C. Resende, L.A.F. Manso, et al, ‘‘Well-being analysis for composite generation and transmission systems’’, IEEE Transactions on Power Systems, 19(4), 1763–1770, 2004 30. R. Billinton, R.N. Allan, Reliability evaluation of Large Electric Power Systems, Kluwer, Boston, 1988 31. R. Billinton, R. Allan, Reliability Evaluation of Power Systems, Plenum Press, New York, 1996 32. X. Wang, J. McDonald, Modern Power System Planning, McGraw-Hill, London, 1994, pp. 108–110 33. R. Billinton, W. Li, Reliability Evaluation of Electric Power Systems using Monte Carlo Methods, Plenum Press, New York, 1994 34. A.M. Leite de Silva, S.M.P. Ribeiro, V.L. Arienti, et al., ‘‘Probabilistic load flow techniques applied to power system expansion planning’’, IEEE Transactions on Power Systems, 5(4), 1047–1053, 1990 35. P. Zhang, S.T. Lee, ‘‘Probabilistic load flow computation using the method of combined cumulants and Gram-Charlier expansion’’, IEEE Transactions on Power Systems, 19(1), 676–682, 2004 36. Z. Hu, X. Wang, ‘‘A probabilistic load flow method considering branch outages,’’ IEEE Transactions on Power Systems, 21(2), 507–514, 2006 37. K. Maurice, S. Alan, The Advanced Theory of Statistics, vol. 1, Macmillan, USA, 1977 38. J.C. Spall, ‘‘Estimation via Markov chain Monte Carlo,’’ IEEE Control System Magazine, 23 (2), 35–45, 2003 39. R. Chen, J.S. Liu, X. Wang, ‘‘Convergence analyses and comparisons of Markov chain Monte Carlo algorithms in digital communications,’’ IEEE Transactions on Signal Processing, 50(2), 255–270, 2002

References

545

40. Reliability Test System Task Force, ‘‘IEEE reliability test system,’’ IEEE Transactions on Power System, 14(3), 1010–1020, 1999 41. R. Von Mises, Mathematical Theory of Probability and Statistics, Academic Press, New York, 1964 42. P.A. Jensen, J.W. Barnes, Network-Flow Programming, Wiley, New York, 1980 43. Y.K. Lin, ‘‘Reliability of a stochastic flow network with unreliable branches and nodes, under budget constraints,’’ IEEE Transactions on Reliability, 53(3), 381–387, 2004 44. D. Maagee, A. Refsum, ‘‘RESIN, A desktop-computer program for finding cut set’’, IEEE Transactions on Reliability, 30(5), 407–410, 1981 45. K. Kobayashi, H. Yamamoto, ‘‘A new algorithm in enumerating all minimal paths in a sparse network’’, Reliability Engineering and System Safety, 65(1), 11–15, 1999 46. S. Ross, Introduction to Probability Models, Academic Press, NewYork, 2006 47. X. Wang, C. Pottle, ‘‘A concise frequency and duration approach to generating system reliability studies’’, IEEE Transactions on Power Amplifier Symposium, 102(8), 2521–2530, 1983 48. F.A. Rahimi, A. Vojdani, ‘‘Meet the emerging transmission market segments’’, IEEE Computer Application in Power, 12(1), 26–32, 1999 49. F.C. Schweppe, M.C. Caramanis, R.D. Tabors, et al, Spot Pricing of Electricity, Kluwer, Boston, 1988 50. J. Carpentier, ‘‘Contribution a’1’etude du dispatching economique,’’ Bulletin de la Societe Francaise des Electricients, 3, 431–447, 1962 51. H.W. Dommel, W.F. Tinney, ‘‘Optimal power flow solutions,’’ IEEE Transactions on Power Apparatus and Systems, 87(12), 1866–1876, 1968 52. A.M. Sasson, ‘‘Combined use of the parallel and fletcher-powell non-linear programming methods for optimal load flows,’’ IEEE Transactions on Power Apparatus and Systems, 88 (10), 1530–1537, 1969 53. A.M. Sasson, ‘‘Decomposition technique applied to the non-linear programming load flow method,’’ IEEE Transactions on Power Apparatus and Systems, 89(1), 78–82, 1970 54. R. Divi, H.K. Kesavan, ‘‘A shifted penalty function approach for optimal power flow,’’ IEEE Transactions on Power Apparatus and Systems, 101(9), 3502–3512, 1982 55. S.N. Talukdar, T.C. Giras, V.K. Kalyan, ‘‘Decompositions for optimal power flows,’’ IEEE Transactions on Power Apparatus and Systems, 102(12), 3877–3884, 1983 56. G.F. Reid, L. Hasdorf, ‘‘Economic dispatch using quadratic programming,’’ IEEE Transactions on Power Apparatus and Systems, 92, 2015–2023, 1973 57. R.C. Burchett, H.H. Happ, D.R. Vierath, ‘‘Quadratically convergent optimal power flow,’’ IEEE Transactions on Power Apparatus and Systems, 103, 3267–3276, 1984 58. D.W. Wells, ‘‘Method for economic secure loading of a power systems,’’ Proceedings of IEEE, 115(8), 606–614, 1968 59. C.M. Shen, M.A. Laughton, ‘‘Power system load scheduling with security constraints using dual linear programming,’’ Proceedings of IEEE, 117(1), 2117–2127, 1970 60. N. Nabona, L.L. Ferris, ‘‘Optimization of economic dispatch through quadratic and linear programming,’’ Proceedings of IEEE, 120(5), 1973 61. Z. Yan, N.D. Xiang, B.M. Zhang, et al, ‘‘A hybrid decoupled approach to optimal power flow,’’ IEEE on Power Systems, 11(2), 947–954, 1996 62. K.R. Frish, ‘‘Principles of linear programming: The double gradient form of the logarithmic potential method,’’ Memorandum, Institute of Economics, University of Oslo, Oslo, Norway, 1954 63. P. Huard, ‘‘Resolution of mathematical programming with nonlinear constraints by the method of centers,’’ Nonlinear Programming, 209–219, 1967 64. I.I. Dikin, ‘‘Iterative solution of problems of linear and quadratic programming,’’ Soviet Mathematics, 8, 674–675, 1967 65. N. Karmarkar, ‘‘A new polynomial-time algorithm for linear programming,’’ Combinatorica, 4, 373–395, 1984

546

References

66. P.E. Gill, W. Murray, M.A. Saunders, et al, ‘‘On the projected newton barrier methods for linear programming and an equivalence to karmarkar’s projective method,’’ Mathematical Programming, 36, 183–209, 1986 67. X. Guan, W.H. Edwin Liu, A.D. Papalexopoulos, ‘‘Application of a fuzzy set method in an optimal power flow,’’ Electric Power Systems Research, 34(1), 11–18, 1995 68. Y.T. Hsiao, C.C. Liu, H.D. Chiang, et al, ‘‘A new approach for optimal VAR sources planning in large scale electric power systems,’’ IEEE Transactions on Power Systems, 8(3), 988–996, 1993 69. K.R. Frisch, ‘‘The Logarithmic Potential Method for Convex Programming,’’ Memorandum, Institute of Economics, University of Oslo, Norway, May, 1955 70. A.V. Fiacco, G.P. MoCormic, Nonlinear Programming: Sequential Unconstrained Minimization Techniques, Wiley, New York, 1968 71. H. Wei, H. Sasaki, J. Kubokawa, et al, ‘‘An interior point nonlinear programming for optimal power flow problems with a novel data structure,’’ IEEE Transactions on Power Systems, 13 (3), 870–877, 1998 72. F.C. Schweppe, M.C. Caramanis, R.D. Tabors, et al, ‘‘Spot Pricing of Electricity,’’ Kluwer, Boston, 1988 73. M.C. Carmanis, R.E. Bohn, F.C. Schweppe, ‘‘Optimal spot pricing: Practice and theory,’’ IEEE Transactions on Power Apparatus and Systems, 101(9), 3234–3245, 1982 74. M.C. Carmanis, R.E. Bohn, F.C. Schweppe, ‘‘The costs of wheeling and optimal wheeling rates,’’ IEEE Transactions on Power Systems, 1(1), 63–73, 1986 75. D. Ray, F. Alvarado, ‘‘Use of an engineering model for economic analysis in the electricity utility industry,’’ The Advanced Workshop on Regulation and Public Utility Economics, Rutgers University, New Jersey, May 25–27, 1988 76. M.L. Baughman, S.N. Siddiqi, ‘‘Real time pricing of reactive power: theory and case study result,’’ IEEE Transactions on Power Systems, 6(1), 23–29, 1991 77. S.N. Siddiqi, M.L. Baughman, ‘‘Reliability differentiated pricing of spinning reserve,’’ IEEE Transactions on Power Systems, 10(3), 1211–1218, 1993 78. A. Zobian, M.D. llic, ‘‘Unbundling of transmission and ancillary services,’’ IEEE Transactions on Power systems, 12(2), 539–558, 1997 79. M.L. Baughman, S.N. Siddiqi, J.M. Zarnikau, ‘‘Advanced pricing in electrical system. Part 1: Theory,’’ IEEE Transactions on Power Systems, 12(1), 489–495, 1997 80. M.L. Baughman, S.N. Siddiqi, J.M. Zarnikau, ‘‘Advanced pricing in electrical system. Part 2: Implication,’’ IEEE Transactions on Power Systems, 12(1), 496–502, 1997 81. K. Xie, Y.H. Song, J. Stonham, et al, ‘‘Decomposition model and interior point methods for optimal spot pricing of electricity in deregulation environments,’’ IEEE Transactions on Power Systems, 15(1), 39–50, 2000 82. C.N. Yu, M.D. Ilic, ‘‘An algorithm for implementing transmission rights in competitive power industry,’’ IEEE Power Engineering Society Winter Meeting, 3, 1708–1714, 2000 83. X. Wang, Y.H. Song, Q. Lu, et al, ‘‘Series FACTS devices in financial transmission rights auction for congestion management,’’ IEEE Power Engineering Review, 21(11), 41–44, 2001 84. R. Bacher, H. Glavitsch, ‘‘Loss reduction by network switching,’’ IEEE Transactions on Power Systems, 3(2), 447–454, 1988 85. R. Baldick, E. Kahn, ‘‘Contract paths, phase shifters and efficient electricity trade,’’ IEEE Power Engineering Society Winter Meeting, 2, 968–974, 2000 86. S.Y. Ge, T.S. Chung, Y.K. Wong, ‘‘A new method to incorporate FACTS devices in optimal power flow,’’ Proceedings of International Conference on Energy Management and Power Delivery, 1, 122–271, 1998 87. X. Wang, Y.H. Song, Q. Lu, ‘‘Primal-dual interior point linear programming optimal power flow for real-time congestion management,’’ IEEE Power Engineering Society Winter Meeting, 3, 1643–1649, 2000 88. G. Hamoud, ‘‘Assessment of available transfer capability of transmission system,’’ IEEE Transactions on Power System, 15(1), 27–32, 2000

References

547

89. X. Luo, A.D. Patton, C. Singh, ‘‘Real power transfer capability calculations using multi-layer feed-forward neutral networks,’’ IEEE Transactions on Power Systems, 15(2), 903–908, 2000 90. M. Pavella, D. Ruiz-Vega, J. Giri, et al, ‘‘An integrated scheme for on-line static and transient stability constrained ATC calculations,’’ IEEE Power Engineering Society Summer Meeting, 1, 273–276, 1999 91. D.S. Kirschen, R.N. Allan, G. Strbac, ‘‘Contributions of individual generators to loads and flows’’, IEEE Transactions on Power Systems, 12(1), 52–60, 1997 92. X. Wang, X. Wang, ‘‘On current trace problem’’, Science in China (E), 30(3), 405–412, 2000 93. J. Bialek, ‘‘Topological generation and load distribution factors for supplement cost allocation in transmission open access’’, IEEE Transactions on Power Systems, 12(3), 1185–1193, 1997 94. D.S. Kirschen, G. Strbac, ‘‘Tracing active and reactive power between generators and loads using real and imaginary currents,’’ IEEE Transactions on Power Systems, 14(4), 1312–1319, 1999 95. Reliability Test System Task Force, ‘‘IEEE reliability test system-1996,’’ IEEE Transactions on Power Systems, 14(3), 1010–1020, 1999 96. Federal Energy Regulatory Commission, ‘‘Open access same-time information system (Formerly Real-time Information Networks) and standards of conduct,’’ Docket no. RM95-9-000, Order 889, 1996 97. North American Electric Reliability Council, ‘‘Available transfer capability definition and determination’’, NERC Planning Standards, June 1996 98. G. Hamoud, ‘‘Assessment of available transfer capability of transmission system,’’ IEEE Transactions on Power Systems, 15(1), 27–32, 2000 99. G.C. Ejebe, J.G. Waight, M. Sanots-Nieto, et al, ‘‘Available transfer capability calculations,’’ IEEE Transactions on Power Systems, 13(4), 1521–1527, 1998 100. X. Luo, A.D. Patton, C. Singh, ‘‘Real power transfer capability calculations using multi-layer feed-forward neural networks,’’ IEEE Transactions on Power Systems, 15(2), 903–908, 2000 101. G.C. Ejebe, J.G. Waight, M.S. Nieto, W.F. Tinney, ‘‘Fast calculation of linear available transfer capability,’’ IEEE Transactions on Power Systems, 15(3), 1112–1116, 2000 102. M. Shaaban, Y. Ni, F. Wu, ‘‘Total transfer capability calculations for competitive power networks using genetic algorithms,’’ Proceedings of International Conference on DRPT, City University, London, April 4–7, 2000 103. A.R. Vojdani, ‘‘Computing available transmission capability using trace,’’ EPRI Power System Planning and Operation News, 1, 1, 1995 104. Y. Xiao, Y.H. Song, ‘‘Available transfer capability (ATC) evaluation by stochastic programming,’’ IEEE Power Engineering Review, 20(9), 50–52, 2000 105. F. Xia, A.P.S. Meliopoulos, ‘‘A methodology for probabilistic simultaneous transfer capability analysis,’’ IEEE Transactions on Power Systems, 11(3), 1269–1278, 1996 106. A.B. Rodrigues, M.G. Da Silva, ‘‘Solution of simultaneous transfer capability problem by means of Monte Carlo simulation and primal-dual interior-point method,’’ Proceedings of PowerCon International Conference, 2, 1047–1052, 2000 107. X.F. Wang, C.J. Cao, Z.C. Zhou, Experiment on fractional frequency transmission system, IEEE Transactions on Power Systems, 21(1), 372–377, 2006 108. N.G. Higorani, ‘‘Power electronics in electric utilities: Role of power electronics in future power systems,’’ Proceedings of IEEE, 76(4), 481–482, 1988 109. L. Gyugyi, ‘‘Dynamic compensation of AC transmission lines by solid-state synchronous voltage source,’’ IEEE Transactions on Power Delivery, 9(2), 904–911, 1994 110. A.A. Edris, R. Aapa, M.H. Baker, et al, ‘‘Proposed terms and definitions for flexible AC transmission system (FACTS),’’ IEEE Transactions on Power Delivery, 12(4), 1848–1853, 1997 111. D.A. Braunagel, L.A. Kraft, J.L. Whysong, ‘‘Inclusion of DC converter and transmission equation directly in a Newton power flow,’’ IEEE Transactions on Power Apparatus and Systems, 95(1), 76–88, 1976

548

References

112. J. Arrillaga, P. Bodger, ‘‘Integration of HVDC links with fast decoupled load flow solutions,’’ Proceedings of IEE, 124(5), 463–468, 1977 113. J. Arrillaga, B. Smith, AC-DC Power System Analysis, The Institute of Electrical Engineers, UK, 1998 114. J. Reeve, G. Fahmy, B. Stott, ‘‘Versatile load flow method for multiterminal HVDC systems,’’ IEEE Transactions on Power Apparatus and Systems, 96(3), 925–932, 1977 115. H. Fudeh, C.M. Ong, ‘‘A simple and efficient AC-DC load flow method for miltiterminal DC systems,’’ IEEE Transactions on Power Apparatus and Systems, 100(11), 4389–4396, 1981 116. J. Arrillaga, C.P. Arnold, B.J. Harker, Computer Modeling of Electrical Power Systems, Wiley, New York, 1983 117. T. Smed, G. Andersson, G.B. Sheble´, L.L. Grigsby, ‘‘A new approach to AC/DC power flow,’’ IEEE Transactions on Power Systems, 6(3), 1238–1244, 1991 118. G.D. Breuer, J.F. Luini, C.C. Young, ‘‘Studies of large AC/DC systems on the digital computer,’’ IEEE Transactions on Power Apparatus and Systems, 85(11), 1107–1115, 1966 119. J.F. Clifford, A.H. Schmidt, ‘‘Digital representation of a DC transmission system and its control,’’ IEEE Transactions on Power Apparatus and Systems, 89(1), 97–105, 1970 120. N. Sato, N.V. David, S.M. Chan, A.L. Burn, J.J. Vithayathil, ‘‘Multiterminal HVDC system representation in a transient stability program,’’ IEEE Transactions on Power Apparatus and Systems, 99(5), 1927–1936, 1980 121. Working Group 38–01, Task Force no. 2 on SVC, CIGRE Report, Static Var Compensators, Ed. by I.A. Erimnez, CIGRE, UK, 1986 122. IEEE Special Stability Controls Working Group, ‘‘Static var compensator models for power flow and dynamic performance simulation’’, IEEE Transactions on Power Systems, 9(1), 229–240, 1994 123. L. Gyugyi, N.G. Hinggorani, P.R. Nannery, et al, ‘‘Advanced static var compensator using gate turn-off thyristors for utilities applications,’’ CIGRE Session, 1990, pp. 23–203 124. Y. Sumi, Y. Harumoto, T. Hasegawa, et al, ‘‘New static var control using force-commutated inverters,’’ IEEE Transactions on Power Apparatus and Systems, 100(9), 4216–4224, 1981 125. C.W. Edwards, P.R. Nannery, ‘‘Advanced static var generator employing GTO thyristors,’’ IEEE Transactions on Power Delivery, 3(4), 1622–1627, 1988 126. C. Schauder, M. Gernhardt, E. Stacey, et al, ‘‘Development of a 100Mvar static condensor for voltage control of transmission systems,’’ IEEE Transactions on Power Delivery, 10(3), 1486–1496, 1995 127. E.V. Larsen, K. Clark, S.A. Miske Jr., J. Urbanek, ‘‘Characteristics and rating consideration of thyristor controlled series compensation,’’ IEEE Transactions on Power Delivery, 9(2), 992–1000, 1994 128. G.G. Karady, T.H. Ortmeyer, B.R. Pilvelait, et al, ‘‘Continuously regulated series capacitor,’’ IEEE Transactions on Power Delivery, 8(3), 1348–1355, 1993 129. L. Gyugyi, C.D. Schauder, K.K. Sen, ‘‘Static synchronous series compensator: A solid-state approach to the series compensation of transmission lines,’’ IEEE Transactions on Power Delivery, 12(1), 406–417, 1997 130. Y.H. Song, A.T. Johns, Flexible AC transmission Systems (FACTS), IEE Press, London, 1999 131. S. Nyati, M. Eitzmann, J. Kappenmann, et al, ‘‘Design issues for a single core transformer thyristor controlled phase-angle regulator,’’ IEEE Transactions on Power Delivery, 10(4), 2013–2019, 1995 132. L. Gyugyi, ‘‘A unified power flow control concept for flexible AC transmission system,’’ Fifth International Conference on AC and DC Power Transmission, London, Sept’17–20, 1991 133. A. Nabavi-Niaki, M.R. Iravani, ‘‘Steady-state and dynamic models of unified power flow controller (UPFC) for power system studies’’, IEEE Transactions on Power Systems, 11(4), 1937–1943, 1996 134. Z.X. Han, ‘‘Phase shift and power flow control’’, IEEE Transactions on Power Apparatus and Systems, 101(10), 3790–3795, 1982

References

549

135. D.J. Gotham, G.T. Heydt, ‘‘Power flow control and studies for systems with FACTS devices,’’ IEEE Transactions on Power Systems, 13(1), 60–65, 1998 136. C.R. Fuerte-Esquivel, E. Acha, H. Ambriz-Prez, ‘‘A thyristor controlled series compensator model for the power flow solution of practical power networks,’’ IEEE Transactions on Power Systems, 15(1), 58–64, 2000 137. C.R. Fuerte-Esquivel, E. Acha, H. Ambriz-Prez, ‘‘A comprehensive Newton-Raphson UPFC for the quadratic power flow solution of practical power networks,’’ IEEE Transactions on Power Systems, 15(1), 102–109, 2000 138. H. Ambriz-Prez, E. Acha, C.R. Fuerte-Esquivel, ‘‘Advanced SVC models for NewtonRaphson load flow and Newton optimal power flow studies,’’ IEEE Transactions on Power Systems, 15(1), 129–946, 2000 139. C.R. Fuerte-Esquivel, E. Acha, ‘‘A Newton-type algorithm for the controlpower flow in electrical power networks,’’ IEEE Transactions on Power Systems, 12(4), 1474–1480, 1997 140. C.R. Fuerte-Esquivel, E. Acha, ‘‘Newton-Raphson algorithm for the reliable solution of large power networks with embedded FACTS devices,’’ IEE Proceedings of Generation, Transmission and Distribution, 143(5), 447–454, 1996 141. S. Arabi, P. Kundur, ‘‘A versatile FACTS device model for power flow and stability simulations,’’ IEEE Transactions on Power Systems, 11(4), 1944–1950, 1996 142. C.R. Fuerte-Esquivel, E. Acha, ‘‘United power flow controller: A critical comparison of Newton-Raphson UPFC algorithm in power flow studies,’’ IEE Proceedings of Generation, Transmission and Distribution, 144(5), 437–444, 1997 143. J.Y. Liu, Y.H. Song, ‘‘Strategies for handling UPFC constraints in steady-state power flow and voltage control,’’ IEEE Transactions on Power Systems, 15(2), 566–571, 2000 144. W. Fang, H.W. Ngan, ‘‘Control setting of unified power flow controllers through a robust load flow calculation’’, Proceedings of Generation, Transmission and Distribution, 146(4), 365–369, 1999 145. H. Sun, D.C. Yu, C. Luo, ‘‘A novel method of power flow analysis with Unified Power Flow Controller (UPFC),’’ IEEE Power Engineering Society Winter Meeting, 4, 2800–2805, 2000 146. A. Blondel, ‘‘The two-reaction method for study of oscillatory phenomena in coupled alternators’’, Revue Ge´ne´rale de Lelectricite´, 13, 235–251, February 1923; 515–531, March 1923 147. R.E. Doherty, C.A. Nickle, ‘‘Synchronous machines I and II’’, AIEE Transactions, 45, 912–942, 1926 148. R.H. Park, ‘‘Two-reaction theory of synchronous machines: Generalized method of analysis Part I’’, AIEE Transactions, 48, 716–727, 1929; Part II, 52, 352–355, 1933 149. C. Concordia, Synchronous Machine, Wiley, New york, 1951 150. G. Shackshaft, P.B. Henser, ‘‘Model of generator saturation for use in power system studies’’, Proceedings of IEE, 126(8), 759–763, 1979 151. G.R. Slemon, Magnetoelectric Devices, Wiley, New York, 1966 152. A.E. Fitzgerald, C. Kingsley, Electric Machinery, 2nd Edn., McGraw-Hill, New York, 1961 153. P. Kunder, Power System Stability and Control, McGraw-Hill, New York, 1994 154. D.W. Olive, ‘‘Digital simulation of synchronous machine transients’’, IEEE Transactions on Power Apparatus and Systems, 87(8), 1968 155. M.K. El-Sherbiny, A.M. El-Serafi, ‘‘Analysis of dynamic performance of saturated machine and analog simulation’’, IEEE Transactions on Power Apparatus and Systems, 101(7), 1899– 1906, 1982 156. D.W. Olive, ‘‘New techniques for the calculation of dynamic stability’’, IEEE Transactions on Power Apparatus and Systems, 85(7), 767–777, 1966 157. T.J. Hammons, D.J. Winning, ‘‘Comparisons of synchronous machine models in the study of the transient behaviour of electrical power systems’’, Proceedings of IEE, 118, 1442–1458, 1971 158. J. Arrillage, C.P. Arnold, B.J. Harker, Computer Modeling of Electrical Power Systems, Wiley, Chichester, 1983

550

References

159. IEEE Committee Report, ‘‘First benchmark model for computer simulation of subsynchronous resonance’’, IEEE Transactions on Power Apparatus and Systems, 96(5), 1565–1572, 1977 160. IEEE Power Engineering Society, IEEE Recommended Practice for Excitation System Models for Power System Stability Studies, IEEE Standards Board, New York, 1992 161. IEEE Working Group Report, ‘‘Hydraulic turbine and turbine control models for system dynamic studies’’, IEEE Transactions on Power Systems, PWRS-7(1), 167–179, 1992 162. D.G. Ramey, J.W. Skooglund, ‘‘Detailed hydrogovernor representation for system stability studies’’, IEEE Transactions on Power Apparatus and Systems, 89, 106–112, 1970 163. M. Leum, ‘‘The development and field experience of a transistor eletric governor for hydro turbines’’, IEEE Transactions on Power Apparatus and Systems, 85, 393–400, 1966 164. IEEE Working Group Report, ‘‘Dynamic models for fossil fueled steam units inpower system studies’’, IEEE Transactions on Power Systems, PWRS-6(2), 753–761, 1991 165. IEEE Committee Report, ‘‘Dynamic models for steam and hydro turbines in power system studies’’, IEEE Transactions on Power Apparatus and Systems, 92(6), 1904–1915, 1973 166. P. Kundur, D.C. Lee, J.P. Bayne, ‘‘Impact of turbine generator overspeed controls on unit performance under system disturbance conditions’’, IEEE Transactions on Power Apparatus and Systems, 104, 1262–1267, 1985 167. M.S. Baldwin, D.P. McFadden, ‘‘Power systems performance as affected by turbine-generator controls response during frequency disturbance’’, IEEE Transactions on Power Apparatus and Systems, 100, 2846–2494, 1981 168. IEEE Task Force on Load Representation for Dynamic Performance, ‘‘Standard load models for power flow and dynamic performance simulation’’, IEEE Transactions on Power Systems, 10(3), 1302–1313, 1995 169. IEEE Task Force on Load Representation for Dynamic Performance, ‘‘Load representation for dynamic performance analysis’’, IEEE Transactions on Power Systems, 8(2), 472–482, 1993 170. IEEE Task Force on Load Representation for Dynamic Performance System Dynamic Performance Subcommittee, Power System Engineering Committee, ‘‘Bibliography on load model for power flow and dynamic performance simulation’’, IEEE Transactions on Power Systems, 10(1), 523–538, 1995 171. T. Dovan, T.S. Dillon, C.S. Berger, K.E. Forward,‘‘A microcomputer based on-line identification approach to power system dynamic load modelling’’, IEEE Transactions on Power Systems, 2(3), 529–536, 1987 172. C.W. Talor, ‘‘Concepts of under-voltage load shedding for voltage stability’’, IEEE Transactions on Power Delivery, 7(2), 480–488, 1982 173. W.S. Kao, C.J. Lin, C.T. Huang, et al, ‘‘Comparison of simulated power system dynamics applying various load models with actual recorded data’’, IEEE Power Engineering Society Winter Meeting, 172–177, 1993 174. W.S. Kao, ‘‘The effect of load models on unstable low-frequency oscillation damping in taipower system experience w/wo power system stabilizers’’, IEEE Transactions on Power Systems, 16(3), 463–472, 2001 175. A. Borghetti, R. Caldon, A. Mari, et al, ‘‘On dynamic load models for voltage stability studies’’, IEEE Transactions on Power Systems, 12(1), 293–303, 1997 176. F. Nozari, M.D. Kankam, W.W. Price, ‘‘Aggregation of induction motors for transient stability load modeling’’, IEEE Transactions on Power systems, 2(4), 1096–1103, 1987 177. P. Kunder, Power System Stability and Control, McGraw-Hill, New York, 1994 178. P.M. Anderson, A.A. Fouad, Power System Control and Stability, The Iowa State University Press, Iowa, 1977 179. J. Arrillaga, C.P. Arnold, Computer Analysis of Power Systems, Wiley, New York, 1990 180. Y.H. Song, A.T. Johns, Flexible AC Transmission Systems (FACTS), The Institution of Electrical Engineers, London, 1999

References

551

181. J.A. Momoh, M.E. El-Hawary, Electric Systems, Dynamics, and Stability with Artificial Intelligence Applications, Marcel Dekker, New York, 2000 182. W.L. Brogan, Modern Control Theory, Prentice Hall, New Jersey, 1991 183. J.J.E. Slotine, W. Li, Applied Nonlinear Control, Prentice Hall, New Jersey, 1991 184. C.W. Gear, Numerical Initial Value Problems in Ordinary Differential Equations, Prentice Hall, New Jersey, 1971 185. L. Lapidus, J.H. Seinfeld, Numerical Solution of Ordinary Differential Equations, Academic Press, New York, 1971 186. J.D. Lambert, Computational Methods in Ordinary Differential Equations, Wiley, New York, 1973 187. J.H. Wilkinson, The Algebraic Eigenvalue Problem, Clarendon Press, Oxford, 1965 188. G.H. Golub, C.F. Van Loan, Matrix Computations, 3rd Edn., The Johns Hopkins University Press, 1996 189. G. Rogers, Power System Oscillations, Kluwer, Dordecht, 2000 190. P. Kundur, G.J. Rogers, D.Y. Wong, et al, ‘‘A comprehensive computer program package for small signal stability analysis of power systems’’, IEEE Transactions on Power Systems, 5 (4), 1076–1083, 1990 191. S. Aribi, G.J. Rogers, D.Y. Wong, et al, ‘‘Small signal stability analysis of SVC and HVDC in AC power systems’’, IEEE Transactions on Power Systems, 6(3), 1147–1153, 1991 192. I.J. Perez-Arriaga, G.C. Verghese, F.C. Schweppe, ‘‘Selective modal analysis with applications to electric power systems, Part I: Heuristic introduction, Part II: The dynamic stability problem’’, IEEE Transactions on Power Apparatus and Systems, 101(9), 3117–3134, 1982 193. J.L. Sancha, I.J. Perez-Arriaga, ‘‘Selective modal analysis of electric power system oscillatory instability’’, IEEE Transactions on Power Systems, 3(2), 429–438, 1988 194. R.T. Byerly, R.J. Bennon, D.E. Sherman, ‘‘Eigenvalue analysis of synchronizing power flow oscillations in large electric power systems’’, IEEE Transactions on Power Apparatus and Systems, 101(1), 235–243, 1982 195. N. Martins, ‘‘Efficient eigenvalue and frequency response methods applied to power system small-signal stability studies’’, IEEE Transactions on Power Systems, 1(1), 217–226, 1986 196. D.Y. Wong, G.J. Rogers, B. Porretta, P. Kundur, ‘‘Eigenvalue analysis of very large power systems’’, IEEE Transactions on Power Systems, 3(2), 472–480, 1988 197. P.W. Sauer, C. Rajagopalan, M.P. Pai, ‘‘An explanation and generalization of the AESOPS and PEALS algorithms’’, IEEE Transactions on Power Systems, 6(1), 293–299, 1991 198. N. Uchida, T. Nagao, ‘‘A new Eigen-analysis method of steady-state stability studies for large power systems: S matrix method’’, IEEE Transactions on Power Systems, 3(2), 706– 714, 1988 199. W.J. Stewart, A. Jennings, ‘‘A simultaneous iteration algorithm for real matrices’’, ACM Transactions on Mathematical Software, 7(2), 184–198, 1981 200. S. Duff, J.A. Scott, ‘‘Computing selected eigenvalues of sparse unsymmetric matrices using subspace iteration’’, ACM Transactions on Mathematical Software, 19(2), 137–159, 1993 201. J.A. Scott, ‘‘An Arnoldi code for computing selected Eigenvalues of sparse, real, unsymmetric matrices’’, ACM Transactions on Mathematical Software, 21(4), 432–475, 1995 202. A. Semlyen, L. Wang, ‘‘Sequential computation of the complete eigensystem for the study zone in small signal stability analysis of large power systems’’, IEEE Transactions on Power Systems, 3(2), 715–725, 1988 203. L. Wang, A. Semlyen, ‘‘Application of sparse eigenvalue techniques to the small signal stability analysis of large power systems’’, IEEE Transactions on Power Systems, 5(4), 635–642, 1990 204. D.J. Stadnicki, J.E. Van Ness, ‘‘Invariant subspace method for eigenvalue computation’’, IEEE Transactions on Power Systems, 8(2), 572–580, 1993 205. N. Mori, J. Kanno, S. Tsuzuki, ‘‘A sparsed–oriented techniques for power system small signal stability analysis with a precondition conjugate residual method’’, IEEE Transactions on Power Systems, 8(3), 1150–1158, 1993

552

References

206. G. Angelidis, A. Semlyen, ‘‘Efficient calculation of critical eigenvalue clusters in the small signal stability analysis of large power systems’’, IEEE Transactions on Power Systems, 10 (1), 427–432, 1995 207. L.T.G. Lima, L.H. Bezerra, C. Tomei, N. Martins, ‘‘New methods for fast small-signal stability assessment of large scale power systems’’, IEEE Transactions on Power Systems, 10(4), 1979–1985, 1995 208. G. Angelidis, A. Semlyen, ‘‘Improved methodologies for the calculation of critical eigenvalues in small signal stability analysis’’, IEEE Transactions on Power Systems, 11(3), 1209–1217, 1996 209. J.M. Campagnolo, N. Martins, D.M. Falcao, ‘‘Refactored bi-Iteration: A high performance eigensolution method for large power system matrices’’, IEEE Transactions on Power Systems, 11(3), 1228–1235, 1996 210. N. Martins, L.T.G. Lima, H.J.C.P. Pinto, ‘‘Computing dominant poles of power system transfer functions’’, IEEE Transactions on Power Systems, 11(1), 162–1170, 1996 211. N. Martins, ‘‘The dominant pole spectrum eigenslover’’, IEEE Transactions on Power Systems, 12(1), 245–254, 1997 212. J.M. Campagnolo, N. Martins, J.L.R. Pereira, et al, ‘‘Fast small-signal stability assessment using parallel processing’’, IEEE Transactions on Power Systems, 9(2), 949–956, 1994 213. J.M. Campagnolo, N. Martins, D.M. Falcao, ‘‘An efficient and robust eigenvalue method for small-signal stability assessment in parallel computers’’, IEEE Transactions on Power Systems, 10(1), 506–511, 1995 214. V. Ajjarapu, ‘‘Reducibility and eigenvalue sensitivity for identifying critical generations in multimachine power systems’’, IEEE Transactions on Power Systems, 5(3), 712–719, 1990 215. T. Smed, ‘‘Feasible eigenvalue sensitivity for large power systems’’, IEEE Transactions on Power Systems, 8(2), 555–563, 1993 216. H.K. Nam, Y.K. Kim, ‘‘A new eigen-sensitivity theory of augmented matrix and its applications to power system stability’’, IEEE Transactions on Power Systems, 15(1), 363–369, 2000 217. K.W. Wang, C.Y. Chung, ‘‘Multimachine eigenvalue sensitivities of power system parameters’’, IEEE Transactions on Power Systems, 15(2), 741–747, 2000 218. F.P. Demello, C. Concordia, ‘‘Concepts of synchronous machine stability as affected by excitation control’’, IEEE Transactions on Power Apparatus and Systems, 88(4), 316–329, 1969 219. P. Kunder, D.C. Lee, H.M. Zein-el-din, ‘‘Power system stabilizers for thermal units: Analytical techniques and on-site validation’’, IEEE Transactions on Power Apparatus and Systems, 100(1), 81–89, 1981 220. M. Klein, G.J. Rogers, P. Kundur, ‘‘A fundamental study of inter-area oscillations in power systems’’, IEEE Transactions on Power Systems, 6(3), 914–921, 1991 221. G. Rogers, ‘‘Demystifying power system oscillations’’, IEEE Computer Application in Power, 9(3), 30–35, 1996 222. J. Hauer, D. Trudnowski, G. Rogers, et al, ‘‘Keeping an eye on power system dynamics’’, IEEE Computer Application in Power, 10(4), 50–54, 1997 223. G. Rogers, ‘‘Power system structure and oscillations’’, IEEE Computer Application in Power, 12(2), 14,16,18,20,21, 1999 224. S.K. Starrett, A.A. Fouad, ‘‘Nonlinear measures of mode-machine participation’’, IEEE Transactions on Power Systems, 13(2), 389–394, 1998 225. F.L. Pagola, I.J.P. Arriaga, G.C. Verghese, ‘‘On sensitivity, residues and participations: applications to oscillatory stability analysis and control’’, IEEE Transactions on Power Systems, 4(1), 278–285, 1989 226. M. Klein, G.J. Rogers, S. Moorty, P. Kundur, ‘‘Analytical investigation of factors influencing power system stabilizers performance’’, IEEE Transactions on Energy Conversion, 7(3), 382–390, 1992

References

553

227. P. Kundur, M. Klein, G.J. Rogers, M.S. Zywno, ‘‘Application of power system stabilizers for enhancement of overall system stability’’, IEEE Transactions on Power Systems, 4(2), 614–626, 1989 228. L. Xu, S. Ahmed-Zaid, ‘‘Tuning of power system controllers using symbolic eigensensitivity analysis and linear programming’’, IEEE Transactions on Power Systems, 10(1), 314–322, 1995 229. J.F. Hauer, F. Vakili, ‘‘A oscillation detector used in the bpa power system disturbance monitor’’, IEEE Transactions on Power Systems, 5(1), 74–79, 1990

Index

AC exciter, 365, 369, 371–374, 379–380 Alternate solution method, 425–427 Ancillary services, 193, 195, 221, 224 Artificial intelligence methods, 201, 202 ASVG, 308–312 Asymptotically stable, 490, 492, 509 ATC1, 248–250, 252–253 ATC2, 248–250 Augmented matrix, 23–26, 67, 93, 98 Automatic generation control (AGC), 76, 250, 251, 406, 534 Automatic voltage regulator (AVR), 333, 352, 357, 363, 375, 377, 378, 534–536 Available transfer capabilities (ATC), 196, 222–224, 241–249, 253 Axiomatic definitions of probability, 131 Bayes’ formula, 132 Binomial distribution, 136, 244, 245 Bipolar system, 258 Blackout, 114, 129, 152, 178, 179, 191 Branch addition method, 56, 57, 64, 65 BX algorithm, 106 Cascading outages, 179–180 CBM, 242 Characteristic equation, 520 Characteristic polynomial, 520 Classical model, 351, 352, 384, 399, 425, 434, 454, 457, 459, 463, 511 Common mode failures, 114 Compensation method, 38, 113–119, 460 Conditional probability, 131, 139, 158 Congestion management, 196, 218, 222–223 Constant eq0 model, 351, 352

Contingency ranking, 123–124, 127 Continuation power flow (CPF), 243 Continuous random variable, 132–137, 162, 173, 244 Convergence characteristic, 73, 89, 106, 112, 216 Convergence condition, 89, 104, 109, 112 Convergence of the Monte Carlo simulation, 146 Convergence property, 100, 112, 113 Converter, 258–262, 264–270, 272, 273, 276–286, 289, 294–301, 326, 329, 331, 392, 479, 480, 505, 507 Converter basic equations, 261, 267, 278, 282, 283, 285, 289 Converter bridge, 259, 260, 280 Converter control, 260, 279, 285, 289, 297 Converter equivalent circuits, 273–276 Converter transformer, 259, 261, 267, 276–286, 294, 296–298 Convolution of random variable, 135 Coordinate transformation, 338, 432, 448, 494, 496, 499 Correction equations, 81, 83, 87, 89, 94, 103–106, 109 Correction equations of fast decoupled method, 104–107 Critically stable, 490, 492 Cumulant method, 162 Cumulative probability, 141–144, 154, 188 Current decomposition, 229, 230, 234 Damping ratio, 523, 536, 537, 540 Damping winding, 335–336, 351, 352, 354, 356, 370, 536

555

556

d axis open-circuit sub-transient time constant, 346, 347, 349, 350, 359, 465–467, 494 d axis open-circuit transient time constant, 346 d axis subtransient reactance, 344–351, 356, 359, 360, 363, 432, 465–467, 475, 485, 494 d axis synchronous reactance, 344–351, 356, 358, 359, 401, 431, 432, 453, 465 d axis transient reactance, 344, 345, 347–352, 356, 359, 370, 374, 401, 432–434, 448, 453, 454, 465–467, 475, 494 DC exciter, 365–367, 369, 370, 378, 379, 494 DC load flow, 114, 119–127, 150, 154 DC network equations, 285, 286, 297 Decaying mode, 523 Decremental bidding prices, 224 Difference equations, 419–423, 464, 466, 470, 472, 474, 475, 483, 484, 487 Differential-algebraic equations, 407, 425, 427–429, 454, 472, 492, 530 Discrete random variable, 132–135, 163, 173 Distribution factor, 229, 230, 235–238, 240, 241, 243 Distribution function, 132–135, 137, 138, 141, 149, 166, 172, 173, 178, 245 Dominant eigenvalue, 527–529 dq0 coordinate system, 335 dq0 transformation, 338 Dynamic load model, 397–403 Dynamic ordering scheme, 46–48 Edgeworth series, 162, 166, 167 Effect of saturation, 357 Eigensolution, 489, 492, 493, 510, 519, 521, 526, 527, 530, 531 Eigensolution analysis, 489, 492, 519 Eigenvalue sensitivity, 524–525, 531–534, 541 Eigenvalue sensitivity analysis, 533–534 Electromagnetic torque, 361–363 Electromechanical oscillations, 492, 493, 536, 540

Index

Energy management system (EMS), 243, 457 Enumeration method, 159, 183, 244 Equilibrium point, 490–492, 509 Equivalent circuits of transformer, 9–11 ETC, 242 Euler method, 409, 412–417, 419, 421, 422, 424, 425, 447, 450, 452 Excitation system, 334, 363–365, 375–381, 407, 430, 446, 463, 464, 469, 472, 473, 494, 495, 498, 511, 535, 536, 540 Expected energy not supplied (EENS), 152, 155, 156, 159 Extended OPF problems, 224 Factor table, 27–31, 34–37, 39, 40, 42–44, 105, 108–111, 114–116, 118, 292, 295 FACTS, 9, 222, 224, 255–258, 260, 301–302, 325, 407, 436, 463, 464, 475, 502, 510, 534 Fast decoupled method, 72, 77, 88, 101–113, 118, 120 Feasible solution, 182–189, 199, 202, 256 Field winding, 335, 336, 350–352, 359, 363, 364, 369–371, 398, 400 Financial transmission right (FTR), 222 Firing angle, 260, 262, 264–269, 271–274, 279, 280, 299–302, 304, 305, 314, 316, 318, 372, 374, 476 Flat start, 89, 289 Fractional frequency transmission system (FFTS), 255 Gauss elimination, 22–27, 31, 43, 44, 89, 93, 94, 434, 447, 448, 450 Genetic algorithm (GA), 73, 201, 243 Gibbs sampler, 157–159 Governing system, 361, 362, 381, 382, 386–389, 391–393, 407, 430, 446, 463, 464, 472, 473, 475, 496, 498 Gram-Charlier series expansion, 162, 166, 170, 173 Harmonic voltages, 259, 260 Hermite polynomial, 167, 168 Hessenberg matrix, 527 Homopolar connection, 258 Householder matrix, 528, 529

Index

HVDC, 255, 256, 258–261, 272, 276, 279, 281, 299, 300, 341, 372, 430, 436, 437, 464, 475, 479, 485–487, 493, 503, 510, 534, 535 HVDC dynamic mathematical models, 299–301 Hydraulic turbine, 381–388, 406, 496 Ideal synchronous generator, 336–338 Implicit integration methods, 419–424 Incidence matrix, 4–7, 9, 122, 247 Incremental bidding price, 224 Independent operator (ISO), 193–196, 222, 223, 226, 227 Independent power producers (IPPs), 228, 235 Initial condition, 143, 157, 234, 315, 354, 408 Initial value problem, 407, 420, 421, 425 Interarea oscillation, 535, 536, 541 Interior point method (IPM), 200–202, 207, 208, 216, 219 Inverse power method, 529–532 Inverter, 259, 260, 266, 272–274, 279, 280, 284–286, 289, 299, 301, 308–310, 319, 320, 326, 437, 438, 481, 483, 484, 503, 504, 506, 507 Jacobean, 72 Jacobean matrix, 72 Lagrangian function, 187, 204 Lagrangian multipliers, 202, 207, 216, 221 Linear optimal excitation controller (LOEC), 364 Linear programming, 151, 198–200, 223, 243 Load flow, 71–127, 130, 150, 151, 154, 155, 161–162, 168–173, 175, 176, 178, 189 Load model, 140, 141, 394–397, 399, 400, 403, 500, 501 Local-mode oscillation, 535, 536 Location marginal price (LMP), 222 Loop current method, 2 Loss allocation, 229, 231–235, 238, 240 Loss of load probability (LOLP), 145, 151, 155, 159–161

557

Lower boundary point, 186, 188 Lyapunov linearized method, 489–491 Market clearing price (MCP), 195 Markov chain, 139, 156, 157, 159, 160 Markov chain Monte Carlo (MCMC) simulation, 156–159 Markov process, 138–140 Mathematical expectation, 134, 173 Mean value, 134, 137 Memory requirement, 71, 72, 89, 149 Midterm and long-term stability, 405 Mixed programming, 198, 200 Modal analysis, 523, 537 Model of load curtailment, 150–151, 153 Modified Euler’s method, 412–417, 421, 425, 447, 450, 452 Monopolar system, 258 Monte Carlo simulation, 130, 145–149, 152, 153, 155, 157, 159, 244–248, 250 MSC, 306 Multi-fold outages, 114 Multiple bridge, 276 Multistep or multivalue algorithms, 411 Mutual admittance, 3, 4, 14, 19, 21 Mutual inductance, 337, 338, 340, 343, 357, 366 Newton-Raphson method, 22, 72, 76, 79, 81, 87–90, 93, 100, 172, 173, 175, 198 N-l checking, 123–124 Nodal admittance matrix, 1, 4, 7, 13–22, 48–50, 72, 104, 247 Nodal impedance matrix, 1, 48–64, 72, 122, 123 Nodal self-admittance, 3, 4 Node power equations, 76–78, 284 Node voltage method, 2 Nondamping winding model, 351, 352 Nonlinear algebraic equation, 71, 426 Nonlinear optimal excitation controller (NOEC), 364 Nonlinear programming, 196, 198–199 Nonoscillatory mode, 523 Non-periodic instability, 523 Normal distribution, 137–138, 140–141, 146, 167, 168, 244–246, 250

558

Numeral characteristics of random variable, 133–135 Numerical stability, 421, 424, 425, 430, 462 One-step transition probability, 139, 140 Open access same-time information system (OASIS), 243–244 Operation risk, 129 Optimal ordering schemes, 43 Optimal power flow (OPF), 1, 38, 196–203, 207, 209, 216–224, 226, 227, 243, 257 Ordinary differential equations, 299, 407, 408, 424 Orthogonal matrix, 499, 526, 527 Oscillation analysis, 493, 534, 537, 540 Outage analysis, 122, 123, 127 Outage table, 142–144, 150, 188 Parallel computing algorithms, 73 Park’s transformation, 335, 338–340 Participation factor, 525–526, 541 Participation matrix, 525 Per unit equations, 282, 340–341, 343, 401 Phase shifting transformer, 1, 11, 13, 16, 18, 20, 38, 301, 322, 325 Piecewise solution method, 72 Polar form of the nodal power equations, 77 Potier voltage, 357, 360 Power exchange (PX), 193–195, 252, 309, 535 Power flow tracing, 196, 228–241 Power market, 114, 129, 193–196, 222–224, 241, 242, 248, 253, 256 Power method, 527–532 Power rectifier, 371, 373, 379, 380 Power system stabilizers (PSS), 364, 365, 377, 407, 468, 469, 471, 495, 533, 535, 541 P-Q decoupled method, 72, 73, 107, 113, 290, 291, 294 PQ nodes, 76, 85, 90, 109, 111 Predictor-corrector method, 462 Prime mover, 142, 360–362, 381, 382, 406, 407, 430, 446–448, 463, 466, 472, 473, 496 Probabilistic load flow, 130, 161–178 Probabilistic model of load, 140–141

Index

Probabilistic models of transformer and generator, 142 Probability density function, 133–135, 137, 138, 171, 177–179 Probability of stochastic events, 130–132 Proportional-integral-differential (PID), 364, 388, 392 Pseudo-random numbers, 149 12-pulse converter, 277, 278 PV nodes, 76, 83, 85, 88, 105, 109, 110 q axis sub-transient reactance, 344 q axis synchronous reactance, 340, 344 q axis transient reactance, 344, 356 QR method, 511, 518, 526–527, 531 Quadratic programming, 198, 199 Quasi-steady state model, 480–484, 493 Random process, 138, 139, 159 Random variable, 130, 132–139, 143, 145–148, 151, 162–164, 166–168, 170–173, 244, 245 Random variable’s moment, 163 Real-time balancing market (RBM), 223, 224 Rectangular form of the nodal power equations, 78 Recursive formula, 143, 144, 410, 411, 419–421 Reliability of transmission system, 188–191 Right eigenvector, 520, 521, 523–525, 531, 533, 537–540 Rotor motion equation, 352, 360–362, 398–400, 464 Round-off error, 412, 424 Round rotor generator, 335, 336, 338, 340 Runge-Kutta method, 417–419, 421, 425, 464 Salient-pole generator, 335, 336, 340, 351 Sampling, 145, 147–150, 152, 155, 157–160, 253, 254 Saturation effect, 336, 337, 359, 366, 369 Saturation factor, 355, 366, 369, 370, 379 Schur decomposition, 526, 527 Self inductance, 337, 338, 340, 365, 366 Semi-dynamic ordering Scheme, 45, 46 Sensitivity method, 114, 246

Index

Simplified models for transient stability, 446 Simultaneous solution method, 425, 427 Single-step algorithm, 411 Slack node, 76, 83, 85, 90, 94, 174, 216, 247, 250 Small-signal stability, 489–493, 500, 506–518, 530, 536 Smoothing reactor, 259, 260, 264, 480 Sparse vector method, 22, 38–43, 460, 461 Sparsity techniques, 1 Spot pricing, 195, 196, 219–221 SPWM, 326 SSSC, 301, 302, 319–322, 326, 329, 331 STATCOM, 301, 302, 308–312, 319, 326, 329, 331 State matrix, 492, 509–511, 518, 519, 533 State space, 139, 145, 150 Static load model, 394–397, 399, 400, 500 Static ordering scheme, 44–46 Static security analysis, 73, 113–114, 189 Stationary exciter, 374, 375 Steady-state equations, 354, 356, 360, 437, 438 Steam turbine, 362, 381, 382, 389–393, 492, 493 Step-by-step integration, 408, 409 Step size, 143, 199, 408, 409, 411, 412, 420, 421, 423–426, 428, 454, 462–464, 466, 467 Stiff differential equations, 425 Stiff rotor, 360–362 Stochastic programming, 244 Stormer and Numerov integration formula, 462 Sub-synchronous oscillation, 493 Sub-transient parameter, 335, 344 SVR, 301, 302 Symmetric, 13, 16, 19, 32, 34, 38, 51, 102, 104, 105, 125, 261, 299, 308, 314 Synchronous generator, 334–340, 343, 344, 347–363, 365, 369–372, 398, 400–402, 493–496, 498, 500, 535 Synthesized impedance matrix, 440, 441, 444, 446, 486 System contingency, 124 System performance index, 124, 126, 127

559

Taylor series, 79, 81, 84, 86, 168, 169 TCPST, 301, 302, 322, 323, 325, 329 TCR. See Thyristor controlled reactors TCSC, 301, 302, 313, 314, 316–319, 322, 325, 436, 437, 464, 477–479, 485, 487, 502, 503, 507, 508, 510 Telegen’s theorem, 124 Three rotor winding model, 351 Three-winding transformer, 10, 11 Thyristor controlled reactors, 302–307, 313, 316, 476, 478 Thyristor switched capacitors, 302 Time domain, 522 Torsional oscillations, 365, 492, 493 Torsional torque, 493 Transformation matrix, 170, 335, 340 Transient parameter, 335, 344 Transient stability, 357, 362, 400, 405–407, 425, 427–431, 435, 444, 446, 447, 450, 453, 457, 459, 463, 464, 480, 484, 537 Transmission open access, 193, 221 Transmission right, 222, 224 Trapezoidal rule, 419–421, 425–427, 464, 466, 470, 471, 474, 477–479, 482, 483 Triangular decomposition, 27–34, 530–532 TRM, 242 TSC, 302, 303, 305, 306 TTC, 224, 242, 243 Two winding model, 351–352 Uniform distribution, 136–137, 148–150, 154, 158, 251 UPFC, 301, 302, 325–331 Usage sharing problem, 234 Variance, 134–137, 140, 141, 145–147, 155, 157, 159, 160, 168, 245, 246, 394 Voltage regulators (AVR), 333, 352, 357, 363, 375, 377, 378, 534–536 XB algorithm, 106

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close