0% found this document useful (0 votes)
152 views224 pages

Watson Thesis

Uploaded by

Shahriar Ahmed
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
152 views224 pages

Watson Thesis

Uploaded by

Shahriar Ahmed
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 224

Fr que cy- ep ndent A.C.

System
Equival nts fo nic Studi s
an
Transie Convertor Simulatio

A thesis
presented for the degree of
Doctor of Philosophy in Electrical Engineering
in the
University of. Canterbury,
New Zealand

by

N.R. Watson, B.E.(Hons)

1987
i.

TABLE OF CONTENTS

Page

List of Figures v
List of Tables x
List of Principal Symbols xi
Abbreviations xii
Abstract xiv
Acknowledgements xv

CHAPTER 1 INTRODUCTION

1.1 Importance of HVDC Transmission


1.2 The Need for Computer Modelling 2
1.3 Types of Studies 2
1.4 Incorporating HVDC Convertors 3
1.5 The need for Equivalent Circuits 4
1.6 Thesis Outline 5

CHAPTER 2 OVERVIEW

2.1 Computational Aids for Dynamic Simulation 7


2.2 Digital Computer Simulation 9
2.3 Development of HVDC Convertor Models 10

CHAPTER 3 MATHEMATICAL MODEL AND COMPUTER IMPLEMENTATION

3.1 Introduction 13
3.2 Representation of Components 13
3.2.1 Static Convertor 13
3.2.1.1 Modes of operation 14
3.2.1.2 Convertor control systems 14
3.2.1.3 Types of control schemes 15
3.2.2 Synchronous Machines 16
3.2.3 Transmission Lines 17
3.2.4 Transformers 17
3.2.5 Static Shunt Elements 18
3.2.6 Static Series Elements 18
3.2.7 A.C. System 19
ii.

Page

3.3 Method of Analysis 20


3.3.1 Trapeziodal Integration 20
3.3.2 Topology Changes 21
3.3.3 Choice of State Variables 22
3.3.4 Per Unit System 23
3.3.5 Initial Conditions 24
3.3.6 Network Equations 25
3.4 Extensions 29
3.4.1 C-type Filter 29
3.4.2 Harmonic Current Source 29
3.4.3 RLC Networks connected between phases 29

CHAPTER 4 ACREP: A FLEXIBLE PROGRAM FOR PROCESSING AND PLOTTING


POWER SYSTEM DATA

4. 1 Capability of the ACREP program 35


4.2 Objectives, Structure and Operation of ACREP 37
4.3 Modular Design 38
4.4 Special Features 40
4.5 Illustrative Examples 41
4.5.1 Spectral Analysis 41
4.5.2 Harmonics Graphics package 43
4.5.3 Frequency Response of an RLC Network 50
4.5.4 General Plot Facility 54

CHAPTER 5 A.C. SYSTEM MODEL

5.1 Historical Review 55


5.2 Obtaining the frequency response of an a.c. system 58
5.3 Synthesis of Frequency-Dependent Equivalent 63
5.3.1 Direct Method 67
Feature Extraction 71
The Scaling Process 74
Correction Filters 83
Derivation of correction branch parameters 83
Matching of correction branches 85
Illustrative Examples 85
iii.

Page

5.3.2 Optimization Method 91


5.3.2.1 Choice of Objective Function 91
5.3.2.2 Optimization Methods 93
5.4 The Philosophy of the Frequency-Dependent a.c. System
Models 97
5.5 Implementation of Frequency-Dependent a.c. System Models 103
5.6 Initialization of Frequency-dependent equivalents 105
5.7 Validation of the Computer Model 106
5.8 Extension to Multi-convertor systems 111
5.9 Frequency-dependent d.c. system model 112
5.10 Negative Resistance Region in Impedance Loci 117
5.11 Discussion and Conclusions 119
5.11.1 Explicit versus Implicit Mutual Representation 119
5.11.2 Comparison between direct and Optimization
techniques for forming frequency-dependent
equivalent circuits. 119
5.11.3 Impedance Magnitude and Phase Angle Match 120
5.11.4 Negative Resistive Component of Mutual Coupling. 120
5.11.5 Dynamic Instabilities. 122
5.11.6 Future Work 123

CHAPTER 6 HARMONIC ASSESSMENT

6.1 Introduction 124


6.2 Harmonic Penetration 124
6.3 Characteristic Harmonics 126
6.4 Incorporation of HVDC Convertors 127
6.5 Illustrative Examples 131
6.6 Computational Efficiency 140
6.7 Harmonic Penetration Example 142
6.8 Conclusions 145
6.8 Future Work 146

CHAPTER 7 DYNAMIC SIMULATIONS

7.1 Introduction 147


7.2 D.C. Line Fault Simulation 148
7.3 Rectifier Side A.C. System Disturbance 150
iv.

Page

7.4 Invertor Side A.C. System Disturbance 155


7.5 Discussion and Conclusions 158

CHAPTER 8 CONCLUSIONS 160

REFERENCES 164

APPENDIX A1 Detailed Synchronous Machine Coefficient Matrix 179


A2 Transmission Line Representation 180
A3 Parameters for RLC circuit analysis example 183
A4 Single Variable Optimization Methods 184
A5 Published Papers 188
v._

LIST OF FIGURES

Figure Page

2.1 Power System Analysis Overview 12

3.1 Steady State characteristics and operating point 15


3.2 Coupled Coil Transformer Model 17
3.3 Static Shunt Elements 18
3.4 A.C. System Equivalents 19
3.5 The fundamental Branches 26
3.6 RLC Networks connected between phases 30
3.7 Separate iterative loop algorithm;
Extra Network inner loop 32
3.8 Separate iterative loop algorithm;
Standard TCS circuit inner loop 33
3.9 Embedded iterative loop approach 34

4. 1 Over-view of the ACREP program 39


4.2 Histogram of current harmonics 42
4.3 Reconstructed current waveform 42
4.4 Histogram of earthquake recording 43
4.5 Continuous version of the earthquake recording 43
4.6 Isometric Plot 44
4.7 Two-dimensional perspective plot 44
4.8 Busbar voltage harmonics throughout the Lower South Island of
New Zealand when one per unit of each harmonic is injected. 45
4.9 Third harmonic Cross-section in isolation 46
4.10 Third harmonic Cross-section 46
4. 11 Multiple cross-sections 47
4.12 Cross-section of the harmonics at a busbar 47
4.13 Point Plot of Current Harmonics
(a) Phase "An 48
(b) Sequence Components 48
4.14 Branch Current Harmonics 49
4.15 Impedance Loci 49
4.16 Impedance versus frequency plot 49
4.17 Ladder Network 50
vi.

Figure Page

4.18 General branches


(a) Series branch 50
(b) Filter branch 50
4.19 Frequently used models 51
4.20 RLC circuit 52
4.21 Impedance loci of RLC circuit 52
4.22 Frequency Response of RLC circuit
(a) Impedance versus frequency plot 53
(c) Resistance versus frequency plot 53
(d) Reactance versus frequency plot 53
4.23 General Plot facility display of six waveforms simultaneously 54

5.1 A.C. System Equivalent proposed by Hingorani and Burbery 56


5.2 Equivalent proposed by Bowles 57
5.3 Data flow and sequence for a Dynamic Study using a frequency-
matched a.c. system equivalent 60
5.4 The Lower South Island of New Zealand test system 61
5.5 Impedance Loci Matrix of the test system 62
5.6 Diagonalized Matrix Impedance Loci 62
5.7 Impedance versus Frequency Matrix derived from figure 5.5 64
5.8 Structure of a frequency-matched a.c. system equivalent 65
5.9 Flow Diagram for the Direct and Optimization Algorithms 66
5.10 Impedance Loci of the equivalent based on intersection with
X axis feature extraction 71
5.11 Impedance Loci of the equivalent based on new feature
extraction 72
5. 12 Equivalent cct based on new feature extraction 73
5.13 Response of modified equivalent cct 73
5.14 Impedance response when the scale factor is optimized
(a) Impedance Loci 75
(b) Impedance versus Frequency 75
5.15 Objective function in conjunction with intersection with
X axis feature extraction 76
5.16 Objective function in conjunction with new feature extraction 76
5.17 Impedance Response
(a) Equivalent based on intersection with x axis feature
extraction 77
vii.

Figure Page

(b) Equivalent based on new feature extraction 77


5.18 Impedance Response
(a) Equivalent based on intersection with x axis feature
extraction and scale factor optimized 78
(b) Equivalent based on new feature extraction and scale factor
optimized 78
5.19 Frequency reponse of equivalent when the scale factor is
optimized from 50 to 1250 Hz 79
5.20 Impedance with the scale factor optimized 79
5.21 Impedance versus frequency 80
5.22 Phase Angle versus frequency 80
5.23 Resistance versus frequency 81
5.24 Reactance versus frequency 81
5.25 Impedance Match of 3x3 Impedance Matrix 82
5.26 (a) Impedance match of (1,2) mutual element 86
(b) Impedance match of (1,2) mutual element when one
correction branch has been added 86
(c) Impedance match of (1,2) mutual element when two
correction branches are used 86
5.27 Matching to the Diagonalized Impedance Matrix
(a) Phase A 88
(b) Phase B 88
(c) Phase C 88
5.28 Impedance Match with a corrective branch for the 50 Hz
parameters
(a) Phase A 89
(b) Phase B 89
(c) Phase C 89
5.29 Phase Angle Match with a corrective branch for the 50 Hz
parameters
(a) Phase A 90
(b) Phase B 90
(c) Phase C 90
5.30 Impedance match obtained using various error criteria
(a) Least Squares 94
(b) Mini-max 94
(c) Mini-average 94
(d) Mini-area 95
viii.

Figure Page

5.31 Pictorial View of the effect of Numerical Noise on Gradient


estimate 96
5.32 Interpretation of Self and Mutual terms 97
5.33 Extension of Hingorani and Burbery's equivalent by
incorporating the mutual voltage as part of the fundamental
frequency source 99
5.34 Full frequency-matched a.c. system equivalent 99
5.35 The self circuit for the frequency-matched a.c. system as
implemented in TCS 100
5.36 Various a.c. system equivalent circuits 102
5.37 Self <-> Mutual Circuit Interface 103
5.38 Two types of interface
(a) Standard TCS circuit <-> torn branches 104
(b) Standard (Self) circuit <-> Mutual circuits 104
5.39 Multi-convertor System 111
5.40 Data Flow in forming a d.c. equivalent circuit 113
5.41 Structure diagram of DCLINK 11 4
5.42 Impedance match of d.c. equivalent 116
5.43 Partitioned equivalent circuit 117
5.44 Busbar voltage, displaying a harmonic instability 122

6.1 Structure diagram of the harmonic penetration program 125


6.2 Structure diagram of the iterative Harmonic Algorithm 130
6.3 Data flow with TCS assessment of convertor currents 132
6.4 Data flow with IHA assessment of convertor currents 133
6.5 Simple test system 134
6.6 Modified test system 136
6.7 Harmonic voltage throughout the Lower South Island of
New Zealand test system 142
6.8 Harmonic Branch currents at sending end for Lower
South Island test system 145

7.1 HVDC Test System 147


7.2 D.C. Line Fault Simulations
(a) Rectifier Current 148
(b) D.C. Voltage 148
7.3 System after Fault application 149
7.4 Rectifier side a.c. system fault 150
ix.

Figure Page

7.5 Dynamic response due to Rectifier side Fault with Thevenin


Representation
(a) A.C. System Current 151
(b) Busbar Voltage 151
7.6 Dynamic response due to Rectifier side Fault with
Frequency-dependent Representation
(a) A.C. System Current 152
(b) Busbar Voltage 152
7.7 Dynamic response due to Rectifier side Fault with Thevenin
Representation
(a) A.C. System Current 153
(b) Busbar Voltage 153
7.8 Dynamic response due to Rectifier side Fault with
Frequency-dependent Representation
(a) A.C. System Current 154
(b) Busbar Voltage 154
7.9 Invertor side a.c. system fault 155
7.10 Faulted Busbar voltage
(a) Thevenin Representation 156
(b) Frequency-dependent representation without mutual coupling 156
(c) Frequency-dependent representation 156
7.11 A.C. System Current
(a) Thevenin Representation 157
(b) Frequency-dependent representation without mutual coupling 157
(c) Frequency-dependent representation 157
7.12 Fault current 158

A2.1 Transmission Line n-segment Model 182


A4.1 Dichotomous Search Technique 184
A4.2 Search Technique used in the Fibonacci
and Golden Section Searches 185
A4.3 Success/Failure step Algorithm 187
x.

LIST OF TABLES

Table Page

5.1 CPU requirements for various error critera 95


5.2 Assessed impedances with phase "Aft current injection 108
5.3 Assessed impedances with phase 118 11 current injection 109
5.4 Assessed impedances with phase "c" current injection 110

6.1 Phase "A" Current Harmonics for simple test system 135
6.2 Phase "A" Voltage Harmonics for simple test system 135
6.3 Phase "A" Current Harmonics for modified test system 131
6.4 Phase "A" Voltage Harmonics for modified test system 131
6.5 IHA/TCS comparison with Thevenin a.c. system representation
used in TCS
(a) Phase "A" harmonic currents 138
(b) Harmonic Voltages 138
(c) Comparison of d.c. Current Harmonics 138
6.6 IHA/TCS comparison with frequency-matched a.c. system
equivalent used in TCS.
(Implicit mutual coupling representation)
(a) Phase "A" harmonic currents 139
(b) Harmonic Voltages 139
(c) Comparison of d.c. Current Harmonics 139
6.1 Comparison of Computational Efficiency 141
6.8 Harmonic Voltages throughout the Lower South Island of
New Zealand's primary transmission network 143
6.9 Harmonic branch currents at the sending end throughout the
Lower South Island (N.Z.) test system 144

A3.1 Parameters for the circuit of figure 4.20 183


xi,

List of Principal Symbols

SYMBOLS

~ - State variable convergence tolerance.


~d - State variable derivative convergence tolerance.
Id - D.C. current.
Vd - D.C. voltage.
I - D.C. current margin,
dm
s - Laplace Operator.
a - Convertor firing angle.
~ - Phase angle,
~ - Convertor commutation angle.
Y - Convertor extinction angle.
f - frequency (Hz)
h - step length,
j imaginary unit (j=i-l)
~ - Inductor flux,
Q - Capacitor charge,
e - Objective function,
S - Electrical angle (S=wt)
o - Penalty function,
A - Lagrange multiplier
p - derivative w,r,t, t (or wt)
w - Angular frequency 2nf)
fpr - frequency of parallel resonance
Ih - Current at harmonic h
V - Voltage at harmonic h
h
V - Convertor terminal voltage
t
Y - Admittance at harmonic h
h
Y - Admittance of the actual a.c, system (required response)
rr
Y - Equivalent circuits admittance
ec
Y - Admittance required by the correction branch
cb
xii.

Subscripts:

a - alpha node
B - beta node
y - gamma node
1· 1.th e 1 emen t 0f vec t or
ij element in ith row and jth column of matrix
~ - inductive branch
r - resistive branch
c - capacitive branch
ec - Equivalent circuit
cb - correction branch
rr - Required response
- denotes a vector

Superscripts :

- derivative w.r.t. time


t - transpose of matrix
j Result of jth iteration
xiii.

Abbreviations

a.c. Alternating current


ABS Absolute value
ACREP Program for processing and plotting Power System Data
BFGS Broyden, Fletcher, Goldfarb and Shanno
C.C.C. - Constant Current Control
CPU Central Processor Unit of the computer
d.c. Direct current
DCLINK - Admittance generating program for d.c. side
DEC Digital Equipment Corporation
E.A.C. - Extinction Angle Control
EFC Equi-distant Firing Control
EMTP Electro-Magnetic Transient Program
F Farads
H Henries
HARMAC - Harmonic Penetration Program
HVDC High voltage direct currect
IHA Iterative Harmonic Algorithm
INTER Interactive data base preparation program
IPC Individual Phase Control
KCL Kirchhoff's Current Law
KVL Kirchhoff's Voltage ,Law
km Kilometres
kV kilovolts
LINE Line constants program for an arbitrary number of conductors
ln Natural logarithm
m Metres
min. Minutes
MW Megawatts
p.u. Per unit
S.C. R. - Short Circuit Ratio
sec. seconds
TCS Transient Convertor Simulator
t Time
TL Line constants program for three phase a.c. lines
TNA Transient Network analyser
w.r.t. - with respect to
xiv.

ABSTRACT

This thesis describes the algorithms developed for synthesizing


frequency matched a.c. system equivalents for use with a Transient
Convertor Simulator program. Two synthesis methods are outlined, with the
merits of each being illustrated by applying them to the lower South
Island portion of New Zealand's primary transmission system.

Validation of the diakoptical technique used to model the


frequency dependent mutual coupling, as well as data preparation, is
achieved by using the current source model to inject harmonic currents
into the frequency-matched a.c. system equivalent.

The need for frequency matched a.c. system equivalents is


demonstrated by using the Transient Convertor Simulator program for
harmonic penetration studies.

Finally the effect of the frequency-dependence of an a.c. system


model on the transient behaviour of a convertor is illustrated by
simulating both a.c. and d.c. disturbances. The difference in transient
behaviour with asymmetric a.c. faults as well as the harmonic assessment
for harmonic penetration studies both demonstrate the need for frequency
matched a.c. system equivalents that more accurately model the actual
system.
xv.

Acknowledgements

I wish to sincerely thank my supervisor, Professor J. Arrillaga


for his great assistance, advice and patience during the course of this
research.

The financial support of the University Grants Committee is also


gratefully acknowledged.

I would also like to thank the staff and postgraduate colleagues


for their helpful discussions and encouragement. The helpful interaction
with A. Joosten, Dr Eggleston, Dr C.P. Arnold, Dr Heffernan, Dr Densem,
C. Callaghan, E. Achadaza, G. Cameron, W. Kennedy, A. Earl, B. Williams
is greatly appreciated.

Finally thanks are due to my family and friends for their support
and understanding.
1.

CHAPTER 1 INTRODUCTION.

1.1 Importance of HVDC Transmission.

The technical advantages of HVDC transmission are well documented


in the literature and were understood from the early days of electricity.
However HVDC transmission had to wait for the development of the Mercury-
arc valves before finding its first commercial application in 1954. This
was the Sweden-Gotland link which transmitted 20 MW at 100 kV d.c. and
was justified because the transmission distance (96 km) was too long for
a.c. technology of the time and cheaper than building a new fossil fuel
generation plant on the island. Since then there has been a rapid
increase in the number and power rating of HVDC schemes.

Mercury-arc valves were exclusively used in the a.c./d.c.


conversion process until 1972 when the Eel River scheme was commissioned.
This was the first use of the thyristor valve which eventually superseded
mercury-arc valves for HVDC transmission schemes. The Kingsnorth HVDC
scheme which was commissioned in 1974 was the last scheme using mercury-
arc and all subsequent schemes used thyristor valves. By the end of 1979
50 percent of the total installed HVDC capacity of 12,000 MW were
thyristor valve systems.

Performances of the HVDC systems throughout the world have been


regularly surveyed (Last and Middleton 1970, 1972, Jarrett and Middleton
1974, Rumpf and Jarrett 1976, Rumpf 1980, Knudsen and Albrecht 1982,
Knudsen et al 1984). These surveys show that HVDC systems are technically
and economically viable due to their high reliability and use of the HVDC
links attributes to meet the particular systems requirements.

Advances in convertor stations, such as the use of air or water


cooling, and developments in thyristor valves have lead to a reduction in
cost relative to a.c. transmission and greater power ratings. The
increased competitiveness of d.c. transmission is reflected in the number
of links presently under consideration. New Zealand is currently
considering a second d.c. link and some third world countries are also
implementing d.c. links. In the early schemes the a.c. systems were
strong compared to the transmitted power. However as the convertor
2,

ratings increase the relative a,c. system strength diminishes and the
problems of dynamic response and harmonics increase greatly.

1.2 Need for Computer Modelling,

The planning, design and operation of a power system requires


several different types of studies to ensure that satisfactory
performance is maintained. If such studies are carried out at the design
stage, solutions to possible problems can be found before they eventuate,
thus preventing costly damage or maloperation of power system components.

However the complexity of power systems makes such studies


difficult and time consuming. A truly analytical solution is virtually
impossible for all except the simplest of problems, The need for
computational aids led to the development of special purpose analog
computers and the use of scaled down models, Due to the inherent
difficulties of such methods and to the rapid development of powerful
digital computers, computer simulation has become the preferred method.
The appeal of the digital computer is its ability to process a vast
amount of data in a systematic way with great accuracy, and to do so in
an extremely short time. This has been enhanced by the development of
efficient numerical techniques for modelling large electrical power
systems, The effectiveness of computer modelling is summarized by the
following quotation from IEEE std. 399-1980 "The modern digital computer
offers power systems engineers a powerful tool to perform more effective
studies to any industrial power system."

1.3 Types of Studies.

The types of studies performed at the planning stage are


(a) Load flow (or Power flow)
(b) Fault and Short-Circuit Studies
(c) Harmonic Studies
(d) Dynamic Studies

Load flow studies determine the voltage, current, power and power-
factor or reactive power at various points in a power system. This is a
3.

fundamental frequency steady state analysis, with all waveforms assumed


to be perfectly sinusoidal. Load flow studies find applications in almost
all other analysis as they supply the initial conditions for the other
studies.

Short-Circuit studies are required to ensure that protection


equipment can isolate faults quickly and minimize damage and personnel
hazard.

Harmonic studies are needed to assess the effect of discrete


harmonic current injections and the resulting voltage distortion at the
various power system busbars.

Many diverse types of studies can be labelled as Transient or


Dynamic Studies. However, they are distinguished by the objective of the
study, which will determine the time scale of interest. The time scale
may range from micro-seconds for fast transients such as lightning
strikes, to many minutes for long-term dynamic studies. The time scale
of interest dictates the type of model required for the various power
system components (Concordia and Schulz 1975).

1.4 Incorporating HVDC Convertors.

The influence of HVDC convertors on system behaviour necessitates


detailed convertor representation for these various types of studies.
Although still relatively few in number, as a result of their large power
ratings and diverse control strategies means they greatly influence the
power systems they are connected to. This has motivated many people world
wide in the development of computer models to analyse the steady state
and dynamic behaviour of HVDC convertors. An early attempt was made at
UMIST with the development of a Transient Convertor Simulation (TCS)
program which models HVDC convertors to a high degree of accuracy.
Further work on TCS has continued at the University of Canterbury by
progressively improving the accuracy of the results by developing better
models for the various power system components. The work of Joosten
(1987) in developing a realistic transformer model that represents both
saturation and hysteresis effects is particularly note-worthy.
4.

The work described in this thesis concentrates on the derivation


of a.c. system models suitable for TCS, with a time scale of interest
from 0.2 to 400 milli-seconds. A three phase a.c./d.c. load flow
developed by S. Harker (1980) is used for the derivation of initial
conditions. The use of the improved TCS program in harmonic studies is
also demonstrated in the thesis.

1.5 The need for equivalent circuits.

The complexity of modern power systems makes it imperative to use


equivalent circuits to ease the computational burden. Normally the areas
of the power system where detailed information is required must be
modelled by explicit representation of each of the power system
components, while the remaining parts of the power system are reduced to
an equivalent circuit. The credibility of the simulation is then
dependent on how well the equivalent circuit performance represents that
part of the actual power system. The use of equivalent circuits has
become well established since Thevenin first published his theorem in
1883. However, since then considerable effort has been put into refining
the equivalents to try and improve their accuracy.

Under fault conditions the a.c. system parameters vary due to the
finite time required for the flux to change in components such as
synchronous machines. A means of obtaining accurate time varying network
equivalents was developed by Heffernan (1980). This involved interfacing
the dynamic simulation program with a multi-machine a.c./d.c. transient
stability program. The network equivalents obtained from the latter
program provided a more accurate indication of the time response of the
a.c. system by virtue of the synchronous machine models used in the
transient stability program.

It has been known for a long time that the frequency response of
the power system components affects the transient response due to the
multitude of frequencies present in transient waveforms. However,
incorporating such frequency-dependence in dynamic simulation programs
has proved difficult, The present work develops methods of obtaining a
practical and computationally efficient equivalent circuit that
accurately represents the frequency-dependence of the actual system being
represented.
5,

1,6 Thesis Outline,

The work in this thesis relates to the development of equivalent


circuits that accurately model the frequency-dependence of the systems
they represent and are computationally efficient, A further constraint is
that the equivalent circuits should be suitable for incorporation into
TCS programs,

The material covered is divided into three separate topics as


follows
(i) the development of a dynamic program with accurate
representation of HVDC convertors,
(ii) incorporating the frequency-dependence of power
system components into dynamic simUlation programs,
(iii) harmonic analysis and dynamic simulations,

Chapter 2 sets the scene by giving a brief overview of various


methods of performing dynamic studies with particular reference to the
representation of HVDC convertors,

Chapter 3 is dedicated to the basic mathematical models by which


each type of system component is represented in the digital computer,
However, discussion of the frequency dependent a,c, system model is left
until chapter 5, The method of network analysis is outlined and the TCS
program improvements that have been implemented are detailed,

Chapter 4 describes the flexible program for processing and


plotting power system data (ACREP) that has been written, The objectives,
structure and operation of the program are discussed as well as its
capability, Illustrative examples of its capability are also given,

Chapter 5 begins by reviewing previous work in the development of


frequency-dependent power system components for use in time domain
studies, Possible methods of obtaining the frequency-response for the
equivalent circuit are discussed, The mathematical basis for synthesizing
an equivalent circuit with a very similar frequency response is
presented, The algorithms are then applied to the lower South Island
6.

(N.Z.) test system and the resulting equivalents examined.

Chapter 6 deals with the specialized use of the developed dynamic


program for steady-state solutions, namely harmonic studies. The existing
techniques for harmonic penetration studies are surveyed and discussed. A
brief portrayal of how the programs developed fit in and interact with
the existing software at the University of Canterbury is given. The
results of a case study are presented and compared with those obtained
from an iterative steady state technique.

In Chapter 7 the results of simulating a.c. and d.c. disturbances


are presented and the effect of a frequency-dependant a.c. system model
discussed. Finally the concluding remarks and suggestions for further
work are presented in chapter 8.
7.

Chapter 2 OVERVIEW.

2.1 Computational Aids for Dynamic Simulation.

The need to predict the dynamic behaviour of electrical power


systems has lead to the development of various computational aids. These
are; (i) Analog computer, (ii) Transient Network Analyser (TNA), (iii)
Digital computer. Each tool has its strengths and weaknesses and they
should be looked on as complementary rather than mutually exclusive.
There is also further subdivision possible depending on the different
implementation of each computational aid. Combined use of these
computational aids has also been proposed to utilize the best features of
each approach.

Mathematical modelling techniques assume that transient events


can be expressed and solved in terms of mathematical equations, whereas
the experimental approaches attempt to simulate the physical behaviour of
components, circuits and systems such that the application of a physical
input to the model results in an output similar to that of the real
system. Transient network analysers (TNA) fall into the latter category,
while digital computers and electronic analog computers model the system
mathematically. TNA is used here in a broad sense and encompasses all the
analysers using the electromagnetic model approach to transient analysis.

The analog computers solve the differential equations that


mathematically represent the transient phenomena. They are well suited
for preliminary sensitivity studies due to the speed and flexibility of
system parameters and control implementation. Their major limitation is
the lack of representation of nonlinear circuit elements such as
saturable transformers and HVDC convertors.

The TNA has been the traditional method of performing transient


power system studies. Peterson (1939) described the TNA developed at GEC
for transient studies. Over the years very elaborate TNA's have been
constructed and the representation of power system components improved.
In the conventional TNA simulators, reactors, capacitors and resistors
are connected to duplicate the behaviour of their actual counterpart;
tapped inductive reactors are used for transformer simulation. Another
8.

type of TNA is the scaled down physical simulator, where miniature


transformers are used to represent the actual ones. An accurate
simulation of transformers is difficult as it requires representation of
magnetization characteristics (saturation curve and losses) as well as
leakage inductance and capacitance. Requirements in different areas often
conflict and compromises have to be made. In particular the transformer
resistance tends to be higher in the model than in practice. While it is
possible to add resistance, inductance or capacitance, to reduce them is
more of a challenge. Transmission lines are represented by lumped
networks made up of TI or T sections. The TNA is limited in the size and
complexity of the system it can simulate due to time and economic
considerations. The system parameters are inflexible making the study of
different configurations time consuming. Once a particular design has
been decided upon, a TNA model can be set up, allowing such parameters as
HVDC control strategies to be optimized for a small extra cost.
Considerable spread in transient voltages develop due to the instant in
the cycle at which a change occurs and the variations between the closing
or opening of the different phases. It is relatively easy to perform the
switch operation on the TNA repeatedly and collect statistical data. The
enormous development in TNA's is evident in the 80 references cited by
Concordia (1956).

The digital computer has been applied to many different


mathematical formulations of the transient phenomena and the most
important ones are outlined in the next section. The digital computer is
well suited to studying different system configurations due to the
flexible manner in which the power system topology can be set up by
software. Greater accuracy is achievable compared to the analog computer
approach, but the optimization of system parameters requires many runs of
digital simulation, which maybe expensive in terms of computer time. The
accurate modelling of nonlinear elements is also easier to accomplish in
the digital computer solution compared to both analog computer or TNA
approaches. The major factor limiting the complexity and accuracy of
digital computer simulations is the computational cost. The type of study
required determines the mathematical model necessary and hence the cost.

The use of equivalent circuits tries to alleviate the computational


burden without significant loss of accuracy. hence the importance of the
present work on a.c. system equivalents.
9.

2.2 Digital Computer Simulation.

Much work has been done to develop ingenious techniques for


digital simulation of transients, as well as methods that allow certain
difficulties to be circumvented or assist in reducing the amount of
computation. Considering the enormous amount of published literature on
the topic only a superficial outline can be given here. For a more
detailed treatment and list of references the reader is directed to the
reviews by Pender (1969), Humpage and Wong (1982), Bickford (1986) and
the books by Greenwood (1971) and Bickford et al (1976).

Travelling wave methods are based on the solution of the


transmission line equations, which maybe expressed as a combination of
forward and backward travelling waves on the line. Two different
approaches exist based on the Lattice diagram of Bewley, and graphical
method initiated by Allievei and developed by Schnyder and Bergeron
(Bickford et al 1976). These techniques were later adapted for solution
by digital computers due to the labour involved in applying travelling
wave methods. The mathematical basis for this method relies inherently on
lossless propagation, but correction terms can be incorporated to account
for the losses in real systems. Although travelling wave methods are
suited towards distributed parameter elements such as lines and cables,
lumped parameter elements like generators, transformers etc., can be
approximated by short line stubs. The travelling wave methods are
attractive for small transmission line problems but they prove rather
cumbersome for realistic power systems studies due to their complexity.
Moreover the represent ion of HVDC convertors in travelling wave methods
has not been attempted.

The importance of correctly modelling the frequency-dependence of


power elements to include its effect on the transient system response has
been recognized for many years. This has led many people to investigate
frequency domain techniques based on Fourier series or Fourier
transforms, due to their easier represent ion of frequency dependence. In
these methods the frequency-dependence is modelled to an accuracy limited
only by the accuracy to which the system parameters are known. The
Fourier frequency-domain method involves formulating the problem in terms
of a repetitive function, transforming, solving and performing the
inverse transform to obtain the time response. However HVDC convertor
10.

representation is a problem that has not been solved with this method.

The z-transform techniques are a class of frequency domain


analysis which seek to avoid going from the frequency domain to time
domain via the inverse Fourier transform, with its inherent difficulties.
As with the other frequency domain analysis techniques, accurate
convertor representation is a problem.

Time domain analysis based on step-by-step numerical integration


of the system's differential equations is the most suitable method for
dynamic studies involving HVDC convertors. The most widely acclaimed
transient program is undoubtedly EMTP (Electro Magnetic Transient
Program). EMTP uses the trapezoidal integration equation to linearize the
differential equations to a set of simultaneous equations, which are then
solved. EMTP logic, as coded by H.W. Dommel (Dommel 1969, Dommel and
Meyer 1974), was designed when switching operations were few and far
between. EMTP features fixed step-length which ensures economic
simulation, as many of the matrices are then constant.

In the TCS program the problem is formulated using state variables


as a set of first order differential equations, and the trapezoidal
integration is then used to solve for the new state variables at each
time step. Variable step length is used in TCS to enable time step
boundaries to fall exactly on switching operations.

A problem associated with EMTP is numerical noise. Campos-Barros


(1985) has demonstrated the superiority of the TCS formulation when
simulating convertor valve switchings, due to the oscillations in EMTP
simulations caused by numerical noise.

2.3 Development of HVDC Convertor Models.

An early method of digital HVDC dynamic simulation used "The


Central Process" technique. In this method HVDC convertor operation is
broken into similar consecutive processes. It requires many subroutines,
with each subroutine solving a set of differential equations arising from
the particular network topology. Only those conduction patterns coded can
be simulated, so prior knowledge of all possible network topologies is
required. The control and processing of the subroutines were included in
the main program. This method is limited in versatility and was applied
11.

to studies where the convertor was connected to an infinite a.c. busbar.


Many different sets of equations became necessary to describe the network
when the a.c. system impedance became significant making the approach
cumbersome and wasteful of computing time (Hay et al 1970a,1970b,1971a,
1971b, Hingorani et al 1966a,1966b,1967,1968).

Williams and Smith (1973) applied the tensor-analysis techniques


proposed by Kron (1939,1959) to the modelling of convertor bridges. This
led to an elegant and superior method of dealing with the periodically
varying topology of the bridge. The main advantages were the logical
procedure for automatic assembly and solution of the network equations
and superior generality. The programmer no longer needed to be aware of
all the reduced sets of equations that described each particular
conduction situation. Mesh analysis formed the basis of the work
performed by Milias-Argitis (1976,1977a,1977b,1978) which was an
extension of Williams and Smith's work.

Kron's techniques are equally applicable to nodal formulation and


this formed the basis of TCS (Arrillaga et al 1977a,1977b). Early dynamic
simulation programs used idealised three phase e.m.f. sources in series
with an impedance to represent the a.c. system, they included explicit
representation of the harmonic filters; but the convertor transformers
were represented by their leakage reactance.

The general dynamic simulation program that resulted from the


developments at UMIST used a nodal formulation and state space theory.
The diakoptical techniques (Kron 1963, Brameller et al 1969) implemented
by AI-Khashali (1976) resulted in an extensive reduction in computational
burden. A detailed synchronous machine model was developed for the UMIST
program by Campos Barros (1976). A more realistic coupled coil
representation for transformers was implemented in conjunction with
existing models for transmission lines, harmonic filters, a.c. system and
convertor. The convertor controller was based on a direct digital control
scheme developed by Arrillaga and Galanos (1969,1970a,1970b,1970c), this
in principle being similar to the phase locked oscillator of Ainsworth
(1968), giving equidistant firing control. Further improvements were
achieved at the University of Canterbury by incorporating the a.c.
system's time response into the model (Heffernan 1980, Turner 1980). By
12.

this stage four different a.c. system models were catered for. A more
realistic d.c. fault model was introduced based on experiments performed
by Kohler (1967). At this stage the two main deficiencies were apparent;
the transformer model, and the representation of the frequency-dependence
of an a.c. system. The development of a transformer model that represents
both saturation and hysteresis effects was accomplished by Joosten
(1987), while this thesis reports on the later problem.

Power System Analysis

Loadflow Short- Circui l Harmonic Penetration Transient

Analog Computer TNA Digital Computer

Travelling Wave Frequency domain z-lransform Numerical Integration

EMTP TCS

Figure 2.1 Power System Analysis Overview


13.

CHAPTER 3 MATHEMATICAL MODEL.

3.1 Introduction.

A power system can be represented by a network of induotive,


capacitive and resistive elements, where each component has a particular
circuit model defined in terms of these three elements. The most suitable
approach is a state space formulation which can be used to treat both
linear and non-linear circuits. This yields a set of non-linear
differential equations. The state variables are related to the energy
stored in the electric fields of capacitors, and in magnetic fields of
inductors, while resistors are non-state elements. This chapter gives a
brief outline of the component models and method of solution with more
detail being given to modifications that have been made to the existing
TCS algorithm (Heffernan 1980).

3.2 Representation of Components.

In a dynamic simulation)consideration of the major area of interest


permits considerable simplification in component models by neglecting
effects outside the area of interest. For example, source e.m.f. 's are
assumed to be of constant frequency as generator-rotor swings are too
slow to have an effect on the convertor or its controller. At the other
end of the scale, the effect of frequencies higher than 2.5 kHz are also
neglected.

3.2.1 Static Convertor.

The Graetz bridge is well accepted as the basic building block for
HVDC schemes and its operation well documented (Adamson and Hingorani
1960, Cory 1965, Kimbark 1971, Uhlmann 1975, Arrillaga 1983). Both 6 and
12 pulse convertor models are available, with the 12 pulse consisting of
two 6 pulse bridges connected in series on the d.c. side and via a
star-star and star-delta convertor transformers respectively. This
configuration of convertor transformers gives the 30 degree phase shift
required for 12 pulse operation.
1 4.

3.2.1.1 Modes of Operation.

In principle any valve conduction is possible as valve switching is


determined purely by the conditions it experiences. If a valve is forward
biased and a firing pulse is present it will attempt to turn "ON".
However the conducting state of one valve influences the conditions and
hence behaviour of the other valves and this limits the conduction
patterns normally experienced. The normal and abnormal conduction states
that can be expected may be categorised by the following modes :

A Non-commutating mode
Two valves on different sides and arms of the bridge are ON.
B Normal commutation
Three valves in ON state ; one on each arm and at least one
conducting on each side of bridge.
C Non-commutating arm short-circuit
Both valves on one of the arms in ON state with no other valves ON.
D Commutating arm short-circuit
Both valves on one of the arms in ON state with one or both valves
on another arm being ON. There is a single commutation process if
one valve on the second arm is conducting, otherwise there are
commutations taking place on both sides of the bridge
simultaneously.
E a.c. short-circuit
Four or more valves conducting, involving all three arms of the
bridge. This has the effect of short-circuiting the convertor
transformer's three secondary terminals.

Although bypass valves are present in existing schemes, their effect


can be modelled implicitly by modes C,D and E.

3.2.1.2 Convertor Control Systems.

The operational details of the control system differ for each


particular application, however their basic principles are the same and
are well documented (Adamson and Hingorani 1960, Cory 1965. Kimbark 1971,
Uhlmann 1975. Jotten 1978, Arrillaga 1983).

The steady state operating characteristics for a basic HVDC link


controller are displayed in figure 3.1. The rectifier operates in
15.

- Invertor

E.A.C.

C.C.C.

Figure 3.1 Steady State characteristics and operating point

constant current control (C.C~C.) mode by varying its delay angle so as


to maintain a link current while the invertor determines the link voltage
by operating with minimum extinction angle (referred to as extinction
angle control E.A.C.). The invertor also has a C.C.C. feature which is
set to operate during transient or abnormal conditions.

High link voltage is required to minimize losses, and to minimize


reactive power demands the control angles (a and Y.) need to be as
r 1
small as practicable. However Y. must be large enough to avoid
1
commutation failures on the occurrence of minor disturbances. Convertor
transformer tap changing is used to alter the level of the effective a.c.
supply voltage, but this is a relatively slow type of control and
therefore not modelled in TCS.

3.2.1.3 Types of Control Schemes.

Individual Phase-Control (IPC) was the norm for early HVDC schemes.
With this scheme valve firing instances are determined independently for
each valve based on the relevant a.c. voltage zero crossing. IPC reflect
any unbalance or distortion in the a.c. supply voltages by producing
variations in valve conduction periods. These in turn produce further
distortion and unbalance, with the result of uncharacteristic harmonic
production in the a.c. system, and harmonic instability if it reinforces
the original distortion (Ainsworth 1968). These phenomena are common with
weak a.c. systems. Filtering the a.c. system waveform entering the
16.

controller can reduce the problem but not remove it.

Equidistant Firing Control (EFC) produces a train of equidistant


firing pulses which are advanced or retarded by the controller. Equal
conduction periods are assured regardless of a.c. voltage waveforms, and
hence uncharacteristic harmonics and harmonic instability problems are
alleviated. Most EFC controllers are variations of the phase locked
oscillator technique developed by Ainsworth (1967,1968). Digital
implementation of this firing control has been proposed by Arrillaga and
Galanos (1969,1970a,1970b,1970c) and Arrillaga and Baldwin (1974), and is
used in TCS.

3.2.2 Synchronous Machines.

When a convertor is present phase-variable representation is


required due to the synchronous machines terminal voltages being non-
sinusoidal and the need to accurately represent the convertor. Since most
manufacturers' data is in the two-axes d-q-o form it must be transformed
to direct phase quantities (a,b,c) (Concordia 1951, Kimbark 1968,
Fitzgerald et al 1983). A phase-variable model allows easy handling of
asymmetries, non-linearities and distortion effects and the extra
computation caused by the time varying inductances is partially offset by
averting the need of transformations at each time step. A d-q axis
representation is retained for the rotor circuit since it fits the actual
geometry and winding arrangement. In writing the differential equations
the synchronous machine is treated as a motor, with the positive current
flow direction being into the positive terminal. In matrix form the
terminal voltage may be expressed as :

vg peL
g
. I )
g
+ Rg . I g

+ I + R • I (3.1)
g g g

where (j) ~~ , the angular velocity.

L is a 6x6 matrix and its elements are given in appendix A1. The
g
stator values can accommodate fourth harmonic terms if available, as
these can be of significance in convertor-generator units (Campos-Barros
1976) .
17.

3.2.3 Transmission Lines.

Wave propagation techniques are classically used to analyse the


transient response of a transmission line. However, the implementation is
difficult where there are multiple a.c. and d.c. lines of differing
lengths, and this has lead to a network approach being adopted. The
transmission line model consists of a number of cascaded n-segments
where the number of n-segments is determined from the highest frequency
of interest (Smith and Bell 1984). Appendix A2 describes the selection of
component values for this model.

3.2.4 Transformers.

The single phase coupled coil transformer model, as shown in


figure 3.2 is the basic building block by which all other three phase
transformers are represented, due to the ease in which winding
configuration and associated phase shifts are accounted for.

LI2
II RI R2 I 2
'" L21
~
t
VI LI I L22
f
V2

I I
0

Figure 3.2 Coupled Coil Model of Single Phase Transformer.

Neglecting the iron losses, the matrix equation for the coupled coil
model is ;

(3.2)

where the relationship between the leakage reactance and inductance in


the model is (in p.u.)

where X~a is the total leakage reactance of the transformer.


18.

Accurate representation of non-linearities in core magnetization


(saturation and hystersis) has been incorporated into the transformer
model by Joosten (1987)

3.2.5 Static Shunt Elements.

Both a.c. and d.c. harmonic filters and surge capacitors greatly
influence the convertor waveforms following a disturbance and even in the
steady state, and hence need to be accurately represented. Fig. 3.3
displays the various shunt elements that can be used to represent plant
components and faults. Three phase a.c. components are assumed to be star
connected with the neutral point earthed.

I I
Figure 3.3 The various static shunt element models.

3.2.6 Static Series Elements.

These models are essentially the same as for the static shunt
elements except that the components are connected between phases. This
allows circuit breakers, delta loads and other such components to be
modelled. Care must be taken to ensure that the topology constraints, as
discussed in section 3.3.6, are not contravened when using these
elements.
19.

3.2.7 A.C. System.

The size and complexity of a.c. power systems necessitates the use
of less complex, yet sufficiently accurate a.c. system models. Explicit
busbar information will generally not be available at all busbars and
hence system components of particular interest must be represented in
detail. A number of a.c. system equivalents have already been developed
and are displayed in figure 3.4. The major feature of the a.c. system
models developed by Heffernan (1980) was their time varying nature. The
present work involves the incorporation of the frequency response of the
system, as expressed by impedance loci, into an a.c. system model. The
details of obtaining the impedance loci and the synthesis of frequency
matched a.c. system models is presented in chapter 5.

E L
s 5

(l-a_
_ IL 5
a

(l-aIL
_ _ 5
a

Figure 3.4 A.C. System Equivalents


20.

3.3 Method of Analysis.

Any lumped network obeys three laws; Kirchhoff's voltage law


(KVL), Kirchhoff's current law (KCL) and the element laws (branch
characteristics). From these a set of first-order differential equations
can be derived that characterize the network and this constitutes the
state variable approach. In the state variable approach the network is
characterized by the following matrix (or state) equations :

[AJ~ + [BJu + [EJ~ C3. 4)

[CJ~ + [D]u C3.5 )

where
u represents the input voltages and currents.
x is the vector of state variables.
¥ represents the output voltages and currents.
z represents a set of dependent variables.

The dependent variables are evaluated by the equation

z = [F]x + [G]u (3.6)


- -
The matrices [A],[B],[C],[D],[E],[F],[G] are the appropriate coefficient
matrices which may be non-linear functions of x or u and/or time varying.

The attractions of the state variable approach are its


amenability to numerical methods of analysis and the ease in which non-
linearities are incorporated. Non-linearities which are functions of
time, voltage or current magnitude are easily handled, which allows most
types of power system non-linearities. The non-linearity not easily
simulated is the frequency-dependence.

3.3.1 Trapezoidal Integration.

The basic trapezoidal method is very well known, having been


established before the advent of the digital computer. More recently an
implicit trapezoidal integration method has gained wide acceptance due to
its good stability, accuracy and simplicity (Gear 1971, Arrillaga et al
1983a).

For each time step the change in a state variable is equal to the
integral of the area under its derivative and the trapezoidal integration
21.

approximates this area by the trapezoid given by

where ~t+h is determined iteratively.

The iterative procedure is as follows


(i) For an initial estimate it is assumed that ~t+h ~t

(ii) An estimate of ~t+h based on ~t+hestimate is obtained.


(iii) ~t+his estimated from the current ~t+hvalue using equation (3.9)

(3.8)

(3.9)

(iv) Steps (ii) and (iii) are performed iteratively until convergence
is reached. Convergence is deemed to have occurred when all the
state variables satisfy

(3.10)

where s is the convergence tolerance. It is sometimes necessary


to specify an additional convergence constraint to ensure the
state variable derivatives have converged sufficiently. i.e.

(3.11)

where sd is the state variable derivative convergence tolerance.

Normally three to four iterations are required with a suitable


step length. If convergence fails the step length is halved and the
iterative procedure is restarted. The integration step length is also
automatically increased or decreased during the simulation based on the
past history of the number of iterations needed to reach convergence.
This greatly improves efficiency of the simulation.

3.3.2 Topology Changes.

The instances of network connection or parameter value changes


need to be determined accurately to achieve an accurate simulation. Two
types of changes can be distinguished; those easily predicted before the
event and those which are detected only after the event. The time for
switching "ON" a convertor valve and fault application are easily
determined before the event. In such cases the simulation step size is
22.

lowered so that the simulation step falls exactly on the required time
the change is to take place. As the state space coefficient matrices are
not functions of step size, changing it does not impose a dramatic
computational burden.

The times of zero current in convertor valves and fault branches


are not easily predictable before they occur. Hence, these conditions are
detected after they occur and linear interpolation (backwards in time) is
used to reach the time of the switching. The state variables are also
obtained by linear interpolation rather than trapezoidal integration.
This is because the linear interpolation gives sufficient accuracy, due
to the small simulation step size, while the trapezoidal integration
would increase dramatically the storage and computation overheads.

3.3.3 Choice of State Variables.

Although inductor current and capacitor charge are the state


variables chosen by most textbooks it is better to use the flux linkage
of the inductor (~) and the capacitor charge (Q) as the state variables.
Regardless of the numerical integration algorithm used in solving the
differential equations, this will result in a better solution due to less
error propagation owing to the presence of local truncation error (Chua
and Lin 1975). Also large step lengths are achievable while maintaining
convergence within a few iterations (Heffernan 1980).

State space analysis requires the number of state variables to be


equal to the number of independent energy storage elements (i.e.
independent inductors and capacitors). Therefore it is important to
recognize when inductors and capacitors in a network are dependent or
independent. For example a dependent inductor is one whose current is a
linear combination of the current in k other inductors in the system.
This is ndt always obvious due to the presence of intervening networks.

The use of capacitor charge or voltage as a state variable


creates a problem when a set of capacitors form a closed loop. In this
case the standard state-variable formulation falls down as one of the
chosen state variables is a linear combination of the others, therefore,
should not be a state variable. This would be a serious limitation as
many power system elements exhibit this situation (e.g. transmission line
model). There are several ways of overcoming this problem; TCS uses the
23.

charge at a node rather than capacitor voltage as a state variable to


circumvent the problem.

A similar problem arises when only inductive branches are


connected to a radial node. If the initialization of state variables was
such that the sum of the current at this radial node was non-zero, then
this error will remain throughout the simulation. The development of a
"Phantom Current source" is one method that has been developed to
overcome the problem (Joosten 1985). There are several other possible
methods for overcoming this current error. One approach is to choose an
inductor at each node with only inductors connected to it, and make its
flux a dependent rather than state variable. Another possibility is to
transform the inductor to its controlled source equivalent (Chua and Lin
1975) .

Each approach has some disadvantages. The "Phantom Current


Source" approach can cause very large voltage spikes when trying to
overcome inaccurate initial conditions. Partitioning the inductor fluxes
into state and dependent variables is hard as it is often difficult to
identify if the inductor flux is dependent or independent. An inductor
can still be dependent even if it is not connected directly to a radial
node consisting of inductive branches, but it has an intervening
resistor/capacitor network.

3.3.4 Per Unit System.

In the analysis of power systems Per Unit quantities rather than


actual values are normally used. This scales voltages, currents and
impedances to the same relative order thus treating each to the same
degree of accuracy. In dynamic analysis instantaneous phase quantities
and their derivatives are evaluated. The variables may be changing
relatively fast causing a large difference between the order of a
variable and its derivative. For example consider the sinusoid:

x A.sin(wt+<jl) (3.12)

wA.cos(wt+<jl) (3.13)

The relative difference in magnitude between x and x is w, which


may be high. Therefore a base frequency Wo is defined. All the state
variables are changed by a factor Wo and this then necessitates the use
of reactance and susceptance matrices rather than inductance
24.

and capacitance matrices. The integration is now with respect


to electrical angle rather than time.

~k= wo·W k = (wolk)·I k <3.14)

Qk= wo·q = (woCk),V k <3.15)

where
lk is the inductance
Lk the inductive reactance
c is the capacitance
k
C the capacitive susceptance
k
Wo the base angular frequency

3.3.5 Initial Conditions.

The initial conditions required for a simulation to be performed


are the state variables at the start of the simulation. An obvious case
is to start from a de-energized state, however, the extensive start-up
interval would be computationally prohibitive. The results from a steady
state analysis such as provided by a load flow program are better,
although it cannot provide exact initial conditions where convertors
and/or detailed synchronous machines are being modelled, by virtue of the
fact that it is a fundamental frequency analysis only.

Therefore a preliminary dynamic simulation, using the appropriate


load flow values as the starting point, is used in order to obtain the
desired steady state operating point. This information is then stored and
used as the initial conditions for subsequent dynamic studies of the
system. This preliminary study will provide the "dynamic initial
conditions", as they are obtained via a dynamic simulation.

The quality of initial conditions is observable in the severity of


the initial transient before the steady state operating point is reached.
The accuracy of the initial conditions is assessed by performing an FFT
on the TCS waveforms once the steady state operating point has been
achieved. The FFT algorithm allows inspection of current and voltages of
both the fUndamental and harmonics. The dynamic initial conditions will
always be better than the user entered load-flow data as the former
25.

inherently contains harmonic information while the latter allows


fundamental magnitudes and angles only.

3.3.6 Network Equations.

The nodes are partitioned into three possible groups depending on


what types of branches are connected to them. The classification of the
three node types are :

a nodes Nodes that have at least one capacitive branch connection.


S nodes Nodes that have at least one resistive branch connection but
no capacitive branches connected.
Y nodes : Nodes that have only inductive branches connected.

The resulting branch-node incidence (or connection) matrices for


t t t
the r,9., and c branches are K ,K nand K respectively. The elements in
rn Nn cn
the branch-node incidence matrices are determined by :

(-i
if node i is the sending end of branch b
t
Kbi if node i is the receiving end of branch b
if branch b is not connected to node i.

Partitioning these branch-node incidence matrices on the basis of


the above node types yields :
t t t
Kt = [ K9.,a,K9.,S,K9.,y
9.,n J <3.15)

Kt = [ Kt ,K t S ,0 J <3.16)
rn ra r

Kt = [ Kt 0 ,0 J <3.17)
cn ca'

The efficiency of solution can be improved significantly by


restricting the number of possible network configurations to those
normally encountered in practice. The restrictions are:

(i) Capacitive branches have no series voltage sources.


(ii) Resistive branches have no series voltage sources.
(iii) Capacitive branches are constant valued (i.e. pC =0).
c
(iv) Every capacitive branch or subnetwork has at least one end as
the system reference (or ground node).
(v) Resistive branch subnetworks have one end connected to the
system reference or an alpha node.
(iv) Inductive branch subnetworks have at least one non-gamma node.
26.

ec I
s

Figure 3.5 The fundamental Branches

Figure 3.5 shows the fundamental branches and the following


branch equations can be derived. Although the equations that follow are
correct as they stand, with L an9 e being inductance and capacitance
matrices respectively and the p operator denoting a derivative w.r.t.
r( (
time, the TeS implementation necessitates that the Land e be intepreted
as inductive reactance and capacitive susceptance matrices respectively.
Also the p operator must now represent a derivative w.r.t. electrical
angle rather than time.

Resistive branches.

0.18)

Inductive branches.

E (L I ) - R I + Kt V + Kt V Kt V
~-p ~ ~ ~ ~ ~a a ~8 8+ ~y y o 0.19)

p~~= E~-pL~I~- R~I~+ K~aVa+ K~8V8+ K~yVy 0.20 )

Capacitive branches.
t V )
ec P(K Ca I 0.21 )
a c
27.

In deriving the nodal analysis technique Kirchhoff's current law


is applied, the resulting nodal equation being :

K I + K I + K 1=0 C3. 22)


nc c nr r ns s
where I ,I ,In and I are the branch current vectors. Applying the node
c r N s
type definitions gives the following three equations ;

KaJl.IJI.+ K I = 0
as s
or KaJl.pIJI.+ K pI = 0
as s I C3. 23)
K I + K~Jl.IJI.+ K I = 0 C3. 24)
~r r ~s s

K I + K I + K I = 0 ,I C3. 25)
'(c c '(r r K,(Jl.IJI.+ '(s s

t
Premultiplying equation (3.21) by K and substituting into
ac
equation (3.25) yields:

pQ = - K I - K I - K I C3. 26)
a aJi. JI. ar r as s

The dependent variables V~ ,V'( and Ir can be entirely eliminated from


the solution so only IJI. 'Va and the input variables are explicit in the
equations to be integrated. This is however undesirable due to the
resulting loss in computational efficiency even though it reduces the
overall number of equations. The reasons for the increased computational
burden are ;

- Loss of matrix sparsity,


- Incidence matrices no longer have values of -1,0 or 1. This
therefore requires actual multiplications rather than simple
additions or subtractions when calculating a matrix product.
- Some quantities are not directly available, making it time consuming
to reoalculate if it is needed at each time step.

Therefore v~,v'(and Ir were retained and extra equations derived to


evaluate these dependent variables. To evaluate V~ equation (3.18) is
premultiplied by K~r and rearranged to give

V = - R (K I + KQnIn+ K R- 1Kt V ) C3.27)


~ ~ ~s S ~N N ~r r ra a

where
28.

Premultiplying equation (3.19) by Ky~ and applying equation (3.23)


gives the following expression for Vy ;
-1
vy = -LyKYSpIS-LyKy~Ly (E~-PL~I~-R~I~+ (3.28 )

where

I is evaluated by use of equation (3.18). The sequence of


r
solution once the trapezoidal integration has converged, for a time step,
is as follows ; the state-related variables are calculated followed by
the dependent variables and lastly the state variable derivatives are
obtained from the state equations.

State-Related Variables.
-1
I~= L~ 'I'~ (3.29)

V = C- 1Q (3.30 )
a a a
Dependent Variables.
1 t
V = - R (K I + Koolo+ K R- K V ) (3.31 )
8 8 8s S ~N N 8r r ra a
1 t
I = R- (K V + KrtoVo) (3.32 )
r r ra a ~ ~

(3.33 )

State Equations.
P'l' ~ = E~- pL ~ I ~- R~ I ~ + Kt~a Va + Kt~8 V8 + Kt~y Vy (3.34 )

pQ = - KNo1o- K I - K I (3.35)
a ~N N ar r as s

where
29.

3.4 Extensions.

3.4.1 C-type Filter.

The use of conventional damped filters for low order harmonics


involves large fundamental power loss in the damping resistor, therefore
the C-type filter has been designed to reduce the power loss (Arrillaga
1983). The shunt filter branch routines in TCS have been changed to
accommodate the C-type filters. The dimension of the capacitive submatrix
representing a C-type filter differs from that of the other components
and some restrictions must be placed on the order in which the various
components are entered.

3.4.2 Harmonic Current Source.

In any work dealing with frequency-dependence, the ability to


inject a controlled amount of harmonic current is important in order to
derive the impedance at the appropriate harmonic frequency. A current
source model has been added to the network equations given in section
3.3.6. For simplicity each current source is designed to inject one
frequency component; hence several current sources are required to inject
a multitude of harmonics. Although this model has been referred to as a
"Harmonic Current Source" the current source can be of any frequency not
necessarily a harmonic frequency.

3.4.3 RLC Networks connected between Phases.

The mutual coupling between phases must be modelled in harmonic


assessment studies as it greatly influences the harmonic levels,
particularly the higher order harmonics. It also needs to be modelled for
accurate dynamic studies. However, to complicate matters, the mutual
coupling is strongly frequency-dependent and exhibits both resistive and
reactive coupling components.

An early attempt to model frequency dependent mutual coupling


between the phases was the connection of frequency dependent networks
between the phases as illustrated in figure 3.6. However the explicit
location of RLC branches between phases would violate a TCS constraint as
the capacitor would not be connected to an alpha node or earth. The
30.

Figure 3.6 RLC Networks connected between phases

removal of this constraint from the TCS formulation would require major
modifications to the program. A possible solution is the addition of a
very small capacitance between earth and each phase in order to
convert the coupling nodes into alpha nodes; however, this would result
in small integration step lengths and thus deterioration of computational
efficiency.

The approach implemented, based on diakoptical techniques,


involves the formation of a set of state equations for the mutual
branches which are solved at each time step to obtain the current in each
branch. This current is then treated as a current source connected to the
appropriate nodes when the main state equations are solved. So as far as
the existing TCS algorithm is concerned the mutual branches appear as
dependent current sources connected between the phases. Therefore these
mutual branches are referred to as "torn" branches as they form an extra
network, separate in identity from the standard TCS circuit, with a
voltage-current source providing the interface between the two. An
example of the state equation for a mutual RLC branch is:
31.

~e [~p,u'J
-R
wL vp.u. 0.36)
p.u. [ /j Zb
wL

p.u.
where V is the per unit voltage across the RLC branch.

Figures 3.7 and 3.8 provide some insight into possible iterative
algorithms. The internal iterative loop estimates the extra network in
figure 3.7 and the standard TCS circuit in figure 3.8. An alternative
approach, and the method adopted, consists of embedding the extra state
variables as part of the standard variable iterative loop as shown in
figure 3.9. The estimates for the new state variables and the state
variable derivatives are evaluated for the standard and extra variables
at the same stage and the convergence criteria reflects the convergence
of both sets of variables. Compared to the use of separate iterative
loops for the standard and extra variables this solution permits
considerable computational saving. However the separate iterative loop
method requires far less modification to the original TCS algorithm.

However the selection of the appropriate parameters for a


frequency matched equivalent circuit including the mutual coupling became
virtually impossible. With reference to figure 3.6, as the six frequency
dependent circuits contribute to each of the self and mutual terms they
must be considered simultaneously. This precludes the use of direct
approaches such as Hingorani and Burbery's as they match only one circuit
to a required frequency response in isolation. Only a multi-variable
optimization technique is applicable and since a scalar objective
function must represent the match of the self and mutual terms
simultaneously the solution will contain significant error due to the
compromises. The computational requirements are also prohibitive. Thus
this approach had to be abandoned, in spite of its obvious advantages in
the modelling of circuit topologies without the restriction of the TCS
formulation. Much further work had to be carried out in this area and is
reported in chapter 5.
32.

Estimate new state variables


for standard TCS circuit

Obtain state variable derivatives


for the standard TCS circuit

Estimate new state variables


for extra network

Obtain state variable derivatives


for extra network

Has
convergence No
been reached for the >---~~---------- __~
extra network
variables

Has
No convergence
been reached for the
standard TCS
variables
?

Yes

Figure 3.7 Separate Iterative Loop algorithm;


Extra Network inner loop
33.

Estimate new state variables


for extra network

Obtain state variable derivatives


for extra network

Estimate new state variables


for standard TeS circuit

Obtain state variable derivatives


for the standard TeS circuit

Has
convergence No
been reached for the)---~~ ____________~
standard TeS
variables
?

Yes

Has
No convergence
been reached for the
extra network
variables
?

Yes

Figure 3.8 Separate Iterative Loop algorithm;


Standard Tes circuit inner loop
34.

Estimate new state variables for


both standard and extra networks

Obtain state variable derivatives


for both standard and extra networks

convergence
No
been reached all
variables (both
&

Figure 3.9 Embedded Iterative Loop approach


35.

CHAPTER 4. ACREP: A FLEXIBLE PROGRAM FOR PROCESSING AND PLOTTING

POWER SYSTEM DATA.

4.1 Capability of the ACREP program.

Although the program derives its name from an acronym of flAC


system REPresentation", this is somewhat misleading as the program is
capable of performing many more functions other than producing AC system
equivalents. The main functions performed by ACREP are :

(i) Synthesis of frequency-matched equivalents for use in dynamic


simulation.
(ii) Calculation and display of the frequency response (Impedance,
phase angle,admittance, ... etc) of RLC networks.
(iii) Spectral analysis. This option allows forward and reverse
Discrete Fourier Transforms to be performed by use of the FFT
algorithm. Two FFT algorithms have been implemented; the
first is the standard, requiring the number of data points to
be some power of two (i.e. 64,128,256,512,1024,2048, ... etc);
the second is a mixed radix FFT allowing less restriction.
Windowing of the data before performing a transformation has
been allowed for. There is also the option to reconstruct the
time waveform from the spectral data by summation of the
Fourier Series.
(iv) A Complex matrix manipulation facility. This allows the series
of complex matrices that represent the frequency response of
a system to be read in and arithmetic calculations to be
performed on them.
(v) Vector arithmetic facility. This allows almost any arithmetic
and/or trigonometric operation to be performed on a set of
real valued numbers stored as vectors on a position by
position basis, where each position represents the response
at a particular frequency. For example this is useful in
derivation of the difference between two waveforms and the
inversion of an admittance vector to obtain the impedance
response.
36,

(vi) General Graphics for displaying data in data files of any


format,
(vii) A Harmonics Graphics Facility which interfaces with the existing
power system programs (i,e, TL, INTER, HARMAC, IHA) to
provide graphical outputs of the content in their data base.

These features of ACREP have already enabled it to be used to


perform many diverse jobs. The flexible graphics facility is probably the
most widely used feature and has been used for displaying a wide variety
of power system data as well as data from electronic experiments such as
noise measurements on a telecommunication line, The ability to display
several waveforms simultaneously has also proved worthwhile.

The spectral analysis facility has been used for the analysis of
earthquake records as well as the time waveforms derived from TCS, The
results may be displayed either in tabular form, as a histogram or as a
two-dimensional plot,

The Complex matrix facility is used to apply the appropriate


transformation to the matrices describing the systems frequency response.

The vector computation facility will be more apparent after reading


chapter 5, where it is used to calculate the required impedance of a
correction filter in the direct method of frequency-dependent equivalent
synthesis, It is also used to provide a direct comparison of Iterative
Harmonic Analysis (IHA) and Transient Convertor Simulation (TCS) in
harmonic assessment, by allowing for differences in their time reference
and p.u. systems.

The calculation of the frequency response of RLC circuits is a


useful feature and has been used (Ireland 1986) in the design of a C-type
filter for a HVDC physical simulator, When constructed, its frequency
response compared very well with the computed estimates.

Examples of graphic displays are two and three dimensional plots of


harmonic voltages and currents, cross-sections of the three dimensional
plots and impedance loci, This is an extension of the work done by Gellen
(Arrillaga et al 1985),
37.

4.2 Objectives, Structure and Operation of ACREP.

At this stage a brief outline of the structure and operation of


ACREP is in order. For a more detailed description refer to the ACREP
MANUAL. The structure of ACREP was designed to provide fast and accurate
operation and minimize the utilization of disc space while being
user-friendly. Although these objectives are not entirely compatible
there is a clear optimum between these constraints. If a user gets
frustrated using the program, it is a clear indication that the happy
medium has not been reached. It is then likely that it suffers from
either not being sufficiently user-friendly, or being too user friendly
and therefore slow in its operation.

The usual practice in the past has been to have separate programs
to perform each distinct major step in the analysis. This therefore
required read and write operations from disc at each step. The speed of
operation can be greatly increased by minimizing the number of read and
write operations. This is achievable by lumping several of the required
steps into one program, as has been done in ACREP. The results of one
step are stored in the program variables (virtual memory), ready to be
used by the next step, rather than writing out a data file and then
reading it in. Virtual memory is the area set aside by the computer for
storing the program variables while the program is executing. When the
program stops execution the contents are lost and this memory is
allocated to store the variables of any another image that starts
execution.

This use of program vectors rather than data files also leads to
increased accuracy and reduced build-up of error. The writing out of a
formatted file windows the data in the program variables by the format of
the write statement, hence causing a loss of accuracy. Where there are
several programs in a chain, each introduces a loss of significance by
writing a data file and the total error builds up. By keeping data in
program vectors maximum accuracy for the given data type is obtained.

However, the danger then exists that by retaining the data in


program vectors allows a large amount of computational work to be lost if
the program stops execution prematurely, due to encountering an error.
Therefore a very comprehensive error-handling capability has been
38.

incorporated into ACREP to make it virtually impossible for it to stop


prematurely, before the results have been stored on disc.

If the error-handler is forced to operate then the program vectors


currently being used are left in an indeterminate state. To circumvent
the problem this causes, there are two sets of program vectors, a data
base and a working set. If the working set is left in an indeterminate
state then it is quickly and easily reloaded from the data base.

4.3 Modular Design.

To produce reliable code which is easily understood and modifiable


the program has been written in a modular manner. Many subroutine calls
are required for each analysis, with each subroutine performing a
specific function. The price of this modular structure is a slower speed
of program execution, however the benefits of tidy and flexible code
out-weight this draw-back. There are 353 subroutines which can be
classified into the following classes, depending on the primary function
they perform :

(i) DCL (Digital Command Language) mimicking.


(ii) Computational.
(iii) Disc I/O.
(iv) User I/O.
(v) Error-handling
(vi) Debugging and Patching.

The DCL mimicking subroutines allow functions normally given in


the computers own command language to be performed from inside the
program. For example editing, printing and deleting files can be done
from within the program, without terminating execution. Other examples
are the changing of the process's priority, the creation of another
process by spawning a process and the listing of files in a directory.

Computational subroutines are those that primarily perform the


required calculations.
39.

The disc 1/0 subroutines are devoted to the reading and writing of
data files. They also perform the additional function of transferring
data between the working set and data base regions of memory.

The User 1/0 subroutines display information on the terminal or


plotter and allow the user to specify commands and parameters. There are
two ways in which the program communicates with the user, the distinction
being made on the basis of whether "text" or "graphics" modes of the
terminal are used for the communication.

An over-view of the ACREP program is given in Figure 4.1.

Local
Program
Vectors

Plotter

DISC =8 Computatioo
Debug
Software --<--

~ ;/
Tenninal

~--
------
Main
Program
Vectors

Figure 4.1 Over-view of the ACREP program


40.

4.4 Special Features.

A versatile program must have the ability to change many


attributes. However to prompt for each characteristic, even with inbuilt
default values present, leads to a program that is slow and laborious to
use. This also increases the probability of entering erroneous numbers.
Therefore the program suppresses the prompts for the characteristics that
are varied infrequently, unless specifically asked for. A vector controls
what attributes the program user is prompted for and hence can vary. The
program user alters the appropriate vector position to enable and disable
the prompting for the various characteristics. Forty-five different plot
characteristics can be varied at present.

Often the coordinates of important points on the plot need to be


determined. Measurement by means of a rule on the plot or screen is less
than satisfactory. This is in part due to the time required to translate
from a distance measurement to the units of the axes, and also to the
very poor accuracy obtained. Therefore the ability to accurately
determine the coordinates of a point by means of a cross-hair cursor has
been incorporated into the program. To display the coordinates of
particular points the cross-hair cursor is first activated, then
positioned on the required point and the space-bar key depressed. The use
of the cross-hair cursor in conjunction with the zoom ability allows very
accurate values to be obtained.

The zoom feature present on many graphic terminals, while allowing


the plot to be enlarged provides no improvement in resolution, as the
enlarged plot is based on the data points originally evaluated. In this
program the zoom recalculates the plot by placing all the data points in
the zoom region, and therefore achieves a far better resolution.

The program has a complete set of default attributes inbuilt, such


that the program user only needs to specify those required to be
different from the default values. These defaults are partitioned into
three sets; those that are data dependent, those that are terminal or
plot device dependent, and those that are independent of both data and
terminal (or plotting device). When using ACREP it is likely that many
attributes need to be different from the default settings and yet only a
few need to be changed at each stage of analysis. The ability to
41.

customize the default settings without editing, recompiling and linking


is an important feature as it would take approximately 33 minutes of CPU
to reform the program. The default settings are customized during program
execution by the use of NAMELIST-directed 1/0 statements (DEC 1984). The
Namelist write produces a namelist file which contains the current
variable values. Next time the program is used the inbuilt defaults can
be superseded by the values used in the last session by reading in the
namelist file.

In a program as large as ACREP, it is not reasonable to expect


every user to know what is going on inside the program, and neither is it
reasonable to expect the program to be entirely free of bugs. For this
reason a PATCH facility has been written to aid debugging. The PATCH
facility performs two functions, the first is to indicate what has gone
wrong, the second is to allow the user to by-pass the problem and
complete the required work. This allows the present user to obtain the
results without requiring a detailed knowledge of the workings of ACREP.
The error should be noted when it appears and given to a person familiar
with ACREP, for correction at a later date. The Patch facility works by
allowing the program variables to be inspected and altered at ~lmost any
stage of the program execution. Besides this, there are very extensive
error detection and notification facilities in almost all facets of the
program's operation.

4.5 Illustrative Examples

4.5.1 Spectral Analysis.

The results can be displayed graphically as a histogram with


either the magnitude, phase angle, real part or imaginary part being the
quantity displayed. The program has the ability to plot the actual values
or the proportion of the maximum value as well as thresholding the data.
42.
0.6

I;

:>
.
0..
. 0.4
v

W
0
:>
I- 0m 2
H
Z
(j
-<
l:

0.0 I I
IZI 5 11Z1 15 2121 25
FREQUENCY (HARMONIC ORDER)

Figure 4.2 Histogram of current harmonics

" . 1. 2
::J
0..
. 0. 8
v

w 121. 4
o
::J
I-
H 121. 121
Z
(j
-< -IZI. 4
l:

I-
Z -IZI. 8
W
!r
!r
:>-1.2
U IZI 6121 12121 18121 24121 3121121 360
T I ME (DEGREES)

Figure 4.3 Reconstructed current waveform

By summing the Fourier series the time reconstruction of a


spectrum is obtained. This is important for two reasons; first, it allows
the time waveform to be inspected when the harmonic data comes from a
table. This is the case with the harmonic assessment programs
(HARMAC<->IHA). Secondly when assessing the harmonics in a time waveform,
the presence of a transient component in the waveform can be tested for
by comparing the time waveform with a time reconstruction of the
harmonics. Figure 4.2 displays a histogram of the current waveform
harmonics while figure 4.3 illustrates its time reconstruction. The
information can also be conveyed as a continuous plot rather than a
1.121

121.8

W
0 121.6
:J
I-
H
Z
!.j 0. 4
-<
2:
0. 2

o. 121 -!I.1.Wl1JJ.W.LLW.Iw.w.

121 5121 100 150


FREQUENCY (HARMONIC ORDER)
Figure 4.4 Histogram of earthquake recording

1. 121 ~----~---r--~------~------r-----~------~------~~

0. 8

w
0 0.6
:J
I-
H
Z
!.j 0. 4
«
2:
0. 2

50 1 00 1 50
FREQUENCY (HARMONIC ORDER)

Figure 4.5 Continuous version of the earthquake recording

histogram as illustrated in figures 4.4 and 4.5, the latter being the
continuous version of the former. In addition to the graphical output the
information can also be written to a data file or terminal screen, in the
form of a table.

4.5.2 Harmonics Graphics Package.

There are normally four files produced by the harmonic penetration


program (HARMAC, section 6.1). Three of them are used for plotting
harmonic voltage, current and impedance respectively. The fourth is a
harmonic admittance file to be used by the IHA program.
44.

Three dimensional diagrams of voltage or current magnitudes, phase


angle, real component and imaginary part can be displayed for either
phase or sequence components. The three possible types of three-
dimensional plots are i isometric, two-dimensional perspective and
three-dimensional perspective. By specifying one of the perspective views
allows the observation point to be varied, which enables the details
previously hidden to be seen.

An example of the isometric view, which is the default display, is


illustrated in figure 4.6 while figure 4.7 shows the same graph from a
different view point using a two-dimensional perspective.

HARMONIC VOLTAGE MAGNITUOES


OEMONSTRATION

w
0 5
~
~
H 4
~
Z rl
~ 3
~
C
~
2
2
W Lru
~
~ 1
~
~
~ 0
0 1 2 3 4 5
>
ORDER OF HARMONIC
Figure 4.6 Isometric Plot

Figure 4.7 Two-dimensional perspective plot


45.
HARMONIC VOLTAGE MAGNITUDES

w
o
::J
f-
H~

Z+l
·rl
19 C
<1::J
L
W L
19 ill
<1~
f-
.J
o
>
ORDER OF HARMONIC

Figure 4.8 Busbar voltage harmonics throughout the Lower South Island
of New Zealand when one per unit of each harmonic is injected

Alternatively a cross-section can be displayed in isolation. With


reference to figure 4.8 the third harmonic cross-section is shown in
figure 4.9 or in a more conventional form in figure 4.10. Many cross-
sections can be over-laid and hence displayed simultaneously as shown
in figure 4.11. Cross-sections of the three-dimensional plot in the xz
plane can also be displayed as illustrated in figure 4.12.

All the phase or sequence components can be displayed


simultaneously or individually. Another option is a point plot which is
illustrated in figure 4.13 (a)&(b) for the three-dimensional harmonic
current plot of figure 4.14.

The harmonics graphics facility allows the plotting of the


impedance loci. In order to produce a smooth loci plot from the data
points in the file the impedance is split into its real and imaginary
components and cubic spline interpolation is used between the data
points. Figure 4.15 shows the impedance loci obtained by this method.
Finally, a magnitude against frequency plot can be plotted as illustrated
in figure 4.16.
46.
HARMONIC VOLTAGE MAGNITUDES
3 RD HARMONIC

w
o 1.0
:J
I-
H~

Z.j.J
.r!
l'J C
i::J 0.5

W L
l'J UJ
<t:-~
I-
-.J 0.0
o
>
ORDER OF HARMONIC

Figure 4.9 Third harmonic Cross-section in isolation

0.5

.........
r 0.4
H
Z
:J
a: 0.3
W
n.
0.2
W
(!)
<t:
r 0.1
-.J
0
>
0.0
1 2 3 4 5 6 7 B 9
BUSBAR NUMBER

Figure 4.10 Third harmonic Cross-section

The magnitude of the following quantities can be displayed against


frequency, impedance, phase angle, resistance, reactance, admittance,
conductance or susceptance. Cubic spline interpolation on each component
is also used for these plots and both phase or sequence components can be
inspected.
47.

0.5

t-
H 0.4
Z
:J
..... .. '
.. ' .. ,
a: 0.3
W .
.....I 1'...,.
[l. . I ..~ I

0.2 ,,~ /'~ :....


/1
W
(!) ./ " .......... /
/" .......... ,
'.~- -- --,
..................'..
:' .-/-.
....~.~ I
4: ~"
t-
-l
0
0.1
,_.__
.-----. "" / . ..........
~. - - .
,':"?
:..--::-:::,.
I

>
0.0
1 2 3 4 5 6 7 8 9

8US8AR NUMBER

Figure 4.11 Multiple cross-sections

HARMOt'-1 I C VOLTAGE MAGNITUDE


1.0 BUS BAA - AOXBURGH-011

W
0 O.B A PHASE
:J
I-
H~

Z.j.J
('Jrl 0.6
<t C
L:J
W L 0.4
('J I]J

<{~
I-
~
0.2
0
>
0.0
1 5 9 13 17 21 25

ORDER OF HARMONIC

Figure 4.12 Cross-section of the harmonics at a busbar


CURRENT MAGNITUDE (Per Un 1 t) CURRENT MAGNITUDE (Per Unit)
o 0 ...,. ...,. f\)

0- o ...,. f\) PJ o ~ 0 ~ 0

C!.l 'V
CD ::s
.0 AOXElUAGH-220
Pl ROxeURGH-Z20
'"rJ C Ul
TO TO
1-'- CD RQXeURGH10lil
CD ROXeURGH:1 0 11
()q ::l
C 0
'J CD ;p
=
ROXeUAGH-O 11 AOXeUFlGH-O 11
CD TO TO
0 AQX15URGH-220
= ROXBUAGH-220
I
.!=" 0 I
a l> »
'D ]] :n
I-IANAPOURI220 MANAPQUR.I220
W 0 3: TO -
3:
TO a
::l MANAPOUR101lA a MANAPOUA" 0 14
CD z Z
'V ...,. H ...,. H
::l
0 cT ....,j [1 ---J [1
1-'- MANAPOUAX:ii:ZO MANAPQUAI2ze
Ul TO rt TO rt
::l MANAPOUR201A :r [1 OJ MANAPOUA201A "J [1
cT OJ
]] c ]] C
'V l> I ]] l> I ]]
i-' Z !NVERCAAG033 l> ]] Z If'oIVERC.a..RG033 l> ]]
[1 TO [1 TO - ]] rn
0 INVERCARGZ20
]] III INVERCAAG220
cT I 3: Z I 3: Z
III a -j III a -j
0 (JJ (JJ Z
....., INVERCARG220
Z :INVERCARG220
H 3:
H 3: TO
TO [1 l>
MANAPOURI220 [1 l> MANAPOURI220
0 (j) (j)
C Z Z
'J H H
'J INVERCARG220 INVEACAF'lC220
-j -j
CD TO TO -
M.4.NAPOURI220 C MANAPOURI220 C
::l 0
cT 0
III III
::r: INVERCA.RG220 INVEI=ICARG220
PJ TO TO -
ROXBUFIGH-220 AOXBUROH-220
'J
a0
::l INVERCARG220 INVERCAAG220
1-'- TO TO -
AOXeUAGH-220 FlOXBU~GH-220
0
Ul

INVERCARG220 INVEF\CA~G220

TO TO -
TII-IA.I----220 TIWAI----220

INVERCARG220 ! g INVERC:AAG220
TO TO -
TlwAI----220 ~ T:IWAI----220
.
"
! "~
MANAPOURI220 MANAPOUAl:220

TO TO
TIWAI----220 TlwAI----220

MANAPOURl220 M.a.NAF'OUAI220 .!="


TO TO en
T.I'rIAI----220 TlwAl----220
49.

HARMONIC CURRENT MAGNITUDES

w
0 3
~
I-
H~

Z-iJ
·rl 2
~ C
<t::J
L
I- L 1
Z QJ
w9::;
II
II
~ 5 9 13 17 21 25
U
ORDER OF HARMONIC

Figure 4.14 Branch Current Harmonics

HARMONIC VARIATION OF IMPEDANCE


jX
600

500

400

300

200

100

o
800
-100 R
5

-200

-300

-400

Figure 4.15 Impedance Loci


8121121 50.

"(j)
E 6121121
..c
0
v

W 4121121
U
Z
-<
0
W 2121121
0..
l:
I-t

121
5121 25121 45121 65121 85121 11215121 125121
FREQUENCY (H",,)

Figure 4.16 Impedance versus frequency plot

4.5.3 Frequency-Response of an RLC Network.

The algorithm developed for calculating the frequency response of


an RLC network is based on the solution of a ladder network. The network
must therefore be representable as a ladder network, as shown in figure
4.17. General filter and series branches are displayed in figure 4.18 (a)
&(b). Other simpler filter branches are modelled by eliminating the

F F F F

Figure 4.17 Ladder Network

(a) Series branch

(b) Filter branch

Figure 4.18 General branches

T
51.

~
I I
Figure 4.19 Frequently used models

appropriate components from the general filter branch. This is achieved


by specifying a zero component value for these components.

The series branches are handled in the same manner. The more
frequently used filter branches (and subnetworks) have their own separate
model and hence data requirements, thereby allowing a simpler and more
economic data input. Figure 4.19 displays these more frequently used
models.

Both impedance loci or "quantity" versus frequency plot can be


displayed, where the quantity to be plotted is either the impedance,
resistance, reactance, phase angle, admittance, conductance or
susceptance. In contrast to when the actual system's response is
displayed, no interpolation is used as the data pOints to be plotted can
be generated at any required frequency spacing.

The program can also plot the response of an RLC network together
with the required response, stored in a file, thereby allowing easy
comparison. This shows the accuracy of a frequency-matched a.c. system
equivalent.

Another feature is the ability to obtain the over-all frequency


response of the parallel combination of the a.c. system response and an
RLC network. This is useful in observing the effect of adding harmonic
filters to the a.c. system's frequency response.
52.

(1) (2 ) (4 ) (5 )

R3 R4

Ll L2 C
L3 L4 s

c C C C C
1 2 3 4 s
I-- -- I I
- -
Figure 4.20 RLC circuit

As an illustration the impedance loci of the circuit displayed in


figure 4.20 (circuit parameters given in appendix A3), is shown in
figure 4.21 and the impedance, resistance and reactance response
displayed in figures 4.22 (a),(b)&(c) respectively.

BOO

600

ADO

200

o ~----~rT--~----~-----+~----~--+--J
400 1200

R
-:200

-400

-600

-800

-1000
Figure 4.21 Impedance loci of RLC circuit
53.

U)
L
I 800
o

UJ
U
z
« 400
o
w
D..
L
H

o
50 450 850 1250
FREQUENCY (Hz)
(a) Impedance versus frequency plot

1200

---UJ
L
I
o 800

UJ
U
z
«
r 400
UJ
H
UJ
W
II
o .J \. .J \... J \...
50 450 850 1250
FREQUENCY (Hz)

(b) Resistance versus frequency plot

800

UJ
L 400
I
o

W 0
U
z
«
r
u -400
«
W
II

-800
50 450 850 1250
FREQUENCY (Hz)
(c) Reactance versus frequency plot

Figure 4.22 Frequency Response of RLC circuit.


54.

4.5.4 General Plot Facility.

A maximum of 25 waveforms can be displayed simultaneously, of


which some can be the sums or differences between waveforms, with freedom
to alter the dimensions of the plot and the drawing and labelling of
axes. The default is a 30x20 cm plot on an A3 page, however there are
five other standard options. Alternatively the program user can enter his
own non-standard plot dimensions. To distinguish between the various
waveforms different pen colours are used in conjunction with differing
line type. There are seven standard line types while the plotter limits
the number of available colours (typically 4 or 6). The user can also
choose between the two core graphics packages available on the computer
system, to obtain the plot. As an example of the capabilities of the
general plot facility, figure 4.23 shows six waveforms plotted
simultaneously.

0.50

0.40

lJJ
0 0.30
:J
I-
H
Z 0.20
t!J
«
:r
0.10

1 2 3 4 56789 10 11 12 13
X Variable

Figure 4.23 General Plot facility display of six


waveforms simultaneously
CHAPTER A.C. SYSTEM MODEL.

5.1 Historical Review.

It is well established that time-domain dynamic simulation is


required for accurate representation of non-linearities such as HVDC
convertor operation, corona losses, transformer hysteresis and
saturation. This makes frequency-dependence modelling difficult because
in a step-by-step numerical integration algorithm all frequencies are
effectively being simulated at each time step.

The transmission line is the predominant frequency-dependent power


system component, and many have concentrated on modelling its
characteristics. Hingorani (1970) showed the importance of the
frequency-dependence of line parameters on the transient by considering a
pole-to-ground fault on a bipolar HVDC scheme.

Budner (1970) developed one of the earliest frequency-dependent


line models by convolving weighting functions with the past history of
node voltages. The weighting functions were derived by applying an FFT
algorithm to obtain the inverse Fourier transform of the frequency
dependent admittance matrices. Snelson (1972), Meyer and Dommel (1974)
and Carroll and Nozari (1975) made further developments in this area. To
increase computational speed the use of recursive convolution was
proposed by Semlyen (1975), Dabuleanu and Semlyen (1975) and Ametani
(1976).

Another approach to modelling frequency-dependence has emerged


based on approximating the transmission line transfer function by
rational functions. Once a rational function approximation has been found
it may be made amenable to direct incorporation into state equations or
realized by a lumped R,L and C network. Semlyen (1982) proposed direct
incorporation as state equations while lumped parameter realization was
suggested by Marti (1982,1983), Yen et al (1982), Semlyen and Deri
(1985), and Naidu anD de Lima (1985).
56.

The problem of modelling the frequency-dependence of transformers


in transient studies has been tackled by Avila-Rosales and Alvarado
(1982). They reviewed seven existing transformer models, none of which
modelled the frequency dependence of the transformer.

Most previous approaches concentrate on modelling the frequency


dependence of individual transmission lines and transformers. The
approaches are based on differential equations, such as the telegrapher's
equations for a transmission line, and are therefore not applicable to an
arbitrary frequency response. The complexity of power systems makes
individual line and transformer representation unreasonable and
frequency-dependent a.c. system models necessary.

Hauer (1981) reported a more general approach to modelling


transmission lines by curve fitting in the frequency domain via
non-linear optimization with a least squares objective function. However,
the details of the fitting process were not elaborated on.

Hingorani and Burbery (1970) were the first to propose the use of
an RLC network to represent the frequency-dependence of an a.c. system.
Their paper described the derivation of the network parameters so that
the modelled frequency response matched the required response and
discussed the numerical problems that were encountered. Figure 5.1 shows
the equivalent circuit proposed by Hingorani and Burbery (1970). The

, - - - - - . - - - -..................... ---.-----1

I I
Figure 5.1 A.C. System Equivalent proposed by
Hingorani and Burbery
57.

stages of the circuit synthesis were ; first the required Land C values
are determined based on a lossless model (ie neglecting the resistance in
the model of Figure 5.1) ; resistances are then added which correspond
with the real part of the impedance at the appropriate resonant
frequencies; finally Land C values are scaled so as to give the correct
impedance level at the peaks while maintaining the correct resonant
frequencies,

Figure 5.2 Equivalent proposed by Bowles

Bowles (1970) proposed the simpler equivalent displayed in figure


5.2 using a T-network which has the same impedance as the short circuit
impedance at fundamental frequency. The values of Rand L were selected
to give the required impedance angle. Bowles claimed that Hingorani and
Burbery's equivalent was too complicated for HVDC simulation as only the
th
impedance at low frequencies (up to the 5 harmonic) are of importance
for most purposes because the harmonic filters swamp the a.c. impedance
at higher frequencies. Other authors have also echoed the sentiment that
Hingorani and Burbery's equivalent is unduly complicated (Reeve and Subba
Rao 1974). Kruempel and Reitan (1970) and Bowles and Clarke (1970)
criticized Hingorani and Burbery's equivalent due to the need for the
impedance-frequency locus, which was difficult to obtain at the time.
However, this is no longer a problem and section 5.2 will discuss its
derivation, Reeve (1970), Bowles and Clarke (1970) commented that the
impedance-frequency characteristic changes significantly due to
switchings in the system on a routine basis, However, any representation
is subject to this draw back. Representation for typical conditions is an
58.

improvement on hypothetically and arbitrarily selected conditions.


Morched and Brandwajn (1983) proposed an equivalent very similar to
Hingorani and Burbery's equivalent and almost identical LC parameter
determination based on a lossless model. The resistances are determined
iteratively with the real part of the impedance in the appropriate
resonant frequency used as the initial estimate, although no information
about the iterative loop was supplied by the authors. Do and Gavrilovic
(1984) extended this work by using least-squares fitting in an iterative
algorithm to obtain an accurate equivalent when system losses were high,
and applied it to zero and positive sequence equivalents. Again no
details of the fitting process were given. Watson et al (1985) presented
Hingorani and Burbery's approach with a few refinements and applied it to
three phase dynamic studies with implicit representation of mutual
coupling using TCS. More recently Watson and Arrillaga (1987) have
applied optimization techniques for synthesizing a.c. equivalents with
explicit representation of the mutual coupling.

5.2 Obtaining the frequency response of an a.c. system.

Before any frequency-dependent model can be produced the system's


frequency-response must be determined. This may be accomplished by
harmonic impedance measurements at a number of frequencies on the system
itself or by calculation on a digital computer over a range of
frequencies (Breuer et al 1982, Densem 1983). The disadvantages of direct
measurement are :

(i) the system must be left in an operating condition and hence the
numerous outage conditions which are liable to give rise to
resonances at harmonic frequencies cannot be studied,
(ii) the need to make measurements while the system is energized
demands the use of a high power source of harmonics,
(iii) the measurements are restricted to a few harmonic frequencies and
do not give a satisfactory picture of the resonances at
intermediate frequencies,
(iv) the accuracy of measurements made on the actual system are prone
to significant experimental error.
59.

The frequency response of a power system at the point of harmonic


injection is normally presented in the form of a single impedance locus
plot. However, to accurately represent network asymmetries and mutual
coupling requires a 3x3 impedance matrix for each harmonic. The Harmonic
Penetration Program HARMAC developed by Densem (1983) has the ability to
generate the required series of 3x3 impedance matrices that describe the
frequency-dependence of the a.c. system. To display the enormous amount
of information that results, a generalization of the impedance loci is
used.

The interaction of the various programs required to generate and


use the impedance loci is illustrated in figure 5.3 . Referring to figure
5.3. the first program in the chain is the TL (Transmission Line)
program. This program reads in the line geometry and conductor data and
calculates the transmission line parameters for each frequency over the
required range using an Equivalent n model. This incorporates Carson's
correction (1926) for earth return and the correction for skin effect in
the conductor. The data base is completed by the INTER (INTERactive)
program. This is achieved by reading the line data produced by TL and
adding it to the other component data. HARMAC (abbreviation of AC
HARMonics) is then used to calculate the three phase system impedances at
the required frequencies. This calculation only uses the first stage of
HARMAC, the second stage is used specifically for harmonic assessment
and is discussed in Chapter 6. The primary function of the ACREP (AC
system REPresentation) program is to input the systems frequency response
and derive an equivalent circuit with a matched frequency response. This
equivalent is then used in Transient Convertor simulation (TCS). The
JAUGPS (Generate Plot Store) program then plots the TCS output.

As an illustration the frequency response of the self and mutual


terms of the New Zealand lower South Island test system in figure 5.4,
observed from the Tiwai bus. are shown in figure 5.5. Because power
systems are generally bilateral, the (n,m) loci are seen to be identical
to loci (m,n). Moreover for the particular line geometry of the test
system, two phases, "A" and "C", are seen to be similar while phase "B"
differs substantially.
Conductor data
and line Tmetry
Transmission Line Interactive 2 Harmonic 3 A.C. System
Parameter Program ---..::;>,,", I Data Program Penetration Program )- Representation Program
(TL) (INTER) (HARMAC) (ACREP)

6
5 1
Power-Flow 7
4 1
Transient Convertor
Program Simulator
(PF) (TCS)

8 1
Graphics and Data
manipulation Program
(JAUGPS)

Parameters of each transmission line over the required frequency range


2 Complete a.c. system data at fundamental and harmonic frequencies of each individual component
3 3x3 impedance loci matrix
4 Frequency-matched a.c. system equivalent
5 a.c. system data at fundamental frequency
6 Remainder of hvdc system data
7 Initial currents and voltages throughout the hvdc system
8 Voltage and current waveform data at unequally spaced intervals

Figure 5.3 Data Flow and sequence for a dynamic study using
a frequency-matched a.c. system equivalent.
0'1
o
61.

14 kV kV

~lanapouri Roxburgh

220 kV 220 kV

90 MW 54 MVAR

220 kV

220 kV

Figure 5.4 The Lower South Island of New Zealand test system

Some dynamic models lack mutual coupling capability and in such


cases it is possible to reduce the full matrix information to three
diagonal terms which contain the mutual impedances implicitly. This is
preferable to using the self terms as at some harmonic frequencies the
mutual terms have a larger effect than the self term. An assumption
regarding the phase sequence must be made in order to perform the
reduction. By way of example, assuming a positive sequence current then
the reduction is achieved by post multiplying each of the 3x3 matrices by
the matrix

1 axa a

a 1 axa where a -1/2 + j 13'12

axa a 1
62.

XW) XWl X(\1)


350 350 350

Zaa Zah Zac

150 15

R(S< )
-50 -5

-250 -260 -250

350 350 350

15 HIO

-5 -50

-250 -250

350 350 350

15 150

-5 -50

-250 -250 -250

Figure 5.5 Impedance Loci Matrix of the test system

X(\1) X((l ) xW)


375 375 375

Za Zb Zc
250 250 250

125 125 125

RW) RW)
0 0 0 --'---'

-125 -125 -125

-250 -250 -250

Figure 5.6 Diagonalized Matrix Impedance Loci


63.

and extracting the diagonal terms. This method is valid for systems with
little phase current asymmetry, as is normally the case with convertor
plant. Applying this method to the above test system results in the three
loci displayed in figure 5.6.

When comparing the frequency response of an equivalent circuit with


the measured or computed response, impedance/frequency plots are more
easily comprehended and used rather than the impedance loci. These are
illustrated in figure 5.7 for the system of figure 5.4.

5.3 Synthesis of Frequency-Dependent Equivalent.

The development of a multi-phase frequency-matched a.c. system


model requires an algorithm that can produce an equivalent circuit
matched to an arbitrary frequency response. Two different techniques have
been developed to perform this; the first is a direct method based on
Hingorani's method, the second on non-linear optimization. Thus the
complex impedance-frequency information contained in the loci matrices
are converted into equivalent circuits. Each element of the 3x3 matrix
must be synthesized into an equivalent circuit, which are then combined
to form the overall frequency dependent equivalent model for the a.c.
system, the structure of which is shown in figure 5.8. The equivalents
matching the self terms are represented explicitly as shunt branches
while the mutual circuits are processed separately by using diakoptical
techniques. The phase currents are first impressed upon the mutual
circuits and the resulting voltages are then incorporated in the overall
circuit as additional voltage sources. The computer implementation of
this frequency-dependent equivalent model is reviewed in section 5.4.

The equivalent circuit topology is selected to suit the requirements


of the time-domain algorithms, such as the use of node-type partitioning
and tensor matrix analysis (Arrillaga et al 1983a), and is discussed in
chapter 3. Although these considerations restrict the possible
topologies, they dramatically increase the solution efficiency in the
case of a.c./d.c. dynamic simulation, which require regular topological
changes, by avoiding involving the whole network during the localized
(n) (n) (n)
500 600 500

500 Z 500 Z 500 Z


aa ab ac
0400 400 400

300 300 300

200 200 200

100 100 100

o 0 a
50 450 850 1250 50 450 8:50 1250 SO 450 650 1250

I II
(n) (nl (nl
sao 600 600

,00 '\,a ,00 ~ '\,b .,0


400 400 400

300

.,0
100

o
00 .4:50 650 12~O
300

.,0
100

a
vV
50 450 SESO
\d'00
300

100

12:50:S0
a
450 B:50 12:10

(nl (nl (nl


~O ~o ~O

500 Zca 500 Zcb 500 Zcc

ADO 400 400

300 300 300

200 200 200

:100 100 100

o a a
!so 450 650 1250 50 450 850 1250 50 450 850 1250

Figure 5.7 Impedance versus Frequency Matrix derived from figure 5.5
0'\
-I::"
65.

I
a +
r-------------~--~ 2:aa~-----<

Zba1a + Zbc1c
Ib
+
2:bb
Z I
ca a + Zcb1b
I
c +
"-
2: cc

Figure 5.8 Structure of a frequency-matched a.c. system equivalent

convertor valve switchings. Since the topology used by Hingorani and


Burbery (1970) is amenable to efficient Transient Convertor Simulation
this form has been retained as the starting pOint in the derivation of
the equivalent circuit.

The basic steps of the equivalent circuit synthesis for both


alternative methods are illustrated in figure 5.9 and discussed in the
next two sections. The acceptability of an equivalent is a subjective
decision based on the following factors
66.

Direct Path optimization Path

Use measurement or analysis to


obtain the system frequency
response (selfs and mutuals)

1 t
Use Hingorani's topology Decide equivalent
and direct synthesis to circuit topology and
derive a multi-phase optimize the circuit
equivalent parameters
~ ~
Is the equivalent .....
optimize
circuit acceptable

No Yes No
~

Design a correction Add another branch to


branch the equivalent circuit
} I
~.

·1 Matching Completed I

Figure 5.9 Flow Diagram for the Direct and Optimization Algorithms

(i) The type of study. For steady state harmonic penetration studies
minimal error at the harmonic frequencies is required while the errors at
intermediate frequencies are unimportant. This is clearly unacceptable
for transient convertor simulations as some intermediate frequencies
could be excited.
(ii) The feasibility of altering automatically the equivalent circuits
to reduce the error to a prescribed accuracy.
(iii) The availability of an economically viable alternative to model
the actual system explicitly. This relates to the amount of CPU time
required to provide the simulation and accuracy required.
67.

The CPU time used in a.c./d.c. simulation is related to the number


of state variables and the time steps of the numerical solution, the
latter being related to the time constants of the equivalent circuit. In
each case the acceptability of a particular equivalent circuit and the
required CPU time are assessed by means of a short-duration dynamic
simulation study.

5 •• 1 Direct Method.

The basic matching philosophy consists of selecting the values of


R,L and C that give the peaks and troughs at the correct frequencies and
the Q (quality factor) of each branch such as to cause the equivalent
circuit's response between these frequencies to approximate the actual
system's response. The mathematical formulation used in initial
assessment of the RLC values to give the peaks and troughs at the correct
frequencies is as follows.

The first step is to obtain values for the inductors and capacitors
of the equivalent circuit based on a lossless model. The resistance of
the branches are ignored as they do not affect the resonant frequencies.
The admittance of an n branch network is
n n
s I {( 1 /L ) II (s 2 + w2 )
n l
s k=l k l=l .. k
yes) I }
n
(5.1)
k=l 2
L (s + W )
2
(s 2 + wi)
2
k k ill1

and s = jw

For an LC network the impedance is zero at a minima and infinite at a


maxima. Impedance minima occur when :

n
II o (5.2)
i =1

Impedance maxima will occur when the numerator of equation (5.1) is


zero. i. e.

n n
s I {(l/L) II o
k=l l=l;tk
68.

Let F1 ,F
, ... F be the frequencies at which the admittance is
2 n
zero. Then equation (5.3) can be written as :

s o (5.4)

where M.=21fF.
1 1

By equating the coefficients of s in equation (5.3) with equation


(5.4) the following set of equations result:

Coefficient for S2n-1

n
I (1/L )
k
(5.5)
k=1

Coefficient for S2n-3

n n n-1
2 M2
I { (1/L )
k I w.
1
I k
(5.6)
k=1 i=1 k=1
.,k

Coefficient for S2n-5

n n n n-1 n-1
2 2
I ( 1 ILk) I I wiw j } I I M~M~
1 J
(5.7)
k=1 i=1 j =1 i=1 j =1
.,i .,i

Coefficient for S1

n n n
2
I {( 1/Lk ) IT wi} IT M2
k
(5.8)
k=1 j=1 k=1
.,k

Thus n equations are formed with 1/L ,1/L , ...• 1/L as the
1 2 n
unknowns to be solved for. i.e.
[A] x b (5.9)

where
t
x [1 IL1 ' 1/L 2 , .... 1 ILn]

n-1 n-1 n-1 n


t
b = [1 ,
- I Mk , I I MkM i ,···· . IT M ]
k
k=1 i=1 k=1 k=1
.,i
69.

n n n-l
2 2 2
I w.
1
I w.
1
I w.
1
i=2 i=l i=l
*2
A

n n n-l
2 2 2
II w. II w. II w.
1 1 1
i=2 i=l i=l
*2

This is solved using Gaussian elimination with partial pivoting


and the result reciprocated to give the inductor values. To obtain the
appropriate capacitor values equation (5.2) is used. The capacitor values
are given by :

(5.10)
where k=l .2 •... n
The admittance of the tuned network at any particular frequency is
given by:

n n R - j (Lk/w)
k
(i- 1 I (LkC ) )
k
Y Y I I 2
(5.11)
k=l k k=l R2+ [(Lk/w)(w - 1 I(LkC )) J2
k k

Branch k is in series resonance when w = I (l/C L ). The


k k
corresponding admittance term for the k branch is l/R . This term is
k
significantly larger than the contribution from the other branches and
hence dominates the admittance. The resistance that dominates at each
impedance minima is inserted in the corresponding branch. This yields
impedance maxima that are higher than those of the actual system. This is
corrected by scaling all inductor and capacitor values. A common scaling
factor is used for inductors and capacitors so that the resonant
frequencies remain unchanged. The appropriate scale factor is obtained by
dividing the calculated impedance of the RLC network at a particular
frequency by the actual systems impedance at this frequency. Some
discretion needs to be exercised in determining what frequency is to be
used to determine the scale factor. The branch resistance dominates the
RLC network's impedance at a minima and maxima, therefore choosing a
match frequency in their close vicinity will produce poor results.
70.

If resistance is ignored Equation (5.11) reduces to

n
w
y - j I (5.12)
k=l

This shows that, as Lk appears in the denominator, to increase the


admittance contribution of each branch, so as to reduce the impedance,
the inductors are divided by the scale factor. In order to maintain the
same resonant frequencies the capacitor values are multiplied by the
scale factor.

A number of numerical errors arise from the direct application of


the above equations. Floating point overflow occurs when trying to
calculate the coefficients of s due to the finite range of numbers a
digital computer can represent. Angular frequency (w) is in the range 0
to 15000 rad/sec and the coefficient of sk is of the order w2 (k-1) , k~l,
where there are n parallel branches whose parameters are to be
determined. To overcome this difficulty some of the coefficients
(elements of matrix A; equation 5.9) are automatically scaled down by a
fixed scale factor before their calculation overflows the number
representation. This ensures that the resulting coefficient remains
within the required range of values for the computer to represent;
knowledge of the scale factors allows the actual coefficient to be
determined.

This crude scaling using an arbitrary scale factor allows the


coefficients (elements of matrix A) to be formed, however, the values of
the coefficients in the matrix equation 5.9 still differ widely between
equations. Solving the system of equations directly as they stand results
in attempts to divide by a number approaching zero and failing to give a
satisfactory solution. This is overcome by scaling each equation, so
that the maximum coefficient in the equation has a magnitude of 10 (This
means scaling each row of matrix A to have a maximum value of 10).
Gaussian elimination using partial pivoting is then used to solve for
l/L ; k=1,2, ... n
k
71.

Feature Extraction.

The feature extraction algorithm used for determining the parallel


and series resonant frequencies differs from that used by Hingorani and
Burbery (1970). Hingorani and Burbery used the frequencies at which the
impedance loci crossed the Real axis (or X axis) as the resonant
frequencies and starting point for the synthesis algorithm. This does not
give satisfactory results when some of the loops in the loci are not
centred about a point on the real axis and completely misses loops that
fail to cross the real axis. Instead the minimum and maximum impedance
pOints of each loop in the impedance loci are taken as the resonant
frequencies.

Figure 5.10 displays the phase "B" loci based on equal current for
an equivalent circuit formed using the intersection of real axis feature
extraction (c.f. figure 5.6). The first loop between 260 to 290 Hz is

HARMONIC VARIATION OF IMPEDANCE


j X
450

300

19

150

480 600
R

-150 5

-300

Figure 5.10 Impedance Loci of the equivalent based on intersection


with x axis feature extraction
72.
HARMONIC VARIATION OF IMPEDANCE
j X
450

300

19

150

480 600
R

-150

-300

Figure 5.11 Impedance Loci of the equivalent based on new feature


extraction

completely missing while the loop consisting of the loci from 500 to 860
Hz does not correspond well with the required loop (displayed in Figure
5.6), as the latter only clips the real axis. Using the new feature
extraction routine the loci of figure 5.11 is obtained. Although the
first loop is too large it can easily be improved by adjusting the
resistance values.

Although the impedance loci display both magnitude and phase


information it is probably more intuitive to display the frequency
response as impedance versus frequency plots. The troughs in these plots
correspond with series resonances and the peaks with parallel resonances.
They provide a better idea of which parameter, and in what direction,
they need to be changed to improve the response. However, two plots
(magnitude and phase plots) are required to display the information
contained in the impedance loci. The impedance versus frequency plot for
the new feature extraction method is illustrated in Figure 5.12, By
inspection the second peak can be seen to be too large while the previous
73.
600

500 REQUIRED RESPONSE


~

W SYNTHESIZED CIRCUIT'S RESPONSE


L
I 400
av
W 300
U
Z
< 200
a
w
~
L 100
H

0
50 450 850 1250
FREQUENCY (Hz)
Figure 5.12 Equivalent cct based on new feature extraction

600

A 500
W
L
I 400
av
W 300
U
Z
< 200
0
W
~
L 100
H

0
50 450 850 1250
FREQUENCY (Hz)

Figure 5.13 Response of modified equivalent cct.

trough too low. By increasing the resistance on the first RLC the
trough's minimum value increases and the parallel resonance peaks on
either side diminish, hence the resistance of the RL branch needs to be
decreased to maintain the level of the first peak. Figure 5.13
illustrates the response after these two resistances have been altered.
This gives a better match across the frequency range, but may cause
slight errors in the fundamental frequency (50 Hz) parameters. This
discrepancy, due to the compromises made to get the troughs and peaks of
the correct size, can easily be corrected by the inclusion of a
correction filter, as will be discussed later.
74.

The Scaling Process.

The original scaling process relied on the user specifying a match


frequency and the impedance at this frequency; the algorithm then scaled
the L's and CiS to achieve the specified impedance level at this
frequency. Although this worked well it required expert judgement of what
frequency point to match to. Usually this involved several attempts at
scaling as once one choice had been made and the response compared to the
actual frequency response, a better choice of match frequency became
apparent. Near the troughs the impedance level is relatively insensitive
to scaling as the resistance dominates in these areas, hence the match
frequency should not be in this region. The peaks are also very dependent
on the resistance, so the match frequency should not be specified to
coincide with or be in close proximity to the peak. Single variable
optimization techniques have been developed and applied to obtain the
best scale factor and this is now the preferred method of obtaining the
scale factor. The three different optimization techniques that have been
applied are
(i) Dichotomous search,
(ii) Golden section search,
and (iii) Success/failure step search.

All three performed equally well, although unlike the first two,
the Success/Failure algorithm has the advantage that it does not need two
initial guesses. A discussion of these three optimization techniques is
given in appendix A4. Although the optimization technique determines the
efficiency of solution, it is the objective function's definition that
determines the optimum point and hence ultimately the extent of success
of the optimization technique. Of the objective functions tried the least
squares function has been found to be the most satisfactory. Applying the
least square technique to the total frequency range is undesirable. A
better fit may be achieved by accepting a large initial error in a
particular region and using a correction branch at a later stage. To stop
regions that are going to be corrected by subsequent use of correction
filters detrimentally affecting the scale factor, the frequency range
can be partitioned into up to five regions, with different weighting
factors for each region. Different weightings can also be given depending
on the sign of the error.
75.

HARMONIC VARIATION OF IMPEDANCE


j X
450

300

19

150

480 600

-150

-300 (a) Impedance Loci

6fZH21

A 5121121 REQUIRED RESPONSE


III SYNTHESIZED CIRCUIT'S RESPONSE
L
I 4121121
a
v

W 3121121
U
Z
< 2121121
0
W
11.
L 1121121
1-1

121
5121 45121 85121 125121
FREQUENCY (Hz)

(b) Impedance versus Frequency

Figure 5.14 Impedance response when the scale factor is optimized


76.
lfZlfZlfZlfZlfZlfZlfZl

Z
0
H 81ZlfZllZllZllZllZl
I-
U
Z
::J 61ZllZlfZllZllZllZl
LL
W
>
H
41Zl1Zl1Zl1Zl1Zl1Zl
I-
U
.,
W 21Zl1Zl1Zl1Zl1Zl1Zl
CD
0
IZl
51Zl lfZ1lZl 151Zl 21Zl1Zl 251Zl
SCALE FACTOR
Figure 5.15 Objective function in conjunction with intersection
with x axis feature extraction

z
8 81Zl1Zl1ZllZlfZllZl
I-
U
Z
:J 61ZlfZllZllZlfZllZl
LL
W
>
H
41ZlfZllZllZlfZllZl
I-
U
~ 21ZlfZllZl1Zl01Zl
rn
o
1ZlL---~--~~~==~==~--~--~--~
51Zl llZllZl 151Zl 21Zl1Zl 251Zl
SCALE FACTOR

Figure 5.16 Objective function in conjunction with new feature


extraction

Figures 5.14 (a)&(b) illustrate the response of a frequency-


dependent equivalent matched to phase "B" of the actual system where
implicit accounting of mutual coupling is present (c.f. loci displayed in
figure 5.6) and when the scale factor is optimized. Care must be taken to
ensure that the global minimum and not just a local minimum is found, as
several minima may exist. This is illustrated in Figure 5.15 which shows
the variation of the least square objective function for different scale
factors for the (1,3) locus of the full 3x3 loci matrix,(in conjunction
with Hingorani and Burbery's original feature extraction technique). The
77.
612HZ!

A 5121121
Ul
2:
I 4121121
0
v

W 3121121
U
Z
-<
0 2121121
W
lL
2: 1121121
H

.. .
'
121
50 45121 85121 1250
FREQUENCY (Hz)
(a) Equivalent based on intersection with x axis feature
extraction

A 5121121
Ul
2:
I 4121121
o
v

W 3121121
U
Z
-<
o 21210
W
lL
2: 11210
H

o
5121 450 850 1250
FREQUENCY (Hz)
(b) Equivalent based on new feature extraction

Figure 5.17 Impedance Response

featUre extraction has a great effect on the objective function as shown


by Figure 5.16, which corresponds with the (1,3) loci in conjunction with
the new feature extraction algorithm. Figure 5.17 (a)&(b) illustrate the
comparison between the frequency responses of the actual system and the
equivalent circuit using the original and new feature extraction
algorithms respectively. The corresponding figures 5.18 (a)&(b) show the
same comparison of frequency response when the scale factor has been
optimized rather than directly assessed.
78.
61Z10

r. 51Z11Z1
U1
2:
I 41Z11Z1
a
v

W 31Z10
U
Z
«
0 21Z10
W
!l
2: l1Z10
H
..........
0
50 451Z1 851Z1 1250
FREQUENCY (Hz)
(a) Equivalent based on intersection with x axis feature
extraction and scale factor optimized

r. 51Z10
U1
2:
I
av 41Z10
l.JJ 31Z11Z1
U
Z
« 21Z11Z1
0
W
!l
2: llZl0
H

...........
IZI
50 451Z1 851Z1 1251Z1
FREQUENCY (Hz)

(b) Equivalent based on new feature extraction and scale factor


optimized

Figure 5.18 Impedance Response

With regard to loci (2,1) of the test system (figure 5.4),


figure 5.19 shows the effect of optimizing the scale factor over the
whole frequency range from 50 to 1250 Hz with a constant weighting
factor. There is a large peak at 884 Hz which requires the use of
correction filters, and this peak has adversely affected the optimal
solution. This results in a large discrepancy for the first peak.
Choosing a weighting factor of zero for the region around the peak stops
this peak influencing the optimum scale factor, giving a more favourable
79.
61210

r.. 51210
U1
L
I 400
0
v

W 31210
U
Z
-< 21210
0
W
(L
L 100
H

.......
0
50 450 850 125121
FREQUENCY (Hz)
Figure 5.19 Frequency reponse of equivalent when the scale factor
is optimized from 50 to 1250 Hz
60121 r----r---.----~--_r--_,----~--_r--~~.~.--T---~---,--~
·.
r.. 500
U1
L
I 41210
ov ·.
·· ..
W 30121
U
Z
-<
o 2121121
W
(L
L 10121
H
... , .
. . '

121 ~---r----~--~--~----~--~----~--~----~~~--~--~
5121 450 85121 125121
FREQUENCY (Hz)

Figure 5.20 Impedance with the scale factor optimized

scale factor. Figure 5.20 shows the frequency response when this new
optimum scale factor is applied.

In matching the frequency-response of an RLC circuit to the


power systems response, not only is the impedance magnitude important,
but also the phase angle match. This information can be inspected by
comparing the match in phase angle or the match of both the resistance
and reactance. A typical frequency-response match of the impedance
80.
6121121

5121121
'"
(f)
L
I 4121121
0
v

W 3121121
U
Z
«
0 2121121
W
0..
L 1121121
H

121
5121 450 850 125121
FREQUENCY (Hz)
Figure 5.21 Impedance versus frequency

180
'"
(f)
W 120
W
0:: .............. .
(j
W 6121
0 ......
v

W 121
-.l
(j
Z -60 .' .
« .... , ..
W
UJ- 1 20
«
I
0..-180
50 450 850 1250
FREQUENCY (Hz)
Figure 5.22 Phase Angle versus frequency

magnitude, phase, resistance and reactance are displayed in figures 5.21,


5.22, 5.23 and 5.24 respectively. The phase "AI! equivalent circuit for
implicit mutual coupling representation was used for this illustration.
If the phase angle match is bad in a region, it may require compromising
the impedance magnitude to improve it or the subsequent use of a
correction filter branch.
81.

r, 500
lD
L
I
ov 400

W 300
U
Z
«
I- 200
lD
H

~ 100
(t

0~~__~~~__~__~~==~~~ __ __
~ ~~~'~"~'"
50 450 850 1250
FREQUENCY (Hz)
Figure 5.23 Resistance versus frequency

r,
lD
L 200
I
0
v

W 0
U
z
«
f-
U -200
«
W
(t

-400 ~--~----~--~--~----~--~----~---r----r----r----r---~
50 450 850 1250
FREQUENCY (Hz)

Figure 5.24 Reactance versus frequency

Figure 5.25 displays the comparison in frequency-response for the


full 3x3 system with the straight application of the direct method. The
(1,2) term obviously requires the use of correction filters.
(n) (n) (n)

::: ... ~ Zaa ;LJ ::: j: \ Zab ::: I ~ Zao

200 J .~ \ j. \ 200 . 200

10: v
50
~
450 850
.\j 10: k!
1250 50
~
450 850
\ld
1250
10:
50 450 550 1250
(n) (Il) (Il)
500 500 500

400

::;00
~
f l · ·
Zba 400

::;00·
J 400

::;00
1 1\ Zbc

..

: : lJ \2JJ : :
a
50 450 850
....
1250
0

50 .. 450 850 :1250


:: a
50 450 850 1250

(Il) (Il) (Il)

::: A Zoa ::: 1 ~ Zob ::: I ~ zoo

300 300 I\ . 300

:: a
50 450 850 1250
: : lJ \2JJ : :
0

50 450 850
. ..

1250
a
50 450 850 1250

Figure 5.25 Impedance Match of 3x3 Impedance Matrix ~


83.

Correction Filters.

When scrutinizing an equivalent circuit it must be kept in mind


that it is easier to lower its impedance level in a region than to
increase it. This is achieved by means of adding correcting RLC filter
branches. Hence it is sometimes desirable to select smaller than optimum
scaling factors as it results in larger impedance levels which can be
compensated for later.

Derivation of Correction Branch Parameters.

From inspection the region of largest deviation from the required


impedance response is found and three matching points (frequencies
Wl'W 2 ,W 3 ) are selected in this region. The impedance of the required
response and equivalent circuits response are inverted and the difference
obtained. This is then inverted to give the required impedance response
of the correction filter. The impedance at the required match frequencies
is then obtained and an RLC branch is synthesized so that its frequency
response passes through these three match points. The admittance of the
required correction branch is given by :

Y
-cb
Y
-rr
Y
-ec (5.13)

where
Y Required admittance response.
-rr
Y Equivalent circuits admittance.
-ec
Y Admittance required by correction branch.
-cb

Y b'Y and Yare complex vectors containing the frequency


-c -rr -ec
response of the admittance of the correction branch, required response
and equivalent circuits response respectively. Let Zl,Z2 and Z3 denote
the impedance of the Z b at the match frequencies W1 ,W 2 and W3
-c
respectively. Then this implies the RLC branch must satisfy the following
three equations

2
IZl 12 = R2 + (w1L - 1/(w 1C» (5.14)
2
IZ21 2 = R2 + (w 2L - 1/(w 2 C» (5.15)
2
Iz31 2 = R2 + (w 3 L - 1/(w 3 C» (5.16)
84.

Equation (5.14) minus equation (5.15) gives

(5.17)

Similarly using equations (5.14) and (5.16) yields

21
1 Z2 - 1Z321 = (2
W 2- 2) L2
W3 + 1 w22 - 1 I"'~)
(I w, IC 2 (5.18)

Rearranging equation (5.18) gives:

( 1 I C2)( 1 I w22 - 1 I W32) = IZ212- 1Z3 12 - (2


w2 - 2)
W3 L (5.19)

Therefore C can be expressed as

(5.20)

Substituting equation (5.20) into equation (5.17) gives

(5.21 )

Hence
2
IZ212- I Z 31
2 2
(1/w 2 - 1/( 3 )

(5.22)

Given three points, equation (5.22) is solved to obtain the


inductance of the correction branch. Equation (5.20) is solved next to
obtain the capacitance value. Lastly equation (5.14) is solved for the
required resistance; this completes the derivation of the correction
filter branch parameters.
85.

Although the above formulation synthesizes a correction branch to


correct for impedance mismatch, other quantities such as resistance,
reactance, conductance, susceptance and phase angle can be matched in a
similar manner.

Matching of correction branches.

The newly derived correction filter is added to the equivalent


circuit to form a new equivalent and the process repeated if necessary by
identifying a new region of discrepancy and another correction filter.
This process continues until the required accuracy is obtained, though
the accuracy may be limited by the increasing computing requirements. The
CPU time used in a.c./d.c. dynamic simulation is related to the size of
the time steps of the numerical solution, which in turn depends on the
time constants of the equivalent circuit. The acceptability of a
particular equivalent circuit and the required CPU time are assessed in
each case by means of a short-duration dynamic simulation study.

Illustrative Example.

Figure 5.26(a) displays the frequency response of the (1,2) mutual


element of the actual system together with that of the equivalent
obtained by the direct method. Although the results are in good agreement
at low frequencies the second and third peaks (at 845 and 876 Hz) show
substantial disparity. The use of a correction filter in the region of
largest discrepancy (862 to 906 Hz) modifies the harmonic response to
that shown in Figure 5.26(b). The new branch reduces considerably the
error in the regions where the impedances of the equivalent circuit were
too high, but no marked improvement is noticeable where the equivalent
circuit impedances are too low. As explained earlier these are best
corrected by a combination of scaling factor and subsequent use of
correction filters.

The region of greatest discrepancy of the new equivalent circuit is


now between 750 and 810 Hz. The addition of a second correction filter in
this region produces a far better match, as is shown in Figure 5.26(c).
86.
6121121 ·· ..
·.
A 5121121 ·· ..
U1
2:
I 4121121
0 ·.
v ·.
·· ..
W 3121121
U
Z
« 2121121
0
W
(l
2: 1121121
H

121
. . ....... .
---'.-:".~
'

5121 45121 85121 125121


FREQUENCY (Hz)
Figure 5.26(a) Impedance match of (1,2) mutual element

A 5121121
U1
2:
I 4121121
0
v

W 3121121
U
Z
« '.
0 2121121 ....
W
(l
2: 11210
H

----:-:-....... ..
121
50 450 850 125121
FREQUENCY (Hz)
Figure 5.26(b) Impedance match of (1,2) mutual element when one
correction branch has been added

61210 r---~----~--~--~----~---r----r---~--~----'---~----,

A 500
U1
2:
I 41210
0
v

W 31210
U
Z
«
0 2121121
W
(l
2: 10121
H

......
121
5121 45121 85121 125121
FREQUENCY (Hz)
Figure 5.26(c) Impedance match of (1,2) mutual element when two
correction branches are used
87.

The matched equivalent circuit with implicit accounting of the


mutual coupling for each of the three phases is displayed in figures 5.27
(a),(b)&(c), The direct synthesis produces an equivalent which compares
well with the required response but the 50 Hz impedance of the equivalent
is too high, Therefore one correction filter needs to be used in each
phase to produce the correct impedance (magnitude and angle) at 50 Hz.
The correction filter is given a high Q to avoid any detrimental effect
on the response at higher frequencies. Figures 5.28 (a),(b)&(c) are an
amplified view of the impedance match in the 50 Hz region showing the
effect of incorporating these correction filters. Figures 5.29 (a),(b)&
(c) shows an amplified view of the phase angle match.
88.
6121121

A 5121121 REQUIRED RESPONSE


(J) SYNTHESIZED CIRCUIT'S RESPONSE
2:
I 4121121
0
v

W 3121121
U
Z
-<
0 2121121
W
0..
2: 1121121
H

0
50 450 850 1250
(a) Phase A FREQUENCY (Hz)

6121121

A 5121121
(J)
2:
I 4121121
0
v

W 3121121
U
Z
-<
0 2121121
W
0..
2: 11210
H

121
50 450 85121 1250
(b) Phase B FREQUENCY (Hz)

600

A 5f210
(f)
2:
I 4f21f21
0
v

W 300
U
Z
-<
0 2f210
W
0..
2: 100
H

0
50 450 850 1250
(c) Phase C
FREQUENCY (Hz)

Figure 5.27 Matching to the Diagonalized Impedance Matrix


89,
612l12l

512l12l
""
Ul
2:
I 412l12l
a
v

W 312l12l
U
Z
« 212l12l
0
W
[L
2: 100
H
........... . ....... . .......
..... . ........... .
0
50 75 112l1Zl 125 1512l
( a) Phase A FREQUENCY (Hz)

"" 51Zl1Zl
Ul
2:
av 41Zl1Zl
I

W 31Zl1Zl
U
Z
« 212.112.1
o
W
[L
2: 112.1!Zl
H ............................. ~
.........................
...... . ......... .
12.1 ~-r--~-r--.--.--.--r--~~--~~--~~--~~--,--,--~~r-~
5!Zl 75 1!Zl0 125 1512.1
(b) Phase B FREQUENCY (Hz)

"" 512.10
Ul
2:
I
a 40!Zl
v

IlJ 3!Zl12.1
U
Z
-< 212.1!Zl
o
W
[L
2: 112.112.1
H
................
..... . ................ .
!Zl +--r--.--r--~~--.-~--~~--,--,--~-,--,-~--~~r--.--r-~
5!Zl 75 101Zl 125 150
(c) Phase C FREQUENCY (Hz)

Figure 5,28 Impedance Match with a corrective branch for


the 50 Hz parameters
90.
180
A
UJ
W 120
W
(t
(j
W 60
0 ............................................................................................ •• •• 0 eo.

W 0
.J
(j
Z -60
«
W
(J)-120
<
I
~-180 +--r--r--r--r--r~~'-~--~-.--~-r--r--r~r-~~--~~~
50 75 100 125 150
(a) Phase A FREQUENCY (Hz)

180
1"\
UJ
W 120
W
(t
(j
W 60
0 ' .......................................... , .................................. .
v

W 0
.J
(j
Z -60
«
W
UJ- 1 20
«
I
~-180 +--r--r--r--r-~~--~-.--~-.--~-r--~~~r-~~--~~~
50 75 100 125 150
(b) Phase B FREQUENCY (Hz)

180
1"\
UJ
W 120
W
(t
(j
W 60
0 ..............................................................................
v

W 0
.J
(j
Z -60
«
W
UJ- 120
«
I
~ - 1 80 +---.---,----,---.---.--r---r--r----r---,-----,---,---,.---,----,----,---,--,---,--!
50 75 100 125 150
(c) Phase C FREQUENCY (Hz)

Figure 5.29 Phase Angle Match with a corrective branch for the
50 Hz parameters
91.

5.3.2 Optimization Method.

5.3.2.1 Choice of Objective Function.

The mathematical problem of optimization is to minimize a scalar


function (known as an objective function), which is dependent on many
parameters and to determine the values of the dependent parameters at
this minima. In the present work the difference between the required
response and equivalent circuits response is to be minimized by
optimizing the R.L and C values of the equivalent circuit. As in a large
number of optimization problems. data is only available at a number of
sample points, regardless of whether the required response is obtained by
measurement or frequency domain analysis. Hence only at these data points
can comparisons between the required and equivalent circuit's response be
made. Let xl ,x •... x denote the n dependent parameters and Al .A •... A
2 n 2 m
the values of the independent variables (namely the values of the
required response) at the m available sample points. Then the form of the
objective function is

(5.23)

However. the values of the independent variables may be


incorporated into the function g(~) to yield equation (5.24)

(5.24)

The form of the h function is termed the error criterion. The


choice of objective function is an important issue in optimization
problems as it greatly influences the optimum point and the ease by which
i t is found.
Four of the more commonly used error criteria are
(a) Least Squares.
(b) Mini-max.
(c) Mini-average.
(d) Mini-area.

Least Squares is the most widely used error criterion. The general
form is :
m 2
Minimize e L (w.G.(x)) }
1 1 -
(5.25)
i=l
92.

where w ,w ' .... w are termed weights or penalties and have the
1 2 m
effect of emphasizing G.(x) in regions of importance.
1 -

Mini-max error criterion minimizes the maximum element in G(x).


Its form is given in equation 5.26. One of the disadvantages with this
error criterion is that the derivatives of 8 with respect to the
parameters are not defined, as 8 jumps discontinuously as one value of i
changes to another during the course of optimization.

Minimize 8 = Max{ Iw.1G.


1-
(x) I } (5.26)

Mini-average minimizes the average, however the problem with this


is that it can allow large deviations from the required response if it is
compensated by a similarly large deviation in the opposite direction.
m
Minimize 8 = L{ w.G.(x) } (5.27)
i=1 1 1 -

Mini-area minimizes the area formed by G(~) (area of discrepancy).


To maintain versatility, numerical integration techniques such as
Simpson's Rule cannot be used as the requirement of an even number of
data points is too stringent. Once a particular frequency range for the
optimization has been chosen, the number of sample points within this
range may be odd or even, and interpolation cannot alter this fact (while
still maintaining equally spaced sample points).

Minimize 8 = f W.G.(x)
11-
(5.28)

In the current problem the function G(~) is

G.(x) = Z.(x) - A. (5.29)


1 - 1 - -1

where Z.(x) is the impedance of the equivalent circuit at the


1 -
frequency of the ith sample point and A. is the required impedance at
1
this point. ~ corresponds to the R,L and C values of the equivalent
circuit. It should be noted that although this chapter deals almost
exclusively with the matching of the impedance response the ACREP
program's algorithm can optimize the following parameters; Admittance,
Conductance, Susceptance, Resistance, Reactance or Phase angle.
93.

5.3.2.2 Optimization Methods.

Once the objective function has been defined it is necessary to


decide on a method of performing the optimization. Although not
completely separate, optimization methods fall into two classes, Search
and Gradient methods. Search methods use objective function evaluations
only while gradient techniques require additional gradient information
(Adby and Dempster 1974).

In general Gradient methods are superior to Search methods if the


functions involved have continuous derivatives which can be evaluated
analytically. Functions for which derivative information is not readily
available normally require the application of Search methods. Recently,
however, a Gradient method that only requires objective function
evaluations (Harwell 1986) has been formulated. It is not necessary to
supply derivative information as an estimate of the Gradient is obtained
numerically from either the Finite or Central Difference formula. This
Gradient method uses a special implementation of the BFGS method to
minimize a Lagrange multiplier penalty function (equation 5.30). The
Lagrange multiplier penalty function with only equality constraints is

k k
Sex) - L (A.C.(x))
11 -
+ (1/2) L (cr.
11 -
C~(x)) (5.30)
'1
1= '1
1=

where the objective function is


S(~) = h{G1 (~) ,G2(~)'" .Gm(~)}

and the constraints are


C(~) = 0 ,i=l ,2, ... k
cr. , i=l , ... ,k are the penalty parameters
1

A. , i=l , ... ,k are the multiplier parameters


1

Two multi-variable optimization techniques have been implemented in


ACREP. The first is the gradient method described above while the second
is the pattern search technique.

Figure 5.30 (a),(b),(c)&(d) show the optimum as found by the


gradient optimization for the four main error criteria mentioned
previously. The optimum using Mini-average and Mini-area are similar as
expected as the sample points are equally spaced. The best is the least
94.
600

500 - - REQUIRED RESPONSE


A '. .. ............ SYNTHESIZED CIRCUIT'S RESPONSE
UJ
2:
I 400
0
v

W 300
U
Z
<
0 200
W
[L
2: 100
H

0
50 450 850 1250
FREQUENCY (Hz)
(a) Least Squares

6121121

51Zl1Zl
"
Ul
::L
I 400
0
v

W 300
U
Z
<
0 200
W
[L
::L 11210
H

......
121
. ' . .. , .
50 45121 85121 1250
FREQUENCY (Hz)
(b) Mini-max

6121121

A 50121
UJ
2:
I 40121
0
v

W 3121121
U
Z
<
0 2121121
W
[L
2: 1121121
H

121
5121 45121 85121 125121
FREQUENCY (Hz)

(c) Mini-average
95.
6lZllZl

A 5lZllZl
ill
2:
I 4lZl121
0
v

W 3lZl121
U
Z
«
0 2lZl121
W
[L
2: llZll2l
H
.. " .........
121
5121 45121 85lZl 125121
FREQUENCY (Hz)
(d) Mini-area

Figure 5.30 Impedance match obtained using various error criteria

squares error criterion, which accounts for its popularity. The various
CPU requirements to reach these solutions from a nominal starting point
are displayed in Table 5.1 . In general the Gradient method takes in the
order of 1 to 4 minutes of CPU time to optimize a typical equivalent
circuit consisting of 11 variables. This compares well with the 20
minutes to 3 hours required for the same optimization problem using the
pattern search.

Error Criteria CPU Time Objective Fcn.

(a) Least Squares 1 Min 8.03 sec 223679.26


(b) Mini-max 31 .17 sec 279.92
(c) Mini-average 38.70 sec -5329.57
(d) Mini-area 1 Min 24. 11 Sec -285796.61

Table 5.1 CPU requirements for various Error Criteria.


96.

Ideal Objective Function

Objective Function with Numerical Noise

Estimated

I
I,
I
I I :

I~
h
J ~h~
h - Optimization Algorithm's Step Size

Figure 5.31 Pictorial View of the effect of Numerical Noise on Gradient


estimate

One question that comes to mind is that if the Gradient Method is


so much better why implement the pattern search method ? The reason being
that in many cases the Gradient method fails to find the minima. The
Gradient search assumes that the objective function can be calculated to
the full precision of the variables, and hence ignores any numerical
noise present. When well away from the minima the step length is large
and hence the required directions are easily determined, regardless of
the numerical error. However, when in the proximity of the minima the
step length is small, hence the numerical error greatly influences the
gradient estimate making it extremely hard to determine in which
direction the minima lie. The nett effect is for the algorithm to
stagnate and then stop before reaching the minima. Figure 5.31 depicts
the effect of numerical noise. When this occurs a search technique such
as the Pattern Search can be applied, starting from where the gradient
search stopped. Search Methods are more robust than the gradient methods
as gradient estimation is always prone to significant numerical errors.
97.

5.4 The Philosophy of the Frequency-Dependent A.C. System Models.

The determination of the self and mutual impedance terms of an a.c.


system can be visualized as the ratio of voltage to current obtained by
shorting every independent voltage source and injecting a current of
frequency f in one phase, as depicted in figure 5.32.

A.C.
system

Figure 5.32 Interpretation of Self and Mutual Terms.

Hence the matrix equation describing this situation is

(5.34)

Then to obtain the frequency response this process is performed


for a range of frequencies. As the impedance is determined with the
independent voltage sources shorted the model must have the fundamental-
frequency source incorporated. Hingorani and Burbery's equivalent (figure
5.1) put the voltage source in the RL branch only. A natural extension of
this is to also include the mutual voltage in this branch. This has the
98.

advantage of requiring no extra circuit elements to be modelled as the


mutual voltage can be incorporated into the existing fundamental-
frequency voltage source. However, it is questionable whether this is
strictly the correct place for the voltage sources to be incorporated.
When the equivalent of figure 5.1 is open circuited the terminal voltage
is less than the source voltage as the RLC branches of the a.c. system
equivalent loads the voltage source causing a voltage drop across the RL
branch. Although this error may be small for a fundamental-frequency
source, by virtue of the high impedance of the RLC branches at 50 Hz
relative to the RL branch impedance, this may not hold where the mutual
voltage source is concerned, due to the higher frequencies present in the
mutual voltage waveform.

Consider the situation depicted in figure 5.33, where the mutual


voltage source is incorporated into only the RL branch. By applying the
Superposition Theorem, consider the effect of the mutual voltage source
in isolation (i.e. shorting the fundamental-frequency source). Let
phases "A" and "c" be open circuited while phase "B" is loaded. From
equation (5.34), under these conditions the terminal voltage of phase "A"
should be ZabIb' however, the circuit of figure 5.33 has the RLC branches
loading the mutual voltage source, therefore the terminal voltage
appearing will be
d I 9.
r
V
a
+ L
r9.
---
d t
(5.35)

where I 9. is the current in the RL branch and Ib the phase "B" current.
r

The Substitution Theorem states that "two subnetworks are


equivalent if the relationships between the terminal voltages and
terminal currents are the same for both and such equivalent subnetworks
can be exchanged without affecting conditions in any part of the network
external to them" (Skilling 1974). Obviously the circuit of figure 5.33
violates this and hence is not a true equivalent of the actual systems
performance as strictly laid down by the Substitution Theorem.

The equivalent displayed in figure 5.34 does however approach this


ideal of being a true equivalent as the fundamental-frequency voltage
sources are now the open circuit voltages.
99.

, - - - - - - , - - -..................... - - . , - - - - l

I I
Figure 5.33 Extension of Hingorani and Burbery's equivalent by
incorporating the mutual voltage as part of the
fundamental frequency source

Zb a I a+Zb I
ce

Figure 5.34 A.C. System Equivalent.

In figure 5.34 there are three parts: a frequency dependent


impedance (Z ,Zbb'Z ) which is implemented by a matched RLC equivalent
aa cc
circuit; a fundamental-frequency voltage source, which is not apparent
when impedances are obtained; lastly a controlled voltage source which
mimics the mutually induced voltage terms, ZabIb+ Zaclc' Zb a I a + Zb c I c
and ZcaIa+ ZCb1b terms.

VNlvmSITY OF
100.

r-------,----..........................- - - - - , r - - - l

Figure 5.35 The self circuit for the frequency-matched a.c. system
as implemented in TCS

The modelling of a voltage source in isolation is not possible in


TCS as each voltage source must be associated with an inductive branch
(Joosten 1985). This difficulty is overcome by moving the fundamental-
frequency and mutual voltage sources into the inductive branches of the
RLC network, as depicted in figure 5.35. This change does not alter the
performance of the circuit. However, this equivalent is computationally
expensive due to the number of voltage sources that need to be updated at
each time step. Since the RLC branches are tuned to exhibit series
resonance above the fundamental frequency their impedance is large at 50
Hz and the RL branch dominates. Under these conditions the fundamental-
frequency voltage sources in the RLC branches can be eliminated. It could
also be argued that the mutual voltage source can be eliminated from the
RLC branches as its value is several orders of magnitude less than the
voltage across the resistor,inductor and capacitor making up the branch;
however this is not strictly valid as the harmonic levels are usually
several orders of magnitude less than the fundamental and therefore
approximations made at harmonic frequencies may result in high relative
errors.
101.

Figure 5.36 displays the various a.c. system equivalent circuits


which may be used for dynamic studies with the corresponding CPU
requirement for a typical 5 cycle simulation. These models represent the
actual system performance with varying degrees of accuracy and in each
case the purpose of the simulation will dictate the most appropriate
model to be used.
THEVENIN FREQUENCY-MATCHED FREQUENCY-MATCHED FREQUENCY-MATCHED FREQUENCY-MATCHED
EQUIVALENT EQUIVALENT No. 2 EQUIVALENT No. 3 EQUIVALENT No.4
EQUIVALENT No. 1

nuplicit Representation
of Mutual Coupling.

r
11-1
1
II~~
1 1
~I ~ 'V~ ~ T

8 min. 1 5 min. 1 5 sec. 26 min. 54 sec. 46 min. 54 sec.


50 sec.

o
Figure 5.36 Various a.c. system equivalent circuits [\)
103.

5.5 Implementation of Frequency-Dependent A.C. System Models.

The solution process for the Self <-) Mutual circuit interface
is depicted in figure 5.37.

With eXisting mutual Voltage estimate


obtain phase currents.

With estimated phase currents obtain


estimate of mutual voltages.

Has No
Convergence
occurred
?

Yes

Figure 5.37 Self <-) Mutual Circuit Interface.

Direct application of this algorithm would result in this


iterative loop being required for each iteration of the integration
algorithm (separate iterative loop solution). This is highly undesirable
due to the computational burden it imposes. Instead, the above loop is
embedded in the existing integration iterative algorithm, as discussed in
section 3.4.3 for the modelling branches between phases, thus achieving
only one level of looping. Compared to the modelling of "torn" branches
(section 3.4.3), the final method for the implementation of
frequency-dependent a.c. system equivalent reverses the roles of the
voltage and current source interface between the standard and extra
circuits. The philosophies 6f the two approaches however differ greatly.
In figure 5.38(a) the extra circuit represents torn branches that are
actually part of the network while in figure 5.38(b) the extra circuits
104.

Standard Torn
TCS Branches
Circuit

(a) Standard TeS circuit <-) torn branches

Standard Mutual
TCS E
m Circuits
Circuit

(b) Standard (Self) circuit <-) Mutual circuits

Figure 5.38 Two types of interface

(mutual circuits) are used to calculate the mutual voltage at each time
step and do not represent actual branches in the network. The early
attempt used tearing to remove branches and the interface consisted of a
voltage source input for the torn branches and an equivalent current
source input for the main part of the network, as depicted in figure
5.38(a). In the frequency-dependent a.c. system equivalent as
implemented, the mutual voltage is the input to the main part of the
network while the current source the input to the mutual networks, as
shown in figure 5.38(b). In this case the voltage source models the
voltage induced due to a current in the other two phases, while the
mutual network calculates the voltage to be applied, based on the phase
currents and frequency-dependent mutual impedance.

The iterative procedure of the integration routine is as follows:


estimate the state variables, hence obtain the phase currents; evaluate
the dependent variables, therefore obtain the mutual voltage ; test for
convergence of all the variables and if required perform another
iteration. This approach however results in a larger number of iterations
prior to convergence and therefore requires the re-tuning of the
parameters governing the step length optimization based on past
convergence history, for satisfactory operation.
105.

5.6 Initialization of Frequency-dependent Equivalents.

The method of initialization is based on what experience has shown


to be the best for starting convertor operation. Disabling the convertor
controller at start-up and enabling it several cycles into the simulation
allows a steady state to be reached in a shorter duration than if the
controller is active from the start of the simulation.

To avoid prolonged initialization transients the mutual circuits


are not excited with the phase currents until sufficient time has been
allowed for the main transient in the self circuit to die away
(approximately two cycles). A further delay is allowed for the main
transient in the mutual circuits to die away before applying the mutual
voltage back into the self network.

At the time of energizing the mutual circuits it is imperative to


initialize the current in the inductors in the best way possible. Firstly
the sum of the currents at a node must be initialized to be zero. This is
very important as the phantom current source (Section 3.3.3, Joosten
1985) developed to over-come the gamma node problem does not operate on
the mutual circuits. Secondly the sum of the initial inductor currents in
the mutual circuit must equal the appropriate self circuit's phase
current. This makes the initialization slightly more difficult as we have
no idea of what this current will be until the program has reached this
point in the simulation. The result of not achieving the correct initial
currents in the branches forming the mutual circuit is a constant offset
between the self circuits phase current and the current applied to the
mutual circuit. This is due to the equation for gamma nodes using rate of
change of current (equation 3.23) rather than current magnitude. Once the
fluxes in the inductors have been initialized, all the changes occurring
during the simulation are governed by the rate of change of current.
106.

5.7 Validation of the Computer Model.

A comparison between the frequency response of the synthesized


RLC circuits and their required responses provides a clear indication of
how well each of the self or mutual terms response is modelled. But in
developing the TCS program to cope with the frequency-dependent a.c.
systems, each stage of the development must be validated to ensure the
correct implementation. Although difficult and time consuming, the
validation exercise is essential for the credibility of the simulated
results.

A preliminary test in the process of building up the model involved


applying a voltage rather than current excitation to the mutual circuits.
By modelling identical RLC networks for both the mutual and self circuits
and applying the same fundamental voltage excitation, while the mutual
voltage interaction with the self circuit removed, allows the transient
responses to be compared directly, hence verifying the mutual circuit.
This tests the software associated with setting up the extra networks
forming the mutual circuits as well as the algorithm for obtaining the
mutual circuit variables at each time step.

When the interaction between the mutual and self circuits is


enabled, if the mutual voltage derived from the phase currents is applied
with the wrong polarity to the self network, an instability may occur
making the error very noticeable. This instability is due to a positive
feedback effect; the mutual voltage is of a polarity that aids the
fundamental-frequency voltage source thus tending to increase the current
magnitude in each phase. The increase in a phase's current due to the
mutual voltage results in an increase in the mutual voltage applied to
the other two phases, hence a larger phase current results. This increase
of phase current in the other two phases produces a bigger mutual voltage
to be applied to the original phase and thus increases its current even
more.

As mentioned in chapter 3 the current source model was first


implemented to allow testing of the complete a.c. system model. Each
current source is capable of modelling a single frequency source,
therefore a separate source is required for each frequency to be
injected. For convenience and to maintain clarity of results, normally
107.

only one frequency at a time was injected. Due to the time considerations
the testing of the complete model by this method, at best, can be
performed only at a few selected frequencies.

By injecting a current in phase "A" and assessing the voltages of


the three phases, the first column of the 3x3 impedance matrix at the
current source frequency is obtained. The other two columns are assessed
by injecting currerit in the other two phase in a similar manner. These
results can then be compared to the expected values from the designed
frequency-response match, to validate both the diakoptical technique used
to implement the frequency dependent mutual coupling and the correctness
of the input data. Tables 5.2, 5.3 and 5.4 summarize the results of
these three current injection tests. The impedances designated by the "A"
are the theoretical result from the designed match while the values
labelled by the "B" are obtained by using an FFT on the TCS derived
waveforms. The results show very good agreement at all the six
frequencies the test was performed at. This clearly demonstrates that the
diakoptical technique has been correctly coded and the input data has
been correctly entered.

A more comprehensive test is the comparison of the harmonic levels


with those derived from frequency domain programs, as outlined in
chapter 6. This effectively tests the frequency-dependent a.c. system
equivalent at all the required harmonic frequencies simultaneously. A
quantitative validation of dynamic simulation results is not possible due
to the lack of an adequate yard-stick. However, credibility for the
dynamic simulations can be provided by the harmonic assessment studies.
If the frequency response of the model is accurate so will be its dynamic
response. The dynamic response can be considered as the convolution of
the input with the impulse response, which corresponds to the
multiplication of the input and system responses in the frequency domain.
Verification of the system response in the frequency domain ensures the
accuracy of the corresponding convolution in the time domain.
108.

Freq. Matrix Impedance Phase


(Hz) Position Magnitude Angle
(Ohms) (Degrees)

50
(1 ,1) A 47.031 46.85
B 47.093 46.81
(2,1) A 15. 122 81.12
B 15.144 81 .09
( 3 , 1) A 12.482 79.21
B 12.488 79.19
150
( 1 , 1) A 160.503 63.10
B 160.941 63.01
(2,1) A 57.962 85.14
B 58.120 85.12
(3,1) A 46.952 84.41
B 47.018 84.40
275
(1 , 1 ) A 474.444 3.86
B 475.135 3.78
(2,1 ) A 499.388 -2.89
B 498.435 -3.50
( 3 , 1) A 397 .566 13.17
B 397.542 12.93
450
(1 , 1 ) A 37.714 -57.18
B 37.575 -57.09
(2,1) A 25.656 -49.13
B 25.470 -48.81
(3, 1) A 21.430 -45.43
B 21.295 -45.13
750
(1 ,1) A 163.500 66.79
B 164.089 66.68
(2,1) A 64.642 45.29
B 64.890 45.32
(3,1) A 49.613 63.65
B 50.183 63.48
1000 Hz
( 1, 1) A 146.361 -6.29
B 145.495 -7.01
(2,1) A 7.202 169.25
B 7.031 162.38
(3,1) A 30.873 -11 .86
B 30.195 -11.15

A Theoretical value obtained from


Frequency domain Software.
B TCS assessed values.

Table 5.2 Current injected into phase "A"


109.

Freq. Matrix Impedance Phase


(Hz) Posi tion Magnitude Angle
(Ohms) (Degrees)

50
(1 ,2) A 15.122 81. 12
8 15.144 81 .09
(2,2) A 47.023 46.69
B 47.085 46.66
(3,2) A 15 . 122 81 .12
8 15.144 81 .09
150
(1 ,2) A 57.962 85.14
B 58. 119 85.10
(2,2) A 164.500 61 .82
8 164.994 61 .71
(3,2) A 57.962 85.14
8 58.119 85.10
275
(1 ,2) A 499.388 -2.89
8 498.435 -3.50
(2,2) A 507.092 5.89
B 507.685 5.79
(3,2) A 499.388 -2.89
8 498.435 -3.50
450
( 1 ,2) A 25.656 -49.13
8 25.468 -48.80
(2,2) A 41 .059 -59.29
8 41.013 -59.23
(3,2) A 25.656 -49.13
8 25.468 -48.80
750
(1 ,2) A 64.642 45.29
8 64.886 45.33
(2,2) A 152.786 65.97
8 153.404 65.70
(3,2) A 64.642 45.29
8 64.886 45.33
1000 Hz
( 1 ,2) A 7.202 169.25
B 7.032 162.35
(2,2) A 155.870 -4.00
B 152.301 -5.29
(3,2) A 7.202 169.25
8 7.032 162.35

A : Theoretical value obtained from


Frequency domain Software.
8 : TCS assessed values.

Table 5.3 Current injected into phase "8"


110.

Freq. Matrix Impedance Phase


(Hz) Position Magnitude Angle
(Ohms) (Degrees)

50
(1 ,3) A 12.482 79.21
B 12,489 79.18
(2,3) A 15. 122 81 .12
B 15.144 81 .08
C3 ,3) A 47.031 46.85
B 47.097 46.80
150
( 1 ,3) A 46.952 84.41
B 47.018 84.39
(2,3) A 57.962 85.14
B 58.120 85.11
(3,3) A 160.503 63.10
B 160.942 63.01
275
( 1 ,3) A 397.566 13.17
B 397.541 12.93
(2,3) A 499.388 -2.89
B 498.435 -3.50
C3, 3) A 474.444 3.86
B 475.135 3.78
450
( 1 ,3) A 21.430 -45.43
B 21 .294 -45.13
(2,3) A 25.656 -49.13
B 25.468 -48.81
C3,3) A 37.714 -57.18
B 37.572 -57.09
750
(1 ,3) A 49.613 63.65
B 50.181 63.49
(2,3) A 64.642 45.29
B 64.886 45.33
(3,3) A 163.500 66.79
B 164.089 66.69
1000 Hz
( 1 ,3) A 30.873 -11 .86
B 30.184 -11.14
(2,3) A 7.202 169.25
B 7.032 162.35
(3,3) A 146.361 -6.29
B 1 45.460 -7.01

A : Theoretical value obtained from


Frequency domain Software.
B : TCS assessed values.

Table 5.4 Current injected into phase "C"


111.

5.8 Extension to Multi-convertor Systems.

Consider now the case when two convertors are connected to the
same a.c. system as shown in figure 5.39. This system requires two
points of observation, i.e. busbars A and B. The convertor connected to
busbar A will affect the impedances of the a.c. system when observed from
busbar B and vice versa. This inter-dependence increases the number of
mutual terms as each phase has a self and five mutual terms.

An extension of the diakoptical technique to this case requires


six self and thirty mutual circuits for a complete frequency-dependent
model. Although extension of the theory to more than two convertors is
straight forward, the number of frequency dependent circuits required
increases with the square of the number of convertors. The complexity of
each frequency dependent circuit is dependent on the number of loops in
the frequency loci. Therefore it becomes impractical to model accurately
a large number of convertors connected to the a.c. system.

D.C.
System

A.C.
System

D.C.
System
2

Figure 5.39 Multi-convertor System


112.

5.9 Frequency-dependent d.c. System model.

The same procedure used in the development of frequency-dependent


a.c. system equivalents can be applied to produce frequency-dependent
d.c. system equivalents. However, because the TL program (figure 5.3) was
designed specifically for three phase transmission lines it cannot
easily be used for modelling d.c. lines. Instead the more general line
constants program (LINE) developed by Prof Dommel (Dommel 1980) is used
to produce the transmission line parameters at all the required
frequencies. Figure 5.40 depicts the new chain of data flow and the
associated programs required to produce a d.c. equivalent circuit which
is frequency-matched.

The LINE program is capable of modelling skin effect and earth


return effects for an arbitrary number of conductors. Carson's equations
are used to implement earth return effects.

The program DCLINK (figure 5.40) forms the overall admittance


matrix by cascading the chain parameter matrices for each different
section forming the d.c. system. The structure of DCLINK is shown in
figure 5.41. The output from DCLINK is a set of 2x2 admittance matrices,
one matrix for each required frequency. The equation describing the
totality of the d.c. system with its different sections is :

(5.31 )

where the subscript 1 denotes quantities associated with the positive


pole and 2 the negative pole.

If a bipolar link is to be represented in the transient program the


set of 2x2 admittance matrices produced by DCLINK result in two frequency
-dependent self and two frequency-dependent mutual circuits being
required. From knowledge of the operating conditions of the link it is
possible to reduce the 2x2 matrices so that only one frequency-dependent
equivalent circuit is required in the transient program, with the mutual
coupling accounted for implicitly. However the use of a monopolar
equivalent of the bipolar link imposes limitations on the d.c. faults
that can be simulated.
Conductor data
and line geometry of
d.c. line

Transmission Line Obtain overall Synthesis of frequency- Transient Convertor


Parameter program >- admittance matrices matched equivalent ~ Simulator
(LINE) (DCLINK) (ACREP) (TCS)
-_._---- ---

Figure 5.40 Data Flow in forming a d.c. equivalent circuit.

w
I·DCLINK .,
,

~
Initialize frequency Repeat until all section C 1
Repeat for each frequency /
Range and number of data has been entered
sections

Prompt for section Form the data C Repeat for each section Convert from
number, type and file for the ABCD to [y]
relevant data section and reduce to
a 2x2 matrix

I
Read the admittance Form the [y] Write the
and impedance data matrix, convert admittance
from data file to ABCD form and data into a
chain with data file
previous section

Figure 5.41 Structure diagram of DCLINK

-!=
115.

If the current in each conductor is identical in magnitude and


opposite in sign then the monopolar equivalent is obtained by; inverting
the admittance matrices obtained from DCLINK, then post multiplying by
the matrix (as 11 = -1 2 ) :

This gives

(5.32)

(5.33)

The last equation is disregarded and the impedances in equation


(5.32) are halved. This is done as Vi must equal V2 in magnitude for the
bipolar link to be represented as a monopolar equivalent. This results in
twice the current that would appear in one pole of a bipolar link
occurring in the monopolar equivalent. As the impedances are halved from
that of a single pole and the losses are I2R, the power loss is twice
that of one pole, or that of the total bipolar link it represents.
However, applying any fault that would cause 11 not to be equal to 12 in
the bipolar link it represents cannot be modelled as it would render this
monopolar equivalent invalid. In general the full bipole operation must
be modelled.
116.

r.
REQUIRED RESPONSE
SYNTHESIZED CIRCUIT'S RESPONSE
~ 91Z11Z1
I
ov
W 61Z1lZl
U
Z
«
o
W 31Z1lZl
(L
L
H

lZl +----r----.---,---~----~--_r----r_--_r--~----~--~----~
lZl 4lZllZl BIZIIZI 121Z11Z1
FREQUENCY <Hz)

Figure 5.42 Impedance match of d.c. equivalent

Figure 5.42 compares the frequency response of a d.c. system with


one pole de-energized, and hence using earth return, with that of the
matching equivalent circuit.

The frequency-dependent d.c. system model is probably best used in


harmonic studies and a.c. fault dynamic studies. The matching equivalent
is correct for the terminal conditions only, with no information being
available on busbars in the system it represents. Therefore when d.c.
faults are being modelled the location must be accessible for TCS to use
its d.c. arc modelling capability. That is, the line must be represented
, by n-equivalent models on each side of the fault.
117.

5.10 Negative Resistance Region in the Impedance Loci.

In the case of power transformers the mutual coupling between


phases is considered to be purely inductive, however, the mutual coupling
between phases in power transmission systems exhibits both resistive and
inductive components. When the mutual coupling is inductive the induced
voltage in one phase lags or leads the current in the inducing phase by
90 degrees. The resistive component of the mutual coupling produces an
induced voltage component which is either in phase or 180 degrees
out-of-phase with the current that caused it.

The mutual coupling between phases often exhibits a frequency range


in which the resistive component is negative. This is predominantly due
to the effect of the transmission lines. In the frequency-dependent
equivalents circuits this is modelled by having a negative resistance in
the appropriate RLC branch. Although this works well for modelling minor
excursions of the impedance loci into the left hand half plane,
difficulties arise where complete loops occur in the left hand half plane
of the actual systems response.

I I
Figure 5.43 Partitioned equivalent circuit
118.

Consider the frequency-dependent equivalent as depicted in figure


5.43. The equivalent has been partitioned into three regions; the first
consists of the RL branch and all RLC branches having a series resonance
below f i the second portion is the branch to exhibit parallel resonance
pr
with the former section; the last section consists of the remaining RLC
branches that do not influence the impedance until a higher frequency is
reached. Let Rl + jX 1 of the
denote the impedance at a frequency f
pr th
first portion and let R2 - jX 2 represent the impedance of the k RLC
th
branch at f Then the k parallel resonance occurs when the
pr
susceptance is zero. (i.e. when the capacitive susceptance of the RLC
branch equals, in magnitude, the inductive susceptance of the first
region). This can be shown to occur when:

(5.34)
( R22 + X2)
2

As the resistive components are considerably smaller than the


reactances involved, this implies the ratio of Xl to X2 (equation 5.34)
is close to unity at the frequency of parallel resonance. Now the total
impedance is given by :

Zl Z2 (R l + jX I )(R 2 - jX 2 )

Zt
Zl + Z2) (R l + R2 ) + j (Xl - X2 )

(R 1 R2 + XI X2 ) + j(X 1 R2 - X2Rl )
(5.35)
(R l + R2 ) + j (Xl - X2 )

From equation (5.35) it can be seen that at a parallel resonance,


if R2 is negative but less than Rl in magnitude the impedance peak will
occur in the right half on the impedance plane. However if (R I + R2 ) is
negative the resistance will be negative at the parallel resonance and
therefore the maximum impedance pOint occurs in the left hand half plane.
This is also accompanied by change to an anti-clockwise progression of
the impedance loci in the region of this resonance. However the frequency
dependence of actual power systems always appear to show a clockwise
progression of the loci. Therefore, although the impedance magnitude
match can be very good, the corresponding phase angle match is bad due to
incorrect direction of the loci progression.
11 9 •

. 11 Discussion and Conclusions.

5.11,1 Explicit versus Implicit Mutual Representation,

The frequency-response of the implicit representation is normally


far better behaved than that of the full 3x3 response, allowing a far
better match between the required and equivalent circuit to be achieved,
However, there are inaccuracies difficult to assess due to the
assumptions used to diagonalize the impedance matrix. With the full
representation, although the frequency-response match is more
complicated, there are no simplifying assumptions and inspection of the
frequency-response match can identify the expected error, The computation
efficiency is another factor in favour of the implicit mutual coupling
representation.

5,11,2 Comparison between Direct and Optimization Techniques

for forming frequency-dependent equivalent circuits,

The CPU time required by the direct method of synthesis is several


orders of magnitude less than that of the optimization technique.
Typically the direct method requires 100 to 200 mill i-seconds of CPU
compared to 10 seconds to 2 hours for the optimization technique,

The topology of the equivalent circuit is fixed with the direct


method whereas the only limitation for the optimization technique is the
ability to evaluate the circuits frequency response. However, the
topology must be decided upon before starting the optimization.

The ability to optimize the equivalent circuit is dependent on how


perverse the objective function is, The application of weighting
functions and/or changing the optimization method are often required to
overcome stagnation problems in the gradient algorithm,
120.

The optimization technique generally finds the best fit for a


given number of components in the equivalent circuit. Therefore if many
simulations are to be performed with an equivalent it is probably worth
the effort of trying to optimize. On the other hand where several models
are to be tried with relatively few simulations of each, the direct
method is the best option.

5.11.3 Impedance Magnitude and Phase Angle Match.

The impedance magnitude and phase angle are both very important in
harmonic assessment studies. This is due to the time waveforms being
dependent on the magnitude as well as phase angle of the harmonics.
Therefore after the circuit has been synthesized the match in both needs
to be inspected. Both the direct and optimization methods can correct for
regions of inadequate matching of phase angle. The direct method can use
correction filter branches while the optimization method can incorporate
both impedance and phase angle as part of the objective function. The
effect of impedance phase angle on the dynamic behaviour depends on the
cause of the transient and the system and is difficult to predict in
advance.

5.11.4 Negative Resistive component of Mutual Coupling.

Conservation of power is one of the fundamental laws of science.


The existence of a negative resistance allows the mutual coupling to
supply power to the load resulting in the load's power dissipation
exceeding the power supplied by the external sources. At first sight it
may appear that energy is coming from no-where, however, this is not the
case. Although negative resistive coupling is not physically realizable
as it amounts to generation, it can occur as a result of the reduction
process required to derive the 3x3 impedance loci for one busbar from the
frequency response of all the system components and their inter-
connection. Therefore it is a mathematically convenient solution to a
non-physical problem.
121.

For instance a transmission line with three phase conductors and


two earthwires has a primitive impedance matrix which is 5x5 in dimension
and there is energy associated with the mutual coupling between the
phases and the earthwires. This primitive impedance matrix is reduced to
a 3x3 by absorbing the terms associated with the earthwire into the
remaining nine terms. Therefore these nine terms embody the mutual
coupling with the earthwires.

[ZJ p . =
[A] [B]]
rlm. [ [C] [0]

where dimensions of the sub-matrices are:


[A] 3x3, [B] 3x2. [C] 2 x 3, [0] 2x2

The reduced impedance matrix is obtained from the following equation:

[Z] [A] - [B][D]-1[C]

Although the elimination of the earthwires from the formulation


may result in a negative real part to the mutual coupling, the
corresponding changes in the self terms ensures that the overall power is
dissipated and power conserved.

Although negative resistive regions require more care in


modelling and are more awkward to handle this is not a major limitation
for HVDC dynamic simulations. The negative resistance regions only result
due to the presence of long transmission lines together with very light
loading conditions. HVDC applications do not include long a.c.
transmission lines as the d.c. line is used to cover the larger
distances; therefore the a.c. system is unlikely to exhibit major
negative resistive regions in its loci. Also the presence of loading
would help to reduce any negative resistance regions.
122.

5.11.5 Dynamic instability.

Dynamic instability can occur with some synthesized equivalent


circuits. This is primarily due to the impedance loci being in the left
hand half plane. Consider for example the case of a small 7th harmonic
current in phase "A" that causes a mutual voltage in the other two phases
to increase their 7th harmonic current. There is then the possibility of
a dynamic instability if the increased current in these two phases tends
to increase the phase "A" 7th harmonic current substantially. The
equivalent with the mutual voltage applied only to the RL branch is
moderately immune to such instabilities as the RLC branches load the
mutual voltage source and hence dampens its effect on the external
network. The equivalents with the mutual voltage applied to all the
parallel branches in the equivalent are susceptible to this form of
instability. An example of a dynamic instability occurring is displayed
in figure 5.44.

When the resistive component of the mutual coupling is positive


the voltages and currents in the system are self regulating due to the
negative feedback formed. With a negative resistive mutual coupling the
feedback is positive and a stable solution depends on the size of the
mutual coupling relative to the self terms. A frequency-matched
equivalent may be unstable if the matching compromises causes the
equivalent impedance matrix, at a particular frequency, not to be
diagonally dominant. If a substantial negative resistive region exists,

3 . 5 , -_ _- .____- .____, -____, -____, -_ _- ,____- .____. -____. -_ _- .

3.0

a.s

-3.0

-:3 ·55.'-0--5,.-'-.5::----=-6"-=.a:-----:s=-'.-=-5---::7...l-.0::----=-7.'"=5-----=6...I...o::---::-e.'-=5--9...1..-0- - - 9L.5----',0.0


TIME (CYCbES)

Figure 5.44 Busbar voltage, displaying a harmonic instability


123.

it must be made smaller, the reduction depending on the values of the


components of the same frequency in other elements of the impedance
matrix. If the negative resistive region is small it can be approximated
by an equivalent with an impedance loci completely contained in the right
half of the impedance plane.

5.11.6 Future Work.

There is still a need for a better understanding of the frequency-


response behaviour of power systems. Although a great deal of work has
been done on the accurate calculation of the transmission line parameters
at any frequency, the combined effect of the interconnection of
transmission lines on the frequency-response is not well understood. More
specifically, an adequate explanation is required of why the impedance
loci always appears to exhibit a clockwise progression with frequency.
Also it needs to be determined if there are any exceptions to this
apparent power system rule.

The optimization techniques allow a great variety of topologies


in the equivalent circuits to be synthesized. There is a need for further
research into the different possible topologies with the aim of finding a
topology that can emulate the power systems frequency-response better.

Rational functions, previously used to model the frequency response


of individual transmission lines, can be extended for an arbitrary a.c.
system frequency response and their performance compared with the methods
presented here. Once a rational function approximation has been made of
the frequency response two possible options are; (i) RLC network
implementation, (ii) a straight incorporation into state equations that
can be implemented directly into the existing state variable solution
method. This requires a system identification process that obtains the
appropriate state equation coefficient matrices from the rational
function.
124.

CHAPTER 6 HARMONIC ASSESSMENT.

6.1 Introduction.

The problems associated with the propagation of harmonic currents


in power systems are many and diverse in nature. They include extra
losses in electrical machines, errors in metering, maloperation of
equipment (e.g. Ripple Control), failure of power system components such
as power factor capacitors, and interference in communication systems.
This has lead to an increasing interest in the calculation of harmonic
effects on supply systems which has been exemplified by a conference
specifically about power system harmonics being held in UMIST (1981), the
introduction of legislation limiting the allowable harmonic levels in
New Zealand and Australia, an IEEE overview (IEEE 1983) and bibliography
(IEEE 1984). The main harmonic source is the static power convertor,
other sources such as transformer magnetizing currents, fluorescent
lights etc. being relatively unimportant. As the use of convertor
installations rapidly grows in size and number the need for A.C. and D.C.
side voltage and current harmonic calculations are becoming an important
part of power system planning.

6.2 Harmonic Penetration.

Harmonic penetr~tion is the propagation of harmonic currents in an


a.c. system from a harmonic injecting source, thus resulting in voltage
distortion at the various busbars. The harmonic Penetration program
(HARMAC), already mentioned (Chapter 3) in connection with the derivation
of a frequency-dependent model of a power system, is now considered in
the context of steady-state harmonic assessment. Therefore a brief
outline of the complete operation of HARMAC needs to be given.

Realistic analysis of harmonic levels requires a three phase


representation of the power system so that important effects such as
system unbalance can be modelled (Densem 1983). Figure 6.1 displays a
structured diagram of the program HARMAC developed by Densem. The basis
HARJViONIC PENETRATION PROGRAM

(HARMAC)

I I Calculate
I
Input current Solve
1
Calculate
Input frequency Read shunt Read line
data for all system injection [Ih)=[Y ] [V ] and output
range and capacitors, h h
frequencies harmonic busbar and for all the line
busbars at transformers,
and include impedance for three phase frequencies current
which harmonic filters and
in harmonic a reduced injection to obtain the flows for
impedances are unbalanced
loads at admittance system data. three phase all
required.
matrices. voltages. frequencies.
fundamental
frequency and Output the

form admittance busbar

matrix. voltages

Figure 6.1 Structure diagram of the harmonic penetration program

[\)
U1
126,

of the harmonic penetration algorithm is the system nodal admittance


matrix [Y ]. calculated at each harmonic h, Each element for a three
h
phase study is represented by a 3x3 block consisting of self and mutual
admittances between the phases and ground, The details of the formulation
of the [Y ] are given by Arrillaga et al (1985), Once [Y ] have been
h h
evaluated the harmonic current injections must be specified and with
them equation 6,1 is solved for the system voltage [V ] at each harmonic
h
of interest, If the individual system element admittances are preserved,
the currents in each element can then be calculated, i,e,

(6,1 )

When HVDC convertors are the sources of the harmonic currents then
determining the injected harmonic currents is a complex problem as they
cannot be considered fixed due to the strong interaction which exists
between the system and a large power convertor, HARMAC requires the
harmonic current injections to be specified in order to perform a
harmonic penetration study, Therefore a program external to HARMAC is
required for the simulation of this interaction and thus determination of
the injected current harmonics, Section 6,4 deals with two methods while
section 6,5 compares their results,

6,3 Characteristic Harmonics,

Idealized conditions such as balanced sinusoidal a,c. voltages at


convertor terminals, perfectly constant d,c, current, equal commutation
inductances and each valve conducting at equal time intervals, have been
in the past commonly used, Under these conditions the harmonics that
appear are known as the characteristic harmonics, For a convertor of
pulse number p the characteristic harmonics on the a,c, and d,c, sides
are given respectively by equations 6,2 and 6,3 '

h pq ± (6.2)

h pq (6.3)

where q is any positive integer.


127.

Since the six pulse Graetz bridge is the basic building block of
an HVDC convertor the pulse number will normally be 6 or 12 or perhaps
some higher multiple of 6. The characteristic harmonics produced by the
three phase bridge are well known and were reported as early as 1945 by
Reid (1945). With the increase in the number and power of convertors the
assumptions used are becoming invalid. The production of non-
characteristic harmonics due to non-ideal conditions have become a
problem. The reasons for non-ideal conditions are :
(a) firing errors.
(b) a.c. voltage unbalance and/or distortion.
(c) Ripple or modulation of d.c. current.
(d) Unbalance commutation inductance.

As the number and power rating of HVDC convertors increase the


operating conditions are becoming less ideal and the problems with non-
characteristic harmonics greater. This has necessitated the development
of more complex methods of analysis, and invalidates some of the
assumptions inherent in the earlier work.

6.4 Incorporation of HVDC Convertors.

The harmonic currents produced by convertors cannot be considered


fixed as convertor operation is influenced by the voltage distortion at
the terminal busbar. There is an inter-dependence between injected
harmonic current and voltage distortion with the a.c. system and filters
providing the link between the two. If the level of distortion is
reinforced by the action of more harmonic current injection due to the
convertor controllers response to the distortion, the convertors
operation is unstable. This phenomena is termed "harmonic instability"
and has been reported by several authors (Ainsworth 1967. Kauferle et al
1 970b) .

The use of physical scaled-down models (Simulators) seems an


obvious method and was indeed used in early studies (Laurent 1962),
however they lacked credibility for quantitative analysis of real
systems. One of the main problems with physical simulators is the
difficulty of modelling the strong frequency dependence of the various
power system components by means of discrete components. With the
128.

limitations of physical models harmonic studies are predominantly being


performed by computer modelling.

Time-domain dynamic programs of the types used to assess the


dynamic response to disturbances can also be used for steady-state
harmonic studies. Without fault application the analysis is used to
generate waveforms converging on the steady-state. The spectrum of the
derived waveforms are obtained via an FFT algorithm once the waveforms
have reached steady-state within sufficient accuracy.

Reeve et al (1974) used this approach to analyse the effect of


frequency deviations on the harmonic levels. The main shortcoming of this
study was the lack of accuracy of the a.c. system model at harmonic
frequencies. The authors acknowledged this problem and settled for a T
representation proposed by Bowles (1971).

Kitchin (1981) proposed the use of state-variable analysis for


evaluating convertor harmonics and mentioned the need for such a program
in conjunction with a Harmonic impedance and penetration program. The
representation of the convertor was via "tearing" and replacing by
equivalent sources rather than by the use of tensor analysis, and the
convertor was assumed to be the only non-linear device. Since a frequency
matched equivalent was not used a correct harmonic assessment is not
achieved.

The dynamic simulation programs use automatic step changing


algorithms to optimize program performance and to force the time
intervals to fall close to valve switching instances. Therefore linear
interpolation is normally used to generate the equally spaced data points
required for the FFT algorithm.

Using dynamic simulation programs for steady-state solutions is


expensive in computer time. This is because the steady-state solutions
are arrived at only after sufficient time has elapsed to ensure that the
initialization transient has decayed. The largest time constant of the
transient can be a few orders of magnitude greater than the integration
step-size. For reasons of numerical stability the step-size has to be
kept smaller than the smallest system time constant (Williams and Smith
1973) and smaller than a quarter of the period of the highest frequency
(Hay and Hingorani 1970).
129.

Methods of reducing the computational burden such as exploitation


of periodicity have been developed (Lipo 1971, Liou 1972, Ooi et al
1980), however, they greatly restrict the generality of such programs. In
these approaches each cycle of the supply frequency is divided into a
number of subintervals and the convertor is treated as a piece-wise
linear problem. Between two successive switchings, the network is
modelled by linear circuit elements, and its behaviour predicted by
solving the standard state space equations. The proponents of this method
acknowledge (Berube 1983) its impracticability for unbalanced conditions
due to the increase in complexity and number of equations. The other
shortcomings are the need for prior knowledge of the sequence of
network topologies, lack of ability to represent large a.c. systems and
the tedious equation formulation.

An alternative to dynamic modelling is to employ an iterative


process. Several authors (Reeve and Baron 1971, Yacamini and de Oliveira
1980a,1980b,1986, Harker 1980) have adopted this approach. A structured
diagram of an iterative harmonic algorithm (IHA) is given in figure 6.2.
Sinusoidal voltage at the convertors a.c. busbar are assumed for the
first iteration and the resulting a.c. harmonic current injections
calculated. The harmonic current injections are then used to assess the
voltage distortion at the convertor terminals. The controller strategy is
used to update the firing instants from the distorted terminal voltage
waveforms at each iteration. The distorted voltage waveform is then used
to reassess the injected harmonic currents. This process continues
iteratively until convergence is reached. Although this iterative
algorithm is based in the frequency domain it is not entirely in this
domain. The calculation of the harmonic current injections requires a
time domain representation of the commutation process as well as an FFT
to convert this to the frequency domain. Early studies assumed a constant
d.c. system, but more recent studies include d.c. system components and
their associated d.c. ripple (Yacamini and de Oliveira 1980b,1986,
Yacamini and Smith 1983).

The iterative harmonic algorithm accurately represents the


frequency-response of the power system by virtue of the independent
individual impedances used at each frequency. The second advantage of the
iterative algorithm is the extremely small computational effort required
compared to a time-domain solution. However, experience with the
Iterative Harmonic Algorithm
(IHA)

-1

I
Read the
1
Repeat Until Converged
harmonic
information
for the a.c.
and d.c. For each convertor The a.c. system The convetor The zero crossings
systems. solve for the a.c. voltages at the voltage and d.c. and firing
current injections convertor system model are instants are
Read the and the d.c. terminal are used to calculate updated because of
initial voltage. found for each the d.c. current the a.c. voltage
power flow harmonic. waveform. distortion
conditions
for the
convertor.

Figure 6.2 Structure diagram of the iterative Harmonic Algorithm

w
o
131.

iterative algorithm developed at the University of Canterbury has shown


that non-convergence can occur and in such cases there is little choice
except to use a time-domain program which contains a frequency-matched
a.c. system model, Recent studies into the non-convergence of the
iterative algorithm and comparisons with time-domain simulation have been
reported by Eggleston (1985), A recent publication (Yacamini and
de Oliveira 1986) also indicates that multiple solutions may exist, the
solutions obtained being dependent on the initial conditions. In view of
these concerns a simple test system was analysed using both the iterative
algorithm and time-domain algorithm in order to verify that they give
results consistent with each other and with theory, The lower South
Island (N,Z.) primary system is used as the test system to demonstrate
the importance of using a frequency matched a,c, system model,

Figure 6,3 shows the data flow and programs used for a complete
harmonic penetration study based on TCS derived assessment of the
convertor interaction. ACREP has been split to show the main functions it
performs and to aid clarity, A corresponding diagram showing data flow
and programs used when the iterative harmonic algorithm is used is given
in figure 6.4,

6,5 Illustrative Examples.

A simple system was devised to verify that the iterative algorithm


and dynamic results are indeed consistent with each other and theory, The
initial test system was chosen to be analytically solvable and is
depicted in figure 6.5, This was necessary to enable easy identification
of incorrect results when comparing the two approaches of harmonic
assessment. The smoothing reactor was chosen to be artificially large
(15.4 Henries or 10 p.u.) so that the d,c. current is virtually constant.
The 6 pulse convertor is fed via a convertor transformer represented by a
leakage reactance X and the a.c. system, which is purely inductive,
2
contains no harmonic filters.

Traditionally the commutating voltage is defined as the sinusoidal


voltage Ec which drives the commutation current (Arrillaga 1983), while
the commutating reactance is the intervening reactance between the
commutating voltage and convertor. This is not practical for a complex
Conductor data
and line geometry

1
Transmission Line Interacti ve I Harmonic 11 Graphical Display
Parameter Program Data Program Penetration Program of the Harmonics
(TL) (INTER) , "" (HARMAC) (ACREP)

5 Synthesis of
frequency-matched
equivalent (ACREP)

4 10

6 Power-Flow 7 Transient Convertor


Program ". Simulation
(PF) (TCS)

Graphics and Data 9 Assess Harmonic


Manipulation Program Levels
(JAUGPS) (ACREP)

1 Parameters of each transmission line over the required frequency range


2 Complete a.c. system data at fundamental and harmonic frequencies of each individual component
3 3x3 impedance loci matrix
4 Frequency-matched a.c. system equivalent
5 a.c. system data at fundamental frequency
6 Remainder of hvdc system data
7 Initial currents and voltages throughout the hvdc system
8 Voltage and current waveform data at unequally spaced intervals
9 Equally spaced waveform data
10 Harmonic levels
11 Harmonic Penetration results. The voltages and currents throughout the a.c. system

Figure 6.3 Data flow with TCS assessment of convertor interaction W


N
Conductor data
and line geometry

Transmission Line Interacti ve 2 Harmonic 8 Graphical Display


Parameter Program Data Program ~ Penetration Program of the Harmonics
(TL) (INTER) (HARMAC) (ACREP)

3
6 71
4
--...;:;....
I Power-Flow
Program
5
;;..
Convertor
Interaction Program
(PF) (IHA)

1 Parameters of each transmission line over the required frequency range


2 Complete a.c. system data at fundamental and harmonic frequencies of each individual component
3 a.c. system data at fundamental frequency
4 Remainder of hvdc system data
5 Initial currents and voltages throughout the hvdc system
6 3x3 impedance loci matrix
7 Harmonic levels
8 Harmonic Penetration results. The voltages and currents throughout the a.c. system

Figure 6.4 Data flow with IHA assessment of convertor interaction

w
w
134.

E
c

Figure 6.5 Simple test system

power system and the iterative algorithm uses the distorted voltage V as
t
the commutating voltage. The delay angle is measured from the zero
crossings of the Ec phase-to-phase voltages and 20 degrees was selected
for this study. The other parameters were; Vc= 13 p.u.,
X =X =0.1
1 2
p.u. and I =1.0 p.u . . The solution of equation 6.4 yields
d
19.07 degrees as the commutation angle. This was indeed verified by
measurement of the TCS waveforms.

Cos(a) - Cos (a + ~) (6.4)

where X
c
A certain amount of data manipulation was required in order to
compare the TCS and iterative harmonic algorithm phase angle results. The
TCS algorithm uses a cosine wave on phase "A" as its reference while the
iterative algorithm uses a sine wave as its reference. The iterative
algorithm also treats a.c. current harmonics as injections while TCS
treats them as a load current. Therefore to compare results the TCS
waveforms were advanced by 1/4 of a cycle and an FFT performed. The
iterative algorithm expresses the harmonics as the coefficients of a sine
series while the TCS results obtained via ACREP are the coefficients for
an exponential or sine series. As, sin(e+90) = cos(e), 90 degrees is
added to the phase angle of the TCS results to obtain the sine series
form.

The results for the harmonic current injected into the a.c. system
for phase "A" are tabulated in table 6.1. The analytic solution assuming
ideal conditions shows a vast disparity with the other results,
135.

Ideal TCS IHA


Harmonic
Order Classical Improved
Magn. I
Phase Magn. I
Phase Magn. Phase Magn. Phase

1 0.7799 149.3 0.7762 149.3 0.7669 149.3 0.7658 149.6


5 0.1557 -153.5 0.1392 -153.7 o . 1384 -152.9 0.1374 -152.3
7 0.1116 145.1 0.0887 144.5 0.0874 145.8 0.0876 146.4
11 0.0707 -157.8 0.0386 -160.7 0.0381 -159.3 0.0383 -157.8
13 0.0602 140.8 0.0244 134.8 0.0246 137.9 0.0243 138.2
17 0.0456 -162.0 0.0072 169.4 0.0073 174.7 0.0070 173.5
19 0.0413 136.6 0.0038 62.4 0.0039 74.7 0.0037 64.9
23 0.0337 -166.2 0.0057 51.6 0.0055 57.4 0.0058 57.2
25 0.0314 132.4 0.0063 -17.5 0.0062 -13.3 0.0064 -11.2

Table 6.1 Comparison of Phase "A" Harmonic Currents

TCS IHA
Harmonic
Order
Magn. Phase Magn. Phase

1 0.96319 -4.1 0.96300 -4.1


5 0.06907 -63.2 0.06949 -62.3
7 0.06099 -124.1 0.06191 -123.6
11 0.04213 -69.0 0.04218 -67.9
13 0.03218 -132.5 0.03134 -131.9
17 0.01257 -95.7 0.01189 -97.2
19 0.00734 165.2 0.00715 154.2
23 0.01245 146.5 0.01341 146.9
25 0.01524 77 .6 0.01597 78.4

Table 6.2 Comparison of Phase "A" Harmonic Voltages.

especially for the higher order harmonics. When the waveshape of the
commutation current is considered in the analytic solution the harmonic
levels obtained are far closer to the simulation results of TCS and the
iterative harmonic algorithm.

Table 6.2 displays the corresponding harmonic voltages at V for


t
phase "At!. The results of Table 6.1 and Table 6.2 show a close agreement
between TCS, the iterative algorithm and theory, once a more detailed
analysis of the commutation period is carried out. However, assessment
136.

E
c

Figure 6.6 Modified test system

assuming the classical ideal conditions leads to considerable errors


th
(e.g. 245% for the 13 harmonic). This error is small for the low order
harmonics and increases with the harmonic order.

Harmonic instabilities generally occur when the a.c. system is


resonant near a low order harmonic and the short circuit ratio is small.
To investigate problems with convergence of the iterative algorithm a set
of typical filters was added to the test system and the a.c. system
reactance was chosen to cause a parallel resonance with the harmonic
filters near the third harmonic. Tes was used to study the same situation
and assess whether a prospective non-convergence was indicative of a
harmonic instability or resulted from the numerical solution. The
modified test system is displayed in figure 6.6. As the power rating of
the convertor was increased the iterative algorithm took longer to
converge and eventually failed to converge. Tables 6.3 & 6.4 show the
results of the Iterative algorithm and TeS with a convertor power rating
low enough to allow convergence. Under these arduous conditions when the
iterative algorithm did converge the results compared well with the TeS
simulation results. However the non-convergence was found to be due to a
numerical instability and thresholding the uncharacteristic harmonics at
each iteration overcame this problem in many cases.

The two previous test systems were balanced and therefore no


uncharacteristic harmonics were expected except those due to the finite
precision of the computer. Also, the potential for modelling the
frequency-dependence of an a.c. system accurately, inherent in the
iterative algorithm's formulation, was not used.
13/

TCS IHA
Harmonic
.order
Magn. Phase Magn. Phase

1 .0.8111 167.2 .0.81.02 167.8


5 .0.1585 -64.1 .0.153.0 -61 .2
7 .0 • .0984 -86.1 .0.1.031 -85.9
11 0.0549 40.9 a.05 L16 44.2
13 .0 . .0402 16.3 0.0407 18.6
17 .0.0222 140.8 0.0219 145.3
19 0 • .0163 113 .5 0.0156 116.3
23 .0 • .0073 -134.8 0.0071 ;.;131.6
25 0.0.052 -171.4 0.0048 -177 • .0

Table 6.3 Comparison of Phase "A" Harmonic Currents

TCS IHA
Harmonic
.order
Magn. Phase Magn. Phase

1 0.91969 0.0 0.92.026 .0.0


5 0.0059.0 -66.8 0.00577 -66.7
7 .0 . .004.08 -94.2 0.00435 -97.1
11 0.00128 33.5 0.0.0125 35.4
13 0 • .00123 46.2 0.00116 43.1
17 .0.00772 -13.0.9 0.0.0768 -126.2
19 0.00743 -157.1 0.0.0714 -154.8
23 0.00468 -45.6 0.00462 -42.3
25 .0.00379 -82.4 0.0.0356 -87.6

Table 6.4 Comparison of Phase "A" Harmonic Voltages.

Next the Lower South Island (N.Z.) system was used as the
frequency-dependent a.c. test system. Two TCS simulations were performed,
the first using a Thevenin equivalent, the second the frequency-matched
equivalent circuit with implicit incorporation of mutual effects. The
iterative algorithm was then used to assess the convertors interaction
with the a.c. system and compared with the TCS results. The results are
summarized in Tables 6.5 (a),(b)&(c) and tables 6.6 (a),(b)&(c) for a
Thevenin and frequency-matched a.c. equivalent respectively.
138.

Harmoni c Phase A
Order
TCS IHA

1 1.5136 1.5094
3 0.0005 0.0014
5 0.2951 0.2742
7 0.1784 0.1760
9 0.0001 0.0011
11 0.0971 0.0825
13 0.0699 0.0555
15 0.0001 0.0006
17 0.0352 0.0260
19 0.0248 0.0187
21 0.0000 0.0003
23 0.0084 0.0143
25 0.0056 0.0130

Table 6.5(a) Phase "A" Harmonic Currents

Harmoni c Phase A Phase B Phase C


Order
TCS IHA TCS IHA TCS IHA

1 1.07265 1.07265 1.07493 1.07493 1.07630 1.07630


3 0.00027 0.00025 0.00063 0.00003 0.00032 0.00027
5 0.01188 0.00975 0.01285 0.00981 0.01149 0.01000
7 0.00769 0.00582 0.00826 0.00554 0.00812 0.00550
9 0.00115 0.00002 0.00057 0.00001 0.00130 0.00002
11 0.00221 0.00140 0.00214 0.00138 0.00243 0.00141
13 0.00207 0.00122 0.00213 0.00120 0.00203 0.00123
15 0.00009 0.00006 0.00007 0.00001 0.00005 0.00005
17 0.00905 0.00601 0.00912 0.00606 0.00908 0.00662
19 0.00804 0.00553 0.00809 0.00473 0.00808 0.00442
21 0.00004 0.00025 0.00008 0.00006 0.00011 0.00023
23 0.00368 0.00156 0.00367 0.00161 0.00355 0.00158
25 0.00269 0.00028 0.00269 0.00036 0.00279 0.00036

Table 6.5(b) Harmonic Voltages.

Harmonic D.C. Current


Order
TCS IHA

0 1.94473 1.94473
2 0.00073 0.00102
4 0.00039 0.00012
6 0.02321 0.02053
8 0.00015 0.00013
10 0.00011 0.00007
12 0.00237 0.00302
14 0.00005 0.00005
16 0.00005 0.00004
18 0.00175 0.00199
20 0.00004 0.00004
22 0.00003 0.00003
24 0.00171 0.00102

Table 6.5(c) Comparison of d.c. Current Harmonics

Table 6.5 TCS/IHA comparison with Thevenin a.c. system equivalent


139.

Harmonic Phase A
Order
TCS IHA

1 1 .4794 1.4782
3 0.0026 0.0031
5 0.2851 0.2849
7 0.1730 0.1708
9 0.0004 0.0009
11 0.0916 0.0900
13 0.0653 0.0632
15 0.0004 0.0006
17 0.0312 0.029 11
19 0.0209 0.0197
21 0.0003 0.0006
23 0.0068 0.0079
25 0.0054 0.0069

Table 6.6(a) Phase "A" Harmonic Currents

Harmonic Phase A Phase B Phase C


Order
TCS IHA TCS IHA TCS IHA

1 1.04677 1.04677 1.04769 1.04769 1.02974 1.02974


3 0.00097 0.00093 0.00178 0.00141 0.00087 0.00060
5 0.00968 0.00985 0.00973 0.00985 0.01006 0.01019
7 0.00331 0.00597 0.00626 0.00574 0.00595 0.00557
9 0.00005 0.00002 0.00011 0.00010 0.00006 0.00005
11 0.00216 0.00160 0.00148 0.00154 0.00144 0.00162
13 0.00188 0.00147 0.00165 0.00146 0.00145 0.00142
15 0.00019 0.00005 0.00036 0.00023 0.00035 0.00018
17 0.01191 0.00797 0.00726 0.00763 0.00936 0.00893
19 0.01699 0.00703 0.00558 0.00607 0.00480 0.00502
21 0.00022 0.00021 0.00119 0.00097 0.00057 0.00104
23 0.00067 0.00097 0.00101 0.00091 0.00071 0.00104
25 0.00014 0.00016 0.00037 0.00024 0.00017 0.00020

Table 6.6(b) Harmonic Voltages

Harmonic DC Current
Order
TCS IHA

0 1.8991 1.8991
2 0.0043 0.0044
4 0.0003 0.0005
6 0.0211 0.0223
8 0.0002 0.0004
10 0.0001 0.0002
12 0.0024 0.0025
14 0.0001 0.0001
16 0.0002 0.0001
18 0.0016 0.0019
20 0.0002 0.0001
22 0.0000 0.0001
24 0.0014 0.0014

Table 6.6(c) Comparison of d.c. CUrrent Harmonics

Table 6.6 TCS/IHA comparison with frequency matched


a.c. system equivalent
140.

The TCSIIHA comparison with Thevenin a.c. system equivalent shows


reasonable agreement between the harmonic current injections, but
considerable disparity between the harmonic voltages at the convertor
terminals. This indicates that the harmonic current is reasonably
insensitive to the frequency response of the a.c. system while the
voltage distortion is greatly dependent on it. It can also be noted that
the TCS evaluated voltage harmonics are larger than those of the IHA,
th rd
e.g. 10 times larger for the 25 harmonic, 2 times for 23 harmonic and
th
1.5 times for the 19 harmonic. The impedance of the Thevenin equivalent
increases with increasing frequency while that of the actual system shows
numerous troughs and peaks as frequency increases. The nett result is
that the Thevenin impedance tends to be larger than that of the actual
system at higher frequencies. Since, as mentioned previously, the
harmonic current level is approximately the same the assessed voltage
distortion is considerably higher and thus the disparity increases with
increasing frequency.

The TCS results using the Frequency-matched a.c. system show


good correspondence with the Iterative algorithm (IHA). In particular the
higher order voltage harmonics show very good agreement.

Thresholding of the harmonics at each iteration was not used as


this made the numerical process less stable and caused non-convergence
in some simulations of unbalanced a.c. systems.

6.6 Computational Efficiency.

The CPU requirements of the TCS algorithm with Thevenin or


frequency-dependent model and the corresponding IHA simulation have a
complex dependence on many factors, such as the relative strength of the
a.c. system and convertor rating, convergence tolerances, initial
conditions etc. Therefore the numerical comparison made in table 6.7 is
only valid for the test cases used to derive them and provides only an
approximate indication of their relative CPU requirements. The frequency
response of the lower part of the New Zealand South Island system was
used for the two cases listed where frequency-dependence was modelled.
The results were obtained on a VAX 11/750 computer.
141.

As mentioned previously TCS requires a number of cycles of


simulation for the initialisation transient to die down before the
harmonic assessment can be made. The number of cycles required depends
largely on the time constants and damping of the circuit as well as the
specified initial conditions. Generally eight to ten cycles are
sufficient.

It should also be noted that TCS has a greatly increased CPU


requirement for the last cycle of simulation. This is caused by the
larger number of data points being written out into a data file, in order
to increase the precision of the FFT process. Moreover, as the TCS
program incorporates an adaptive step-size selection algorithm, the
program JAUGPS must derive a set of equally spaced data points from TCS's
unequally spaced information.

By way of example the TCS solution takes about 32 minutes to


assess the harmonic levels with a frequency-dependent a.c. system
equivalent. In contrast the IHA program only takes 4 minutes to produce
the same information.

A.C. System model used TCS IHA


in both algorithms.
per cycle total total

Thevenin 106
283 Clast cycle) 11 31 148

Frequency-dependent 183
439 Clas t cycl e) 1903 241

Table 6.7 Comparison of Computational Efficiency (in sec.)


1 42.

6.7 Harmonic Penetration Example.

To complete the harmonic penetration study the assessed convertor


currents, obtained from either TCS or IHA, are supplied to HARMAC, which
then calculates the branch harmonic currents and busbar harmonic
voltages. When TCS is used to assess the steady state convertor currents
the harmonic penetration study is self checking; when the frequency
response of the a.c. system has been correctly modelled, the harmonic
voltages at the terminal busbar derived from HARMAC will agree with the
harmonic voltages obtained from the TCS simulation for this busbar.

To illustrate the use of TCS in conjunction with HARMAC for


harmonic penetration studies the assessed current using a frequency-
matched a.c. system equivalent (Table 6.6(a)) were used in HARMAC to
obtain the harmonic penetration throughout the a.c. system. The busbar
voltages and branch currents are summarized in Tables 6.8 and 6.9 while
figures 6.7 and 6.8 graphically display these results.

HARMONIC VOLTAGE MAGNITUDES


TCS ASSESSMENT

w
o 0.03
::J
f- TIWAI----220
H~

z+l
·rl O. 02
c.9 [
<t::J
2:
wL 0.01
c.9 OJ
<tf!:,
f-
..J 0 . 0 0 ..JLLb.LL.'L.L.}.L.~e:.L.LI...<,;.L..L-L..J,L-.L..'-""""""""'~
o 1 5 9 13 17 21
> BUSNAMES

ORDER OF HARMONIC

Figure 6.7 Harmonic voltage throughout the Lower South Island


of New Zealand test system.
143.

Busbar
Harmonic
INVER CARGILL MANAPOURI ROXBURGH
Order
33 kV 220 kV 14 kV a 14 kV b 220 kV 11 kV 1 11 kV g 220 kV

5 0.00716 0.00992 0.00517 0.00518 0.00897 0.00642 0.00419 0.00692


0.00737 0.00992 0.00516 0.00517 0.00911 0.00653 0.00431 0.00712
0.00735 0.01032 0.00531 0.00532 0.00925 0.00624 0.00438 0.00704

7 0.00449 0.00741 0.00485 0.00485 0.00895 0.00510 0.00332 0.00577


0.00423 0.00698 0.00516 0.00517 0.00873 0.00488 0.00361 0.00578
0.00434 0.00670 0.00487 0.00488 0.00833 0.00470 0.00345 0.00543

9 0.00007 0.00002 0.00015 0.00015 0.00003 0.00012 0.00007 0.00001


0.00014 0.00026 0.00018 0.00018 0.00050 0.00022 0.00009 0.00025
0.00007 0.00020 0.00033 0.00033 0.00048 0.00010 0.00016 0.00022

11 0.00029 0.00087 0.00398 0.00399 0.00641 0.00073 0.00061 0.00106


0.00025 0.00084 0.00358 0.00359 0.00622 0.00061 0.00058 0.00108
0.00034 0.00037 0.00371 0.00371 0.00708 0.00078 0.00048 0.00070

13 0.00044 0.00103 0.00146 0.00146 0.00266 0.00228 0.00150 0.00296


0.00038 0.00103 0.00149 0.00149 0.00253 0.00192 0.00188 0.00286
0.00040 0.00091 0.00138 0.00138 0.00237 0.00181 0.00153 0.00231

15 0.00002 0.00001 0.00007 0.00007 0.00003 0.00016 0.00011 0.00006


0.00004 0.00008 0.00010 0.00010 0.00027 0.00029 0.00014 0.00037
0.00002 0.00009 0.00017 0.00017 0.00025 0.00013 0.00025 0.00036

17 0.00084 0.00203 0.00536 0.00537 0.00857 0.00359 0.00378 0.00497


0.00094 0.00290 0.00457 0.00458 0.00800 0.00417 0.00327 0.00610
0.00093 0.00320 0.00507 0.00508 0.00963 0.00414 0.00380 0.00673

19 0.00184 0.00654 0.00292 0.00292 0.00629 0.00539 0.00446 0.00871


0.00164 0.00569 0.00339 0.00339 0.00555 0.00478 0.00517 0.00811
0.00158 0.00436 0.00303 0.00303 0.00451 0.00465 0.00458 0.00641

21 0.00017 0.00016 0.00032 0.00032 0.00013 0.00041 0.00036 0.00013


0.00032 0.00116 0.00037 0.00037 0.00114 0.00076 0.00041 0.00122
0.00015 0.00083 0.00068 0.00068 0.00093 0.00035 0.00077 0.00096

23 0.00028 0.00124 0.00095 0.00095 0.00142 0.00060 0.00085 0.00122


0.00032 0.00116 0.00070 0.00070 0.00125 0.00070 0.00063 0.00110
0.00038 0.00138 0.00082 0.00082 0.00164 0.00081 0.00074 0.00134

25 0.00015 0.00052 0.00045 0.00045 0.00081 0.00031 0.00032 0.00049


0.00015 0.00069 0.00049 0.00049 0.00091 0.00033 0.00035 0.00064
0.00014 0.00058 0.00051 0.00051 0.00080 0.00029 0.00037 0.00056

Table 6.8 Harmonic Voltages Throughout the Lower South Island


of New Zealand test system. Based on TCS assessed harmonic
currents using implicit accounting for frequency-dependent
mutual coupling.
144.

Jlarmonic BRANCHES

Order
A B C D E F G H I

1 0.73145 0.36648 0.27142 0.25781 0.22882 0.22913 0.09768 0.42806 0.04876


0.73054 0.36602 0.26945 0.25624 0.23264 0.23275 0.09683 0.43494 0.04874
0.72705 0.36427 0.26812 0.25460 0.22835 0.22905 0.09901 0.42760 0.04783
5 0.02643 0.01324 0.00730 0.00548 0.00927 0.00915 O.OlDon 0.01303 0.00674
0.02645 0.01325 0.00797 0.00608 0.00949 0.00936 0.01030 0.01312 0.00685
0.02743 0.01374 0.00733 0.00556 0.00953 0.00939 0.01027 0.01330 0.00655

7 0.01917 0.00960 0.01390 0.00723 0.00632 0.00615 0.00627 0.00791 0.00536


0.01792 0.00898 0.01445 0.00683 0.00621 0.00605 0.00592 0.00753 0.00512
0.01749 0.00876 0.01303 0.00662 0.00580 0.00567 0.00607 0.00727 0.00494

9 0.00005 0.00003 0.00010 0.00010 0.00001 0.00001 0.00010 0.00001 0.00013


0.00082 0.00041 0.00099 0.00093 0.00024 0.00023 0.00019 0.00028 0.00023
0.00079 0.00040 0.00093 0.00087 0.00019 0.00019 0.00009 0.00021 0.00010

11 0.00806 0.00404 0.01175 0.01323 0.00102 0.00096 0.00041 0.00112 0.00077


0.00849 0.00425 0.01186 0.01338 0.00093 0.00086 0.00035 0.00129 0.00064
0.00982 0.00492 0.01293 0.01341 0.00043 0.00041 0.00047 0.00046 0.00082

13 0.00304 0.00152 0.00408 0.00474 0.00227 0.00210 0.00062 0.00228 0.00239


0.00293 0.00147 0.00411 0.00,174 0.00217 0.00199 0.00053 0.00194 0.00201
0.00257 0.00128 0.00362 0.00423 0.00170 0.00157 0.00056 0 .. 00165 0.00190

15 0.00003 0.00002 0.00002 0.00009 0.00004 0.00004 0.00003 0.00004 0.00017


0.00026 0.00013 0.00009 0.00066 0.00026 0.00023 0.00005 0.00022 0.00031
0.00024 0.00012 0.00009 0.00057 0.00025 0.00022 0.00002 0.00023 0.00014

17 0.00730 0.00365 0.03762 0.04156 0.00327 0.00285 0.00118 0.00216 0.00376


0.00680 0.00340 0.03974 0.04354 0.00418 0.00364 0.00132 0.00383 0.00437
0.00851 0.00426 0.03495 0.04767 0.00435 0.00367 0.00130 0.00395 0.00434

19 0.00559 0.00280 0.00341 0.00326 0.00550 0.00441 0.00257 0.00483 0.00566


0.00422 0.00211 0.00362 0.00366 0.00530 0.00.131 0.00230 0.00405 0.00502
0.00306 0.00153 0.00384 0.OU351 0.00394 0.00314 0.00221 0.00266 0.00488

21 0.00018 0.00009 0.00011 0.00007 0.00007 0.00005 0.00024 0.00014 0.00043


0.00089 0.00045 0.00104 0.00060 0.00079 0.00055 0.00045 0.00063 0.00080
0.00057 0.000213 0.00086 0.00049 0.00061 0.00039 0.00021 0.00035 0.00037

23 0.00087 0.00044 0.00214 0.00127 0.00046 0.00080 0.00039 0.00049 0.00063


0.00085 0.00042 0.00196 0.00115 0.00043 0.00070 0.00045 0.00051 0.00074
0.00105 0.00052 0.00245 0.00145 0.00053 0.00082 0.00053 0.00052 0.00085

25 0.00049 0.00024 0.00143 0.00105 0.00023 0.00023 0.00020 0.00016 0.00033


0.00055 0.00028 0.00174 0.00123 0.00031 0.00030 0.00021 0.00027 0.00035
0.00045 0.00022 0.00148 0.00109 0.00027 0.00027 0.00019 0.00021 0.00031

Branch Designation
WINAPOUR1014 TO HlINI\POURI220 A
HIINAPOUR2014 '1'0 /-IIINAPOURI220 B
~IANAPOURI220 TO TlI'iAI----220 C
HANAPOURI220 TO INVERCARG220 D
ItlVERCARG220 'ro ROXBURGII-220a E
INVERCARG220 TO ROXBURGH-220b f'
HNERClIRG220 TO INVERCIIRG033 G
ROXDURGlIl0l1 TO ROXDURGII-220 II
ROXBURGII-220 '1'0 HOXBURGII-Ol1 I

Table 6.9 Harmonic Branch Currents at sending end. throughout the Lower South Island
of New Zealand test system. Based on TCS assessed harmonic currents using
implicit accounting for frequency-dependent mutual coupling.
145.

HARMONIC CURRENT MAGNITUDES


TCS ASSESSED

w
o .08
:::::J
I-
H-:;.06
Z .r!
l'J c
i:::::J .04

I- L
Z ~. 02
W~
0::
0:: . 0 0 .JL:.':;;"";:;'.L!...)L..I.-'-<:::L....:"""-"""'-L....:::.LL""",-,L.:>.L:"::"""'-"'-",,,-,,-
:::::J 1 5 9 13
U
ORDER OF HARMONIC

Figure 6.8 Harmonic Branch currents at sending end for Lower


South Island test system

6.8 Conclusions.

The computational cost clearly makes it desirable to use the


iterative algorithm (IHA) wherever possible. However, the possibility
of false solutions and numerical instability exist with the iterative
algorithm and in such cases a TCS simulation program must be used to
obtain reliable results.

In general, increasing the number of sample points used for the FFT
improves the accuracy. However beyond a certain point the accuracy
deteriorates as the round-off errors and errors associated with
obtaining equally spaced data points, build up. Although the results of
the previous sections were all obtained using 2048 sample points,
subsequent tests have shown that the use of 1024 sample points provides
better results.
146,

6,9 Future Work.

Further work is required into the numerical instabilities of the


iterative algorithm. Two factors can contribute to the IHA failing to
converge, they are: (i) a very weak a.c. system and/or (ii) a resonance
between the a.c. system and harmonic filters near a low order harmonic.
It appears that thresholding the harmonics at each iteration helps
convergence in the case of balanced systems but hinders convergence if
the a.c. system is unbalanced. With reference to the recent work by
Yacamini and de Oliveira (1986) the question must be asked of whether
multiple solutions can exist depending on the initial conditions. This
possibility must be investigated before any faith can be placed on the
IHA results.

The elusive goal of studying the effect and interaction of multiple


harmonic sources, such as two convertors, needs to be addressed. This is
particularly relevant to New Zealand's South Island system where there
exist two large convertors, the HVDC link with convertor station at
Benmore and an aluminium smelter at Tiwai. At present the TCS approach
looks the best for achieving this as it can already model multiple
convertor systems and has proved reliable. However, the accuracy of the
frequency-dependent model may be the limiting factor. Although the
extension of the frequency-dependent equivalent for multiple harmonic
sources was detailed in chapter 4 the alterations required to implement
this in TCS have yet to be made,
147.

CHAPTER DYNAMIC SIMULATIONS

7.1 Introduction.

To demonstrate the effect of the a.c. system representation on the


dynamic performance of HVDC convertors various faults are applied to the
a.c./d.c. system depicted in figure 7.1.

The lower South Island of New Zealand is again used as the test
system and comparisons are made between a Thevenin equivalent of this
system and a frequency-matched equivalent. The filter data used is the
monopolar equivalent of the New Zealand hvdc link filters (Robinson
1966). A simple Thevenin representation is used for the a.c. system at
the receiving end of the d.c. link (system 2) in sections 7.2 and 7.3.
Finally, in section 7.4, the detailed a.c. representation and the a.C.
fault location are placed at the receiving end (system 2) in order to
assess the effect of the disturbance at the invertor end.

Transmission Line

III,
I JlI
A.C.
System

Figure 7.1 HVDC Test System


148.

7.2 D.C. Line Fault Simulation.

Figures 7.2 (a)&(b) show the rectifier current and d.c. voltage on
the convertor side of the rectifier's smoothing reactor, when a d.c. line
fault is applied midway between the two convertors. The control strategy
on the rectifier is such that on detection of the fault the convertor is

7.5

;J.~~
7.121

6.5

6.121 ---- THEVENIN EQUIVALENT


................... - FREQUENCY-MATCHED EQUIVALENT No. 1
5.5

~!.~
--._- FREQUENCY-MATCHED EQUIVALENT No. 3
5.121 ----- FREQUENCY-MATCHED EQUIVALENT No. 4

4.121

3.5
~
~\
3.121

2.5

2.121 .~
1.5

1.121
~
121.5

121. 121
~
-121.5
51214121.121 522121.121 54121121. 121 558121.121 576121.121
T I ME (DEGREES)

(a) Rectifier Current

4.5r-------.-------~--------r_------,_------~--------r_------._------_,

1\
3. 121

-;0;~12I-.~12I---5~2~2~12I~.~12I~~5~4~12I~12I~.-12I--~5~5~8~12I-.~12I---5~7~6~12I-.-I2I----5-9-4~12I-.-I21----6-1-2LI2I-.I21----6-3-12I~12I-.-I2I----6-4~8121.121
T I ME (DEGREES)

(b) D.C. Voltage

Figure 7.2 D.C. Line Fault Simulations


149.

Transformer Smoothing Transmission


Reactance Reactor Line Reactance

A.C.
System

Figure 7.3 System after Fault application

ordered to operate in the inversion mode. This speeds up the de-


energizing of the line and hence arc ~xtinction. Once the arc has been
extinguished the convertor is in the blocked (non-conducting) state.
During this period the d.c. voltage oscillates with the natural frequency
of the line. After a delay larger than the deionization time the line can
be re-energized and normal operation resumed.

Figure 7.2 shows that the convertor currents estimated by the


different models differ only slightly. This can be explained with
reference to figure 7.3 which displays the circuit looking from the
rectifier end immediately after fault occurrence. Due to the position of
the fault, the transformer, smoothing and d.c. line reactances limit the
rate of change of current during the fault.

The d.c. voltage waveforms show (figure 7.2 (b)) greater variation
between the various alternative models. This is due to its sensitivity to
the trapped energy in the d.c. line once blocking of the convertors has
occurred. The d.c. voltage is primarily determined by the inductor
voltage and hence dependent on the rate of change of current.
150.

7.3 Rectifier Side a.c. System Disturbance.

The fault is a single phase short-circuit applied to phase 'IA" and


the single line diagram of figure 7.4 depicts the fault location. Figures
7.5 (a)&(b) display the a.c. system current and busbar voltage waveforms
respectively, when a Thevenin equivalent is used. Figures 7.6 (a)&(b)
show the corresponding currents and voltages when a frequency-matched
a.c. system equiv~lent is used. There is considerable difference between
the two sets of results.

The interaction of the power system and convertor controller, after


a disturbance, complicates the transient behaviour. Thus to observe
exclusively the effect of the different a.c. system representations the
convertor's C.C.C. and E.A.C. controllers are disabled. Figures 7.7
(a)&(b) and figures 7.8 (a)&(b) display the a.c. system currents and
busbar voltage waveforms with the Thevenin and the frequency-matched
equivalents respectively. There are considerable differences between the
two sets of results; for instance the peak voltage in phase "B" for the
frequency-matched equivalent (figure 7.8(b)) is 28% greater than the
corresponding peak for the Thevenin alternative (figure 7.7(b)). With
the Thevenin model the a.c. system current (figure 7.7(a)) on the faulted
phase becomes sinusoidal as the fault forms a simple RL series circuit
driven from the fundamental-frequency voltage source, with no possibility
of resonances occurring. However the corresponding current with the
frequency-matched a.c. system equivalent (figure 7.8(a)) does show some
distortion due to the resonances present. Moreover the frequency-matched

A.C. system
CUrrent
A.C.
System

Figure 7.4 Rectifier side a.c. system fault


151.
30.---------------r--------------,---------------,--____________-,
25_ Phase A
- . _ - Phase B
20
- - - - Phase C
15

10

-5

-10

-15

-20

-25

-30L-__________
5040.0
~~~~--------~~~~--------~~~----------~~
5220.0 5400.0 5580.0 5760.0
T I ME (DEGREES)
(a) A.C. System Current

1.4

1.2

1.0

0. 6

0.6

0. 4

0.2
.J
0.0
! \ '\
-0.2
~ ,
v\ .
-0. 4
, ~\(\ / .\ I
-0.6
1/ . J ~ .

-0.8
'.,..1\
..- \1\1 I'j .
\/-V
~

-1. 0

-1. 2

-1.4L-__________~~~~--------~~~~--------~~~----------~~[
5040.0 5220.0 5400.0 5580.0 5760.0
T I ME (DEGREES)

(b) Busbar Voltage

Figure 7.5 Dynamic response due to Rectifier side Fault with


Thevenin Representation

a.c. system equivalent also models the coupling between phases and thus
any distortion present in the two unfaulted phases is reflected into the
faulted phase. The fault effectively shorts out the harmonic filters on
the faulted phase and thus the dynamic response is that of the a.c.
system following the sudden application of a short circuit.

It is also interesting to compare figures 7.8 (a)&(b) with figures


7.6 (a)&(b), as this clearly demonstrates how the convertor's controller
action has considerable influence on the dynamic performance. The busbar
voltage peak in phases "B" & "e" are substantially reduced due to the
effect of the controller.
152.

301r----------------,----------------1I----------------r---------------~

25

20

15

10

\//
-1121 ."-..,.. X.... _____ / /
-15

-2121

-25

-301L-__________~~~~--------~~~~--------~~~~--------~~
512140.121 522121.121 54121121.121 558121.121 576121.121
T I ME <DEGREES)

(a) A.C. System Current

1.4.-______________. -______________- ,______________- .______________

~
1.2

1.121

121.8

121.6
\
121.4 \

121.2
\
121. 121

-121.
\
\
-121.8 \ .~.
/'
-1. 121 -V
-1. 2

-1.4L-____________~~~----------~~~----------~~~~--------~~
512140. IZI 522121.121 54121121.121 SSB0.121 S76121.121
TIME <DEGREES)

(b) Busbar Voltage

Figure 7.6 Dynamic response due to Rectifier side Fault with


Frequency-dependent Representation
153.

3121r-______________- .________________, -________________, -______________- ,

25

2121

15

1121
,
5 \

\
-5
'\ \)(.-r--~
-}121 '----..../ '--- - /

-15

-2121

-25

-3121L-____________~~~~----------~==~----------~~=_=_----------~~
51214121.121 522121.121 54121121.121 558121.121 576121.121
T I ME (DEGREES)

(a) A.C. System Current

1.4

1.2

1.121

121. 8

121.6

121. 4

121.2
J
121.121

-121.2
I I
I

-121. 4

-121.6
\

'V"'" ,../
I
I \ ~. )
-121.8

-1. 121
" \."-J.\-1
-I. 2

-1.4L-______________~______________=_~-----------------L-------------~~
51214121.121 522121.121 54121121.121 558121.121 576121.121
T I ME (DEGREES)

(b) Busbar Voltage

Figure 7.7 Dynamic response due to Rectifier side Fault with


Thevenin Representation
154.

30'~-----------r------------,------------r----------~

25

20

15

I /

''I' /

, ~--
-15

-20

-25

-30~------------~~~~----------~5~4~0~0~.~0~----------!5~5518~0a..~0;------------;577660.0
5040.0 5220.0 TIME <DEGREES)

(a) A.C. System Current

/ \
-0.8
\
--..,./
/
/

\ /
-1.0
\ Ii
-I. 2 \ .r
-1.4L-----------------L-------------~5~4~0~0~.~0~----------~5~5~8~0;.~0;------------5~776~0.0
5040.0 5220.0 TIME <DEGREES)

(b) Busbar Voltage

Figure 7.8 Dynamic response due to Rectifier side Fault with


Frequency-dependent Representation
155.

7.4 Invertor Side a.c. System Disturbance.

In this series of simulations it is the representation of a.c.


system 2 that is varied and faulted, as depicted in figure 7.9, An HVDC
system is more sensitive to disturbances in the invertor side of the link
than the rectifier due to the possibility of commutation failures. Hence
any differences in the dynamic response should be more observable. Three
alternative a.c, system representations were used i,e. a Thevenin
equivalent and a frequency-matched equivalent (No.3, figure 5,36) with
and without mutual coupling. Figures 7,10, 7.11 & 7.12 display the busbar
voltage, a.c. system current and fault branch current respectively, for
each of the representations.

A.C. System
Current
A.C.
System
2

Figure 7.9 Invertor side a.c. system fault

Figures 7.10 (b)&(c) show that the resonances following fault


application cause numerous multiple zero crossings, which always pose a
problem for convertor controllers.

Following the voltage collapse in the faulted phase, the


corresponding convertor valves cannot conduct, hence with the Thevenin
equivalent, phase "A" is a simple RL circuit uncoupled to the other
two phases. Therefore the phase "A" a.c. system current (figure 7,11 (a))
is sinusoidal for the Thevenin equivalent. Figure 7.11(b) shows that with
the frequency-dependent model (without mutual coupling) the phase "A"
a.c. system current is distorted immediately after fault application but
the distortion dies away with time. This distortion is due to the dynamic
156.
1.4r----------------.-----------------r----------------.---------------~

1.2
- - - Phase A
1. '" _

"'.9
\

-"'.4 \\ /
-0.6

-121.8
\. ).~
-1. 121
\. J .

\.
-1.4L-____________
51214121_ t2I
~~L--------------~~~------------~~~------------~
S22121~
121 5400. 121 5580. 0 576121. 12)

TIME <DEGREES)

(a) Thevenin Representation


1.4.-______________. -______________-r______________-,______________ ~

1.2

1.121
(\
" ,
J \
/ /\ \
. 1'"'/
I
\
.
.~.
\ \

~ l\, I \
\

\ I ..... ------- " -' I


\ I
\ ! \ .",--.)
-1. 0

-1. 2
\)
-1.4L-______________-L____________~~~~----------~~~=_----------~~
5040. I2l 5220.0 5400.0 5580.121 576121.121
T I ME (DEGREES)

(b) Frequency-dependent representation without mutual coupling


1.4r----------------r----------------.-----------------.----------------,
1.2

1.121
1'.
121.8
.'\
/ .\ / /\ ,....F' \
.

\
-0. 2

-121. 4

-121.6

-0.8

-1. 0

-1. 2

-1.4L-______________ ____________
~ ~~~~------------~--------------~_:".

504121.121 522121.121 54121121.121 558121.121 576121.121


T I ME (DEGREES)
(c) Frequency-dependent representation

Figure 7.10 Faulted Busbar voltage


24 __--______________T -________________. -________________i r________________- , 157.
20

16

12
.r--" ____... ~'\
/
8

'

/I/\,,_V\,
-8

-12

-16

-20

-24L-________________L-________________ ~ ______________~~~------------~=
5040.0 5220.0 5400.0 5580.0 5760.0
TIME (DEGREES)
(a) Thevenin Representation

24~----------------.-----------------._----------------.-----------------,

20

16

12

-16

-20

-24L-________________L -________________L-________________ ~ ______________ ==~

5040.0 5220.0 5400.0 5580.0 5760.0


T I ME (DEGREES)
(b) Frequency-dependent representation without mutual coupling
24.-________________,-________________.-________________,-________________ ~

20

-24L-________________L-________________ ~ ________________ ~ ________________ ~

5040.0 5220.0 5400.0 5580.0 5760.0


T I ME (DEGREES)
(c) Frequency-dependent representation

Figure 7.11 A.C. System Current


158.
40

35

30

25

20

15

10

0
-5

-10

-15

-20

-25

-30

-35
-40
5040.0 5220.0
T I ME (DEGREES)

Figure 7.12 Fault current

response of the circuit that models the frequency-response and hence the
resonances of the a.c. system. The remaining two phases experience
enormous distortion due to the upset in the valve conduction pattern.
When a frequency-matched e·quivalent with mutual coupling is used (figure
7.11 (c)) this distortion is reflected into phase "A" causing the faulted
phase current to be grossly distorted.

It is also interesting to note the permanent current offset that


occurs in the fault branch current, which is dependent on the point of
fault application, when either the Thevenin model or frequency-matched
model without mutual coupling are used. Although the offset also appears
when the frequency-matched and coupled equivalent is used, in this case
it appears to decay with time (figure 7.12).

7.5 Discussion and Conclusions.

The choice of gains for the convertor controller requires


considerable experience as dynamic instability can occur due to the
interaction between the a.c. system and convertor. The strength of the
a.c. system connected to the convertor is one factor that influences this
choice. The Thevenin equivalent can suffer from this type of instability,
although it is less prone than the frequency-matched equivalent. Such
instabilities are also experienced in practice (Gunn 1966) and have been
159.

the subject of much investigation. Although it is not the purpose of the


thesis to deal with this problem, it has been demonstrated that the
controller does significantly influence the dynamic response. Moreover
with a valve in a bridge being fired nominally every 60 degrees the
convertor has the ability to respond quickly to changes by advancing or
delaying the firing instant.

The rectifier current is relatively insensitive to the a.c. system


representation when d.c. line faults are simulated. The reasons for this
are; the presence of transformer, smoothing and line reactances, the
symmetry of the fault and the presence of the harmonic filters. The d.c.
voltage cannot be influenced by the a.c. system representation when the
convertor is blocked; however, in the interval between fault application
and convertor blocking the a.c. system equivalent will have considerable
effect on the dynamic behaviour and hence on the line oscillations that
occur after blocking.

The a.c. system representation has a greater influence on the


dynamic behaviour following an a.c. system disturbance. The effect of the
frequency-dependent self impedances is to distort the faulted phase
current due to the dynamic response caused by an abrupt change, the
amount of distortion, and its duration, being determined by the
characteristics of the a.c. systems frequency-response. Finally the
addition of mutual coupling also shows the current distortion induced by
the unfaulted phases, which is more realistic.
160.

CHAPTER 8 CONCLUSIONS

With the rapid growth in both size and number of HVDC schemes there
is a greater need for the accurate simulation of transient phenomena,
particularly in cases where the d.c. convertor station feeds into a
relatively weak a.c. system. Although explicit modelling of essential
a.c. components (e.g. convertor transformer, harmonic filters) has been
retained, due to the size and complexity of modern power systems some
form of equivalent circuit must be used for the remainder of the a.c.
system. Towards this end the main aim of this thesis has been to present,
and show the effect of, a three phase a.c. system model that accurately
represents the a.c. system at harmonic frequencies. Both the
frequency-dependence of the self and mutual terms associated with each
phase have been accurately modelled. Several frequency-matched a.c.
system models have been presented with any inherent assumptions outlined.

The accuracy of the proposed frequency-matched a.c. system models


has been verified by the use of a Harmonic current source model to inject
a specified harmonic component and applying an FFT to the resulting time
domain waveforms. These results compared very well with the expected
values derived from a harmonic penetration study of the complete system
in the frequency domain. Further verification of the proposed a.c. system
equivalent was made by comparison between an Iterative Harmonic Algorithm
and Transient Convertor Simulation for assessing convertor harmonics.

Both a.c. and d.c. system disturbances have been presented and the
response of the proposed models show considerable difference with the
Thevenin representation, particularly in the case of an asymmetrical
fault.

The use of a Thevenin model based on the S.C.R. of the a.c. system
only represents the system accurately at fundamental frequency and
therefore gives inaccurate results at harmonic frequencies. The
resonances between the a.c. system and harmonic filters as well as
resonances within the a.c. system must be modelled for accurate dynamic
simulation. It is the combined impedance response of the a.c. system in
parallel with harmonic filters that needs to be accurately modelled. The
resonances in the parallel combination may be excited by disturbances,
161.

resulting in waveform distortion and possible overvoltages, which will


affect the convertor controllers and their response. If the harmonic
filters dominate the combined response at the high order harmonics then
the a.c. system model may only need to be accurate at the lower order
harmonics, below the high pass filter's resonant frequency. Bowles (1970)
and Giesner (1971) both proposed equivalents that more realistically
represented the a.c. system at low order harmonics on this basis.
However, a general model should be capable of representing a.c. system
resonances at relatively high frequencies (such as the case of the New
Zealand Lower South Island system), when these are not swamped out by the
presence of filters.

The most significant difference with respect to the models proposed


by Hingorani and Burbery (1970), Bowles (1970) and Giesner and Arrillaga
(1971) is the modelling of the mutual effect between phases.

With reference to the New Zealand HVDC link the initially installed
th th th
filters were tuned to the 5 , 7 , 11th & 13 . The N.Z. Electricity
Department did this as some HVDC authorities recommended a "wait and see"
policy about harmonic problems, for there is no justification in
providing expensive filter branches to guard against trouble that might
not occur. However, early operating experience showed a need for a
th
high-pass filter and 9 harmonic filters to reduce the wide spread
th
telephone interference that occured. The 9 harmonic problem was a
result of a resonance between the filters and a.c. system (Robinson
1966). The development of the ACREP program described in this thesis
allows the a.c. system, filter and combined reponse to be inspected and
any resonance problems forecasted. This allows remedies to be found
before the troubles occur. Also the combined response of "a.c. system
equivalent" and filters can be compared to the combined response of
"actual" a.c. system and filters in order to assess the sophistication
needed in the a.c. system equivalent for accurate steady state and
dynamic studies.

In steady state harmonic penetration studies, frequency-matched


a.c. system equivalents are needed as the small levels of higher order
harmonics are greatly affected by small errors in the combined response.
A table of harmonics can easily display relatively large discrepancies at
high order harmonics without any noticeable difference in the waveform.
162.

The mutual coupling between phases has a considerable influence on


the harmonic level, particularly at the higher order harmonics. Most
power systems try to balance the loading on the phases as best as
practicable, and therefore even if mutual coupling is not to be
explicitly represented it is still better to diagonalize the 3x3
impedance matrices than to simply ignore the mutual coupling and use the
self terms.

The differences between the alternative a.c. system models have


been clearly illustrated with reference to an asymmetrical fault applied
to the a.c. side of an a.c./d.c. system.

Future Work.

The response of a d.c. link to sudden changes in system conditions


or requirements is dependent on the convertor controller's response. At
present a simple algebraic relationship is used to simUlate what is in
practice a complex circuit. The TeS convertor controller needs to be
enhanced so that its dynamic behaviour more realistically models the
controller it represents. As the type and operation of convertor
controllers differ for each application, a modular approach is required
which allows the controller's transfer fUnction to be built up of
standard transfer blocks. Once this has been achieved extension to the
fault detection and protection mathematical models can be made. For
example the advantages of altering the normal firing sequence once a
fault has been detected (Graham 1984) can be demonstrated.

Although the modelling of d.c. line faults has been carried out in
detail, further work is needed for the a.c. fault Simulation, in
particular the arc characteristics of the a.c. fault and more flexible
circuit breaker models are required.

The proposed extension of the frequency-matched a.C. system


equivalent for multi-convertor systems needs to be implemented and
tested. This would permit investigation of the interaction of two
harmonic sources connected to the same a.C. system. This is relevant to
163.

the Lower South Island of New Zealand where Tiwai and Benmore busbars
both have large a.c./d.c. convertors connected.

With the recent advances in transformer and a.c. system


representation, when the improvements in the convertor controller are
completed, the Transient Convertor Simulation program will be a very
powerful tool for both steady-state and dynamic HVDC system studies.
164.

REFERENCES

Adamson C. and Hingorani N.G. (1960)


"High Voltage Direct Current Power Transmission"
Garraway Ltd.

Ainsworth J.D. (1967)


"Harmonic instability between controlled static convertors and
a.c. networks"
Proc. lEE., Vol. 114, No.7, pp 949-957

Ainsworth J.D. (1968)


"The Phase-Locked Oscillator - A new Control system for controlled
static convertors"
IEEE Trans., Vol. PAS-87, No.3, pp 859-865

Al-Khashali (1976)
"Generalised Dynamic Modelling of High Voltage AC-DC Transmission
Systems"
Ph.D. Thesis, University of Manchester Institute of Science and
Technology, Great Britian.

Ametani A. (1976)
"A highly efficient method for calculating transmission line
transients"
IEEE Trans., Vol. PAS-95, No.5, pp 1545-1551

Arrillaga J. and Galanos G. (1969)


"Fault-development control in a.c.-d.c. convertors"
Proc. lEE., Vol. 116, No.7, pp 1202-1207

Arrillaga J. and Galanos G. (1970a)


"Fault-detection scheme for a direct digital control of a.c.-d.c.
interconnections"
Proc. lEE., Vol. 117,No. 4, pp 785-793
165.

Arrillaga J. and Galanos G. (1970b)


"Theoretical Basis of a Digital Method of Grid Control for HVDC
Convertors"
IEEE Trans., Vol. PAS-89, No.8, pp 2049-2055

Arrillaga J., Galanos G. and Powner E.T. (1970c)


"Direct Digital Control of HVDC Convertors"
IEEE Trans., Vol. PAS-89, No.8, pp 2056-2065

Arrillaga J., Arnold C.P., Heffernan M.D. and Campos-Barros J.G. (1977a)
"Stability of Isolated Generator-HVDC Convertor Units"
IEEE/PES Summer Power Meeting, Mexico City, Paper A77-508-5

Arrillaga J., Al-Khashali H.J. and Campos-Barros J.G. (1977b)


"General Formulation for dynamic studies in power systems
including static convertors"
Proc. lEE., Vol. 124, No. 11, pp 1047-1052

Arrillaga J., Arnold C.P. and Harker B.J. (1983a)


"Computer Modelling of Electrical Power Systems"
John Wiley & Sons Ltd.

Arr illaga J. (1 983b)


"High Voltage Direct Current Transmission"
lEE Power Engineering Series 6, Peter Peregrinus Ltd.

Arrillaga J., Bradley D.A., and Bodger P.S. (1985a)


"Power System Harmonics"
John Wiley & Sons Ltd., London.

Arrillaga J., Bradley D.A., Gellen S.J. and Densem T.J. (1985b)
"Computer Graphics as a Tool for Understanding the Harmonic
Behaviour of Power Systems"
Journal of Electrical and Electronic Engineering, Australia,
Vol. 5. No.1, pp 1-7
166.

Arrillaga J., Watson N.R., Eggleston J.F. and Callaghan C.D. (1987)
"Comparison of steady-state and dynamic models for the calculation
of ACIDC System Harmonics"
lEE Proc., Vol. 134, Pt. C, pp 31-37

Avila-Rosales J. and Alvarado F.L. (1982)


"Nonlinear Frequency Dependent Transformer Model for
Electromagnetic Transient Studies in Power Systems"
IEEE Trans., Vol. PAS-101, No. 11, pp 4281-4288

Berube G.R. and Ooi B.T. (1983)


"Fast Periodic solution for twelve-pulse Convertor"
IEEE Trans., Vol. PAS-102, No.7, pp 1994-2002

Bickford J.P., Mullinex N. and Reed J.R. (1976)


"Computation of Power-system Transients"
lEE Monograph series 18
Peter Peregrinus Ltd

Bickford J.P. and Heaton A.G. (1986)


"Transient Overvoltages on Power Systems"
lEE Review.
lEE Proceedings, Vol. 133, Pt. C, No.4, pp 201-225

Bowles J.P. and Clarke C.D. (1970)


Discussion of Hingorani and Burbery (1970) paper.
IEEE Trans., Vol. PAS-89, No. 5/6, pp 826-827

Bowles J.P. (1970)


"AC System and Transformer representation for HV-DC Transmission
Studies"
IEEE Trans., Vol. PAS-89, No.7, pp 1603-1609

Brameller A., Scott M.R. and John M.N. (1969)


"Practical Diakoptics for Electrical Networks"
Chapman and Hall, London.
167.

Budner A. (1970)
"Introduction of Frequency-Dependent Line Parameters into an
Electromagnetic Transients Program"
IEEE Trans., Vol. PAS-80, No.1, pp 88-95

Breuer G.D., Chow J.H., Gentile T.J., Lindh C.B., Numrich F.H.,
Lasseter R.H., Addis G. and Vithayathil J.J. (1982)
"HVDC - AC Harmonic Interaction.
Part II - AC System Harmonic Model with Comparison of Calculated
and Measured Data tl
IEEE Trans., Vol. PAS-101, No.3, pp 709-718

Campos-Barros J.G. (1976)


"Dynamic Modelling of Synchronous Machines Connected to HVDC
Transmission Systems"
Ph.D. Thesis, University of Manchester Institute of Science and
Technology, Great Britian.

Campos-Barros J.G. and Rangel R.D. (1985)


"Computer Simulation of Modern Power Systems: The elimination of
Numerical Oscillation caused by valve action"
4th International Conference on AC and DC Power Transmission.,
London, pp 254-259

Carrol D.P. and Nozari F. (1975)


"An Efficient Computer Method for Simulating Transients on
Transmission Lines with Frequency Dependent Parameters"
IEEE Trans., Vol. PAS-94, No.4, pp 1167-1174

Carson J.R. (1926)


"Wave Propagation in Overhead Wires with Ground Return"
Bell System Technical Journal, Vol. 5, pp 539-554

Chua L.O. and Lin P.M. (1975)


"Computer aided analysis of electronic circuits:
algorithms & computational techniques"
Prentice-Hall Inc.
168.

Concordia C. (1956)
"The Transient Network Analyzer for Electric Power System
Problems"
Supplement to CIGRE Committee No. 13 Report for 1956.

Concordia C. and Schulz R.P. (1975)


"Appropriate Component Representation for the Simulation of Power
System Dynamics"
IEEE/PES 1975 Winter Power Meeting, Symposium on Adequacy and
Philosophy of Modeling: Dynamic System Performance

Cory (1965)
"High Voltage Direct Current Convertors and Systems"
Macdonald & Co. Ltd.

Dabuleanu A. and Semlyen A. (1975)


"Modelling of Transpositions and Double Circuit Transmission
Lines in Switching Surge Calculations"
IEEE Trans., Vol. PAS-94, No.2, pp 242-251

DEC (1984)
"Programming in VAX FORTRAN" (Version 4.0)
Digital Equipment Corporation, Maynard, Massachusetts.

Densem T.J. (1983)


"Three Phase Power System Harmonic Penetration"
Ph.D. Thesis, University of Canterbury, New Zealand.

Do X.D. and Le Van A.D. (1976)


"A New Technique for Computer Simulation of Transients on
Overhead Transmission Lines with Frequency Dependent Parameters"
IEEE/PES Winter Meeting & Tesla Symposium, New York.

Dommel H.W. (1969)


"Digital Computer Solution of Electromagnetic Transients in
Single and Multiphase Network"
IEEE Trans., Vol. PAS-88, No.4, pp 388-399
169.

Dommel H.W. and Meyer W.S. (1974)


"Computation of Electromagnetic Transients"
Proc. IEEE, Vol. 62, No.7, pp 983-993

Dommel H.W. (1978)


"Line Constants of Overhead Lines and Underground Cables"
Notes used for course E.E. 553 at the University of British
Columbia

Dommel H.W. (1980)


tlOverhead Line Parameters Program User's Manual"
Department of Electical Engineering,
The University of British Columbia, Canada.

Eggleston J.F. (1985)


ttHarmonic Modelling of Transmission Systems containing Synchronous
Machines and Static Convertors"
Ph.D. Thesis, University of Canterbury, New Zealand.

Fitzgerald A.E., Kingsley C. and Umans S.D. (1983)


"Electrical Machinery" (4th Edition),
McGraw-Hill International Book Company.

Gear C. W. (1 971 )
"Numerical Initial Value Problems in Ordinary Differential
Equations"
Prentice-Hall, Inc., New Jersey.

Giesner D.B. and Arrillaga J. (1971)


"Behaviour of HVDC Links Under Balanced AC Fault Conditions"
Proc. lEE, Vol. 118, No. 3/4, pp 591-599

Graham J.C. (1984)


"Microprocessor based system for the development of control and
protection of HVDC convertors"
Ph.D. Thesis, University of Canterbury, New Zealand.

Greenwood A. (1971)
"Electric Transients in Power Systems"
J. Wiley & Sons, Inc.
170.

Gunn H. R. (1966)
"Commissioning and Early Operating Experience with the Benmore-
Haywards H.V.D.C. Transmission Scheme"
New Zealand Engineering, Vol. 21. No.3. pp 93-101

Harker B.J. (1980)


"Steady State Analysis of Integrated a.c. and d.c. Systems"
Ph.D. Thes{s, University of Canterbury, New Zealand.

Harwell (1986)
"Harwell Subroutine Specification, VF04"
Computer Science and Systems Division.
Atomic Energy Research Establishment,
Harwell, Oxfordshire, England.

Hay J.L. and Hingorani N.G. (1970a)


"Dynamic Simulation of Multiconvertor HVDC Systems by Digital
Compu ter. Part I : Mathematical Model"
IEEE Trans., Vol. PAS-89, No.2, pp 218-222

Hay J.L. and Hingorani N.G. (1970b)


"Dynamic Simulation of Multiconvertor HVDC Systems by Digital
Computer. Part II Computer Program"
IEEE Trans., Vol. PAS-89, No.2, pp 222-228

Hay J.L., Bhatti J.S. and Hingorani N.G. (1971a)


"Simplified Dynamic Simulation of HVDC System by Digital Computer
Part I : Computer Program"
IEEE Trans., Vol. PAS-90, No.2, pp 859-864

Hay J.L., Bhatti J.S. and Hingorani N.G. (1971b)


"Simplified Dynamic Simulation of HVDC System by Digital Computer
Part II : Design of Controller and Test Results"
IEEE Trans., Vol. PAS-90, No.2, pp 865-875

Heffernan M.D. (1980)


"Analysis of AC/DC System Disturbances"
Ph.D. Thesis, University of Canterbury, New Zealand.
171.

Hingorani N.G., Hay J.L. and Crosbie R.E. (1966a)


"Dynamic Simulation of HVDC Power Transmission Systems on Digital
Computers"
Proc. IEE., Vol. 113, No.5, pp 793-802

Hingorani N.G. and Hay J.L. (1966b)


"Dynamic Simulation of an HVDC System on a Digital Computer"
Proc. IEE International Conference (No. 22) on HVDC, pp 119-124

Hingorani N.G. and Hay J.L. (1967)


"Representation of Faults in the Dynamic Simulation of HVDC
Systems by Digital Computer"
Proc. IEE., Vol. 114, No.5, May 1967, pp 629-638

Hingorani N.G., Kitchin R.H. and Hay J.L. (1968)


"Dynamic Simulation of HVDC Power Transmission Systems on Digital
Computers - Generalised Mesh Analysis Approach"
IEEE Trans., Vol. PAS-87, No.4, pp 989-996

Hingorani N.G. (1970)


"Transient Overvoltages on Bipolar HVDC Overhead Line Caused by
DC Line Faults"
IEEE Trans., Vol. PAS-89, No.4, pp 592-602

Hingorani N.G. and Burbery M.F. (1970)


"Simulation of AC System Impedance in HVDC System studies"
IEEE Trans., Vol. PAS-89, No. 5/6, pp 820-828

Humpage W.D. and Wong K.P. (1982)


"Electromagnetic Transient Analysis in EHV Power Networks"
Proc. of the IEEE, Vol. 70, No.4, pp 379-402

IEEE (1983)
"Power System Harmonics : An Overview"
IEEE Working Group on Power System Harmonics,
IEEE Trans., Vol. PAS-102, No.8, pp 2455-2460
172.

IEEE (1984)
"Bibliography of Power System Harmonics", Parts I and II,
IEEE Trans., Vol. PAS-103, No.9, pp 2460-2479

Ireland B.D. (1986)


"Steady State and Dynamic Performance of C-type filters for HVDC"
M.E. Thesis, University of Canterbury, New Zealand.

Jarrett and Middleton (1974)


"Operational experience of HVDC systems throughout the world
during 1971-1972"
CIGRE Conference, Paris, Paper 14-09

Joosten A.P.B., Arnold C.P. and Arrillaga J. (1985)


"State Variable Dependence in State Space Analysis of Electric
Power Systems"
Modelling, Simulation & Control, A, AMSE Press, Vol. 2, No.3,
pp 51-63

Joosten A. P. B. (1987)
"Convertor Transformer Simulation for T.C.S. Studies"
Ph.D. Thesis to be submitted, University of Canterbury,
New Zealand.

Kimbark E.W. (1968)


"Power System Stability - Volume III:
Synchronous Machines"
Dover Publications Inc., New York.

Kimbark E.W. (1971)


"Direct Current Transmission"
Volume 1
J. Wiley and Sons, Inc.

Kitchin R.H. (1981)


"New method for digital-computer evaluation of convertor
harmonics in power systems using state-variable analysis"
Proc lEE., Vol. 128, Pt. C, No.4, pp 196-207
173 •

Knudsen J. and Albrecht P.F. (1982)


"A Survey of the Reliability of HVDC systems throughout the world
during 1979/1980"
CIGRE Conference, Paris, Paper 14-06

Knudsen J., Skogheim O. and Albrecht P.F. (1984)


"A Survey of the Reliability of HVDC systems throughout the world
during 1981-1982"
CIGRE Conference, Paris, Paper 14-06

Kohler A. (1967)
"Earth Fault Clearing on an HVDC Transmission Line, with Special
Consideration of the Properties of the DC Arc in Free Air"
IEEE Trans., Vol. PAS-86, No.3, pp 298-304

Kron G. (1939)
"Tensor Analysis of Networks"
J. Wiley

Kron G. (1959)
"Tensors for Circuits" (2nd Edition)
Dover Publications.

Kron G. (1963)
"Diakoptics - the Piecewise Solution of Large Scale Systems"
Macdonald, London.

Kruempel K.C. and Reitan O.K. (1970)


Discussion of Hingorani and Burbery (1970) paper.
IEEE Trans., Vol. PAS-89, No. 5/6, pp 827

Last F.H. and Middleton R.M. (1970)


"A Survey of the Performance of HVDC Schemes throughout the world
for 1968/1969"
CIGRE Conference, Paris, Paper 14-05
174.

Last F.H. and Middleton R.M. (1972)


"A Survey of the Performance of HVDC Schemes throughout the world
for 1970"
CIGRE Conference, Paris, Paper 14-07

Liou M.L. (1972)


"Exact Analysis of Linear Circuits containing Periodically
Operated Switches with Applications"
IEEE Trans., Vol. CT-19, No.2, pp 146-154

Lipo T.A. (1971)


"The Analysis of Induction Motors with Voltage Control by
Symmetrically Triggered Thyristors"
IEEE Trans., Vol. PAS-90, No.2, pp 515-525

Marti L. (1982)
"Accurate Modelling of Frequency-dependent Transmission Lines in
Electromagnetic Transient Simulations"
IEEE Trans., Vol. PAS-101, No.1, pp 147-155

Marti L. (1983)
"Low-Order Approximation of Transmission Line Parameters for
Frequency-Dependent Models"
IEEE Trans., Vol. PAS-102, No. 11, pp 3582-3589

Meyer W.S. and Dommel H.W. (1974)


"Numerical Modelling of Frequency-Dependent Transmission Line
Parameters in an Electromagnetic Transients Program"
IEEE Trans., Vol. PAS-93, No.5, pp 1401-1409

Milias-Argitis J., Giannakopoulos G. and Galanos G. (1978a)


"Dynamic Simulation for Multiterminal HVDC Systems"
IEEE Trans., Vol. PAS-97, No.2, pp 587-593

Milias-Argitis J., Giannakopoulos G. and Galanos G. (1978b)


"Power Flow Control in Multiterminal HVDC Systems"
lEE Conf. Publ. 154, Second International Conference on Power
Electronics,"Power Semiconductors and Their Applications", London.
175.

Marched A.S. and Brandwajn V. (1983)


"Transmission Network Equivalents for Electromagnetic Transient
Studies"
IEEE Trans., Vol. PAS-102, No.9, pp 2984-2994

Naidu S.R. and de Lima F.N. (1985)


"A Frequency-Dependent Transmission Line Model for
Electromagnetic Transient Studies"
Proc. lEE., Vol. 132, Pt. C, No.6, pp 294-297

Pender J. T. (1 969)
"Fast Transients in Electric Power Systems"
Int. J. elect. Engng Educ, Vol. 7, pp 419-429

Peterson H.A. (1939)


"An Electric Circuit Transient Analyzer"
General Electric Review.
Vol. 42, No.9, pp 394-400

van Que Do and Gavrilovic M.M. (1984)


"An Iterative Pole-removal Method for synthesis of Power System
Equivalent networks"
IEEE Trans., Vol. PAS-103, No.8, pp 2065-2070

Reeve J. (1 970)
Discussion of Hingorani and Burbery (1970) paper.
IEEE Trans., Vol. PAS-89, No. 5/6, pp 827

Reeve J. and Baron J.A. (1971)


"Harmonic interaction between HVDC converters and AC power
systems"
IEEE Trans., Vol. PAS-90, No.6, pp 2785-2791

Robinson G.H. (1966)


"Harmonic Phenomena associated with the Benmore-Haywards h.v.d.c.
transmission scheme"
New Zealand Engineering, Vol. 21, No.1, pp 16-29
176.

Rumpf E. and Jarrett G.S.H. (1976)


"A Survey of the Performance of HVDC systems throughout the world
during 1973-1974"
CIGRE Conference, Paris, Paper 14-07

Rumpf E. (1 980 )
"The Operational Performance of HVDC Systems throughout the world
during 1975~1978"

"Incorporating HVDC Power Transmission into Systems Planning", a


Symposium Sponsored by the U.S. Department of Energy,
Phoenix, Arizona, March 1980.

Semlyen A. and Dabuleanu A. (1975)


"Fast and Accurate Switching Transient Calculations on
Transmission lines with Ground Return using Recursive
Convolutions"
IEEE Trans., Vol. PAS-94, No.2, pp 561-569

Semlyen A. and Abdel-Rahman M.H. (1982)


"Transmission Line Modelling by Rational Transfer Functions"
IEEE Trans., Vol. PAS-101, No.9, pp 3576-3584

Semlyen A. and Deri A. (1985)


"Time Domain Modelling of Frequency Dependent Three-Phase
Transmission Line Impedance"
IEEE Trans., Vol. PAS-104, No.6, pp 1549-1553

Skilling H.H. (1974)


"Electric Networks"
J. Wiley and Sons, Inc., New York.

Smith J.R. and Bell W.W. (1984)


"Selection of Transmission circuit models for Power-system
transient studies"
Proc. lEE., Vol. 131, Pt. C, No.1, pp 1-4
177.

Snelson J.K. (1972)


"Propagation of Travelling waves on Transmission Lines -
frequency-dependent parameters"
IEEE Trans., Vol. PAS-91, No.1, pp 85-91

Turner K.S. (1980)


"Transient Stability Analysis of Integrated AC and DC Power
Systems"
Ph.D. Thesis, University of Canterbury, New Zealand.

Uhlmann E. (1975)
"Power Transmission by Direct Current"
Springer-Verlag Berlin/ Heidelberg

UMIST (1981)
"Harmonics in Power Systems"
An International Conference on Harmonics in Power Systems,
UMIST, Manchester.

Watson N.R., Arrillaga J. and Joosten A.P.B. (1985)


"AC system equivalents for the dynamic simulation of HVDC
convertors"
4th International Conference on AC and DC Power Transmission.,
London, lEE Conf. Publ. 255, pp 394-399

Watson N.R. and Arrillaga J. (1987)


"Frequency-Dependent A.C. System Equivalents for Harmonic Studies
and Transient Convertor Simulation"
lEEE/PES Winter Power Meeting, Paper 171-2

Williams S. and Smith I.R. (1973)


ttFast digital computation of 3-phase thyristor bridge circuits"
Proc. lEE., Vol. 120, No.7, pp 791-795

Yacamini R. and de Oliveira J.C. (1980a)


"Harmonics in Multiple Convertor Systems: A Generalised Approach"
Proc. lEE., Vol. 127, Pt. B, No.2, pp 96-106
178.

Yacamini R. and de Oliveira J.C. (1980b)


"Instability in h.v.d.c. schemes at low-order integer harmonics"
Proc. lEE., Vol. 127, Pt. C, No.3, pp 179-188

Yacamini R. and Smith W.J. (1983)


"Third-harmonic current from unbalanced AC/DC convertors"
Proc. lEE., Vol. 130, Pt. C, No.3, pp 122-126

Yacamini R. and de Olveira J.C. (1986)


"Comprehensive Calculation of Convertor Harmonics with System
Impedances and Control Representation!!
Proc. lEE., Vol. 133, Pt. B, No.2, pp 95-102

Yen C., Fazarinc, Zvonko and Wheeler R.L. (1982)


"Time-Domain Skin-Effect Model for Transient Analysis of Lossy
Transmission lines"
Proc. of IEEE, Vol. 70, No.7, pp 750-757
179.

APPENDIX A1

DETAILED SYNCHRONOUS MACHINE COEFFICIENT MATRIX.

This matrix, L is a 6x6 matrix which can be partitioned into


g
stator, stator-rotor, and rotor terms as follows:

L
g [
~ss ~srJ
rs rr
where

-L
aa -L aa 200S(28» -L ao -L ao 2008(2(8+150)

-L OC -L 00 200S(2(8+120»

LakdCOS(e) -L a kq sineS) ]
L Lbkdcos(S-120) -L sin(S-120)
sr bkq
Lckdcos(S+120) -L c k q sin(S+120)

L Lt
rs sr

Lfd Lfkd 0

L Lfkd Lkd 0
rr
0 0 L
kq

The torque matrix is a 6x6 matrix derived from Lg

[
d L
d L
Ci8g
aEi ss a-e
aL ]
sr

d L aL
a-e rs a-e rr
a Lrr a L d L }t
where
aS
0 and
a-e rs ae sr
180.

APPENDIX A2

TRANSIMISSION LINE RESPRESENTATION.

Overhead transmission lines or cables are defined by their series


impedance (R + X~) and shunt susceptance (Bc) matrices per unit length.
The elements of the series impedance and shunt susceptance matrices are
derived from the geometry of the tower configuration, the characteristics
of the conductors and from the earth resistivity. A method for
calculating these matrices is described by Arrillaga et al (1983) and
summarized here ;

To obtain the capacitance matrix, Maxwell's potential coefficient


matrix [p] is first obtained from the geometry of the tower
configuration and from the conductors' radii. The diagonal element of [p]
is
2h.
1
p . .= In (km/F)
11 21fEo r.
1
and the off-diagonal element is
Dik
P. = In (km/F)
1k 21fEo d
ik
where
r. - radius of conductor i
1

Dik - distance between conductor i and image of conductor k (in m)

d - distance between conductors i and k (in m)


ik
h. - average height above ground of conductor i (in m)
1
Eo - Permittivity constant

Note that the potential coefficient matrix is real and symmetric.


The capacitance matrix is obtained by inverting matrix P.

When only the phase quantities are of interest the capacitance


matrix [e'] can be reduced to a matrix [C] called the capacitance matrix
for the "equivalent phase conductors". Details of the elimination of
ground wires and bundling of conductors are given in the following two
references; Dommel (1978) and Dommel (1980). The reduction process for
the series impedances and Maxwell's potential coefficients are identical.
181.

The diagonal elements for the series impedance is given by;


-4 2hi
Zii= (R ii + ~Rii) + j(2w.10 In GMR. + ~Xii (S1/km)
1

The mutual impedance terms are given by

(S1/km)

where
R .. - a.c. resistance of conductor i (S1/km)
11
GMR.- geometric mean radius of conductor i (S1/km)
1
f - frequency in Hz,
w = 21ff
~Rii' ~Xii - are Carson's correction terms for earth return effects

Note for the self impedance terms that the geometric mean radius
(GMR) instead of the actual radius r is used to account for the
contribution which the internal inductance makes to the total inductance.
The series impedance matrix is also symmetric as Zik= Zki

The equivalent circuit of a typical transmission line segment is


illustrated in figure A2.1. As the choosen state variable is the charge
at a node rather than capacitor charge, the capacitance matrix [C"] is
required. The elements of [C"] are related to the capacitor matrix for
the equivalent circuit [C] by ;

a' b' c'


a' C Cab C
aa ac
[C"] b' Cba Cbb C
bc
c' Cca C C
cb cc

L Lab L L
aa ac ae
Lba Lbb Lbc Lbe
[L]
L LCb L L
ca cc ce
L Leb L L
ea ec ee
182.

a at

Lab (
Lbb
b b'

Lee
c et

~ f..-
C
Lee s

~ I--- >-- f..-


C C4
6

C
1

Figure A2.1 Transmission Line ~-segment Model


183.

Appendix A3

Parameters for RLC circuit analysis example.

R L C C
s
(rt) ( H) ( llF) (llF)

(1) 18.310 0.58000 0.69500


(2 ) 20.190 0.45780 0.44990
(3) 11.066 0.16000 0.52200
(4 ) 13.564 0.16667 0.36227
(5) 24.888 0.44444 0.27225
(6 ) 200.000 0.02667 1 .01250 1 .01250

Table A3.1 Parameters for the circuit of figure 4.20


184.

Appendix A4

Single Variable Optimization Methods.

Let x and x be two x values that bracket the actual minima. Then
1 2
the method that successively halves the interval known to contain the
minima by placing two test points close together at the centre of the
interval, is known as a Dichotomous Search. The dichotomous search is
illustrated in figure A4.1 . A pair of test pOints are required at each
step in order to be able to distinguish which half of the interval being
divided contains the minima. The small difference ~x between the
evaluations of each test point is limited by the accuracy in which the
objective function can be evaluated and sets a limit on the resolution of
the optimization.

--

Figure A4.1 Dichotomous Search Technique

The dichotomous search is inefficient in terms of objective


function evaluations as at each interval reduction two search points that
are very close together are evaluated. By using one of the two search
points in the next interval to be searched enhances efficiency as only
one additional search point is needed for each successive interval
reduction. Figure A4.2 illustrates this process. At the first step the
two test points xa,k and xb,k are required. This discriminates whether
R
the minimum is in the left or right hand region (I~+1 or I + ).
k 1
If say the minima lies in the latter interval then only one new test
L
point (x b ,k+1) is required for the next interval reduction. where I k +2
and I~+2 are the two possible regions.
185.

lk ----------~
I
I
I L
~I~~-----------Ik+l
I
R I
I~~~--------+--Ik+l ----------~)~I

: L
I ~ Ik+2 I >- !
I! I

I I R '
,
1-1.«.«---, Ik+2 ---.;>--1
I ,
I I I
I L ' :
~Ik+3~
I R
r--:
,,
Ik+3~
I
,I ,,
I

'L I I ,R >-
x xa,k I xb,k x
k , I k
L R
x x b ,k+l xk+l
,xk+l I
a,k+l
I I
I I
L R
x k +2 x a,k+2 x ,k+2
b xk+2

Figure A4.2 Search Technique used in the Fibonacci


and Golden Section Searches

The Fibonacci Search requires the placement of the first two


evaluations so that the search ends when the last two evaluations
coincide and halve the final interval. If the final interval is I then
n
the length of all previous intervals are given by :

(A4 . 1 )

Note that ~t the final step the interval In+l is created equal in
length to I n but is not used. However, it is needed in equation (A4.1) to
determine I n- l' The ratio of the kth interval length to the final
interval length then forms the well known Fibonacci sequence.

I n+l I
n

I 21
n-l n

I 31
n-2 n

I 51
n-3 n

I = 81 etc ...
n-4 n

The Fibonacci search achieves the greatest interval reduction of


methods requiring a fixed number of function evaluations.
186.

The main drawback of the Fibonacci Search is the need to specify


the number of objective function evaluations in advance. The Golden
Section Search overcomes this by placement of the test points so that a
constant ratio between successive interval lengths exists. The ratio is
1.618034, which is obtained by solving the appropriate quadratic equation
which results from imposing this condition.

The problem with the previous methods is the need to specify two
points that bracket the true minima. The Step Success/Failure Search,
like the dichotomous search, uses two closely positioned test points.
However, only one initial starting point is required. The algorithm then
steps in the direction of the minima and evaluates a new pair of test
points. Based on these results, the algorithm either: steps again in the
same direction; halves the step length and steps in the new direction
indicated; or remains at the same point and reduce the distance between
the two test pOints. A flow diagram for the Step Success/Failure search
is displayed in Figure A4.3.
187.

Evaluate the objective function at


two test poInts that are closely
spaced and straddle the current
point under consideration

Do
these two tes t Ves
poInts IndIcate the0-____________________-,
minIma lies
at higher
vaiue
?

Do Was
the last No
step taken In this
dIrectIon
?

Was No
the last
step taken In this step In direction
direction of Minima
?

Halve the distance


between the two
tes t points

step than /~~Ye~s------------_{~~~~~~~~~~~~


length Issmaller Halve test point spacing
test point
spacing
?

lIave
too many Yes
halvlngs of test
point spacing
occurred
?

Is Evaluate objective function at


the step Yes the mid-pOint between the two
leng th smaller than test points
convergence
tolerance
?
Is
Yes mid-pOint
smaller than both
test points
Has ?
No the max Imum
L-__________________________$ -_ _- - ( number of Iterations
been reached
? No

Figure M.3 Success/Failure Step Algorithm


188.

Appendix A5

PUblished Papers.
189.

394

A. C. SYSTEM EQUIVALENTS FOR THE DYNAHIC SHlULATION OF HVDC CONVERTORS

N.R. Watson, J. Arrillaga, A.P.B. Joosten

University of Canterbury, Christchurch, New Zealand.

INTRODUCTION Analytic impedance loci can be obtained from


a computer algorithm recently developed at
The behaviour of hvdc convertors following the University of Canterbury makes use of the
small or large disturbances can be assessed admittance matrix and linear transformations
by means of scaled down physical models or to interconnect the various system components
computer analysis in the time domain. In represented by appropriate equivalent
both cases, models with various degrees of circuits. Transmission lines generally have
complexity are already available to represent the greatest influence on the impedance loci
the convertor configuration, control, filters and need to be accurately represented i.e. by
and convertor transformers; in contrast the means of coupled multi conductor Equivalent-n
degree of representation of the a.c. network circuits with frequency dependent distributed
leaves much to be desired. parameters.(3)(4)

Given the size and complexity of a power As an illustration, Fig. 1 shows the New
system it is not practical to model each of Zealand South Island transmission system in
its components individually. Equivalent the vicinity of an aluminium smelter (at
circuit models are usually derived to suit Tiwai) and the three phases of the
the fundamental frequency, the simplest being corresponding impedance locus is illustrated
the short circuit impedances used in fault in Figure 2.
studies. Such models, however, are
inadequate to represent the system behaviour Derivation of Harmonic Equivalents
at harmonic frequencies.
In an earlier model (1) the selected minima
AC networks are generally inductive for the and maxima frequencies used for the
lower harmonics. Since the convertor filters derivation of the equivalent circuit are
are often capacitive for such frequencies, a those where the impedance locus crosses the
parallel resonant condition may exist, real axis (refer to Figure 2). However the
typically at around the third or fourth impedance versus frequency plots derived from
harmonic, or even at higher frequencies the harmonic impedance locus provide more
depending on the a.c. system strength. explicit information of the maxima and minima
~Ioreover a complex system/f i 1 ter combination harmonic impedances.
will produce several oscillatory harmonic
frequencies of relatively low orders. The equivalent circuit will consist of a
Disturbances may excite these natural number of single-tuned parallel branches as
frequencies and produce considerable waveform shown in figure 3 and the first step is to
distortion and overvoltages which will in obtain values for the inductors and
turn interact with the convertor control. capacitors of the equivalent circuit that
will produce the same minima and maxima
Symmetrical impedance loci are currently resonance frequencies. The resistances of
measured or calculated for the complete the branches are ignored as they have very
harmonic spectrum at various operating little influence on the magnitude of the
conditions. These can be used to derive admittance at frequencies that are
accurate harmonic equivalents (1)(2) for the sufficiently higher or lower than the series
a.C. system. However it has been shown (3) resonant frequency of the branch.
that there is considerable diversity between
the resonant frequencies seen by different The admittance of an n branch network is;
phases and thus in general the use of
three-phase loci and corresponding n n
independent phase equivalent circuits will be
n
s L .l.. Ti(s'+w')
more appropriate. I s
k=l Lk J;i k J
Y (s) (1)
k=l Lk(s'+w~) n
This paper describes the derivation of TI(s'+w')
three-phase harmonic equivalent circuits to 1=1 I
represent a.c. networks and illustrates their
effect on the dynamic behaviour of hvdc where:
convertors.
1
w'I LIC 1=1,2 ... etc and s = jw
HARMONIC MODELLING OF THE A.C. NETWORK r
For an LC network the impedance is zero at
System Harmonic Impedances minima and infinite at maxima frequencies.

Information of the system harmonic impedances Impedance minima occur when:


is often presented in the form of impedance
loci. These can be derived either
experimentally or analytically. The latter -IT (s'+w~) 0 (2)
provides a more flexible alternative, which 1=1
can assess any specified critical operating
condition and can derive the loci at any Impedance maxima will occur when the
number of interharmonic frequencies without numerator of equation (I) is zero, 1. e.
the need for interpolation.
190,

395

n 1 n n n
s L -
k=l Lk
n
J~
(s '+w ') = 0
J
(3 ) Y
jk!lL k (W 2 =r/L k C )
k
-j \"
L L
k=l k
(",2'" _",2 )
k
(11 )

Let 1'1' 1'2' ... F n be the frequencies at which As Lk appears in the denominator of equation
the admittance is zero, then equation (3) can (11), to increase the admittance contribution
be written as: of each branch, so as to reduce the
impedance, the inductors are divided by the
scaling factor. In order to keep the same
s n
n-l

k=l
(s'+M k) where Mk = 21lFk (4 ) resonance frequencies the capacitor values
are multiplied by the scaling factor.

By equating the coefficients of s in (3) and These equivalent circuits, developed in the
(4) the following equations result: frequency domain, can also be used in the
time domain according to standard circuit
For s2n-l laws. When using transfer functions in the
frequency domain, multiplication by the
n Laplace transform of the input and
1
L 1 (5 ) application of the inverse Laplace transform
k=l Lk obtains the time solution.

For s2n-3 Moreover, in order to prove the general


applicability of the equivalent model it is
n n n-l necessary to match the system harmonic
1
L
Lk
L w'J l: M'
k
(6 ) impedances in magnitude and phase. Figures
k=l J=k k=l and 5 compare the real and imaginary (or
resistance and reactance) magnitudes of the
For s2n-5 actual and equivalent circuits of the system
in Figure 1; they show very good agreement.
n n n n-l n-l It can thus be expected that the time
l: (.1:...
l: l: w~"'~) l: l: M~M~ (7 ) response of the proposed equivalent circuit
k=l Lk 1=1 J=I+l 1=1 J=I+l will follow that of the actual system,
regardless of the input, both on steady and
For sl transients states.

n n n-l TRANSIENT CONVERTOR SIMULATION (TCS)


1
L TT w' 11 M'
k
(8 )
k=l Lk J=k J k=l An existing TCS programme (5) solves the
following set of state space equations:
Thus n equations are formed which are then
solved for the n unknown inductances t t t
p \j!£ E f - R£ If + Kfa Va + Kf8 Ve + Kiy Vy
(Ll,L2, ... Ln ). The appropriate capacitor
values are found using: p Q -K a £ If Kar Ir
a
1 where
w'k k=l, 2 ... n (9 )
LkCk
\j!i(magnetic flux in inductive branches) and
where w are the required minima frequencies. Qa(electric charge at capacitive nodes) are
k
used as state variables and the network nodes
The admittance of the tuned network at any are divided into a (with capacitive
particular frequency is given by: branches),8 (with resistive, but not
capacitive branches) and y (with only
Lk inductive branches) nodes to provide a
__1_)
n n Rk - (w'
LkC k
diakoptical solution for efficient
Y L Yk l: '" (10 )
computation.
2
k=l k=l R' + Lk 1
k [- (",' - LC)] The dependent variables are interrelated by
'" k k the following equations:
-1
Branch k is in series resonance when w~ = If L£ I/!f
l/(Ck Lk). The corresponding admittance term
for the kth branch is l/Rk. This term is V C- l Q
significantly larger than the contribution a a a
from the branches not in series resonance and -1 t
V8 -R (K £ 1£ + KSr Rr K V )
hence dominates the admittance. The S 8 ra a
-1 t
resistance that dominates at each minima is V -L Kyi L£ (E£ - R I f + K£a V + Kt V 8)
inserted in the corresponding branch. This Y Y f a i8
yields maxima point impedances for the I R- l (K t t
V + K
equivalent circuit which are much higher than r r ra a r8 V 8)
those of the actual system. The problem can In the above equations the R, L, C
be corrected by the scaling of all inductor coefficients are assumed constant at each
and capacitor values. A common scaling step of an implicit integration procedure
factor is used for all inductors and based on a trapezoidal approximation. Any
capacitors so that the resonance frequencies attempt to make these parameters
remain unchanged. The appropriate scaling frequency-dependent would make the solution
factor is obtained by dividing the calculated extremely complicated.
impedance of the network at a particular
frequency by the actual systems impedance. With the use of the harmonic matched a.c.
Equation (10) with R ignored yields: system equivalents described in the last
section it is possible to derive realistic
191.

396

waveform information from the TCS programme 2. Van Que Do and Gavrilovic, M.M. 'An
without altering the basic algorithm. iterative pole-removal method for
synthesis of power system equivalent
Illustrative Example networks', IEEE Trans. Power apparatus
and systems, Vol. PAS-103, No.8, August
Figure 6 illustrates the test system used to 1984, pp. 2065-70.
demonstrate the effect of alternative a.c.
system representations during a d.c. short- 3. Densem, T., Badger, P.S. and Arrillaga,
circuit. J. 'Three-phase transmission system
modelling for harmonic pentration
The a.c. system on the rectifier side is studies', Trans. IEEE, PAS-103, 1984, pp.
first represented as a Thevenin equivalent to 310-17.
give the specified Short Circuit Ratio. The
dynamic response prior to, during, and after 4. Arrillaga, J., Densem, T. and Bodger,
fault clearance is illustrated in Figures 7 P. S. 'Zero sequence harmonic current
and 8 (continuous line). generation in transmission lines
connected to large convertor plant',
The dynamic study is then repeated with the Trans. IEEE, PAS-I02, 1983, pp. 2357-63.
a.c. system represented by the tuned
equivalent circuit developed above with all 5. Heffernan, M.D. 'Analysis of AC-DC
other parameters, controls and specifications system disturbances', Ph.D. Thesis,
remaining unchanged. The results are plotted University of Canterbury, School of
in the same figures (dash and dot). Engineering, Electrical Engineering
Department, 1980.
It is not the purpose of this paper to
discuss the fault behaviour in any detail,
but rather to show that the predicted
behaviour can be inaccurate in the absence of
a frequency-dependent a.c. system model.

Figure 7 shows that the fault current peak


calculated with the improved a.c. model is
only slightly lower than that of the Thevenln
equivalent. However the voltages across the
convertor d.c. terminals, illustrated in
Figure 8, show considerable dif f erences. In
particular the difference in the first
voltage peaks following rectifier current
extinction is of the order of 50%.

CONCLUSIONS

A general method has been described to derive


three-phase a.c. system harmonic equivalents.
It has been shown that the equivalent
circuits derived match very accurately the
actual system impedances, both in phase and
magnitude.

The use of this equivalent circuit for the


a.c. system provides adequate representation
of system resonances, the effect of their
associated overvoltages and waveform Figure 3 Single-phase network equivalent
distortion on the convertor controller is matching n resonant frequencies
then catered for.

The proposed circuit does not give explicit


information at the individual a.c. system
busbars, but it provides an extremely
efficient simulation in comparison with a
full a.c. system representation.

The harmonic equivalent circuit has been used


as part of a Transient Convertor Simulation
programme to assess the effect of frequency
dependence on the development of hvdc
disturbances. Results obtained under typical
d.c. line fault conditions have indicated
substantial differences with respect to
earlier simulations based on a simplified
Thevenin equivalent a.c. representation.

REFERENCES

1. Hingorani, N.G. and Burbery, M.F.


'Simulation of A.C. system impedance in
hvdc system studies', IEEE Trans. Power
apparatus and systems, Vol. PAS-89, No.
5/6, May/June 1970, pp. 820-26. Figure 6 Equivalent h.v.d.c. representation
for d .c. faults
192.

397

--r-r-~~r-ROXBURGH

nWAI 90 MW 54 MVAR

Figure 1 Test system

X(rI)
25121

2121121

150

1121121

5121

121
5121121

-5121
/
-11210 / /
/ (/
-150

-2121121

-25121
.
----
~
--._._.--- ./
//
.

Figure 2 Three-phase impedance loci of the a.c. system - - - - phase A


- phase B
- - - - phase C
193.

398

R (12)
500

400

300

200

100

....
a
50 250 450 650 850 1050 1250
(f)Hz

Figure 4 Resistive component of the test system impedance versus frequency


(i) Equivalent model
(ii) Actual system

X(I1)
300

(ii )

200

.....
100

f(Hz)
a
50 250

-100

-200

-300

Figure 5 Reactive component of the test system impedance versus frequency


(i) Equivalent model
(ii) Actual system
194.

399

ld (per unit)

10

o
-1
6 6.25 6.5 6.75 7.25 7.5
cycles

Figure 7 Variation of the d.c. current following a line short-circuit


(i) Improved equivalent model
(ii) Thevenin equivalent

Vd (per unit)
7

6.25 6.5 6.75 7.5


cycles

Figure 8 Variation of the voltage at the conVertor d.c. terminals following a line short-circuit
(i) Improved equivalent model
(ii) Thevenin equivalent
195.

87 WM 171-2
FREQUENCY-DEPENDENT A.C. SYSTEH EQUIVALENTS FOR HARHONIC STUDIES AND TRANSIENT CONVERTOR SIHULATION

N.R. Watson J. Arrillaga, Sen.Mem.


University of Canterbury (New Zealand)

Abstract A generalization of the impedance SYSTEH FREQUENCY RESPONSE AT A CONVERTOR TERHINAL


loci is first descrihed capable of displaying the
complete frequency response of any given power system The frequency response of a power system at the
including multi-phase related effects such as phase point of harmonic injection can be derived either from
asymmetries and mutual couplings. This information is measurements or from frequency-domain studies, and the
used to derive frequency dependent equivalent circuits results are normally presented in the form of a single
suitable for integration in the time-domain solutions impedance-locus plot. However the exact frequency
of power system waveforms. Direct and optimization response must include network asymmetries and mutual
techniques are used in the derivation of the couplings, thus requiring a 3 x 3 matrix impedance for
equivalent circuits. Their effect on impedance each harmonic. A generalization of the impedance loci
frequency matching and computational efficiency are is used in this paper to display the enormous amount
compared. of information that results. As an illustration, the
self and mutual impedance loci of the system of Figure
INTRODUCTION 1 (the lower part of the New Zealand network) observed
from the TIWAI Bus are shown in Fiqure 2.
Harmonic studies are beginning to play an
important part in power system analysis. The harmonic
currents produced by non-linear components are
normally derived on the assumption of a strong, i.e.
perfectly sinusoidal, voltage supply. These nominal
harmonic currents are then injected into the a.c. ""T"T".I---,-ROXtlURGH
network (assumed linear) to determine the levels of
voltage distortion.
However, when the non-linear load is a large
power convertor and the a.c. system harmonic TI ....," 90 M'J ~'HVAR

impedances are large, the supply voltage waveforms are ./


not perfectly sinusoidal and the derivation of the
injected harmonic currents requires an iterative
algorithm. Moreover, if the injected harmonic
frequency is close to a parallel resonant the Fig. 1. Test System
algorithm often diverges. In such cases there is no
alternative to the use of transient convertor Because power systems are generally bilateral,
simulation (TCS) for the derivation of harmonic the (n,m) loci are seen to be identical to loci (m,n).
levels.[l] However the acceptability of TCS harmonic Moreover for the particular line symmetry of the test
studies depends on the existence of a.c. system system, two phases, A and C, are seen to be similar
equivalents responding accurately to power and while phase B differs substantially.
harmonic frequencies. Some dynamic models lack mutual coupling
These frequency-dependent equivalents are also capability and in such cases i t is possible to reduce
required to analyse the behaviour of a.c./d.c. the full matrix information to three diagonal terms
interconnections following convertor or system which contain the mutual impedances implicitly. The
disturbances. The scaled-down simulators and TCS method is valid for systems with little phase current
programs currently used to investigate the dynamic asymmetry, as is normally the case with convertor
behaviour of h.v.d.c. convertors are by necessity very plant. This is achieved by post multiplying each of
limited in their frequency dependence representation. the 3 x 3 matrices by the matrix
An equivalent circuit consisting of tuned RLC
branches has been advocated by N.G. Hingorani {2} as a
possible solution for aVDC studies and a three-phase 1 a2 a
extension of the method has been proposed by the
authors at a recent conference.[3) The accuracy of a 1 a2 where a
- 2+ j ~2
these equivalent circuits is restricted by the lack of
mutual coupling representation and their 'matching
l
a2 a 1
capability is limited to one quantity, the harmonic
impedance, and to a few selected harmonic frequencies.
This paper presents a generalization of the and extracting the diagonal terms. This is important
above techniques, capable of matching the impedances because for some harmonics the mutual terms have as
of mutually coupled and asymmetrical systems for any much effect as the self terms. The resulting loci for
required continuous range of frequencies. the system of Figure 1 are displayed in Figure 3.
The frequency response is probably better
assessed with reference to impedance/frequency plots
87 WM 171-2 A paper recommended and approved rather than impedance loci. These are illustrated in
by the IEEE Transmission and Distribution Committee Figure 4 for the system of Figure 1.
of the IEEE Power Engineering Society for presenta-
tion at the IEEE/PES 1987 Winter Meeting, New Orleans, Implementation in Time Domain Studies
Louisiana, February 1 - 6, 1987. Manuscript submit-
ted August 28, 1980; made available for printing Generally time domain solutions will be required
November 17, 1986. to derive the voltage waveforms resulting from the
interaction between a static convertor and the 9upply
system. Thus the complex impedance-frequency
information contained in the loci matrices (such as

1
© 1987 IEEE
196.

XII] ) x III \ X(\l)


350 350 350

Zaa Zah Zac

150 15

RIll )
-50 -5

-250

350 Zba '50

15 150

-5 -50

-250 -250 -250

350 350 350


ZCu Zcb Zec

150

-50

-250 -2~O -250

Fig. 2. ImpeoilncQ Loci Hatrix of Test System

X(1l ) XIIl ) X(\l)


375 375 375

Za Zb Zc
250 250 250

125 126 125

Rln) R (Q)
0 0 0

-125 -125 -125

-250 -250 -250

Fig. 3. Diagonalized Matrix Impedance Loci


197.

WI (1/ I (1/1

aoo
Zac
.00

000

'00

12f10 00

Zbb

'00

.00 .00 '00

eoo

'00 eoo
Zca

'250

Fig. 4. Impedance/Frequency Matrix Derived from Figure 2.

Self circuits
+

that of Figure 2) must be converted into equivalent


circuits. Every element of the 3 x 3 matrix must be
synthesised into an equivalent circuit.
The equivalents matching the self terms are
represented explicitly as shunt branches. The mutual
terms are processed separately by using diakoptical
techniques; the phase currents are first impressed
upon the mutual circuits and the reSUlting voltages
are then incorporated in the overall circuit as
additional voltage sources per phase. Thus the
overall frequency dependent model of the a.c. system
to be used in the time domain solution is as shown in
Figure 5.
Mutual circuits

Extension to f.1ul tlconvertor Systems

Let us now consider the case of a two-convertor


system as shown in Figure 6. This system needs two
points of observation, i.e. busbars A and B. The
convertor connected to busbar B will affect the
impedances of the a.c. system when observed from
busbar A and vice versa. This interdependence
Equivalent circuits
increases the number of mutual terms, i.e. each phase
has a self and five mutuals.
An extension of the diakoptical technique to
this case requires six self and thirty mutual circuits ~--Q~:·-·~:·r
for a complete frequency-dependent model.
Although extension of the theory to more than
two convertors is straight forward, the number of
frequency dependent circuits required increases with
it· ...........
-=!= ~ ~
the square of the number of convertors and the
complexity of each frequency dependent circuit is Fig. 5. Structure of the Frequency Dependent Equivalent
dependent on the number loops of the frequency loci. Hodel
198.

functions of frequency).
and h is the Error Criterion Function
Among the various types of Error Criterion, the e"",
Least Squares method has been found to be the most
suitable in our problem. Its general form is: 5"'..
minimize 0 = Sum{(wi·g(~))2) ..."
(a)
where wI,---w m are weighting factors to emphasize 3"'"
9i{~) in regions of importance.
2 ....
The optimization method is more flexible in the
selection of equivalent circuit topology; the only ",,.,
limitation in this respect is the possibility of
evaluating its frequency response. However, the m f
circuit topology must be decided before the 50 250 ."0 6"01 SS0 112150 1250
optimization process.
The CPU time requirements of the multi-variable
optimization techniques is several orders of magnitude aD ..
greater than those of the direct approach but they can
be substantially reduced if the circuit derived by the 5""
direct method is used as a first approximation in the
optimization approach. ..." (b)
EFFEC~'IVENESS OF THE IILGORITHH 3""
2 ....
Figure 8(a) displays the frequency response of
one of the mutual elements of the 'actual' test system
1 ....
of Figure 1, together with that of Hingorani's [21
harmonic equivalent. Although the results are in good
agreement at low frequencies the second and third "5111 2"0 .5" 65111 85" lUSt!! 12513
peaks (at 845 and 876 Hz) show substantial disparity.
The use of a correction filter in the region of
largest discrepancy (862 to 906 Hz) modifies the
harmonic response as shown in Figure 8(b). The new 6""
branch reduces considerably the error in the regions
where the impedances of the equivalent circuit were 5""
too high, but no marked improvement is noticeable
where the equivalent circuit impedances are too low. 4""
As explained earlier these are best corrected by a (e)
combination of scaling factor and subsequent use of 3""
correction filters.
The region of greatest discrepancy of the new 2""
equivalent circuit is now between 750-810 Hz. The
addition of a second correction filter in this region '"''
produces a far better match as shown in Figure 8(c).
With reference to multi-variable optimization "5"
techniques, the effectiveness of the synthesised
equivalent circuit is very dependent on the choice of
error criteria. If the optimization criterion is the
minimization of the maximum deviation, the response, Fig. 8. Direct Synthesis of a Mutual Element of the
illustrated in Figure 9(a), is unsatisfactory, i.e. a Impedance/Frequency Hatrix
large discrepancy is introduced in the first peak. In Actual system response (continuous line)
this case it is necessary to use more elaborate Initial equivalent (dotted line (a))
weighting functions which are a function of the With one additional correction filter
discrepancy as well as the frequency. On the other (dotted line (b))
hand the use of the least squares as the error With two additional correctioll filters
criterion, illustrated in Figure 9(b), shows a marked (dotted line (e))
improvement. Although the sum of the errors squared
is minimized in this case, there are larger In order to illustrate the effectiveness of the
discrepancies at Some individual frequencies when frequency-matched equivalents described above, the
compared with the direct solution. test system of Figure 1 Was modified to include an
Thus an interactive approach based on the direct h.v.d.c. convertor at the TIWAI Bus, Figure 10 shows
method enhanced by the use of correction filters the d.c. voltage waveform, derived from Transient
generally has les& error over most of the frequency Convertor Simulation, when a short-circuit occurs on
range as compared to an optimization derived the d.c. line. The continuous line shows the response
equivalent. However, each correction filter increases with a freqUency-matched a.c. system equivalent and
the complexity of the equivalent circuit and hence the the dotted line that of a Thevenin equivalent derived
computational burden when carrying out time domain from the short-circuit capacity of the a.c. system.
studies. On the other hand the CPU time required for There is a significant difference between the two
a multi-variable optimization study is several orders responses which shows the importance of using accurate
of magnitude greater than the direct approach. Hence frequency-dependent equivalents for the a.c. system.
if a relatively small number of time domain The time domain simulation algorithm includes
simulations are required the direct. approach is accurate modelling of the non-linear characteristics
preferable while for many, the optimization technique of the convertor plant, such as convertor transformer
is probably best as a comparable accuracy equivalent saturation, surge arresters, etc.
can be obtained with fewer circuit elements.
199.

Therefore it becomes impractical to model accurately a


large number of convertors connected to the a.c.
Direct Path Optimization Path
system.

Use measurement or analysis to


obtain the system frequency
response (selfs and mutuals)

A.C. Use Hingorani's topoloqy Decide eauivalent


System and direct synthesis to circuit topology and
derive a mUlti-phase optimize the circuit
equiva.lent parameters

---1 optimize

I
I
No

branch to
Fig. 6. Two-convertor System circuit

EQUIVALEN'r CIRCUIT SYNTIIESIS

An initial equivalent circuit topology is


selected to suit the requirements of the time-domain
algorithms, such as the usc of node-type partitioning Fig. 7. Flow Diagram for the Direct and optimization
and tensor matrix analysis.{41 These restrictions Algorithms
increase dramatically the solution efficiency in the
case of a.c.-d.c. simulation, which require regular peaks and troughs, to obtain the best fit between the
topological changes, by avoiding involving the v,'hole equivalent and real system responses.
network during localised convertor valve switchings. The selection of small scaling factors results
Since the topology used by flingorani,[2j i.e. in larger impedance levels, which are easier to
RLC shunt branches, is amenable to Efficient Transient compensate. The compensation is achieved by means of
Convertor Simulation (4J this form has been retained correcting RLC (filter) branches which drop the
as the starting point in the derivation of the impedance levels in the regions of large discrepancy.
equivalent circuit. From inspection the region of largest deviation from
The l)asic matching pllilosophy consists of the required response is found and three matching
selecting the values of R, (, and C that give the peaks points are selected in this region. The formulation
and troughs at the correct frequencies and the Q of Appendix B is then used to fit the correcting RLC
(quality factor) of each branch such as to cause the branch to these three points and thus obtain the
equivalent circuit response between these frequencies values of RLC required. This correction filter is
to approximnte the actual 5ystem response. The added to the equivalent circuit to form a new
analytical part of the alqorithm is described in equivalent and the process repeated if necessary to
Appendix A. find a new region of discrepancy and another
An acceptable error bet\.;een the actual system correction filter.
and the synthesised cireui t depends mainly on the This process continues until the required
following factors: accuracy is obtained, though the latter may be limited
(i) The type of study. For steady state harmonic by the increasing computing requirements. The CPU
penetration stUdies minimal error at the harmonic time used in a.c./d.c. dynamic simulation is related
frequencies is required while the error at to the size of time steps of the numerical solution,
intermediate frequencies are unimportant. This is which itself depends on the time constants of the
clearly unacceptable for transient convertor equivalent circuit. In each case the acceptability of
simulations as some intermediate frequencies could be a particular equivalent circuit and the required CPU
excl ted. time, are assessed by means of a short-duration
(ii) The feasibility of altering automatically the dynamic simulation study.
equivalent circuits to get the error down to a
prescribed accuracy. Optimization Method
(iii) The availability of an economically viable
alternative to model the actual system explicitly. Optimization methods {S][6) use either the
This relates to the amount of CPU time required to Search or Gradient approaches. The latter being more
provide the simUlation and accuracy required. difficult to implement due to the need for derivative
The basic steps of the equivalent circuit informa tion.
synthesis are illustrated in the flow diagram of A gradient optimization method based on a
Figure 7. This diagram includes two alternative special implementation of the
paths, i.e. the direct and optimization methods, Broyden-Fletcher-Goldfarb-Shanno (BSGF) technique has
which are described in the next two sections. been used to minimize the difference(scalar quantity
Ie') between the actual system and equivalent circuit
Direct Nethod responses. Its form is:

The mathematical formulation used to make an


initial assessment of the RLC values to give the peaks
and troughs at the correct frequencies is described in where 9 is tenned the objective fUJlction
Appendix A. Scaling of circuit parameters is then Al---Am are the values of the independent
carried out, without altering the position of the variables at the m sample points (which are complex

4
200.

individual a.c. system busbars. It does, however,


eo",
provide accurate assessment of the voltage and current
wav~forms at the convertor terminals, which is
511J11I
essential in Transient Convertor Simulation. This
information can also be used in harmonic penetration
4"''' studies and filter design for the accurate derivation
of harmonic levels throughout the oystem. For this
3"0
purpose the steady state harmonic currents finally
obtained from the Time Domain Simulation must be
201111
injected into the actual network (represented in the
frequency domain) to derive the voltage distortion
"'111
levels at remote buses. \'1e have already tried this
I1J
technique with promising preliminary results.

5"
REFERENCES
8,m
[ 1) Kitchin, R.II. "New mothod for digital comlluter
5"''' evaluation of convertor harmonics in power
4<111J
systems using state variable analysis n , Proc.
lEE, Pt. C, Vol. 128, No.4, pp. 196-207.
3111111 (b) [ 2) }lingorani, N.G. & Burhery, M.P. "Simulation of
A.C. system impcdnnce in IIVDC system studies",
2111" IEEF. Trans. Vol. PAS-89, No. 5/6, May/Juno 1970,
pp. 820-826. .
1"111 [ 3) watson, N.R., Arrillaga, J. & Joosten, A.P.B.
"A.C. system equivalents for the dynamic
I1J simulation of HVDC convertors", IER Conference
5" Publication 255 on AC and DC Power Transmission,
Sept. 1985, London, pp. 366-371.
Fig. 9. Optim~zed Mutual Element of the Impedance/ [ 4) Arrillaga, J., Arnold, C.P. llarker, B.J.
Frequency Hatrix "Computer Modelling of Electrical Power
Actual system response (continuous line) Systems", J. Wiley (1983).
Using maximum deviation criterion (dotted [ 5) Adby, P.R. and Dempster, M.A.H. "Introduction
line (a)) to optimization Methods", Chapman and Hall
Using least squares criterion (dotted line (1974).
(b) ) [6) Fletcher, R. Practical Methods of Optimization,
Vol. 1 Unconstrained Optimization, John Wiley &
Sons (1980). Vol. 2 Constrained Optimization,
John Wiley & Sons (1981).

APPENDICES

A. Derivation of Equivalent Circuit Parameters

The basic equivalent circuit contains a number


of single-tuned parallel branches as shown in Figure 5
and the first step is to obtain values for the
\ inductors and capacitors of the equivalent circuit

\ that will produce the same minima and maxima resonance


frequencies. The resistances of the branches are
\ ignored as they have very little influence on the
magnitude of the admittance at frequencies that are
sufficiently higher or lower than the series resonant
frequency of the branch.
The admittance of an n branch network is:

Fig. 10. Dynamic Simulation of the d.c. Voltage at the


n 1 n
Convertor Terminals following a Line Short-
Circuit
s r - n
k=l Lk Jrk
(s2+w2)
J
(i) Frequency-dependent equivalent y(s) (1)
n
(ii) Thevenin equivalent
rT (s2+wi)
1=1
CONCLUSIONS
where:
Direct synthesis and multi variable optimization
techniques have been described for the derivation of _1_ I=l,2 •.• etc. and s = jw
LrC r
three-phase frequency dependent a.c. system
equivalents. These equivalents model the frequency For .an LC network the impedance is zero at minima and
dependence of the self and mutual impedance terms and infinite at maxima frequencies.
are amenable for implementation into time-domain Impedance minima occur When:
programs. Both methods have proved capable of
achieving high levels of matching accuracy, but the n
optimization techniques require much greater computer n
]=1
o (2 )

requirements.
It should be noted that the equivalent circuit
does not give explicit harmonic information at the Impedance maxima will occur when the numerator

6
201.

of equation (1 ) is zero, i.e. n n

n n
y )'
k=1
-w
Lk (wLI/L C )
k k
-j r
k=l
W
Lk(w2-w~)
( 11)

L
k=1 Lk J;ik
.l.. n (s2+wJ) 0 (3 )
As Lk appears in the denominator of equation
(11) ,to increase the admittance contribution of each
Let FI' F2, ... F n be the frequencies at which the branch, so as to reduce the impedance, the inductors
admittance is zero, then equation ()) can be written are divided by the scaling factor. In order to keep
as: the same resonance frequencies the capacitor values
n-l are multiplied by the scaling factor.
5 f! (5 2HI~) where Mk (4 )
k=l B. Derivation of Correction Circuit Parameters
2
By equating the coefficients of 5 in (3) and (4) 1 zl1
2 R2 + (wl L - _1_) (12 )
wJC
the following equations result:

For 5 2n - 1 IZ212 R2 + (w2 L --.L) 2 (13)


w2 C
2
(5 ) I Z312 = R2 + (w3 L - _1_) (14 )
w3 C
Equation (12 ) - equation (13 ) gives:
For 5 2n - 3
1 zll2-1 Z21 2 = (wIL- -l...) 2_ (w2L- -l...)
wlC w2 C
2=(wi-w~) L+:h( ~- ~)
C wJ w
n-l 2
L M~ (6) (15 )
k=l
similarly using equation (13) and (14) yields;
For 5 2n - 5
1 Z21 L 1Z312=(w~-w~ )L2+;\(~_~) (16)
n-l n-l C w2 w3
I L (7 ) Rearranging equation (16) gives:
1=1 J=1+1

n n n-l
)' .l.. IT w~ iT (8 ) ( 17)
k=l Lk J=k k=l

Thus n equations are formed which are then substituting equation (17) into equation (15)
solved for the n unknown inductances (Ll,L2, ... Ln l. gives)
The appropriate capacitor values are found using:

k=1,2 ... n (9)

where wk are the required minima frequencies.


The admittance of the tuned network at any
particular frequency is given by:

n n Rk - ::x
w
(w 2
1
- LC)
k k
y = I. Yk I. (10)
k=l k=l
R~ + [.~ (w 2 - LC)]1 2
w
k k
Hence

Branch k is in series resonance when w~ = l/(Ck 1 )


Lk)' The corresponding admittance term for the kth ~
branch is l/Rk' This term is significantly larger
than the contribution from the branches not in series (19)
resonance and hence dominates the admittance. The
resistance that dominates at each minima is inserted
in the corresponding branch. This yields maxima point
impedances for the equivalent circuit which are much Given three points, equation (19) is solved to
higher than those of the actual system. The problem obtain the inductance of the required correction
can be corrected by the scaling of all inductor and branch. Equation (17) is solved next to obtain the
capacitor values. A common scaling factor is used for capacitance value. Lastly equation (12) is 50lved for
all inductors and capacitors 50 that the resonance the required resistance, this completing the
frequencies remain unchanged. The appropriate scaling determining of the correction filter parameters.
factor is obtained by dividing the calculated
impedance of the netHork at a particular frequency by
the actual systems impedance. Equation (IO) with R
ignored yields:

7
202.

Comparison of steady-state and dynamic models


for the calculation of ACiD system harmonics

Prof. J. Arrillaga, DSc, FlEE, FRSNZ


N.R. Watson, BE
J.F. Eggleston, BE, PhD
C.D. Callaghan, ME

llldexi'IY terms: Power transmission and distribution, Compllter Simulation, hfodelling

However, the AC/DC conversion process involves


Abstract: The steady-state and dynamic simula- multiple point-on-wave switching operations within each
tion models currently proposed for the derivation cycle and these cause a periodic sequence of transients,
of AC/DC system waveforms are described. Their which needs modelling in the time domain.
application to the prediction of harmonic distor- Thus in general the modelling of AC/DC power
tion is compared with reference to accuracy, systems in periodic stady-state requires three basic steps
general applicability and computational efficiency. and an iterative process. The steps are:
It is concluded that generally the use of an iter- (i) time domain derivation of the alternating-current
ative steady-state algorithm provides the most and direct voltage waveforms
accurate and efficient solution. However, such an (ii) time-to-frequency conversion of the current wave-
algorithm often presents convergence problems forms
and in such cases dynamic simulation provides the (iii) harmonic power flow.
necessary back up to obtain a solution.
2.1 Single-frequency AC/DC power flow
The conventional AC power-flow algorithm assumes
Introduction
undistorted voltage and current waveforms, whereas the
The highly nonlinear and waveform-dependent behaviour AC/DC conversion process produces a spectrum of har-
of AC/DC convertors cannot be modelled purely in the monic frequency components. Therefore, when carrying
frequency domain. In approximate harmonic studies the out AC/DC power flows, three reasonable assumptions
convertor harmonic currents are calculated from ideal- are normally made to simplify the convertor voltage and
ised waveforms derived from ACjDC load flows [IJ, and current relationships and to comply with the single-
the harmonic voltages are then obtained by injecting frequency power-flow requirements. These are:
each of the harmonic currents independently into the AC (a) the presence of perfect filtering at the convertor ter-
system, which is assumed perfectly linear. minals (behind the convertor transformer)
Two basically different techniques have been proposed (b) the presence of infinite smoothing inductance on
recently to improve the harmonic calculations. The first the DC side (thus providing perfect DC)
is an extension of the approximate steady-state model, (e) the convertor transformer leakage (commutation)
but uses time and frequency domain studies in an iter- impedance being purely inductive (thus ensuring that the
ative manner and claims to provide exact solutions upon commutating currents are sinusoidal).
convergence of the algorithm [2, 3J. The second tech- With reference to step (i) of the iterative process
nique uses only dynamic simulation and thus claims to described in Section 2, once the firing angle is known, the
provide a more realistic representation of the physical commutation current waveform and its limits are clearly
behaviour [4]. defined, the rest of the phase-current waveform being
However, hitherto, no attempt has been made to vali- perfect DC.
date the results and compare the computational efficiency The Fourier series of the current waveform (step (ii)),
of the proposed algorithms. This paper presents the basic contains the main power frequency and nontriplen, odd
characteristics of the steady-state and dynamic simula- harmonics, which are assumed to be filtered. Thus only
tion models and compares their performance in terms of the first term of the Fourier analysis is used to derive the
general applicability, accuracy and efficiency. current-phasor injection at the convertor terminal.
The explicit use of the current phasor leads to a
2 Analysis of AC/DC systems in periodic sequential power-flow solution; alternatively the nonlin-
steady-state
ear convertor equations can be used together with the
The steady-state behaviour of purely AC systems is nor- rest of the AC system in a unified solution. Finally, the
mally analysed lIsing single-frequency phasors, i.e. in the AC/DC power flow (step (iii)) wiII require either a single-
frequency domain. phase or a three-phase algorithm [lJ depending on the
degree of symmetry of the AC and/or DC system configu-
Paper 1455C (1'9). first received 6th March and in revised rorm 12th ration and operation.
June 1986
Dr. Eggleslon is with GEC Transmission and Distribution Projects
Lid., Lichfield Road. SlnlTord, United Kingdom. the remaining authors
2.2 Conventional harmonic power-flow analysis
are with the Department or Electrical and Electronic Engineering, Uni- The Fourier analysis carried out to derive the current
versity or Canterbury, Christchurch. New Zealand phasor in single-frequency power flows can also provide
lEE PROCEEDINGS. Vol. 134. PI. C. No. I, JANUARY 1987 31
203.

the frequency components of the phase-current wave- As a result of initial mismatches between the real and
forms. In conventional harmonic analysis these are the assumed voltages and currents throughout the system,
harmonic currents used for the derivation of harmonic the dynamic simulation must be run for a number of
voltages throughout the (otherwise linear) power system. cycles until the solution reaches a steady state.
Each of the harmonic currents is injected in turn into The structure of the TCS algorithm is described in
the AC system to derive the individual levels of harmonic Appendix 12.2 and a more detailed formulation and the
voltage distortion. numerical solution can be found in Reference 1.
Finally the steady-state time-domain current wave-
3 Iterative harmonic analysis (IHA) forms are subject to a fast Fourier transform to obtain
their harmonic components.
The direct solution described in the previous section
assumes that the convertor-current waveform derived 5 Verification of the algorithms
from the single-frequency power-flow solution is not
affected by voltage distortion. However, the approx- The IHA and TCS programmes have been developed
imations used in the power-flow solution can introduce independently using different conventions, per-unit
considerable error in the derivation of the harmonic cur- systems etc. It is thus essential to validate their per-
rents. formance before attempting to compare their relative
On completion of the three-phase AC/DC power-flow merits. For a reliable verification, the results of IHA and
solution, which provides reasonable initial information TCS must be compared with an exact solution.
for the harmonic analysis, the approximations used can The AC/DC power-flow algorithm provides an exact
then be removed and an iterative solution carried out solution when the system under consideration satisfies
involving the following steps: the approximations set out in Section 2.1. The idealised
(i) Input information is read consisting of AC and DC circuit diagram of Fig. 1 represents such a system which
system harmonic impedances as well as (initial) convertor includes a six-pulse bridge rectifier.
operating conditions derived from a three-phase AC/DC

"~~L
load flow.
(ii) The alternating (harmonic) current injections and
the direct (harmonic) voltage are calculated.
(iii) For each harmonic, the AC system voltages at the
convertor terminals are found.
(iv) The convertor direct voltage and DC system har-
monic impedances are used to calculate the direct-current 1.73p.u.
waveform.
Fig. 1 Basic equivalent AC/DC sysrem
(v) The voltage zero crossings and firing instants are
updated with the latest information of AC convertor- The AC system voltage VG (the commutating voltage)
voltage waveforms. behind inductance X, is assumed perfectly sinusoidal. On
Steps (ii) to (v) are repeated until convergence is achieved. the DC side the smoothing inductance is very large so
In the calculation of the alternating-current injections that the current is perfect DC. The actual values of the
from the convertor (step (ii» the effects of waveform test-system components are indicated (in p.u.) in the
asymmetry and distortion are included, both during and Figure. Details of the per-unit system are given in Appen-
outside the commutation intervals. The basic formulation dix 12.3.
is given in Appendix 12.1. For a specified DC current of say 1 p.u., the power-
The AC and DC system models used in the derivation flow solution provides all the information required to
of the alternating-voltage and direct-current waveforms derive the phase-current waveform (Le. IX = 20°, Jl. = 19°)
(steps (iii) and (iv» are frequency dependent, i.e. the trans- and with it the current harmonics. Under these condi-
mission lines are represented by their equivalent-It cir- tions the conventional (direct) solution can be used to
cuits. calculate the voltage harmonic distortion and the results
are shown in Table 1.
4 Transient convertor simulation (TCS)
5.1 Verification of IHA
State-space and nodal analysis can be used to derive a set Instead of the voltage source, the intermediate node VT ,
of dynamic equations which allows a systematic solution shown in Fig. 1, is used as the commutating voltage in
of the network in the time domain. the iterative solution.
At each step of the solution the instantaneous values The fundamental component of the phase current,
of the network voltages and currents are checked against derived in the previous section, is used to calculate the
specified logical constraints, to decide whether a topo- corresponding voltage component at VT and this value,
logical change is required. together with those of the direct current Id and firing

Tabla 1 : Comparison of phase 8 harmonic currents for the system of


1
Tes Exact IHA
Harmonic
order magnitude phase magnitude phase magnitude phase
1 0.7666 149.3 0.7762 149.5 0.7658 149.6
5 0.1384 -152.9 0.1392 -152.6 0.1374 -152'3
7 0.0874 145.8 0.0887 164.1 0.0876 146.4
11 0.0381 -159.3 0.0386 -158.3 0.0383 -157.8
13 0.0246 137.9 0.0244 137.7 0.0243 138.2

32 lEE PROCEEDINGS, Vol. /34, Pro C, No.1, JANUAR Y 1987


204.

angle ct. constitute the initial conditions for the iterative parallel resonance between the filters' capacitance and
process described in Section 3. At each iteration the latest the system inductance
value of VT is used as the commutating voltage and the (ii) the relative strength of the AC system, normally
process is repeated until convergence. represented by the SCR (short circuit ratio).
Preliminary results obtained differed markedly from
those of Section 5. The deviations were traced to different The sensitivity of the IHA algorithm to these two factors
interpretations of the specified conditions, and in particu- is investigated in Sections 6.1 and 6.2.
lar the implementation of the zero crossing points used as
a reference for the delay angle. Once the above anomalies 6.1 Symmetrical system
were corrected, the results, illustrated in Table 1, show Fig. 4 represents a six-pulse rectifier connected to a sym-
very good correspondence with the exact solution. metrical system (represented by its Thevenin equivalent)
La= 00

5.2 TCS verification


The initial conditions for the dynamic simulation are also
derived from the results of the AC/DC load flow. With
this information the test system required a IS-cycle run
_..J"0y,JyJ-..,P,'U_,--'....J'0-v.O,5rP~.u'-.--' [>f
to reach a steady-state condition. Voltage and current
waveforms for the last cycle of the TCS solution are
shown in Figs. 2 and 3 respectively.
Fig.4 Basic AC/DC system includinojilters

and characteristic harmonic filters (Le. the 5th, 7th, 11 th


and 13th orders). The filters' ratings and characteristics
are those of the single-tuned branches in the New
Zealand HVDC link.
The system inductance has been chosen to resonate
with the filters at a frequency close to the third harmonic
and the short circuit ratio can be controlled by altering
the convertor power rating.
Fig. 5 shows the effect of altering the short circuit ratio

Fig. 2 Wal'ejimllS derived will. tJoe TCS model


(i) Direct current
(ii) Direct voltage before the smoothing reactor 100 (i)
{iiH Phase 1..1 voltnge at the convertor
(iv) Phase h voltage at the convertor
(v) Phase c voltage at the convertor
oil
c
~
-0
~ 50

10 (ii)

1.5 1.8 2.5


5 CR
Fig. 5 Effect of COllvertor loadillg all the numerical solutioll of the
IlfC algorit/.m
(i) Without thresholding
(ii) With thresholding
Fig. 3 PJoase-cllrrelll waveform derired witJo tJoe TCS model
on the number of iterations to reach convergence. The
Application of an FFT to the current waveform of Fig. continuous line illustrates that when the SCR was
3 produced the harmonic currents listed in the last two reduced to about 1.8 the algorithm failed to converge and
columns of Table 1. the results showed considerable third-harmonic content.
It is apparent that the harmonic levels derived from This was an unexpected result because in a perfectly sym-
the TCS model are practically the same as those obtained metrical system (such as that of Fig. 4) no third-harmonic
with the IHA and exact solutions. current can be generated by the convertor. However a
closer look at the numerical result of the first iteration
6 Limitations of the iterative solution showed that small levels of third and all harmonics were
generated by the computer as a result of random numeri-
The previous section has shown that upon convergence cal error and these were amplified by the paraliel reson-
the iterative harmonic analysis provides the right solu- ant system. It could be argued that this numerical
tion. The next question is to determine the range of appli- problem has a replica in a real system, caused by small
cability of the IHA algorithm. random variations between the individual firings.
Two main factors affect the level of voltage distortion However, to isolate this problem from the investiga-
at the convertor terminals, i.e. tion, the number of significant figures of the harmonic
(i) the presence of a harmonic frequency close to the results was reduced by introducing 'thresholding'. The
lEE PROCEEDINGS, Vol. 134, PI. C, No.1, JANUARY 1987 33
205.

new results, illustrated by the dotted line in Fig. 5, show lion does not converge. This conclusion is reached by
that the algorithm converges in two or three iterations .observing that the convertor alternating-voltage wave-
irrespective of the SCR. forms settle down to a steady state. For instance, Fig. 6
shows two cycles of the waveforms predicted by TCS
6.2 Asymmetrical system conditions (with an SCR of 5.3) after ten cycles of simulation.
The system of Fig. 7 represents the New Zealand South
Island network in the vicinity of a large power convertor
(at TlW AI) which, for simplicity, is assumed to be a six-
pulse rectifier. The convertor is fed from a long-distance
untransposed transmission line which introduces some
asymmetry into the test system, and therefore produces
triplen harmonic currents.
Fig. 8 illustrates the three-phase impedance loci of the
test system, seen from the convertor busbar; these loci
have been derived by injecting 1 p.u. harmonic currents
at that busbar (with the convertor disconnected).
With this test system the IHA algorithm failed to con-
verge when the short circuit ratio was reduced to 5.3.
Earlier studies carried out with iterative algorithms had cycles
concluded that computational divergence reflects a har- Fig. 6 TCS results Jar a case olltside t"e converging region oj tlte
monic instability in the real system or at least a weakness IHC algoritltm (SCR = 5.3)
which may result in unreliable convertor operation (i) Phase a voltage at the filters
(ii) Phase b voltage at the filters
[5, 7]. So far, however, no definite proof has been offered (iii) Phase c vohage at the filters
to substantiate such claims. Should they be correct, the
use of dynamic simulation should also fail to settle into a 7 Limitation of TeS in a practical system
steady state.
Therefore the system of Fig. 7, with a six-pulse conver- The complexity of a power system cannot be accurately
tor at TIW AI, was analysed in the time domain using the represented by means of a single Thevenin equivalent and
frequency-dependent model described in Section 7.1. therefore the algorithmic verification of TCS carried out
The TCS results show that it is possible to operate the in Section 5 gives no indication about its accuracy in a
system with SCR levels ill the region where the IHA solu- real system. Even with a more comprehensive equivalent
circuit the state-space model makes no provision for the
frequency dependence of the individual parameters
involved.
To assess the error incurred by ignoring such fre-
quency dependence, the test system of Fig. 7 was used
instead.
In this case the results of an accurate IHA model, with
the transmission lines represented as equivalent-n
matrices were used as a reference. These results are
compared in Table 2 with a dynamic simulation using
a Th6venin equivalent of the AC network and they
Table 2: TCS/IHA comparison for the system of Fig. 7 with
Fig. 7 Practical AC system witlt a 6-plllse bridge rectijier at THV AI ThBvenin AC system equivalent
(a) Harmonic current injections into AC system (phase a)
350
TCS IHA
300
1 1.5136 1.5098
250 5 0.2951 0.2822
7 0.1784 0.1661
11 0.0971 0.0791
13 0.0698 0.0522
17 0.0351 0.0223
19 0.0248 0.Q155
23 0.0083 0.0131
25 0.0056 0.0119

(b) Harmonic voltages


500
Rfl Phase a Phase b Phase C

TeS IHA TeS IHA TeS IHA


1 1.0726 1.0726 1.0749 1.0749 1 1.0762 1.0762
5 0.0118 0.0097 0.0128 0.0098 5 0.0114 0.0099
7 0.0076 0.0058 0.0082 0.0055 7 0.0081 0.0055
11 0.0022 0.0014 0.0021 0.0013 11 0.0024 0.0014
-250 13 0.0020 0.0012 0.0021 0.0012 13 0.0020 0.0012
17 0.0090 0.0060 0.0091 0.0060 17 0.0090 0.0066
Fig. 8 Tllree-phase impedance loci oJthe test system 0.0080 0.0044
19 0.0080 0.0055 0.0080 0.0047 19
phase a 23 0.0036 0.0015 0.0036 0.0016 23 0.0035 0.0015
phase b 0.0003
25 0.0026 0.0002 0.0026 0.0003 25 0.0027
phase c

34 lEE PROCEEDINGS, Vol. 134, Pt. C, No. I, JANUARY 1987


206.

clearly show that, in the absence of detailed three-phase Table 3: TCS/IHA comparison for the system of Fig. 7 with
and frequency-dependence representation, the TCS solu- frequency matched AC system equivalant
tion cannot provide accurate information. (a) Harmonic current injection into the AC system (phase A)
TCS IHA
7.1 Modified res to include the effect of frequency
dependence 1 1.4794 1.4782
5 0.2850 0,2849
It is not practical to represent the frequency dependence 7 0.1729 0.1707
of each individual system component in the dynamic 11 0,0916 0,0900
simulation model. A more realistic solution is to derive 13 0,0652 0,0631
the simplest AC system equivalent that can provide a 17 0,0311 0,0294
19 0,0208 0.0196
reasonable match at the main resonant frequencies. 23 0,0068 0.0078
A simple algorithm based on the impedance loci of 25 0,0054 0,0068
Fig. 8 has recently been developed by the authors [6] to
derive a three-phase dynamic equivalent with reasonably (b) Voltage harmonics
accurate harmonic-frequency dependence. The applica- Phase a Phase b Phase c
tion of this algorithm to the system of Fig. 7 results in the
equivalent system of Fig. 9 which provides a greatly res IHA Tes IHA res IHA
improved frequency match, as shown in Fig. 10. 1 1,0467 1,0467 1,0476 1.0476 1,0297 1,0297
5 0.0096 0.0098 0,0097 0.0098 0,0100 0.0101
7 0,0033 0,0059 0,0062 0.0057 0.0059 0.0055
11 0.0021 0.0015 0.0014 0.0015 0,0014 0.0016
13 0.0018 0.0014 0,0016 0,0014 0.0014 0,0014
17 0.0119 0,0079 0,0072 0.0076 0.0093 0.0089
19 0,0169 0,0070 0.0055 0.0060 0.0047 0.0050
23 0.0006 0,0009 0.0010 0,0009 0.0006 0.0010
25 0,0001 0,0001 0,0003 0,0002 0,0001 0,0002

Table 4: CPU comparison (in seconds) of TCS and IHA


derived results
Fig. 9 Frequency·clepelldeflt equiLJQie11l circuit
TeS IHA (total)
500
per cycle total
400 With Thilvenin 106 1131 148
equivalent 283 last cycle
c:! With frequency· 183 1903 241
~. 300
c dependence 439 last cycle
d
"tJ
2. 200 isation transient to die down before the harmonic assess-
§ ment can be made. The number of cycles required to
reach a steady state depends largely on the time con-
stants and damping of the circuit as well as the specified
a 450 ~-6~50--B'~50~---'1~05~0-1"'2~50 initial conditions. In general, however, eight to ten cycles
50 250
frequency. Hz are usually sufficient.
Fig. 10 Impedallce match affrequellcy.dependellt equ;I1aleru circuit
By way of illustration the TCS solution takes about 32
minutes to assess the harmonic levels with the frequency-
dependent AC system equivalent. In contrast the IRA
When the circuit of Fig. 9 is used (instead of the programme only takes 4 minutes to produce the same
TIH!venin equivalent) the harmonic information derived information.
from TCS is very close to that of the IHA solution. The It should also be noted that TCS requires a greatly
results of the IHA and TCS alternatives are illustrated increased CPU time in the last cycle of simulation. This
in Table 3. is caused by the larger number of data points being
written out into a data file, in order to increase the preci-
8 Computational efficiency sion of the FFT process. Moreover, as the TCS pro-
gramme has automatic step-size selection, the FFT
The CPU requirements of the two models described in process has to derive a set of equally-spaced data points
previous sections have a complex dependence on many from the programme's unequally-spaced information.
factors, such as the relative strength of the AC system
and convertor rating, convergence tolerances, initial con- 9 Conclusions
ditions etc. Therefore the numerical comparison made in
Table 4 only provides an approximate indication of the The relative merits and limitations of harmonic analyses
relative computational efficiencies. The test system under in the time and frequency domains have been compared
consideration relates to the lower part of the New and the positive aspects of each model have been used to
Zealand South Island System and the two cases listed crossverify and improve the accuracy of the two alterna-
refer to (i) a simplified Thevenin equivalent of the AC tive methods.
system and (ii) an accurate frequency-dependent AC In general the use of the steady-state iterative harmo-
system equivalent. The results were obtained on a nic analysis is to be recommended both in terms of com-
VAXII/750 computer. putational efficiency and accuracy of system
It has already been explained that the TCS results representation at harmonic frequencies. It has been
require a number of cycles of simulation for the initial- shown, however, that when the relative strength of the
lEE PROCEEDINGS, Vol. 134, Pt, C, Na, 1, JANUARY 1987 3S
207.

AC system (determined by the short circuit ratio) is low, ~J H~ -l/T;..


Sh = 2 2 2
the iterative algorithm presents convergence problems. In ItwLnm It w + l/T.m
such cases there is no alternative to the use of transient
convertor simulation to derive harmonic information. y = RmlJ
The results derived from transient convertor simula- R.m
tion are unrealistic if the AC system equivalent is not fre-
quency dependent. However, a preliminary study of the and /I is the incoming phase and m is the outgoing phase;
AC system in the frequency domain provides the neces- Rnm = R. + R",
sary information to derive an accurate AC system har-
monic equivalent, and thus improve the accuracy of the L.rn = L. + Lm
TCS model. 7;,rn = L."'/R.m
10 Acknowledgments Ah = v". cos (1r8. + rPh.) - Vhrn cos (1r8. + rPhm)
The authors wish to acknowledge the financial help
Bh = v"n sin (IrO. + rPh.) - V.m sin (IrO. + rPhm)
received from the UGC and the NZERDC to carry out Hh = IIWAJBh
the investigation reported in this paper.
rPh = tan-I (1/hw7;,"') - tan-I (:~)
11 References

1 ARRlllAGA, J., ARNOLD, e.P., and HARKER, B.1.: 'Computer


where Vhn and rPhn are the magnitude and phase of the hth
modelling of electrical power systems' (J. Wiley & Sons, london, harmonic of the phase to neutral voltage of the incoming
1983) phase, v"m and rPhm are the same, but for the outgoing
2 YACAMINI, R., and DE OLIVEIRA, J.e.: 'Harmonics in multiple phase, and Cnm is the integration constant such that
conYertor systems: a generalised approach', lEE Proc. B, Eleclr. i.(O) = O.
Power Appl., 1980. 127, (2), pp. 96-106
3 EGGLESTON. J.F.: 'Harmonic modelling of transmission systems However, for a more general transformer model,
containing synchronous machines and static conyertors·. Ph.D. capable of representing the star-delta connection, the
Thesis, University of Canterbury (NZ), 1985 foregoing equation is rewritten in terms of the phase-to-
4 KITCHIN, R.H.: 'New method for digital computer evaluation of phase voltages, i.e.
convertor harmonics in power systems using state variable analysis',
lEE Proc. C, Gell. TrailS. & Dislrib, 1981, 128, (4), pp. 196-207 50
5 REEVE, J., and BARON. J.A.: 'Harmonic interaction between V.rn = L (Ah sin Irwl + Bh cos Itwt)
h.y.d.c. convertors and a.c. power systems', IEEE TrailS., 1971, h=1
PAS-90, (6), pp. 2785-2793 using the first firing angle 0 t as a time reference.
6 WATSON, N.R., ARRILlAGA, J., and JOOSTEN, A.P.B.: 'A.e.
system equivalents for the dynamic simulation of H VDC convertors', For a star/star transformer
I EE Calif. P"bl. 255, 1985, pp. 366-371
7 YACAMINI, R., and DE OLIVEIRA, T.e.: 'Comprehensive calcu- Ah = Vh• cos (IrO I + rPhn) - Vhm cos (1r0 1 + rPhm)
lation of convertor harmonics with system impedances and control
representation' lEE Proc. B, Electr. Power Appl., 1986, 133, (2), Bh = v"n sin (1r0 1 + rPhn) - v"m cos (hOI + rPhm)'
pp.95-102
For a star-gldelta transformer (DY11) and a com-
mutation from valve 1 to valve 3
12 Appendix
Ah = -.j(3)Vl cos (ItO I + rPhtl
12.1 Derivation of the alternating currents at the
convertor terminals Bh = -.j(3)Vt sin (IrO I + rPhl)
When only two valves conduct, say valves 1 and 2, the
instantaneous phase-current injections are derived from 12.2 res formulation
the expression: The following set of state equations has been found to
provide a very efficient computer solution [1J:
I = [; [ ~] ~d] pQ. = -(K.,I. + K",I r)
13 -Id
pljl, = E, - R,I, + KT.V. + K[PVp + KTrV,
During the commutation from valve 1 to 3 the current
injections are where Q and IjI are state variables representing the node
charge and inductive branch flux respectively.
_[It]_[
I - 12 -
I in] d
.-
I.
The rest of the symbols and suffixes in the above equa-
tions represent the following:
13 -ld
I = inductive branches
where in'
the commutation current, has the following r = resistive branches
expression [7J: c = capacitive branches
50 IX = node with at least one capacitive branch con-
in(l) = L (Xhe-r/T'ffl + Sh sin (II wi + rPh)) nected
h=1 f3 = node with at least one resistive but no capacitive
branches connected
y = node with inductive but no capacitive or resistive
where branches connected
The above definitions give the topological or branch-
node incidence matrices, their general elements being:
36 lEE PROCEEDINGS. Vol. 134, P,. C, No. I, JANUARY 1987
208.

I if node i is the sending end of branch p 12.3 p.u. system


Kp; = - I if node i is the receiving end of branch p Base quantities AC side DC side
1 o if node i is not connected to branch p Power, MVA 100 100
Voltage, kV 220 105
Details of the formulation and numerical solution can be Current, kA 0.264 0.55
found in Reference 1. Impedance, 0 484 191
Hz 50 50

lEE PROCEEDINGS, Vol. 134, Pt. C, No. I, JANUAR 1'1987 37

You might also like