Ac DC Power System Analysis Arrillaga, - Jos

IEE Power and Energy Series
AC-DC
POWER
SYSTEM
ANALYSIS
Jos Arrillaga
and Bruce Smith
The Institution of Electrical Engineers
Published by: The Institution of Electrical Engineers, London,

United Kingdom
1998: The Institution of Electrical Engineers
This publication is copyright under the Berne Convention and the
Universal Copyright Convention. All rights reserved. Apart from any fair
dealing for the purposes of research or private study, or criticism or
review, as permitted under the Copyright, Designs and Patents Act, 1988,
this publication may be reproduced, stored or transmitted, in any forms or
by any means, only with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms
of licences issued by the Copyright Licensing Agency. Inquiries
concerning reproduction outside those terms should be sent to the
publishers at the undermentioned address:
The Institution of Electrical Engineers,
Michael Faraday House,
Six Hills Way, Stevenage,
Herts. SG1 2AY, United Kingdom
While the authors and the publishers believe that the information and
guidance given in this work is correct, all parties must rely upon their own
skill and judgment when making use of it. Neither the author nor the
publishers assume any liability to anyone for any loss or damage caused
by any error or omission in the work, whether such error or omission is
the result of negligence or any other cause. Any and all such liability is
disclaimed.
The moral right of the authors to be identified as authors of this work has
been asserted by him/her in accordance with the Copyright, Designs and
Patents Act 1988.
British Library Cataloguing in Publication Data

A CIP catalogue record for this book
is available from the British Library
ISBN 0 85296 934 1
Printed in England by Short Run Press Ltd., Exeter
Preface
Although the early decision to generate electric power at 50/60 cycles and
three phases is practically irreversible, the transmission and utilisation of
energy is not necessarily tied to these conditions. The choice between
transmission alternatives is made on the basis of cost and controllability.
The original justification for HVDC transmission was its lower cost for
long electrical distances which, in the case of submarine (or underground)
cable schemes, applies to relatively short geographical distances. At present,
the controllability factor often justifies the DC alternative regardless of cost,
as evidenced by the growing number of back-to-back links in existence.
The merits of HVDC over AC transmission have been explained in
several books by Adamson and Hingorani, Uhlmann, Kimbark, Arrillaga,
and Paddiyar, listed in chronological order.
In earlier days, the dynamic performance of the DC link was assessed with
the help of scaled-down physical simulators. These provided a reasonable
representation of the converter control and protection functions, but were
very restricted in AC-network representation.
With the expansion of HVDC transmission throughout the world, and
particularly the increasing numbers of interconnections between different
countries, few power systems can continue to escape the effect of this
technology in their planning and operation. Such expansion has encouraged the development of analytical models to represent the behaviour of the
AC-DC power system.
An early attempt to describe the HVDC link as a power system component
was made in the book 'Computer modelling of electrical power systems'.
Although the book's main objective was conventional power-system analysis,
it did propose algorithmic modifications for the incorporation of HVDC
transmission. Since then the experience of many years of HVDC operation
has produced more advanced models to represent the behaviour of both the
AC and DC systems.
In particular, the availability of the EMTP (electromagnetic-transient
program) with detailed representation of power-electronic components and,
more recently, its implementation in the RTDS (real-time digital simulator)
xii
Preface
has practically eliminated the need for physical simulators. Consequently,

the impact of the EMTP techniques is given prominence in this book.
Although steady-state waveforms and their harmonic components can also
be derived using the EMTP method, such information can be obtained
more accurately and efficiently in the frequency domain. Therefore, the
present book also contains several chapters describing frequency-domain
techniques with reference to the AC-DC converter.
The primary object of this book is the incorporation of HVDC converters
and systems in power-system analysis, but the algorithms described can
easily be extended to other industrial components such as drives and
smelters, and to the FACTS (flexible AC transmission systems) technology.
Conventional AC power-system concepts and techniques are only included in as much as they are required to explain the incorporation of the
HVDC link behaviour.
The book only deals with system studies, influenced by converter control,
whether steady state or transient. Fast transients, such as lightning and
switching events (in the ns or fj,$ region) are not considered, as they are
beyond the influence of HVDC controllers and can be analysed by conventional power-system methodology.
Contents
Preface
Introduction
.1 Basic AC-DC configuration
.2 AC-DC simulation philosophy
.3 Steady-state simulation
.4 Fault analysis
.5 Harmonic analysis
.6 System stability
The AC-DC converter in steady state
2.1 Introduction
2.2 Power frequencysymmetrical operation
2.2.1 Analysis of the commutation circuit
2.2.2 Rectifier operation
2.2.3 Inverter operation
2.2.4 Power factor and reactive power
2.3 Power frequencyunbalanced operation
2.3.1 Terminology and waveforms
2.3.2 Variables and equations
2.4 Characteristic harmonics
2.5 The converter as a frequency modulator
2.5.1 The modulation process
2.6 Harmonic transfer generalisation
2.6.1 From the AC to the DC sides
2.6.2 From the DC to the AC sides
2.6.3 Effect of switching-instant variation
2.6.4 Transfer across the DC link
2.7 Harmonic instabilities
2.8 Generalised harmonic domain converter model
2.8.1 Analysis of the commutation
Star-connected bridge
Delta-connected bridge
2.8.2 Valve-firing control
Current control
Commutation margin control
XI
1
1
2
3
4
4
5
7
7
8
10
12
14
15
17
17
18
23
26
27
29
30
31
32
33
36
41
41
41
43
44
45
46
vi
Contents
2.8.3 Direct voltage
Star-connection voltage samples
Delta-connection voltage samples
Convolution of the samples
2.8.4 Phase currents
Converter side
System side
2.9 Summary
2.10 References
48
48
49
51
54
54
54
59
60
The power flow solution

3.1 Introduction
3.2 Specification of the operating condition
3.3 Formulation of the algorithm
3.4 Decoupled Newton techniques
3.5 Incorporation of AC-DC buses
3.6 DC-system model
3.6.1 Converter variables
3.6.2 DC per unit system
3.6.3 Derivation of equations
3.6.4 Incorporation of control equations
3.6.5 Control of converter AC terminal voltage
3.7 Unified AC-DC solution
3.7.1 Multiconverter systems
3.7.2 Programming considerations
3.8 Convergence properties
3.9 Modification of the power flow for use with the unit connection
3.9.1 An equivalent inverter model
3.10 Components-related capability
3.10.1 HVDC test system
3.10.2 Type of HVDC constraint loci
3.10.3 Constraint equations
3.10.4 Loci of operating constraints
3.10.5 Complete capability chart
3.11 System-related HVDC capability
3.11.1 Converter powercurrent characteristics
3.11.2 Converter power limits
3.11.3 DC link power-transfer capability
3.12 Summary
3.13 References
61
61
61
63
66
70
71
72
74
74
77
78
78
78
83
84
86
87
90
92
93
94
95
98
100
102
102
104
106
107
108
The
4.1
4.2
4.3
4.4
4.5
109
109
109
110
113
116
117
117
121
harmonic solution
Introduction
Basic AC-DC system
Functional notation of the converter equations
Mismatch equations
Newton's method
4.5.1 The Jacobian matrix
Numerical differentiation
Analytical derivation
Contents
4.6
4.7
4.8
4.9
With respect to AC phase-voltage variations

With respect to direct current ripple variation
With respect to end-of-commutation variation
With respect to firing-instant variation
4.5.2 Sequence-components model
Computer implementation
4.6.1 Initialisation
4.6.2 The switching system
4.6.3 Harmonic solution
4.6.4 Convergence tolerance
Validation and performance
Summary
References
Three-phase power and harmonic flow

5.1 Introduction
5.2 The three-phase power
5.2.1 Mismatch equations
5.2.2 The power-flow Jacobian
5.2.3 Newton's method
5.2.4 Performance of the power
5.2.5 Zero-sequence blocking
5.3 Converter harmonic model
5.4 Combined solution
5.4.1 Sequential method
5.4.2 Unified Newton method
5.4.3 Test system
5.4.4 Convergence characteristics
5.4.5 Power flow/converter interaction
5.5 Summary
5.6 References
flow
flow
Electromagnetic transient simulation

6.1 Introduction
6.2 The state-variable solution
6.2.1 Structure of TCS
6.2.2 Integration method
6.2.3 Choice of state variables
6.2.4 Forming the network equations
6.2.5 Valve switchings
6.2.6 Effect of automatic time-step adjustments
6.2.7 TCS converter control
6.3 The EMTP method
6.3.1 Discretisation of system components
6.3.2 Error analysis
6.3.3 Switching discontinuities
6.3.4 Voltage and current chatter due to discontinuities
6.4 Subsystems
6.5 The EMTP program
6.6 The NETOMAC program
vii
122
123
123
124
124
128
130
130
132
133
134
141
141
143
143
143
145
146
153
153
158
161
162
162
165
167
167
169
171
172
175
175
177
178
180
181
182
184
186
189
192
193
199
201
204
209
215
216
viii
Contents
6.7
The PSCAD/EMTDC program

6.7.1 Program structure
6.7.2 DC valve groups
6.7.3 Transmission-line model
Frequency-dependent transmission-line models
Curve fitting for Zc and ,4(0
Line constants
6.7.4 Converter-transformer model
Single-phase UMEC model
UMEC Norton equivalent
UMEC implementation in PSCAD/EMTDC
Three-limb three-phase UMEC
6.7.5 Future developments
6.8 Examples of PSCAD/EMTDC simulation
6.9 Modelling of flexible AC transmission systems (FACTS)
6.9.1 Simulation of the SVC in PSCAD/EMTDC
6.9.2 Dynamic voltage control at an inverter terminal
6.10 Real-time digital simulation
6.11 Summary
6.12 References
217
217
220
223
225
228
229
230
230
233
235
237
241
241
258
259
262
263
269
271
Electromechanical stability
7.1 Introduction
7.2 Dynamic model of the synchronous machine
7.2.1 Equations of motion
7.2.2 Electrical equations
7.2.3 Synchronous-machine controllers
7.2.4 Generator representation in the network
7.3 Load representation
7.4 The AC transmission network
7.4.1 System faults and switching
7.5 Static power conversion models
7.5.1 Single-converter loads
Abnormal modes of converter operation
Converter representation in the network
7.5.2 DC links
DC power modulation
DC link representation in the network
7.6 , AC-DC transient stability programs
7.6.2 Trapezoidal integration
7.6.3 Initial conditions
7.6.4 Test of operating mode
7.6.5 Converter program interface
7.6.6 Test for commutation failure
7.7 Test system and results
7.7.1 Minor disturbance
7.7.2 Major disturbance
7.8 Summary
7.9 References
275
275
276
276
278
280
282
285
286
287
287
288
290
293
296
298
301
301
301
301
304
306
307
309
312
315
316
320
320
Contents
8
Electromechanical stability with transient converter simulation

8.1 Introduction
8.2 Description of the hybrid algorithm
8.2.1 Individual program modifications
8.2.2 Data
flow
8.3 TS/EMTDC interface
8.3.1 Equivalent impedances
8.3.2 Equivalent sources
8.3.3 Phase and sequence-data conversions
8.3.4 Interface variables derivation
8.4 EMTDC to TS data transfer
8.4.1 Data extraction from converter waveforms
8.5 Interaction protocol
8.6 Interface location
8.7 TSE hybrid algorithm
8.8 Test system and results
8.8.1 Electromagnetic transient response
8.8.2 TSE hybrid response
8.9 Summary
8.10 References
ix
323
323
324
326
326
328
329
331
331
332
335
339
343
346
351
355
356
356
361
362
Appendices
I
Newton-Raphson method
363
1.1 Basic algorithm
363
1.2 Techniques to make the Newton-Raphson solution more efficient 364
Sparsity programming
365
Triangular factorisation
365
Optimal ordering
366
1.3 References
366
II
The short-circuit ratio (SCR)

Definitions
Derivation of short-circuit ratios
Reference
367
367
368
370
III
Test systems
III. 1 CIGRE HVDC benchmark model
111.2 Simplified test system
111.3 Test systems used in the stability chapters
System A
371
371
371
373
373
System B
375
IV
State-space analysis
379
Numerical integration
383
VI
Curve-fitting algorithm
387
Index
391
Chapter 1
Introduction
1.1 Basic AC-DC configuration

The three-phase bridge, shown in Figure 1.1, is the basic switching unit
used for the conversion of power from AC to DC and from DC to AC. The
valve numbers in the Figure (1, % 3, 4, 5, 6) indicate the sequence of their
conduction with reference to the positive sequence of the AC-system phases
(R, Y, B).
Two series-connected bridges constitute a 12-pulse converter group, the
most commonly used configuration in high-voltage and large-power applications. Figure 1.2 illustrates schematically the main components involved
in a typical AC-DC converter station and Figure 1.3 shows the standard
circuit of a monopolar high-voltage direct-current transmission scheme.
Although the analysis described in this book relates specifically to the
12-pulse converter and a point-to-point DC link, the proposed algorithms
can easily be extended to higher pulse converters and multiterminal
AC-DC interconnections.
k 4
.z
A^ 3
_
ii 5
Figure 1.1
Three-phase bridge configuration
i^ 6
'f 2
AC-DC power system analysis

Converter
busbar
Figure 1.2
Converter
transformers
6-Pulse
bridges
D.C. line
|W
12-pulse converter configuration
2
1
filters
filters
Figure 1.3 Basic HVDC interconnection
1.2 AC-DC simulation philosophy

Considering the relatively small number of HVDC transmission links in
existence, as compared with their HVAC counterparts, there may appear to
be no need for altering substantially the conventional power-system analysis.
However, the ratings of the HVDC links are often considerably higher, and
their behaviour extremely different. In practice, even the presence of a
single DC link can have a big influence on overall system performance, and
must be adequately represented.
Introduction
Being under continuous electronic control, DC links are free from

synchronous restrictions, but their waveform-related behaviour is more
difficult to predict. Consequently, much present-day design is still carried
out with the help of physical simulators. As the name indicates, the physical
simulator involves real (although scaled down) converters and their actual
controllers. Thus, the electromagnetic behaviour of the DC link is realistically represented, at least qualitatively. On the other hand, the physical
models are very restricted in AC-system representation, particularly for
frequency-dependent and electromechanical behaviour.
By emulating the converter point-on-wave dependent behaviour, recent
developments in electromagnetic-transient simulation provide a digital
alternative to the physical simulator, with the additional advantage of a
better representation of the AC-system components. In particular, the
recent addition of real-time processing capability (the RTDS) has made
electromagnetic-transient simulation even more realistic, particularly in
the hybrid mode where the actual converter controllers are interfaced with
the RTDS simulation via analogue to digital and digital to analogue
conversions.
1.3 Steady-state simulation

A steady-state condition implies that all the variables and parameters
remain constant for the period of observation. The most commonly used
steady-state study is the load flow or power flow, which is carried out to
ensure an adequate power supply under a specified load condition. Being a
purely fundamental-frequency study, the power flow requires no waveform
information.
In its simplest form, the DC link can be represented like other loads in
the power-flow solution, i.e. by active (P) and reactive (Q) power specifications at the converter terminals; however, at the solution, checks must be
made to ensure that the DC-link controls remain within their capability
limits.
For greater flexibility and improved convergence, some of the DC-link
variables must be explicitly represented in the power-flow solution, replacing the P and Q specifications. This alternative, requiring either sequential
AC and DC iterations or a unified AC-DC Newton-Raphson solution, is the
main object of Chapter 3.
Three-phase steady-state simulation, discussed in Chapter 2, is rarely
needed in conventional power systems. However, the DC-link behaviour, and specifically the generation of uncharacteristic harmonics, is greatly
affected by voltage or components' unbalance. Therefore, there is a greater
need for three-phase modelling in the presence of AC-DC converters.
Moreover, in the absence of perfect filters, there can be substantial interac-
tion between the fundamental and other frequencies. This interaction is

considered in some detail in Chapter 5.
1.4 Fault analysis

Apart from the values of the system parameters to be used, which will
depend on the time frame of the study, fault simulation also comes under
the category of steady-state solutions, although it is often referred as
quasisteady state. Like the power-flow solution, conventional fault analysis is
only concerned with the fundamental frequency and is mostly used for
design and relay-setting purposes. Regarding AC-DC power systems, unless
the fault is sufficiently remote from the DC link for it to maintain normal
operation, the inverter will develop commutation failures and the DC-link
behaviour cannot be represented in such programs. Instead, a detailed
waveform analysis needs to be carried out using electromagnetic-transient
simulation. The most popular transient-simulation methods are based on
the EMTP programs, and these include detailed models of the HVDC
converter. In particular, the PSCAD-EMTDC program has been specially
designed with HVDC transmission in mind, and is given particular consideration in Chapter 6.
The small integration step (typically 50 fis) used by the EMTP programs
gives an accurate prediction of the distorted voltage and current waveforms
following a disturbance. Realistic electromagnetic-transient simulation of
HVDC systems always requires that the switching instants are accurately
determined as they occur. Invariably, they fall between time steps of the
simulation, and much work has been directed towards removing any error
resulting from this mismatch. Solutions to this problem first appeared in
state-variable converter models, where the simulation step itself was varied
to coincide with switching instants and to follow any fast transients subsequent to switching. More recently, interpolation techniques have been
successfully applied in electromagnetic-transients programs. Both of these
approaches are discussed in Chapter 6.
1.5 Harmonic analysis

The levels of the so-called characteristic harmonics, i.e. those related to the
pulse number, can easily be calculated from the symmetrical steady-state
converter model. Moreover, these harmonics are normally absorbed by local
filters and have practically no effect on the rest of the system.
On the other hand, the presence of unbalance or distortion on either side
of the link produces a plethora of frequency components as a result of the
converter's crossmodulation process. An accurate prediction of these effects
requires very sophisticated modelling, and often a complex iterative algo-
Introduction
rithm which also involves the AC and DC systems. Important contributions

have already been made in this respect and these are discussed in Chapters
2 and 4.
1.6 System stability

Dynamic and transient-stability studies use quasisteady-state component
models at each step of the electromechanical solution. Similarly to fault
simulation, the presence of DC links prevents the use of these steady-state
models for disturbances close to the converter plant. Unlike fault simulation, however, the stability studies require periodic adjustments of the
generator's rotor angle and internal e.m.f.'s, information which the electromagnetic-transient programs cannot provide efficiently. Thus, in general, the AC-DC stability assessment involves the use of the three basic
programs discussed in previous sections, i.e. power flows, electromagnetic
transient and multimachine electromechanical analysis. The interfacing of
the component programs, by no means straightforward, is considered in
Chapters 7 and 8.
Chapter 2
2.1 Introduction
The transfers of voltage and current across the AC-DC converter are
completely specified by the switching instants of the bridge valves, being
both the firing and end of commutation instants. On the assumption of a
balanced, undistorted AC-terminal voltage, and infinite smoothing reactance, the converter is readily analysed by Fourier methods.1 Under these
conditions, closed-form expressions can be obtained for the firing angles,
commutation duration, fundamental-frequency voltage and current, characteristic phase-current harmonics and DC-voltage harmonics.
However, under realistic conditions, there is some asymmetry and
distortion, the switching instants of the bridge valves are not equispaced
over one cycle owing to control action, and the transfer function between
the AC and DC system is modified. Even a small modulation of the
switching instants can lead to current components in the AC system at the
modulation frequency sidebands. This effect is equally important to both
commutation duration and to firing-angle variation.
The incorporation of switching-angle modulation in the converter model
permits an accurate derivation of the individual switching instants; their
effect on transfers between the AC and DC systems can then be quantified,
and all causes influencing the modulation accounted for. An early cause of
firing-angle modulation was the use of individual firing control, which was
responsive to harmonic distortions in the terminal voltage.2 The adoption
of equidistant firing control, with its much longer time constant, effectively
eliminated this type of firing-angle modulation. However, the firing angle
is still modulated as a result of harmonics in the DC current, via the
current-control loop, sometimes resulting in harmonic instability.3 The
commutation duration is modulated by terminal-voltage harmonics, DC
ripple and firing-angle variation.4 All these variations can have a significant
effect and must be included in the converter model if good accuracy is
required. Modulation of the commutation-period duration by terminalvoltage harmonics, in particular, has a significant impact on the AC-side
harmonic response of the converter.
Moreover, in response to an applied harmonic distortion, sideband

frequencies will be present that are phase reversed, and which correspond
to frequency-shifted negative-frequency components. If the converter is to
be linearised, this effect should be represented. This is an important factor,
especially if converter impedances are being derived; however, this effect
is not commonly appreciated.
Imbalance in the AC system at the fundamental frequency owing to load
and transmission-line asymmetries also leads to the generation of abnormal
harmonics by the converter, and even small levels of negative sequence at
the fundamental will promote the injection of odd triplen harmonics by the
converter.5
Accurate modelling of converter transformers is necessary because of
their direct influence on the commutation process, the effect of tap change
on firing angle and the effect of imbalance. Also, core saturation has often
led to harmonic instability and must be represented.
The small resistive voltage drops of the commutation circuit (converter
transformer and thyristor stacks) slightly alter the average firing angle
derived by the converter controller. This, in turn, phase shifts the harmonic
injections by an angle proportional to the harmonic order. Therefore, the
phase angle of high harmonic orders injected by the converter, such as the
49th, is sensitive to this effect.
Finally, the effect of stray capacitance on the DC side can be very
significant, causing odd triplen harmonics.6
This Chapter describes the converter models required for different
power-system studies in the steady state; these range from fundamentalfrequency power flow to harmonic crossmodulation assessment in order of
increasing modelling complexity.
2.2 Power frequencysymmetrical operation

For fundamental-frequency studies, the following assumptions are normally
made in the development of the converter model:
(i)
The forward-voltage drop in a conducting valve is neglected so that

the valve may be considered as a switch. This is justified by the fact
that the voltage drop is very small in comparison with the normal
operating voltage. It is, furthermore, quite independent of the current
and should, therefore, play an insignificant part in the commutation
process since all valves commutating on the same side of the bridge
suffer similar drops. Such a voltage drop can be taken into account by
adding it to the DC-line resistance. The transformer-windings resistance is also ignored in the development of the equations, although it
should also be included to calculate the power loss.
(ii) The converter-transformer leakage reactances as viewed from the

secondary terminals are identical for the three phases, and variations
of leakage reactance caused by on-load tap changing are ignored.
(iii) The direct current ripple is ignored, i.e. sufficient smoothing inductance is assumed on the DC side.
(iv) All the current-harmonic content is filtered out at the converter
terminals and, therefore, the converter-terminal voltage is perfectly
sinusoidal.
(a)
(b)
(c)
- j _
'j>
(d)
T\ r/
Figure 2.1
Six-pulse converter waveforms

a Phase-to-neutral source voltages
b-d Phase currents on the converter side
e Phase current (phase R) on the system side of a delta-star connected
transformer
10

B
(c)
Figure 2.2
Typical six-pulse rectifier operation
a Positive and negative direct voltages with respect to the transformer

neutral
b Direct bridge voltage Vd
c AC line current of phase R
Under these assumptions, Figure 2.1 shows the three-phase system voltages
and (prefiltered) converter currents of the bridge rectifier of Figure 1.1 for
star and delta-connected bridges. The currents in continuous trace correspond to ideal commutations, and the dotted lines show the effect of
commutation overlaps.
Figure 2.2 shows the AC voltages on the secondary side of the converter
transformer, the rectified voltage on the converter side of the smoothing
reactor and the AC current (phase R) on the converter side of the
transformer.
2.2.1 Analysis of the commutation circuit1

With reference to Figure 2.3, the commutation from valve 1 to valve 3 can
start (by the firing of 3) any time after the upper voltage crossing between
D
e
11
1^1 -
(a)
'3
(c)
Figure 2.3 The commutation process
a Equivalent circuit of the commutation from valve 1 to valve 3
b Voltage waveforms showing early (rectification) and late (inversion)
commutations
c The commutating currents
eR and eY (and must be completed before the lower crossing of these two
voltages).
Since eY > eR, a commutating current ic (=i3) builds up at the expense of
ix so that at all times
h +H = h
(2.2.1)
As the rates of change of i3 and ix are equal (provided that the commutation reactances are balanced), the voltage drops across XCR and XCY are
the same and, thus, during the overlap period, the direct voltage Vd is the
mean value of eY and eR.
12
From the circuit of Figure 2.3a and assuming XcR = XCY = Xc we can
write
eY - eR = 2(Xc/co) d(ic)/dt
(2.2.2)
Taking as a reference the voltage crossing between phases R and Y
where Vterm is the phase-to-phase r.m.s. voltage, referred to the secondary

(converter) side and a is the transformer tap position.
Eqn. 2.2.2 can also be written as
-â Vtermsin(a)0 d(wt) = XA
(2.2.3)
and integrating from cot = a
- L ^ aVtermsin(cot)d(cot) = Xc f * d(ic)
(2.2.4)
The instantaneous expression for the commutating current is thus

aV
ic = -j^-
[cos a - cos(cot)]
(2.2.5)
and substituting the final condition, i.e. ic = I d at cot = a + fi yields

aV
h = -j^-
[cos a - cos(a + fi)]
(2.2.6)
2.2.2 Rectifier operation

In Figure 2.2, P indicates a firing instant (e.g. PI is the firing instant of
valve 1), S indicates the end of a commutation (e.g. at S5 valve 5 stops
conducting) and C is a voltage crossing (e.g. Cl indicates the positive
crossing between phases blue and red).
Graph a illustrates the positive (determined by the conduction of valves
1, 3 and 5) and the negative (determined by the conduction of valves 2, 4
and 6) potentials with respect to the transformer neutral, graph b the direct
voltage output waveform and graph c the current in phase R.
The following expression can easily be derived for the average output
voltage with reference to the waveform of Figure 2.2&
V, = (i)Vc0[cos a + cos(a + /x)]
(2.2.7)
13
where Vc0 is the maximum average DC voltage (i.e. at no load and without
firing delay); for the three-phase bridge configuration Vc0 = (5y/2/n)aVteTm
and aVterm is the phase-to-phase r.m.s. commutating voltage.
Eqn. 2.2.7 specifies the DC voltage in terms of aVterm, a and fi. However,
the value of the commutation angle is not normally available and a more
useful expression for the DC voltage, as a function of the DC current, can
be derived from eqns. 2.2.6 and 2.2.7, i.e.
Vd = ^L
aVterm cosa
- ^ 1 ,
(2.2.8)
The r.m.s. magnitude of a rectangular current waveform (neglecting the

commutation overlap) is often used to define the converter transformer
MVA, i.e.
I J
Irms
w P'P I2d(cot)\ = y/njj*
rms = Jhw
(2.2.9)
Since harmonic filters are assumed to be provided at the converter terminals, the current flowing in the AC system contains only the fundamental
component frequency and its r.m.s. magnitude (obtained from the Fourier
analysis of the rectangular waveform) is
(2.2.10)
If the effect of commutation reactance is taken into account and using eqns.
2.2.5 and 2.2.6, the currents of the incoming and outgoing valve during the
commutation are defined by eqns. 2.2.11 and 2.2.12, respectively
L,(cos a cos cot)
_j*
L for a < G * < a + ix
(2.2.11)
l =
cos a cos(a + fi)
.
cos a cos(cot - 2n/3) _
2%
2n
i = Id-Id
^
- ^ fora + <cot<(x + - + // (2.2.12)
3
3
cos a - cos(a + fi)
In between commutations, the current is
i = Id
for a + / i < cot < + a
(2.2.13)
The fundamental component of the current waveform defined by eqns.

2.2.11, 2.2.12 and 2.2.13 is
(2.2.14)
14
where
k = V{[cos 2a - cos 2(a + fi)]2 + [2/x + sin 2a
- sin 2(a + /*)]2}/{4[cos a - cos(a + //)]}
(2.2.15)
and taking into account the transformer tap position, the current on the
primary side becomes
(2.2.16)

With reference to Figure 2.4a and b, a commutation from valve 1 to valve 3
(at P3) is only possible as long as phase Y is positive with respect to phase
R. Furthermore, the commutation must not only be completed before C6,
but some commutation margin angle yx (>y0) must be left for valve 1, which
has just stopped conducting, to re-establish its blocking ability. This puts a
limit on the maximum angle of firing a = n (/i + y0) for successful
inverter operation. If this limit were exceeded, valve 1 would pick up the
current again, causing a commutation failure.
cof
Figure 2.4
Six-pulse inverter operation

a Positive and negative direct voltage with respect to the transformer
neutral
b AC line current of phase R
15
Moreover, there is a fundamental difference between rectifier and inverter operations which prevents an optimal firing condition in the latter case.
Although the rectifier delay angle, a, can be chosen accurately to satisfy a
particular control constraint, the same is not possible with respect to angle
y because of the uncertainty of the overlap angle fi. Events taking place after
the instant of firing are beyond predictability and, therefore, the minimum
commutation margin angle, y0, must be sufficient to cope with reasonable
uncertainties (values between 15 and 20 are typically used).
The analysis of inverter operation is not different from that of rectification. However, for convenience, the inverter equations are often expressed
in terms of the commutation margin angle y(y = /} n, where /? = n a).
Thus, omitting the negative sign of the inverter DC voltage, the following
expressions apply
(2.2.17)
The expression for the direct current is
l
aV
cos
2.2.4 Power factor and reactive power

Owing to the firing delay and commutation angles, the converter current in
each phase always lags its voltage (refer to Figure 2.2c). The rectifier,
therefore, absorbs lagging current (consumes VARs).
In the presence of perfect filters, no distorting current flows beyond the
filtering point and the power factor can be approximated by the displacement factor (cos (/)) where (j) is the phase difference between the fundamental-frequency voltage and current components.
Under these idealised conditions, with losses neglected, the active fundamental AC power (P) is the same as the DC power, i.e.
P = JSaVUimI
cos <f> = V,I,
(2.2.19)
and
cos <t> = VdId/(V3aVtermI)
(2.2.20)
Substituting Yd and ld from eqns. 2.2.7 and 2.2.10 into eqn. 2.2.20, the
following approximate expression results
cos (j) = (|)[cos oc + cos (a + /i)]
(2.2.21)
16
The reactive power is often expressed in terms of the active power, i.e.
Q = P.tan<
(2.2.22)
where tan <f> (derived from eqns. 2.2.15 and 2.2.20) is

sin (2a + 2/0 - sin 2a - 2/i
cos 2a - cos (2a +
t a n <f> =
(2.2.23)
Similarly to eqn. 2.2.21, the following approximate expression can be

written for the power factor of the inverter
cos (f) |[cos y + cos /?]
(2.2.24)
Referring to the AC voltage and valve-current waveforms in Figure 2.4, it is

clear that the current supplied by the inverter to the AC system lags the
positive half of the corresponding phase-voltage waveform by more than
90, or leads the negative half of the same voltage by less than 90. It can
either be said that the inverter absorbs lagging current or that it provides
leading current, both concepts indicating that the inverter, like the rectifier,
acts as a sink of reactive power. This point is made clearer in the vector
diagram of Figure 2.5.
Eqns. 2.2.19, 2.2.21 and 2.2.22 show that the active and reactive powers
of a controlled rectifier vary with the cosine and sine of the control angle,
Figure 2.5
Vector diagrams of current and power, suffix Hi for rectification and I

for inversion
17
respectively. Thus, when operating on constant current, the reactive-power

demand at low powers (< 90) can be very high.
However, such an operating condition is prevented in HVDC converters
by the addition of on-load transformer tap changers, which try to reduce the
steady-state control angle (or the commutation margin angle) to the minimum specified.
2.3 Power frequencyunbalanced operation

23 A Terminology and waveforms
The converter model for unbalanced analysis is considerably more complex
than that developed for balanced operation. The additional complexity
arises from the need to include the effect of the three-phase convertertransformer connection and of the converter firing-control strategies. Under balanced conditions, the converter transformer modifies the source
voltages applied to the converter and also affects the phase distribution of
current and power. Each bridge operates with a different degree of unbalance, due to the influence of the converter-transformer connections and
must be modelled independently.
The converter, whether rectifying or inverting, is represented by the
circuit of Figure 2.6.
As for symmetrical operation, in fundamental-frequency studies the
converter is assumed to be connecting a system with perfect filtering on the
AC side and perfect current smoothing on the DC side. By using one of the
converter angles (e.g. 0\ttm in Figure 2.6) as a reference, the mathematical
coupling between the AC system and converter equations is weakened and
the rate of convergence of the power-flow solution improved. Computational simplicity is achieved by using common power and voltage bases on
Jd
2 /2
Primary
Figure 2.6
Basic converter unit
Secondary
18

Phase 1
////#////
Phase 3
Phase 2
Figure 2.7 Star-star transformer connection
both sides of the converter. The phase-neutral voltage is used as the base
parameter and, therefore
MVAbase = base power per phase
Vbase = phase-neutral voltage base
The current base on the AC and DC sides is also equal. Therefore, the p.u.
system does not change the form of any of the converter equations.
The phase-to-phase source voltages referred to the transformer secondary
are found by a consideration of the transformer connection and off-nominal
turns ratio. For example, consider the star-star transformer of Figure 2.7.
The phase-to-phase source voltages referred to the secondary are
--i-V 1
"~
^ ter
term /^term
tet
(2.3.1)
(2.3.2)
(2.3.3)
which in terms of real and imaginary parts yield six equations.
23.2 Variables and equations7

With reference to Figures 2.6 and 2.8 and eqns. 2.3.1 to 2.3.3, the converter
model uses 26 variables, i.e.
Ev E 2 , E 3 , (j)v (j>2, </>3, Iv I 2 , I 3 , (ov a)2, co3
U 1 2 , U 1 3 , U 2 3 , Cl9 C 2 , C 3 , <xl9 a 2 , a 3 , av a2, a3
19
Phase 1
Phase 3 / ^ \
Phase 2
(a)
(b)
wmm
I n
(c)
jj
j i l l j i ill
jj!
iji
I
C3+OL
Figure 2.8
Unbalanced converter voltage and current waveform

a Phase voltages
b DC-voltage waveform
. .
,
c Assumed current waveshape for Phase I (actual waveform is indicated
by dotted line)
Since the terminal voltage is assumed to be undistorted apart from negative

and zero sequence at fundamental frequency, every third delay angle and
commutation duration is the same, i.e. a4 = alf ju4 = fit etc. and consequently only three commutation durations and three delay angles need be
considered.
Therefore, 26 equations have to be found involving these variables.
Starting from the DC side, the average DC voltage, found by integration
of the waveforms in Figure 2.86, may be expressed in the form
+ OLX - C 3 + n) - cos(C2 + a2 - C 3 4- n)]
+ U 13 [cos(C2 + a2 - CO - cos(C3 + a3 - Cx)]
+ U 2 3 [cos(C3 4- a3 - C2) - c o s ^ + ^ + n - C2)]
-I,(Xcl+Xc2+Xc3)}
where X d is the commutation reactance for phase i.
(2.3.4)
20 AC-DC power system analysis

Another equation is derived from the DC-system topology relating the DC
voltages and currents, i.e.
/(V,, I,) = 0
For example, the system shown in Figure 1.3 provides the following
equation
V + Vd2 + V, 3 + Va4 - I d l . Rd = 0
(2.3.5)
The three-phase converter transformer is represented by its nodal admittance model, i.e.
* node
Ypp
YpS
YSP
Yss
where p and 5 indicate the primary and secondary sides of the transformer
on the assumption of a lossless transformer (i.e. Ypp =jbpp etc.).
Turning to the AC side, on the assumption of a lossless transformer, the
currents at the converter-side busbar are expressed as follows
I,**8' = - Z [jK^kem + / X ^ " ~ e '" J ]
(2.3.6)
By subtracting the 6lerm in the above equation, the terminal busbar angles
are related to the converter-angle reference.
Separating this equation into real and imaginary components, the following six equations result
3
11 cos cox = [blsEk sin0 fc + blpVkterm$in(6*erm
- C,)]
(2.3.7a)
k= l
12 cos co2 = f; [b2jEk s i n ^ + b^Yktermsm(dkterm - 0 ^ ) ]

1 3 cos o)3 = I
[b'sEu sin $k + bfp V ^ s i n (dkterm - Cm)]

3
lt sin CO! = - ^ [6"E k cos^ k + ^Vjermcos(0ferm - 0 ^ ) ]
(2.3.76)
(2.3.7c)
(2.S.7d)
k=l
12 sin o>2 = - [Êj cos 0, + ^Vfermcos (0ferm - e,1^,)]
(2.3.7)
13 Sin W3 = - [fc*E4 COS </>, + ^VfermCOS ( < L ~ Cm)]
(2.3.7/)
fc=l
21
Three equations are derived from approximate expressions for the fundamental r.m.s. components of the line-current waveforms as shown in Figure
2.8c, i.e.
(2.3.8a)
li|sin(T 2 /2)
(2.3.86)
l3 = - - - k s i n ( T 3 / 2 )
(2.3.8c)
where T is the assumed conducting period of each phase.

Six further equations can be derived from the transformer connection,
interrelating the secondary (converter-side) currents and the primary and
secondary phase voltages, i.e. from the components of the transformer's
nodal admittance matrix equation.
As an example, the nodal admittance matrix (in per unit) relating these
variables for the star-delta connected transformer (assuming the nominal
tap position a = 1) is
y/y/i
II
y>/s
P
1
-ylj*
y/y/3
~P
-y/y/3
-v
1
-v
2
~P
1
2
3^
1
V
v
terir
V2
V3
v
terir
(2.3.9)
22
The sum of the real powers on the three phases of the transformer
secondary may be equated to the total DC power, i.e.
Vd.Id = X %ih cos(4>t ~ ,)
(2.3.10)
To obtain a reference which may be applied to all transformer secondary

windings, an artificial reference node is created corresponding to the
position of the zero-sequence secondary voltage. This choice of reference
results in the following two equations
E;cos<^ = 0
(2.3.11)
Efsin<^ = 0
(2.3.12)
The nodal admittance matrix for the star-connected transformer secondary

is now formed for an unearthed star winding. The restriction on the
zero-sequence current flowing on the secondary is, therefore, implicitly
included in the transformer model for both star and delta connections.
A total of 20 equations have been selected so far. The remaining six are
obtained from the control strategies. For three-phase inverter operation it
is necessary to retain the variable cc in the formulation, as it is required in
the specification of the symmetrical firing controller. Therefore, the restriction upon the extinction advance angle, y, requires the implicit calculation
of the commutation angle for each phase.
Using the specification for y, as defined in Figure 2.8, the following
expressions apply
c o s / / + cosa, - ld
(Xcl
c o s / / + cosa 2 - ld
(Xc
+ Xc3)
=0
(2.3.13a)
ltXc3) = 0
(2.3.13*)
V2U13
V ^
^23
cos yf + cosa 3 - Id ( X c l"t T X c l ) = 0

v 2 u2i
(2.3.13c)
Two further equations result from assuming that the off-nominal tap
position is the same in the three phases, i.e.
a1=a2
(2.3.14a)
a2=a3
(2.3.14*)
23
and a final equation relates to the constant-current or constant-power

controller, i.e.
Id = r /
(2.3.15a)
VA, = 1 7
(2.3.156)
or
Summarising, the formulation of the three-phase converter model involves

the following 26 equations
(2.3.4) 1
(2.3.5) 1
(2.3.7) 6
(2.3.8) 3
(2.3.9) 6
(2.3.10) 1
(2.3.11)1
(2.3.12)1
(2.3.13) 3
(2.3.14) 2
(2.3.15)1
2.4 Characteristic harmonics

Under the assumptions made in section 2.2 for power-frequency symmetrical operation, the AC-DC converter only produces the so-called characteristic harmonics. Neglecting the commutation overlap, the Fourier
components of the AC current in phase a of a star-star transformer
connected bridge are
I = y- \d I COS (Ot - COS 5(Ot + 7T COS 7(Dt COS 1 l(JOt
11
)
(2.4.1)
+ cos 13co^ - cos 17co^ + cos \9cot
From eqn. 2.4.1 the r.m.s. magnitude of the fundamental frequency is
V^ I ( ) = v^ I(j
(2A2)
Eqn. 2.4.1 also shows the existence of characteristic harmonics of orders

k = 6w 1 (for integer values of n) and of magnitudes lk = I,/A. It also
shows a total absence of triplen harmonics.
24
Although not evident from the single-phase equation, the harmonics of

orders 6w + 1 are of positive sequence and those of orders 6^1 are of
negative sequence.
To maintain the same primary and secondary voltages as for the star-star
connection, a factor of y^3 is introduced in the transformer ratio of the delta
connection and the resulting Fourier series for the current in phase a on
the primary side is
ia =
2v/ 3
/
1
1
1
ld I cos cot + - cos 5cot - cos 7 cot cos llcot
n
\
5
7
11
+ cos \3cot + cos I7cot
cos \9cot
(2.4.3)
This series only differs from that of a star-star connected transformer by

the sign of harmonic orders 6n 1 for odd values of n, i.e. the fifth,
seventh, 17th, 19th etc. Therefore, the AC-current components of the
standard HVDC converter are
(i 12 ) = 2 "
TC
~ ld( cos0) cos 1 Icot + cos 13co

\
11
1 3
- cos 23cot + cos 25cot
(2.4.4)
This series only contains harmonics of order \2n 1. The harmonic currents
of orders &n + 1 (with n odd), i.e. n = 5, 7, 17, 19 etc., circulate between the
two converter transformers but do not penetrate the AC network.
The effect of commutation overlap, represented by eqns. 2.2.11 and
2.2.12, is to reduce the magnitude of the harmonic currents but does not
alter the half-wave symmetry of the waveform and, therefore, the harmonic
orders remain the same.1 Figure 2.9 illustrates the variation in 11th
harmonic with delay and overlap angles.
The DC voltage contains the following three different functions with
reference to voltage crossing Ct in Figure 2.2
vd = y/2aVXmQos\cot
+I
vd = >/2Vtermcos cot + I
= ^ aV term cos [cot]
vd = y/t aVterm
cos II mt
mt -- | |
term cos
for
0 < cot < oc
(2.4.5)
+ -J2aVtermsm[cot]
for
a < cot < oc + fi

for
a + ft < cot < |
(2.4.6)
(2.4.7)
25
10
p.
>
1'
E
c
3
6\
54-
a 40^vv^X \
or - SO^X
a -
o-
10
20
30
40
Angle of overlap |i degrees)
Figure 2.9
Variation in 11th harmonic current with delay and overlap angles
where aVterm is the (commutating) phase-to-phase r.m.s. voltage, and a and

fi the firing and commutation angles, respectively.
From eqns. 2.4.5, 2.4.6 and 2.4.7, the following expression is obtained for
the r.m.s. magnitudes of the harmonic voltages of the DC voltage waveform
v2c0
J2(k - 1
(k + I) 2
- I) 2
-2(k - 1)(^ + 1) cos I (k + 1)^ I cosUk -
l)^l cos(2a +
(2.4.8)
Figure 2.10 illustrates the use of eqn. 2.4.8 to derive the variation of the
12th harmonic as a percentage of Vc0, the maximum average rectified
voltage, which for the six-pulse bridge converter is 3y/2aVterm/n. For a = 0
and fi = 0, eqn. 2.4.8 shows that
ir
V c0
/o
/Q
(2.4.9)
k2 - 1 ~ k2
whereas for a = n/2 and fi = 0, that ratio becomes

(2.4.10)
26
30
Angle of overlap JJ.
Figure 2 JO
Variation of the 12th harmonic voltage as a percentage ofVco
This means that the higher harmonics increase faster with a. Eqn. 2.4.10
represents the maximum proportion of harmonics in the system, particularly when it is considered that at a = 90, /i is likely to be very small.
2.5 The converter as a frequency modulator

The harmonic transfers through an HVDC converter are best explained
using modulation theory.8'9
The function of the HVDC converter as a modulator is twofold. First, it
takes a three-phase positive-sequence AC voltage waveform and, by switching consecutively through the phases, ensures that a DC voltage is always
applied on the DC side of the converter. In a twelve-pulse converter, every
30 degrees there is a thyristor switching which connects a combination of
phases that maintains a constant average voltage on the DC side. In this way,
the fundamental-frequency waveform is modulated down to DC.
Secondly, the same switching pattern takes the DC current on the
DC side, and switches it onto the AC-side phases in such a way that
a fundamental-frequency positive-sequence AC current exists on the AC
side.
The switching pattern is synchronised with the AC-side fundamental
frequency and, as such, contains a large fundamental component. As
27
switching is an on-off process, harmonics of the fundamental are present

as well. As shown in section 2.4, these manifest themselves as the characteristic harmonics on both sides of the converter, i.e. on the DC side,
harmonics 12n, and on the AC side, harmonics 12w + 1 in positive sequence, and \2n 1 in negative sequence, where n is an integer. These
components are always present, even under ideal (undistorted AC voltage
and DC current) operating conditions.
Any noncharacteristic frequencies on the AC or DC sides are subjected to
exactly the same modulation process by the converter.
The analysis of section 2.4 assumed a steady firing angle, a fixed
commutation duration and undistorted AC-system voltage conditions. In
practice this is rarely the case. Harmonic voltages and/or unbalance will
exist on the AC side, and current ripple will exist on the DC side. Through
constant-current control, the firing angle will not be steady and the commutation-period duration will also be varying. Therefore, not only will harmonic voltages and currents be transferred through the converter but these
may be amplified through the variation of thyristor switching instants.
These interactions have particular relevance for noncharacteristic harmonics, i.e. of different order to those discussed in the previous section.
Noncharacteristic harmonics arise from a number of causes, typically a
system imbalance, presence of fundamental-frequency current on the DC
link and AC system nonlinearities. Harmonically unrelated frequencies exist
whenever the fundamental supply frequencies of the interconnected AC
systems are not identical. This effect is more critical in the case of
back-to-back interconnectors because of the close coupling which exists
between the two AC systems.
The situation is further complicated by the fact that the switching pattern
of the HVDC converter bridges is affected by distortion in the AC voltage
and DC current waveforms. Usually, DC current directly affects the firing
instants of the thyristors through the converter control, and the commutation period is directly affected by both AC voltage and DC current.
2.5.1 The modulation process

A direct method of analysis in the frequency domain is the transfer function,
a concept which provides a general linearised solution of the converter for
small levels of distortion.
The voltage and current relationships across the converter can be expressed as follows
V = Y #AC .i DC
(2.5.2)
where Y , ^ and Y^c are transfer functions for the voltages and current,
respectively, and \// = 0, 120, 240 for each of the three phases.
28

'p DC
Figure 2.11 Transfer functions YILDcand Y*AC

a Transfer function to DC voltage
b Transfer function to AC current
In the absence of commutation overlap (Figure 2.1 1 in dotted lines), the

transfer functions for the voltage and current modulation are rectangular
waveforms and can be expressed aslo
m
Y,, =
1 A, c o s n q t
n=l
co
+ $)
(2.5.3)
An = 41,, '1" . sin ""Iz. cos "=I6
(2.5.4)
00
Y,, =
A, cos(w,t
n= 1
where
This process is illustrated in Figure 2.12, which shows the modulated output
current on the AC side of the converter in response to a DC current which
contains a ripple frequency.
29
I,, /phase
*t
u+
Figure 2.12 Idealised switching jknction and modulating jknction giving

modulated AC output
OIEEE 1992 Reproduced by permission
When the commutation angle is taken into account, the voltage transfer
functions (Figure 2.1 la-continuous line) are shown as in eqn. 2.5.3 but
with the value of A, replaced by
A,,,, = A,. cos nw12
(2.5.5)
where p is the commutation angle.

Regarding the current-transfer fimction, the assumption of a linearly
changing commutation current during the commutation (Figure 2.1 l b continuous line) leads to the following alternative expression for the
coefficient A,
Ani= A,.
sin ( 4 2 )
nd2
(2.5.6)
For an accurate quantitative assessment, the analysis must include the effects
of pulse-position and pulse-duration modulation as affected by the modulating frequencies as well as explicit representation of the converter controller.'
2.6 Harmonic transfer generalisation"

The main transfer relationships in the modulation process have been
collected together in Figure 2.13 for the case of a 12-pulse HVDC link
interconnecting two AC systems of frequencies fi and f2, respectively.
30

AC2-
AC?
DC
AC,*
AC!
(12^,U
ir,t
{ 1 2 n * 1 ) f , / r , -1
(12m+1)W^2
12n
(12m-1)(12/>'*i)/1 /i
-!
(I2m+1)^
(12m+1)(12ny
(12m-1)(12nU)
Figure 2.13 Harmonic transfers across a 12-pulse HVDC link. The encircled
elements indicate harmonic sources and m,ne(l,
2,3...)
The exciting harmonic sources, surrounded by a circle, are multiples,

integers or nonintegers, of the frequency in system 1.
(k])i$ a current harmonic source, whereas(i~l)and(^ +"T)are voltage
harmonic sources.
The resulting harmonic orders in system 2 are related to the frequency of
system 2.
The DC column refers to the DC side of the link, the ACj" and AC^
columns represent the positive and negative sequences of system 1 and AC2
and AC2 represent the positive and negative sequences of system 2,
respectively.
The justification for the transfer frequencies of Figure 2.13 is considered
next.
2.6.1 From the AC to the DC sides

The largest component of the thyristor switching function is at the fundamental frequency. This component modulates positive-sequence voltage at
harmonic multiple k + 1, and negative-sequence voltage at harmonic
multiple k 1, to multiple k on the DC side. The voltages are switched
across in the same ratio as the fundamental is, except that the firing-delay
angle does not reduce the magnitude but merely phase shifts the voltage.
Thus, if the maximum DC voltage (zero firing-delay angle and no voltage
drop due to commutation) is 550 kV and the AC voltage 345 kV, with a
commutation period ju, then harmonic voltages will be increased by
550
/345Cos(M/2) across the converter. To a good approximation, the percentage voltage distortion on the AC side is preserved in its transfer to the DC
side. This mechanism is not unique to HVDC converters: in fact any
three-phase rectifier or inverter demonstrates similar characteristics.
A secondary mechanism comes from the AC-voltage distortion affecting
the commutation period, which modifies the harmonic voltage level on the
DC side by up to ten per cent, and the phase angle of that distortion by up
31
to 0.3 radians. This effect is frequency dependent, as well as being closely

related to variables such as firing-delay angle and commutation-period
duration.
Further noncharacteristic harmonic voltages appear on the DC side,
associated with the other components of the thyristor switching pattern. A
positive-sequence voltage at harmonic multiple H I on the AC side will
appear as a voltage at multiples k and \2n k on the DC side; k may or may
not be an integer. The same frequencies will appear for an AC-side
negative-sequence voltage at harmonic multiple k 1. These higher order
noncharacteristic harmonics would be expected to be at a level reduced by
a factor of approximately 1/(12n) from the terms described in the first
paragraph of this section, if the switching instants remained unaffected by
the distortion.
However, the same secondary mechanism applies to these harmonics as
well. Unfortunately, the spectrum that appears as a result of commutationperiod variation does not reduce nearly as quickly with increasing order.
Thus, for the terms 12 k, commutation-period variation is as important a
mechanism as the direct transfer, and for higher orders it is substantially
more important. This variation is more difficult to describe and generalise.
However, these harmonics can be expected to be present at levels of up to
20% of the terms described in the first paragraph of this section.
In summary, the frequencies on the DC side from a positive-sequence
harmonic multiple k + 1, or a negative-sequence harmonic multiple k 1,
can be written as
fDC = (l2nk)f0
forne(0, 1,2,3,...)
(2.6.1)
2.6.2 From the DC to the AC sides

The fundamental component of the switching function modulates current at
harmonic multiple k to multiple k + 1 in positive sequence and k 1 in
negative sequence on the AC side. Considering just DC on the DC side, it
becomes multiple -I-1 in positive sequence, and multiple 1 in negative
sequence on the AC side; negative frequency in negative sequence is
equivalent to positive frequency in positive sequence, and the two components add together to give the total AC current. Harmonic currents are,
therefore, switched across at half the ratio as the direct current, i.e. if the
DC current is 2 kA and the fundamental frequency AC current 2.6 kA, then
harmonic currents will be changed by (1/2) (2.6/2) across the converter.
A secondary mechanism comes from the DC-current distortion affecting
the commutation period, which modifies the harmonic level on the AC side
by up to 15%, and the phase angle of that distortion by up to 0.2 radians.
This effect is frequency dependent, and is closely related to variables such
as firing-delay angle and commutation-period duration.
32
Further noncharacteristic harmonics appear, associated with the other

components in the thyristor switching pattern. For the DC-side current
harmonic multiple k , harmonics (12%+ 1) f k appear on the AC side in
positive sequence, and (12n - 1) f k appear on the AC side in negative
sequence. Once again, these higher order noncharacteristic harmonics
could be expected to be at a level reduced by a factor of approximately
1/(12n f l), as appropriate, from the terms described in the first paragraph
of this section.
The same secondary mechanism also applies to these harmonics, and the
spectrum which appears as a result of commutation-period variation reduces
more quickly than that for DC voltage, but not as quickly as do the
characteristic harmonics. Thus, for the terms where n = 1, commutationperiod variation is as important a mechanism as the direct transfer, and for
higher orders it is more important. These harmonics can be expected to be
present at levels of up to 20% of the first-order terms described in the first
paragraph of this section.
In summary, the AC-side frequencies resulting from a DC-side harmonic
multiple k , can be written
fAc
= ((12m
f 1) f k ) f ,
for mG(0, 1, 2, 3 , . . .)
(2.6.2)
in phase sequences as described.
2.6.3 Effect of switching-instant variation

The effect of firing-angle or commutation-period variation consequential to
AC-voltage or DC-current distortion has been described above in a limited
way; the spectrum which appears from this sort of variation encompasses all
the frequencies described above. Thus, if there is a negative-sequence
fundamental-frequencyimbalance, even if no second-harmonic current can
flow on the DC side, there will still be third-harmonic positive-sequence
current on the AC side as a direct result of the variation of the commutation
period, and the full DC-side DC current. Further to this, additional
frequencies appear but in rapidly diminishing magnitudes. For a harmonic
multiple k on the DC side, if m is an integer, additional DC-side harmonic
multiples of 12n f mk can be expected, and on the AC side, multiples
(12%f 1) f mk will appear. In most cases, these will be at very low levels.
Figure 2.14 shows the DC-side voltages and AC-side currents which could
be expected for a firing-angle-order sinusoidal variation of k 3 degrees, at
a frequency of four and a half times the fundamental. The proliferation of
frequencies, at low and high orders, should be noted. This shows how the
spectrum generated by switching-instant variation alone diminishes only
slowly with increasing order.
15
Figure 2.14
20
25
30
35
40
33
45
50
a DC voltage harmonic spectrum, modulation frequency = 4.5 c%

b AC current harmonic spectrum, modulation frequency = 4.5 co0
2.6.4 Transfer across the DC link

When the DC link interconnects separate power systems, either of the
same nominal (but in practice slightly different) frequency or having
different nominal frequencies, there will be a wide range of harmonic and
34
nonharmonic frequency transfers. These can be divided into two groups:

1 Frequencies at terminal 1 caused by the characteristic DC-voltage harmonics (I2nf2) and their consequential currents from terminal 2. These are
represented in the expression
/ AC1 = (12m l)fx Unf2
(2.6.3)
where w, we(0, 1, 2, 3...) which can have any frequency including

frequencies below the fundamental.
The frequency-conversion schemes, and particularly the back-to-back
schemes such as Sakuma, represent the worst condition for noninteger
harmonic frequencies. In this case, with small smoothing reactors, the
DC-side coupling is likely to be strong which means that the flow of
harmonically unrelated currents on the DC side can be large. In six-pulse
operation such schemes can produce considerable subharmonic content
even under perfect AC system conditions. However, 12-pulse converters do
not produce subharmonic content under symmetrical and undistorted
AC-system conditions. These will produce interharmonic currents as defined by eqn. 2.6.3, which for m =n = 1, f2 = 60 and/! = 50 results in a
70 Hz AC current and, through the AC-system impedance, voltage. The
latter will beat with the main frequency / x producing some 20 Hz flicker.
However, the levels of 70 Hz expected from this second-order effect will
normally be too small to be of consequence.
When the link interconnects two isolated AC systems of the same nominal
frequency such as in the New Zealand scheme, and the two ends of the DC
link are different in frequency by a smaller frequency Af0, then the
characteristic harmonics are different by \2nAf0. If a DC-side voltage at
frequency 12n(/0 + A/o) is generated by one converter, then at the other
converter this will be modulated down again by a characteristic frequency in
the thyristor switching pattern as per eqn. 2.6.3, i.e.
( I 2 m l ) / O 1 2 n ( / O + A/O)
which on the AC side, among other frequencies, includes (for m =n)
f0 12nA/0
The latter will beat with the fundamental component at a frequency 12nA/o,
which at some values of n will allow flicker-inducing currents to flow.
2 Frequencies caused in system 1 by unbalance or distortion in the supply
voltage of system 2.
35
Negative-sequence voltages at frequencies (k l)/ 2 produce the following

noncharacteristic frequencies on the DC side
/Dc = (12n*)/ 2
(n = 0,1,2,...)
(2.6.4)
Crossmodulation of these current components produces the following

frequencies in system 1
/ACI = (12m 1)/! (12n k)f2
(2.6.5)
Let us first consider a frequency-conversion scheme with a sinusoidal but

negative-sequence unbalanced voltage in system 2, i.e. (k 1) = 1 (and
therefore k = 2). Substituting m = n = 0 and k = 2 into eqn. 2.6.3 yields
currents (and therefore voltages) at frequencies
/ACI = / i 2/2
(2.6.6)
One of these frequencies (ft 2/2) will beat with the fundamental-frequency voltage of system 1 at a frequency
/2)
(2.6.7)
which for a 50/60 Hz conversion scheme becomes 20 Hz. This is a flickerproducing frequency. This same frequency will be referred to generator
rotor-shaft torque at 20 Hz, which may excite mechanical resonances.
Again, this type of crossmodulation effect is most likely to happen in
back-to-back schemes due to the stronger coupling between the two converters, although it is also possible with any HVDC scheme in the presence of
a suitable resonance. However, for the McNeill back-to-back scheme in
Alberta, which has no DC smoothing reactor, the calculated and measured
effects were small and, therefore, buried in the continuous random variations found in the power system.
Now consider two AC systems of the same nominal (but slightly different)
frequency.
Substituting m n = 0, and k = 2, for fundamental frequency f0 into eqn.
2.6.5, a current and resultant voltage (through the AC-system impedance)
of frequency
/ A C I = / O 2 ( / O + A/ O )
(2.6.8)
which leads to/ 0 2A/0 is induced on the AC side. This will either beat with
the fundamental frequency f0 at frequency 2A/0 or produce generator/motor
shaft torques at 2Af0. This frequency is generally too low to produce flicker
but may induce mechanical oscillations.
36
Substituting m = n = 1 and k = 2 in eqn. 2.6.5 gives, among others, a

current (and thus voltage) at the frequency
(12 + l)/ 0 - (12 + 2)(/o - A/)
and for / 0 = 50 Hz and A/= 1 Hz the resulting AC current (and thus
voltage) in system 1 is
13 x 5 0 - 14 x 49 = 36 Hz
This distorting voltage will, therefore, beat with the fundamental producing
14 Hz flicker. However, the subharmonic levels expected from this secondorder effect will normally be too small to be of consequence.
2.7 Harmonic instabilities

AC-DC systems with low short-circuit ratios (SCR) often experience problems of instability in the form of waveform distortion. The low SCR indicates
a high AC system impedance, with an inductance which may resonate with
the reactive compensation capacitors and the harmonic filters installed at
converter terminals. These resonant frequencies can be low, possibly as low
as the second harmonic. The resonances can be excited under certain
operating conditions or in the event of fault, and the small initial distortion
may develop to an instability. Instability related to the interaction of
harmonics (or any frequencies) has been customarily referred to as harmonic instability.
The most common type of instability involves the converter transformer
under core saturation. It is characterised by extremely slow growth and,
owing to the difficulty of its prediction, it is usually discovered during or
after project commissioning.
The mechanism of the phenomenon can easily be explained using the
block diagram of Figure 2.15.3 If a small-level positive-sequence secondharmonic voltage distortion exists on the AC side of the converter, a
fundamental-frequency distortion will appear on the DC side. Through the
DC-side impedance, a fundamental-frequency current will flow resulting in
a positive-sequence second-harmonic current and a direct current flowing
on the AC side. The direct current flowing on the AC side will begin to
saturate the converter transformer, resulting in a multitude of harmonic
currents being generated, including the positive-sequence second-harmonic
current. Associated with this current will be an additional contribution to the
positive-sequence second-harmonic voltage distortion, and in this way the
feedback loop is completed. The stability of the system is determined by the
characteristics of this feedback loop.

positive-sequence
second-harmonic
voltage distortion VaCp
AC-side
second-harmonic
impedance Zacp
convertor
switching
action
positive-sequence
second-harmonic current
distortion I
convertor
switching
transformer
action
core
saturation negative-sequence
dc distortion lacn
AC side
37
fundamentalfrequency voltage
distortion
DC-side
fundamentalfrequency
impedance Zdch
fundamentalfrequency current
distortion l d c h
DC side
Figure 2.15 Mechanism of core saturation instability
The small-signal linearisation method described in section 2.5.1 can be

used to analyse the instability mechanism illustrated in Figure 2.15. For this
purpose, the AC-DC system is simplified to the equivalent circuit of Figure
2.16 consisting of a converter block interconnecting the AC and DC-side
impedances at the relevant frequencies. On the AC side, the positivesequence second-harmonic current, IACp, generated by the converter, flows
into the AC-system second-harmonic impedance, ZACp, and the negative
sequence DC, IACll, flows into a parallel circuit of transformer magnetising
positive-sequence
second harmonic
fundamental
frequency
acp
negative-sequence
DC harmonic
Va
Racn
AC side
converter
DC side
Figure 2.16 Equivalent circuit for study of transformer core saturation instability
38
inductance, Lm, and AC-side impedance at the low frequency IACn variation.
Assuming that the AC-side impedance remains fairly constant around 0 Hz,
the impedance can be simplified to the AC-system resistance at 0 Hz of
RAOl. On the DC side, the presence of fundamental-frequency voltage
distortion causes an equivalent current distortion to flow through the
DC-side impedance.
The converter block accounts for the transfer of AC-side voltage distortion to the DC side and the DC-side current distortion to the AC side, as
described in section 2.6. Considering only the most significant low-order
harmonics, these interactions can be represented by eqn. 2.7.1. The elements within the matrix describe the amplitude change and phase shift
introduced during the transfer of distortions from either side of a converter.
The response of the converter controller and the signal transducers is
included in the matrix
^DCh
ÂCp
= d
h i
JÂCn __
vAC;
b c
*ACn
(2.7.1)
'DCh
Implicit in this expression is that the VACp and IACp are the higher-frequency
positive-sequence components on the AC side at one harmonic higher than
the DC-side components of VDOl and IDC/I, which are in turn one harmonic
higher than the negative-sequence components of VACn and IACn on the AC
side. In the analysis of core saturation instability, VACp and IACp are at the
second harmonic, Vj^h a n d IDC/I a t the fundamental frequency and VACw and
IACll at DC. The elements within the matrix are calculated from average
values of the converter operation in the steady state. The responses of the
converter controller and the signal transducers are incorporated. The
response of a constant-current controller is included in elements c, f and i
which accounts for the transfer to the three left-hand-side harmonics
through the controller loop.
In addition to the linearised converter transfer function, eqn. 2.7.1
describing the converter operation, a set of simultaneous complex equations
is written to describe this equivalent circuit
YACp I A cpZ ACp + 1 2 + Z AC|J
(2.7.2b)
I,+ = - X L ,
X
(2.7.2fl)
ACn
ACn
dt
(2.7.2c)
(2.7.2d)
(2.7.2e)
39
Expressing IAClI in exponential vector form

*ACn ~* 1 A C B e
V* ' ' 3 a /
RAC
1 4- AZACp + XCZ AQ ,
Lm (1 +R A C w D)(l + AZ A C p )~R A O l BCZ A C p
**
A= d+ ^
(2.7.3c)
= *+ ^
(2.7.M)
^
D= A + ^
(2.7.3.)
(2.7.3/)
The term a in eqn. 2.7.3a is defined as the saturation stability factor (SSF)
which basically defines the susceptibility of the system to the development
of this type of instability. A positive SSF value indicates that the harmonic
sequence will decay over time and, hence, the system is stable. On the other
hand, a negative SSF suggests negative damping and the development of
instability as the distorting harmonic sequence increases over time. The
value and sign of the p term determine the speed and the direction of the
variation of these harmonic sequences.
This saturation stability-factor technique has been verified against dynamic simulation,10 and is included as an example of electromagnetictransient simulation of HVDC schemes in Chapter 6, section 6.8. In that
section, electromagnetic simulation is used to verify several solutions to a
core saturation instability, derived by consideration of the SSF.
From Figure 2.15, the contribution from the transformer saturation to the
instability feedback loop comes in the form of an additional positivesequence second-harmonic current resulting from the saturation. This effect
is modelled as a positive-sequence second-harmonic current injection I 2 +
into the second-harmonic part of the equivalent circuit. The depth of the
saturation is calculated according to the amount of negative-sequence DC
from the converter that is flowing into the transformer magnetising inductance, hm (i.e. I o _).
The transformer saturation related to this instability can be regarded as
asymmetrical saturation since the transformer is only saturated in one half
of a fundamental cycle. As the transformer is coming in and out of
saturation, the magnetising inductance, LM, is nonlinear, but it is only in
saturation for a short period of each cycle as indicated by the width of the
distorted magnetising current pulse. Therefore, it is reasonable to assume
40
Lm as the unsaturated magnetising reactance, as indeed it has this value

most of the time.
When the magnetisation characteristic of Imag/flux approaches infinity in
the saturated region there is a one-to-one linear relationship between the
resulting positive-sequence second-harmonic current I 2 + and the level of
saturating negative-sequence DC I o . In practice, the relationship between
I 2 + and I o _ will realistically be less than one, depending on the magnetisation characteristic of the transformer, i.e.
I2+ = - X . I 0 _
0<X<l
(2.7.4)
For each particular transformer, the values of X must be calculated, either

experimentally or using dynamic simulation.
A detailed analysis using the saturation stability factor, has revealed11 that
a vulnerable HVDC rectifier system is likely to have the following impedance profile
a low and predominantly capacitive DC-side impedance at the fundamental frequency with the presence of a series resonance near to but
higher than the fundamental frequency;
a high and predominantly inductive AC-side second-harmonic impedance with the presence of a parallel resonance near to but higher than
the second harmonic frequency;
a high AC-side resistance near 0 Hz.
On the other hand, a susceptible inverter system will possess opposite

reactive characteristics, with inductive DC-side impedance at fundamental
frequency and capacitive AC-side second-harmonic impedance. A high 0 Hz
resistance is also observed at the unstable inverter station but the two
reactive components have the dominant role in determining the system
stability.
The common use of HVDC back-to-back interties to interconnect large
and weak AC networks has resulted in low-order resonances at the converter
terminals, making them prone to core saturation instability. However, with
comparable network sizes at both the rectifier and inverter ends, this
harmonic instability is most likely to occur only at one end of the scheme.
This is due to the opposite reactive requirements of the impedances for the
instability to occur at either end. Moreover, the high resistance at the
unstable end will be reflected onto the DC side as additional damping which
tends to stabilise the opposite end system. Therefore, for the back-to-back
scheme, it is necessary to consider the consequential impact on the stability
of the remote end system when undertaking any modification at the local
end.
Apart from the system impedances, the stability of the AC-DC system is
strongly dependent on the response of the converter controller. Considering the stringent reactive requirements for the instability to develop, the
onset of this harmonic instability almost certainly involves a destabilising
41
contribution from the converter controller. This suggests the possibility of

preventing the onset of the instability through proper tuning of the
converter controller.
2.8 Generalised harmonic domain converter model12

In section 2.4 the transfer functions by means of modulation theory have
been expressed in terms of switching instants which are themselves
modulated as a result of applied distortions. The modulation of the
switching instants and the transfer-function shapes involves approximations
valid for small levels of distortion, and low-order harmonic, subharmonic
and interharmonic frequencies. These approximations can be removed by
modelling the converter in the time domain, but at the expense of solution
speed, since time-domain simulations must run until transients have
decayed.
This section describes a powerful steady-state alternative. The proposed
formulation convolves periodic sampled quantities in the harmonic domain
with their sampling functions, so that no Fourier transform is required,
resulting in substantial computational savings. The sampling functions are
defined in terms of the exact switching instants, which are obtained as part
of the overall iterative procedure discussed in Chapter 4. As described here,
the model takes one cycle of the AC voltage as the fundamental, and so only
harmonics are analysed. However, an extension to the steady state over
several cycles would allow interharmonics to be solved.
2.8A Analysis of the commutation

The commutation process is modelled here allowing for imbalance and
harmonic distortion in the terminal voltage and DC current. To simplify the
analysis of the interaction between the converter and the AC system, the
commutation resistance is placed between the AC-system terminal and the
converter-transformer primary windings.
The angle reference used in the analysis is arbitrary, and possibly
unrelated to the converter at all. For example, all angles may be referenced
to the most recent positive zero crossing of the fundamental internal e.m.f.
of the slack bus generator. At present, the first equidistant timing reference
has arbitrarily been assigned an angle of zero. All firing angles and
end-of-commutation angles are referenced to this angle, as opposed to the
converter-terminal voltage fundamental-frequency component, as is the
usual practice.
Star-connected bridge
In the commutation circuit shown in Figure 2.17, Vfl, Vb, lc and Id are sums
of harmonic phasors. In this diagram phase a is commutating off, phase b
is commutating on and the commutation ends when Ic = ld. Considering a
Figure 2.17
Circuit for star-glstar commutation analysis
periodic steady state, and summing voltage drops around the commutation
loop at harmonic order k
Vak +jkXaIck -jkXaldk
+jkXblck
- Vbk = 0
(2.8.1)
Solving eqn. 2.8.1 for the commutation current

l k
'
- vabk
jk(xa + xb)
(2 8 2)
- -
The periodic commutation current in the time domain is, therefore
(2.8.3)
where
(2.8.4)
.k = l
D can be considered to be either a constant of integration, an initial

condition, or equivalently a circulating DC current in the commutation
43
loop. Assigning this value to D ensures that at the moment of firing a valve,
6i9 the current in it is zero.
The resulting commutation current is the steady-state solution of the
commutation circuit and, since there is no resistance in the circuit, the
steady state is achieved instantaneously when the appropriate valve is fired.
The commutation ends when the instantaneous commutation current is
equal to the instantaneous DC current. This angle, the end of commutation
0t., cannot be solved directly, as eqn. 2.8.3 is transcendental. Instead, the
end of each commutation is determined by the zero crossing of a differentiate mismatch equation, solvable by Newton's method. The mismatch
equation is easily constructed by substituting cot = ${ into the Fourier series
for the DC and commutation currents, and taking the difference
(I* - IoJ^j
(2-8.5)
Eqn. 2.8.5 completes the commutation analysis for a bridge connected to a

star-connected source via an inductance. This equation is suitable for
modelling the connection to an unbalanced star-g/star-connected transformer, if the leakage reactances and terminal voltages are referred to the
secondary side after scaling by off-nominal tap ratios on the secondary or
primary windings.
Delta-connected bridge
The circuit to be analysed is that of Figure 2.18, which corresponds to a

particular commutation. The objective is to solve for the commutation
current, Ic, in terms of the voltage sources. Proceeding directly with a
phasor analysis in the steady state, a series of loop and nodal equations can
be obtained for this circuit at each harmonic k
*dk ~~" *cfc + hk ~~ hk ~ 0
I** + hk - hk = 0
hk ~~ lck ~~ ^dk ~ 0
=0
vck -jiCK*ck + y* = o
V^-AA-VÔ
(2.8.6)
where for an inductance Xk = kXv This set of equations is readily solved to

yield the DC voltage during the commutation, and the commutation
current itself
44
Figure 2.18 Circuit for star-gldelta commutation analysis
VckKak
-h]IdkXckXak
(2.8.7)
(2.8.8)
A similar analysis holds for every separate commutation, with appropriate

modifications to the phase subscripts, and the direction of the DC current.
The end-of-commutation mismatch equation for the star-g/delta connection
is obtained as for the star-g/star connection above.
2,8.2 Valve-firing control

There are two aspects to modelling the valve-firing process: the firing
controller and the converter controller. In modern schemes, the firing
controller consists of a phase-locked oscillator (PLO) tracking the fundamental component of the terminal voltage, and generating essentially
equispaced timing references. A well designed phase-locked oscillator is
unaffected by harmonics in the terminal voltage, since its time constant is
of the same order as the fundamental. Consequently, the PLO is not
modelled and the timing pulses are assumed perfectly equidistant, spaced
by 60.
In general, the effect of a nonideal PLO would be to introduce a coupling
between terminal-voltage harmonics, and the firing-mismatch equation to
45
be derived below for current control.

Current control
A valve firing occurs when the elapsed angle from a time pulse is equal to
the instantaneous value of the alpha order. The alpha order is a command
variable received from the converter controller. The controller modelled
here is a constant-current control of the proportional integral type (Figure
2.19), which will respond to harmonics in the DC current. From Figure 2.19,
the alpha order can be expressed as a sum of harmonic phasors
(2.8.9)
where
(2.8.10)
1 +jk<oT
With reference to Figure 2.20, it can be seen that firing occurs when the
elapsed angle from the equidistant timing reference is equal to the instantaneous value of the alpha order i.e. a = 0i pt. The equidistant timing
references are represented by /?,. = (i 1)TT/3. The firing-mismatch equation
is, therefore
= 0
(2.8.11)
k=l
This analysis of the firing process is also valid for a bridge connected via a
star-g/delta bridge to the AC system, in which case the equidistant timing
references should be advanced by 30.
controller
current transducer
p
Id
current order
Figure 2.19
Current controller
46
A
a
Figure 2.20
Method of finding the firing instants; the timing instants are assumed
perfectly equidistant (7 3 )
The constant component of the alpha order, oc0, cannot be solved directly
as the PI control has a pole at zero frequency. However, in the steady state
the average delay angle, <x0, takes on a value which causes the DC component of the DC current to be equal to the current order. This requirement
is easily expressed as a mismatch equation with a zero crossing at the
current order
jr
ry
4V
E ^Y
(2 H 19^
where Vfwd is the constant forward voltage drop through a group. This
equation states that the DC voltage, when applied to the DC system, causes
the current order to flow in the DC system. The DC voltage is obtained from
eqn. 2.8.35. Note that eqn. 2.8.12 is not a function of a0. The average delay
angle does, however, feature in the firing-angle mismatch eqns. 2.8.11.
When the converter is solved in Chapter 4, the average alpha order emerges
from the Newton solution.
A similar equation to eqn. 2.8.12 can also be written to describe a
constant-power control
= Pspeoified
- 4Vfwd)I,dO
(2.8.13)
This equation has a zero crossing at a value of average DC current which

causes the power order to be satisfied.
Commutation margin control
The commutation margin angle (also called gamma or extinction angle) is
defined as being the angle from the end of the valve-commutation period
to the first commutating voltage-crossover point (i.e. when the voltage
47
driving the commutation has fallen to zero)

(2.8.14)
yn = dn-4>n
where
y = angle of extinction
3 = voltage-crossover angle
(f> = angle of end of commutation
A common inverter-control mode is extinction angle (or gamma) control, to
minimise reactive-power demand with little chance of commutation failure.
Under normal levels of distortion, the twelve gammas will differ, and it is
the smallest of these which is controlled to a specified minimum, typically
15 to 20 degrees. As the switching with the minimum gamma does not
change from cycle to cycle when in the steady state, the firing order remains
constant and unresponsive to harmonics.
Since the commutation voltage is assumed to be distorted with harmonics,
an iterative solution is required for the crossover points, 3n. A suitable
equation for solution by Newton's method is just the statement that the
commutating voltage is zero at 3
i
dek)e
y^.O.lJ)
where Vdbk and Vdek are the relevant DC voltage samples derived in section
2.8.3 corresponding to the noncommutating states before and after the
commutation, respectively. The zero crossing being solved for S is illustrated
Vdc
Figure 2.21
Voltage samples forming crossover
48
in Figure 2.21. The use of the DC-voltage samples to define the 8 is

necessary since the effect of DC-side ripple current flowing through the
commutating reactance is accounted for. With the crossover angles and,
hence, gamma angles determined, the control constraint is simply that
Tspecified = ?min = 0
(2.8.16)
for minimum gamma control, or

1 12
rspecified-T^ I li = 0
(2.8.17)
for average gamma control.
2.83 Direct voltage

The six-pulse bridge passes through 12 states per cycle. Six of these are
commutation states, and six are direct conduction states. During direct
conduction the positive and negative rails of the DC side are connected to
the AC side via two conducting thyristors. Similarly to the commutation
analysis, either state can be modelled by the immediate steady state of a
simple linear circuit. The circuit consists of a star or delta-connected AC
voltage source with inductance connected to a current source representing
the DC system. The particular configuration of each circuit depends upon
the conduction pattern of the valves in the bridge.
Although it is straightforward to solve the representative linear circuits,
the outcome of each steady-state solution is a harmonic spectrum which,
when transformed into the time domain, matches the DC voltage during the
appropriate conduction interval only. The objective is a single spectrum that
is valid for one complete cycle of DC voltage, not 12 spectra each valid for
only one twelfth of a cycle. The complete spectrum is obtained by convolving each of the 12 sample spectra with the spectrum of a periodic square
pulse which has a value of one during the corresponding conduction
interval, and a value of zero everywhere else. This yields 12 DC voltage
sample spectra, the sum of which is the spectrum of the DC voltage across
the bridge.
Star-connection voltage samples
During normal conduction, the positive and negative rails of the DC side
are directly connected to different phases of the AC terminal via the
commutating reactance in each phase. The kth harmonic component of the
DC voltage is, therefore
y,pk = V + - V_ -j(ok(L+ + L_)Id(i
(2.8.18)
The AC-DC converter in steady state 49

Table 2.1 Construction of DC-voltage and AC-phase current samples
Phase currents
Sample (p)
2
3
4
5
6
7
8
9
10
11
12
DC voltage
Icl
-Id
Id Icl
Id
-Id
Id
el-h
hi
0
I c3
Id
h-h
C
B
Id
*d
Ic4
Id
~Ic4-Id
Id
c5
"Id
-he-
Id
Id
0
I c5
ld
B
,
C
A
.
B
A
C
-Id
A
C
Id
Id
-Id
-Id
b
C
eqn.
2.8.19
2.8.18
2.8.20
2.8.18
2.8.19
2.8.18
2.8.20
2.8.18
2.8.19
2.8.18
2.8.20
2.8.18
During a commutation on the positive rail, analysis of Figure 2.17 yields
Vw-L-V' + L ' ( V r - i t o W - V . - ; t o L , U
(2.8.19)
and
i ^ + y-jfca.w
(2 8 20)
for a commutation on the negative rail. In these equations e refers to a

phase-ending conduction, b to a phase-beginning conduction and o to the
other phase.
From the known conduction pattern in each of the 12 states, eqns. 2.8.18,
2.8.19 and 2.8.20 are used to assemble the 12 samples of the DC voltage.
These samples are summarised in Table 2.1.
Delta-connection voltage samples
The DC voltage during a particular commutation has already been derived
in section 2.8.1 with reference to Figure 2.18. The general result is
V,p* = PflpkVflk + bpkVbk + PcpkYck + P ^ I *
(2.8.21)
where p = 1, 3, 5, 7, 9, 11 refers to the conduction interval number. The
50
coefficient matrix P is constant, and need only be calculated once. During

a commutation on the positive rail
= 0
*V
(2.8.24)
,x
(2.8.^5)
where, if is{ 1... 6} is the number of a commutation on the positive rail,

p % 1, then the subscripts {e, b, o} are a permutation of {a, b, c}
according to i. A similar result holds for a commutation on the negative rail.
During a normal conduction period all three phases of the voltage source
contribute to the DC voltage. Figure 2.22 shows the representative linear
circuit of a particular conduction period. This circuit is analysed by first
writing nodal and loop equations at harmonic k
âk
^ck
*dk
hk - hk ~hk =
z
ck = hk
Vak jiakXak V dk = 0
V^-AA+V^ + V^-iuXÔ
(2.8.26)
This system is readily solved for the DC-voltage sample at harmonic k

V
ak
~^âfc^<*fe)(Xfefc + Xcfc) Xflfc(Vftfc + Vcfc)

Xflk + Xbk + Xck
As for the star-connected source, the solution for the DC-voltage samples is
generalised over all 12 conduction periods into a matrix of coefficients of
the DC and AC sources, i.e.
Vm = VaikVak + PwkV* + PclkVck + VmIA
where
I = 2, 4, 6, 8, 10, 12
(2.8.28)
51
Figure 2.22
Representative linear circuit for a particular conduction period with a

delta-connected source
Convolution of the samples

Having obtained 12 DC-voltage samples as a function of DC and AC-side
sources, the overall solution for the DC voltage spectrum is constructed by
convolving each sample with a square-pulse sampling function (Figure
2.23). The convolution described here is positive frequency only, and so
generates phase-conjugated terms. The square pulse is periodic at the
fundamental frequency, and delimited alternately by the firing and end-ofcommutation angles as listed in Table 2.2. The complex Euler coefficient
for the sampling function at harmonic k is
Wpk = [cos (kap) - cos (kbp)] +j [sin (kbf) - sin (ka,)]
[ # P - ap)/2rt
bp>ap
otherwise
j[l (ap bp)l2n\
(2.8.29)
(2.8.30)
Since the end of one conduction interval is the beginning of the next, all
of the trigonometric evaluations are used in two consecutive sampling
functions, thus halving the number of calculations. The DC voltage can now
be written as
vd=
12
(2.8.31)
52

A
sampling function p
2K
Figure 2.23
Sampling functions used for convolutions
The convolution of two phasors is given by

if Jb>z
otherwise
(2.8.32)
The conjugate operator makes the convolution nonanalytic, and so not

differentiable in the complex form. It avoids the need for negative harmonics, however, and it is still possible to obtain partial derivatives by
decomposing into real and imaginary parts. Sum and difference harmonics
are generated by the convolution, and since it is required to calculate
voltage harmonics up to nh, the sampling function spectra must be evaluated
up to 2nh. As the convolution operator is linear, the 12 convolutions
in eqn. 2.8.25 can be decomposed into convolutions of the component
Table 2.2
Limits of converter states for use in sampling functions
Sample (p)
1
2
3
4
5
6
7
8
9
10
11
12
ap
02
4>2
02
4>2
03
<t>3
05
4>5
06
4>6
04
04
05
05
06
06
01
03
53
Time domain dc vol
voltage sample convolved with square pulse
complete harmonic solution for dc voltage
-1 cycle-
Figure 2.24
Construction of the DC voltage and validation against time-domain

solution
phasors
(2.8.33)
fe=l
1= 0
This equation generates voltage harmonic components of order above nh

which are discarded. By using eqns. 2.8.31, 2.8.32 and 2.8.33, the Ath
54
harmonic phasor component of Vd is
1=1
=k
1=0
(2.8.34)
V.0 = kj I j E O V * ) - 2/(V<Jp0p0)}
P=I
u=o
(2.8.35)
This completes the derivation of the DC-voltage harmonics in terms of a

DC-side harmonic source, and a three-phase AC-side voltage source connected in star or delta, with source inductance. The process of sampling and
convolution to obtain the DC voltage is demonstrated graphically in Figure
2.24. This sequence of plots was obtained by first simulating a six-pulse
rectifier to the steady state in the time domain. The spectra of a terminal
voltage and DC current were then used to calculate the DC voltage, allowing
a validation against the time-domain solution.
2.8.4 Phase currents

Converter side
The derivation of the DC voltage involves the convolution of 12 different
DC-voltage samples; therefore, by using the same sampling functions, 36
convolutions would be required to obtain the three-phase currents. However, referring to Table 2.1, and using the linearity of the convolution, the
phase A secondary current can be written as
j - Ic3 <g> V5 - I c4 7 + Ic6
(2.8.36)
and similarly for one of the other two phases. The third phase must always
be the negative sum of the first two, since there is no path for zero sequence
into a bridge. This leads to a total of eight convolutions to calculate the
three-phase currents. As evident in eqn. 2.8.36, the periodic samples for the
phase-current calculation are just the DC-side current, and the commutation
currents derived in section 2.8.1. The calculation of the phase current
flowing into the transformer primary is addressed in the next section. The
derivation of a phase current is illustrated graphically in Figure 2.25.
System side
Unbalance in the tap-changer setting between the two six-pulse groups will
lead to imperfect cancellation of six-pulse harmonics on the AC and DC
sides of the converter. If the impedances in each phase of a three-phase
bank are not all equal, the converter will generate positive and negative-
55
time domain phase current
commutation current sample
convolved dc current
convolved dc current + convolved commutation currents

-1 cycle-
Figure 2.25
Construction of the phase current and validation against time-domain

solution
sequence odd triplen harmonics. The unbalanced star-g/delta connected

transformer also acts as a sequence transformer, causing the converter to
both respond to, and generate, zero-sequence harmonics.
In this section the transformer is modelled as a series connection of ideal
tap-changing transformers on the primary and secondary sides, a conduction resistance, a leakage reactance and a star-delta connection; core
56
Figure 2.26
Equivalent circuit for star-glstar transformer
saturation and hysteresis are not modelled. The resistance and reactance
may be unbalanced, but the tap settings are assumed to be the same on all
phases. The tap-change controller is not modelled as it does not respond to
harmonics. The magnetising current injection is approximated by a shunt
to ground at the primary terminal.
The outcome of this analysis is a transfer model of the transformer; the
primary currents are related directly to the secondary currents and the
secondary voltage to the primary voltage. This is easily achieved for the
star-g/star connection, shown as a single-line diagram in Figure 2.26. The
transformer and thyristor resistances have been referred to an equivalent
AC-side resistance, RAC
(2.8.37)
The leakage reactance has been referred to an equivalent on the secondary

side
X
(2.8.38)
Since all impedance has be^n removed from the transformer, the secondary
voltage is now independent of the current through the transformer
v
(2.8.39)
and similarly for the phase current

I
~
*p
I1 s
(2.8.40)
These equations are repeated over all harmonics, and all three phases.
57
star-g-delta
primary to secondary
Figure 2.27
a Equivalent circuit for star-g/delta transformer

b Transfer from star to delta
c Transfer from delta to star
The star-g/delta connected transformer is considerably more difficult to

model and, in fact, requires two separate analyses for transfers from the star
to delta side and vice versa. As shown in Figure 2.27, the transfer from star
to delta is primarily concerned with setting up the delta-connected source
for the voltage sampling and commutation analyses. The secondary-side
delta-connected voltage source is scaled by
(2.8.41)
and the equivalent reactance is
(2.8.42)
58
The J3 scaling for the delta winding does not affect the transfer of thyristor
resistance through the transformer, since it is not connected in delta. Thus,
the referred AC-system resistance is the same as that given by eqn. 2.8.37.
Calculation of the primary-side phase currents in terms of the secondary
currents is complicated by the circulating current in the delta winding. If
the transformer is unbalanced, some of this appears as a positive or
negative-sequence current on the primary side.
The admittance matrix for an unbalanced star-g/delta transformer is
readily obtained
I
where
(2.8.44)
(2.8.45)
(2.8.46)
Eqn. 2.8.43 is used to calculate the primary current, by assuming that Vp

and 1, are known, and eliminating V,. The admittance matrix in eqn. 2.8.43
is not invertible, as the delta winding is floating; there are an infinite
number of possible potentials of the delta winding which are consistent with
a given current injection into the transformer. One such potential is that
obtained by grounding phase c on the secondary so that Vsc = 0. This
permits the removal of the last row and column from eqn. 2.8.43
IPb
IPC
h a
A B
=
[c
D]
(2.8.47)
59
where
a2Ya
A=
c=
a Yfc
0
0
B=
0
2
0
a Y,
-ajSY a
ajSY,
-ap\
a^Yc
-a.pYa
aj8Yfl
(2.8.48)
-a/T
D=
(2.8.49)
(2.8.50)
(2.8.51)
Eliminating Vs, the primary-phase currents are

I P = [A - BD~ 'CJVp + BD" %
def
(2.8.52)
(2.8.53)
YD is a shunt admittance to ground at the converter terminal that is added

to the filter shunt. TD is a transfer matrix across the transformer, of size
3 x 2 , indicating that there is no zero-sequence current on the secondary
side, and that only the phase a and b currents need he calculated. If the
transformer is balanced, then YD is a zero-sequence shunt, and so is not
invertible. This also implies that if the transformer is nearly balanced, YD
has a high condition number and should not be inverted into an impedance
without first being combined with an admittance which offers a path to
positive and negative-sequence currents.
2.9 Summary
Following an introductory discussion on model accuracy, this Chapter has
described a variety of steady-state AC-DC converter models with varying
degrees of complexity.
The starting case is the fundamental-frequency model based on symmetrical operation, with perfect AC-current filters and DC-voltage smoothing.
This model is used in the power flow solution considered in Chapter 3. The
removal of the symmetry constraint is considered next, and a three-phase
model is developed for use in more detailed power-flow analysis, a subject
60
described in Chapter 5. The perfect filter and smoothing assumptions are

then removed and the AC-DC converter is considered as a frequency
modulator. A small-signal transfer-function model is described, capable of
direct analysis of crossmodulation effects at any frequency. This technique
is used to simulate the mechanism of harmonic instabilities. The use of
convolutions in the harmonic domain avoids the problems of aliasing
associated with numerical FFT calculations, or the complexity of Fourier
analysis. An additional advantage of the convolution analysis is that all of
the equations are differentiable when decomposed into real and imaginary
components, a feature which enables a straightforward implementation of a
Newton's method solution in Chapter 4.
2.10 References
1 ARRILLAGA, J.: 'High voltage direct current transmission', IEE Power Eng. Ser.,
1988, pp. 246
2 AINSWORTH, J.D.: The phase locked oscillator-a new control system for
controlled static converters', IEEE Trans. Power Appar. Syst., May 1968, 87, (3),
pp. 859-865
3 CHEN, S., WOOD, A. R., and ARRILLAGA, J.: 'HVDC converter transformer
core saturation instability: a frequency domain analysis', IEE Proc, Gener. Transm.
Distrib., January 1996, 143, (1), pp. 75-81
4 WOOD, A.R.: 'An analysis of non-ideal HVdc converter behaviour in the
frequency domain, and a new control proposal'. PhD thesis, University of
Canterbury, New Zealand, 1993
5 GIESNER, D.B., and ARRILLAGA, J.: 'Behaviour of HVDC links under unbalanced ac faults', Proc. IEE, 1972, 119, (2), pp. 209-215
6 SHORE, N.L., ANDERSSON, G., CANELHAS, A, and ASPLUND, G.: 'A
three-pulse model of dc side harmonic flow in hvdc systems', IEEE Trans. Power
Deliv., July 1989, 4, (3), pp. 1945-1954
7 HARKER, B.J.: 'Steady state analysis of integrated ac and dc systems'. PhD thesis,
University of Canterbury, New Zealand, 1980
8 SWARTZ, M., BENNETT, W.R., and STEIN, S.: 'Communication systems and
techniques' (McGraw-Hill, 1966)
9 PERSSON, E.V.: 'Calculation of transfer functions in grid controlled converter
system', IEE Proc, May 1970, 117, (5), pp. 989-997
10 HU, L., and YACAMINI, R.: 'Harmonic transfer through converters and HVdc
links', IEEE Trans. Power Electron., July 1992, 7, (4), pp. 514-525
11 CIGRE SC14, WG 14.25, 'Harmonic cross-modulation in HVDC transmission'.
HVDC colloquium, Johannesburg, South Africa, September 1997
12 SMITH, B.C.: 'A harmonic domain model for the interaction of the HVDC
converter with ac and dc systems'. PhD thesis, University of Canterbury, New
Zealand, 1996
Chapter 3
3.1 Introduction
Power-flow analysis is used to determine the steady-state operating characteristics of the power-generation/transmission system for a given set of
busbar loads. In this context, steady state means a time-independent
condition which implicitly takes into account the final adjustment of the
parameters involved in maintaining the specified operating condition at
each bus, and the component-related capability constraints.
The solution provides information on voltage magnitudes and angles,
active and reactive power flows in the individual transmission units, losses
and the reactive power generated or absorbed at voltage-controlled buses.
The Newton-Raphson method,1 complemented with sparsity programming and optimally-ordered Gaussian elimination, has become the cornerstone of modern power-flow programmes. The method is described in
Appendix I.
Most of this Chapter describes the operating state of the combined AC
and DC systems under the specified conditions of load, generation and
DC-system control strategies. The superiority of the fast decoupled AC-load
flow2 is now generally accepted and, therefore, the incorporation of HVDC
transmission is described with reference to this algorithm.
In short-term studies, such as the system response to a change in load
specification, the power-flow assessment is made without the assistance of
generator-control adjustment. The term quasisteady state is used here when
referring to such a condition. An assessment of the power-transfer capability
of the DC link also considered in this Chapter comes under this category.
3.2 Specification of the operating condition

The power-flow solution requires knowledge of four variables at each bus k
in the system
P fc - real or active power
Q k - reactive or quadrature power
62
Vk - voltage magnitude
6k- voltage phase angle
Two of these variables are specified at the outset and the aim of the power
flow is to find the remaining two variables at each bus.
Three different bus conditions exist:
(i)
Voltage-controlled bus. The total injected active power, Pfe, is specified,

and the voltage magnitude Vk is maintained at a specified value by
reactive-power injection. This type of bus generally corresponds to a
generator where Pfc is fixed by turbine governor setting and Vk is fixed
by automatic voltage regulators acting on the machine excitation; or a
bus where the voltage is fixed by supplying reactive power from static
shunt capacitors or rotating synchronous compensators, e.g. at substations.
(ii) Nonvoltage-controiled bus. The total injected power, Pfc +jQk, is specified at this bus. In the physical power system this corresponds to a load
centre such as a city or an industry, where the consumers demand their
power requirements. Both Pfc and Qk are assumed to be unaffected by
small variations in bus voltage.
(iii) Slack (swing) bus. This bus arises because the system losses are not
known precisely in advance of the power-flow calculation. Therefore,
the total injected power cannot be specified at every single bus. It is
usual to choose one of the available voltage-controlled buses as slack,
with its power components as the unknown variables. The slack-bus
voltage is usually assigned as the system phase reference and its
complex voltage
is, therefore, specified. The analogy in a practical power system is the

generating station which has the responsibility of system frequency
control.
The power flow solves a set of simultaneous nonlinear algebraic equations
for the two unknown variables at each node of the system by means of an
iterative method.
The basic algorithm of the power-flow solution is illustrated in Figure 3.1.
System data, such as busbar power conditions, network connections and
impedance, are read in and the admittance matrix formed. Initial voltages
are specified at all buses; these are 1 H-jO for P,Q buses and V + jO for P,V
buses.
The iteration cycle is terminated when the busbar voltages and angles are
such that the specified conditions are satisfied at all buses. This condition is
achieved when the power mismatches are less than a small tolerance rjlf and
63
input
read system data and
node specifications
use system data to form
the system admittance
matrix
initialise voltages at all
system busbars
update voltages and angles
to satisfy the specified
conditions of load and
generation
are
specified
conditions
satisfied
9
output busbar voltages,

generation and line power
flowsjosses, etc.
Figure 3.1 Flow diagram of basic power-flow algorithm
the voltage errors less than rj2. Typical figures for r\x and Y\2 are 0.01 p.u.
and 0.001 p.u., respectively.
When the solution is reached, the terminal conditions at each bus are
calculated and with that information the line power flows and system losses
are determined.
3.3 Formulation of the algorithm

The network nonlinear equations are
lk =
for all k
mek
(3.3.1)
64
where Ik is the current injected into bus k. The power at the k bus is then
given by
S* = P* + jQ* = E*I
= Ek I J&.E*
(3.3.2)
mek
As the power-flow equations cannot be differentiated in complex form, the

problem is separated into real equations and variables using polar or
rectangular co-ordinates for the bus voltages. This produces the following
two equations per bus
or
P(e,f)
or
QLe,f)
In polar co-ordinates, the real and imaginary parts of eqn. 3.3.2 become
P* = I V,V m (G tm cos 6km + Bkm sin 0km)
(3.3.3)
mek
Ok = E V,V m (G, m sin0 km - B k m cos<U
(3.3.4)
mek
where
The resulting linear relationships, obtained for small variations in the

variables 8 and V by forming the total differentials, are
for a P Q busbar
AP^lg^+XjÂV,,,
mek oum
oy
mzk
(3.3.5)
and
^
me* ^ m
V
mek
Oy
(3.3.6)
for a PV busbar
only eqn. 3.3.5 is used, since Q^ is not specified
for a slack busbar
no equations
The voltage magnitudes appearing in eqns. 3.3.5 and 3.3.6 for PV and slack
65
busbars are not variables, but are fixed at their specified values. Similarly,
9 at the slack busbar is fixed.
The set of linearised equations consists of two for each PQ busbar and
one for each PV busbar. The problem variables are V and 9 for each PQ
busbar and 9 for each PV busbar. The basic Newton solution is then
expressed by the equation
P mismatches
for all PQ and
PV busbars
Q mismatches
for all PQ and
PV busbars
AP""1
AQ* *
HP-1
A9P
Np-1
Lp-1
AVP
yp-i
^ 9 corrections
I for all PQ and
J PV busbars
1 V corrections
> for all PQ
J busbars
(3.3.7)
For busbars k and m the Jacobian matrix terms are
H km = H * = V t V m (G km sin 9km - B km cos6 km )

N t a = Vm * = V,V m (G km cos0, m + B t m sin0 t m )
= Vm
and for m k
_ P _ r v2
A basic flow diagram of the Newton-Raphson algorithm is shown in

Figure 3.2.
66

input data
read all input data, form

system admittance matrix,
initialise voltages and angles
at all busbars
calculate real and reactive

power mismatches
yes
form Jacobian matrix
solve Jacobian equation for

voltage and angle updates
update voltages and angles

output
results
Figure 3.2
Flow diagram of the basic Newton- Raphson power-flow algorithm
3.4 Decoupled Newton techniques

A characteristic of AC-power transmission in the steady state is the strong
interdependence between active powers and bus-voltage angles, and between reactive powers and voltage magnitudes. Correspondingly, the coupling between the P 9 and Q V components is relatively weak. Various
algorithms have been proposed to exploit this decoupling.3"5
67
The basic decoupled equations are

(3.4.1)
(3.4.2)
The elements of matrices [T] and [U] are
(3.4.3)
(3.4.4)
mcok
(3.4.5)
(3.4.6)
u tt = - I u ta
where Zkm and Xkw are the branch impedance and reactance, respectively.
Eqns. 3.4.1 and 3.4.2 can be solved using Newton's method, by expressing
the Jacobian equations as
AP
A0
AQ/V
AV
(3.4.7)
or
(3.4.8)
[AP] = [T][A0]
[AQ/V] = [U][AV]
(3.4.9)
where
[AP] = [AP]
and
[AQ/V] = [AQ]
The decoupled power flow based on the Jacobian matrix equation of the
formal Newton method is
AP
Ad
AV
(3.4.10)
If the submatrices N and J are neglected, since they represent the weak
coupling between P 6 and Q - V, the following decoupled equations

result
[AP] = [H][A0]
(3.4.11)
[AQ] = [L][AV]
(3.4.12)
The latter equation can be unstable at some distance from the exact solution
and improved convergence is obtained by dividing the right-hand side of
both eqns. 3.4.11 and 3.4.12 by the voltage magnitude V, i.e.
[AP/V] = [A][A0]
(3.4.13)
= [C][AV]
(3.4.14)
These equations are solved successively using the latest values of V and 6
available. [A] and [C] are sparse, nonsymmetric in value and are both
functions of V and d.
The efficacy of the decoupling solution is further improved by making the
following assumptions:
(i) V t > V m =1.0p.u.;
(ii) Gkm BkOT, and hence can be ignored;
(iii) cos(6k-em) = h0
since angle differences across transmission lines are small under normal
loading conditions.
This leads to the following decoupled equations
[AP/V]==[B*][A0]
(3.4.15)
[AQ/V] = [B*][AV]
(3.4.16)
Further refinements to this method are:

(a) Omit from the Jacobian in eqn. 3.4.15 the representation of those
network elements which predominantly affect the reactive power flow,
e.g. shunt reactances and off-nominal in-phase transformer taps. Neglect also the series resistances of lines.
(b) Omit from the Jacobian of eqn. 3.4.16 the angle-shifting effects of
phase shifters.
The resulting fast decoupled power flow2 equations are then
(3.4.17)
[AQ/V] = [B"][AV]
(3.4.18)
no
no
solve eqn. 3.4.17
and update (0)
calculate (AQ/V)
/reactive
powers \
converged
yes
solve eqn. 3.4.18 and update (V)
output results
Figure 33
Flow diagram of the fast-decoupled load flow
where
Bklfl --
mmk
Bkm = Bkm
m<ak
for m # k
for m # k
69
70
The equations are solved alternatively using the most recent values of V and
Q available as shown in Figure 3.3.6
The matrices B' and B" are real and are of order (N 1) and (N - M),
respectively, where N is the number of busbars and M is the number of PV
busbars. B" is symmetric in value and so is B' if phase shifters are ignored;
it is found that the performance of the algorithm is not adversely affected.
The elements of the matrices are constant and need to be evaluated and
triangulated only once for a network.
Convergence is geometric (between two and five iterations are required
for practical accuracies) and more reliable than the formal Newton's
method. This is because the elements of B' and B" are fixed approximations
to the tangents of the defining functions AP/V and AQ/V, and are not
susceptible to any humps in the defining functions.
3.5 Incorporation of AC-DC buses

The operating state of an AC-DC power system is defined by the vector
[V, 9, xf
where
V is a vector of the voltages magnitudes at all AC system busbars
9 is a vector of the angles at all AC-system busbars (except the reference bus
which is assigned 9 = 0)
x is a vector of DC variables
Earlier sections have described the use of V and 9 as AC-system variables
and the selection of DC variables x is discussed in section 3.6.
The modified Newton-Raphson solution requires the formulation of n
independent equations in terms of the n variables.
The modified real and reactive power mismatches at the converter
terminal busbars become
P&m - Pterm(AC) - Pterm(DC) = 0
(3.5.1)
Q&m ~ Qterm(AC) - Qterm(DC) = 0
(3.5.2)
where the subscript term refers to the converter AC-terminal busbar.

Pterm(AC) is the injected power at the terminal busbar as a function of the
AC-system variables
Pterm(DC) is the injected power at the terminal busbar as a function of the
DC-system variables
Pjfrm is the AC-system load (if any) at the busbar
and similarly for Qterm(DC) and Qterm(AC).
71
The injected powers Qterm(DC) and Pterm(DC) are functions of the converter AC terminal busbar voltage and of the DC-system variables, i.e.
Pterm(DC) =/(V term) x)
(3.5.3)
Qterm(DC) =/(V term) x)
(3.5.4)
The equations derived from the specified AC-system conditions may,

therefore, be summarised as
AP(V, 0)
AP term (V ( 6, x)
AQ(V, 0)
=0
(3.5.5)
AQ term (V, 6, x)
where the mismatches at the converter terminal busbars are indicated

separately.
A further set of independent equations are derived from the DC-system
conditions. These are designated
R(v t e r m , x)k = o
(3.5.6)
for k = 1, to the number of converters present.

The DC-system eqns. 3.5.3, 3.5.4 and 3.5.6 are made independent of the
AC-system angles 8 by selecting a separate angle reference for the DCsystem variables. This improves the algorithmic performance by effectively
decoupling the angle dependence of AC and DC systems.
The general AC-DC power flow problem may, therefore, be summarised
as the solution of
AP(V, 6)
AP term (V, 0, x)
AQ(V, d)
=0
(3.5.7)
AQ term (V, 9, x)
R(V term , x)
3.6 DC-system model

The selection of variables x and formulation of the equations require several
basic assumptions which are generally accepted in the analysis of steadystate AC-DC converter operation. These are:
72
(i)
The three AC voltages at the terminal busbar are balanced and

sinusoidal.
(ii) The converter operation is perfectly balanced.
(ill) The DC-side current and voltage contain no AC components,
(iv) The converter transformer is lossless and the magnetising admittance
is ignored.
3.6.1 Converter variables

Under balanced conditions, similar converter bridges attached to the same
AC-terminal busbar will operate identically regardless of the transformer
connection. They may, therefore, be replaced by an equivalent single bridge
for the purpose of single-phase load-flow analysis. With reference to Figure
3.4a, the set of variables illustrated, representing fundamental-frequency or
DC quantities, permits a full description of the converter-system operation.
An equivalent circuit for the converter is shown in Figure 3.#, which
includes the modification explained in section 3.5 as regards the position
of angle reference.
EUL
a.c. terminal
busbar
QBf
(a 2 -a) Bt
Id
JS
(\-Q)Bf
(b)
Figure 3.4
Basic AC-DC converter terminal

a Transformer-converter unit (with the angles referred to the ACsystem reference
b Single-phase equivalent circuit (with the angles referred to the DC
reference)
73
The variables, defined with reference to Figure 3.46 are:

Vterm L 0
converter terminal busbar nodal voltage (phase angle referred

to converter reference)
E L ifr
fundamental-frequency component of the voltage waveform at
the converter transformer secondary
fundamental-frequency component of the current waveshape
Ip, Is
on the primary and secondary of the converter transformer,
respectively
a
firing-delay
angle
a
transformer off-nominal tap ratio
average DC voltage
Vd
converter direct current
ld
These ten variables, nine associated with the converter, plus the ACterminal voltage magnitude Vterm, form a possible choice of x for the
formulation of eqns. 3.5.3, 3.5.4 and 3.5.6.
The minimum number of variables required to define the operation of
the system is the number of independent variables. Any other system
variable or parameter (e.g. PDC and QDC) m a y be written in terms of these
variables.
Two independent variables are sufficient to model a DC converter,
operating under balanced conditions, from a known terminal-voltage
source. However, the control requirements of HVDC converters are such
that a range of variables, or functions of them (e.g. constant power), are the
specified conditions. If the minimum number of variables is used then the
control specifications must be translated into equations in terms of these two
variables. These equations will often contain complex nonlinearities, and
present difficulties in their derivation and programme implementation. In
addition, the expressions used for PDC and QDC in eqns. 3.5.1 and 3.5.2 may
be rather complex and this will make the programming of a unified solution
more difficult.
For these reasons, a nonminimal set of variables is recommended, i.e. all
variables which are influenced by control action are retained in the model.
This is in contrast to AC loadflowswhere, owing to the restricted nature of
control specifications, the minimum set is normally used.
The following set of variables permits simple relationships for all the
normal control strategies7
[*] = [Vd, Id,a, cos a, <f>f
Variable $ is included to ensure a simple expression for QDC and cos a is
used as a variable rather than a, to linearise the equations and thus improve
convergence.
74
3.6.2 DC per unit system

To avoid per unit to actual value translations and to enable the use of
comparable convergence tolerances for both AC and DC-system mismatches, a per unit system is also used for the DC quantities.
Computational simplicity is achieved by using common-power and voltage-base parameters on both sides of the converter, i.e. the AC and DC
sides. Consequently, in order to preserve consistency of power in per unit,
the direct current base, obtained from (MVAB)/VB, has to be ^/3 times larger
than the AC current base.
This has the effect of changing the coefficients involved in the AC-DC
current relationships. For a perfectly smooth direct current and neglecting
the commutation overlap, the r.m.s. fundamental component of the phase
current is related to ld by the approximation
IS = ^ . I ,
%
(3.6.1)
Translating eqn. 3.6.1 to per unit yields
and if commutation overlap is taken into account, this equation becomes

Is(p.u.) = k ^ - . ld(p.u.)
n
(3.6.2)
where k under normal steady-state operating conditions is above 0.99.
3.6.3 Derivation of equations

The following per unit relationships are derived for the variables defined in
Figure 3.4:
(i)
The current magnitude on the converter AC side is related to the

direct current by the equation
(ii)
The current magnitudes on both sides of the lossless transformer are

related by the off:nominal tap i.e.
I, = .I.
(3.6.4)

(iii)
The DC voltage is expressed in terms of the AC-source commutating

voltage referred to the transformer secondary, i.e.
Vd = ^ . a. Vtermcos a - ^ . I,. Xc
(iv)
75
(3.6.5)
where Vterm, the commutating voltage, is the busbar voltage on the

system side of the converter transformer.
The relationship between DC current and voltage is determined by
the DC-system configuration
f(Vd, ld) = 0
(3.6.6)
e.g. for a simple rectifier supplying a passive load
(v)
Owing to the assumptions listed at the beginning of section 3.6, only

the real power equation relates the DC power to the transformer
secondary power in terms of fundamental components, i.e.
(3.6.7)
(vi)
As the transformer is lossless, the primary real power is equal to the

DC power, i.e.
V.-I^V^.W
(vii)
(3.6.8)
The fundamental component of current flow across the converter

transformer is expressed as
(3.6.9)
where jBt is the transformer leakage susceptance.
So far, seven equations have been derived and no other independent

equation may be written relating the total set of nine converter variables.
However, variables I p , I s , E and \j/ can be eliminated as they play no part
in defining control specifications.
76
Thus, eqns. 3.6.3, 3.6.4, 3.6.7 and 3.6.8 can be combined into
VÂ1.a.Vtcrmcos^ = 0
(3.6.10)
where
The final two independent equations required are derived from the specified control mode.
The DC model may thus be summarised as follows
R(x, V term ) t = 0
(3.6.11)
where
R(l) = Vd-k1.a.VteTm.cos<t>
R(2) = Vd - k , .a .V t e r m .cos + ^ . X ,
R(4) = control equation

R(5) = control equation
and
x = \Vdi Id9 a, cos a, <]T
Vterm can either be a specified quantity or an AC-system variable. The

equations for Pj^ and Q ^ may now be written as
= Vterm. Ip. sin 4
(3.6.12)
= Vterni.Ip.cos4>
(3.6.13)
and
or
V (( .I d
(3.6.14)
77
3.6.4 Incorporation of control equations

Each additional converter in the DC system contributes two independent
variables to the system and thus two further constraint equations must be
derived from the control strategy of the system to define the operating state.
For example, the basic two-terminal DC link has two converters and,
therefore, requires four control equations. The four equations must be
written in terms of the ten DC variables (five for each converter).
Any function of the ten DC-system variables is a valid (mathematically)
control equation so long as each equation is independent of all other
equations. In practice, there are restrictions limiting the number of alternatives. Some control strategies refer to the characteristics of power transmission (e.g. constant power or constant current), others introduce
constraints such as minimum delay or extinction angles.
Examples of valid control specifications are:
(i)
Specified converter transformer tap

a-
(ii)
sp
a
= 0
Specified DC voltage
V, - Yf = 0
(iii) Specified DC current
(iv) Specified minimum firing angle

cosa~-cosocmin = 0
(v)
Specified DC power transmission

d ' ld "~ * D C -
These control equations are simple and are easily incorporated into the
solution algorithm. In addition to the usual control modes, nonstandard
modes such as specified AC terminal voltage may also be included as
converter-control equations.
During the iterative solution procedure the uncontrolled converter variables may go outside prespecified limits. When this occurs the offending
variable is usually held to its limit value and an appropriate control variable
is freed.
78
3.6.5 Control of converter AC terminal voltage

There are two alternative specifications for the converter terminal voltage:
(a) By local reactive power injection at the terminal. In this case, no
reactive power-mismatch equation is necessary for that busbar and the
relevant variable (i.e. AVterm) is effectively removed from the problem
formulation. This is the situation where the converter terminal busbar
is a P V busbar.
(b) As a DC-system constraint. That is, the DC converter must absorb the
correct amount of reactive power so that the terminal voltage is
maintained constant, i.e.
Vm-Vterm = 0
(3.6.15)
In this case, the above equation becomes one of the two control
equations and some other variable (e.g. tap ratio) must be specified
instead.

The equations presented so far are equally applicable to inverter operation.
However, during inversion it is the extinction-advance angle (y) which is the
subject of control action and not the firing angle (a). For convenience,
therefore, equation R(2) of eqn. 3.6.11 may be rewritten as
V, - Kt .a.V term .cos (n - y) - - Xc.ld = 0
n
(3.6.16)
This equation is valid for rectification or inversion. However, under inversion the value of Vd, calculated from eqn. 3.6.16, will be negative.
To specify operation with constant extinction angle, the following equation is used
cos (n~y)~
cos (n - ysp) = 0
(3.6.17)
where ysp is usually y minimum for minimum reactive-power consumption

of the inverter.
3.7 Unified AC-DC solution

The solution techniques are discussed here with reference to a single
converter; its extension to multiple or multiterminal DC systems is discussed
in section 3.8.
79
The unified method gives recognition to the interdependence of AC and

DC-system equations and simultaneously solves the complete system. Referring to eqn. 3.5.7, the standard Newton-Raphson algorithm involves repeat
solutions of the matrix equation
AP(V, 0)
Ad '
APterm(V, 6, x)
AQ(V, S)
AQterm(V, 8, x)
A0 t e r m
AV
(3.7.1)
AV t e r m
Ax
R ( V t e r m , *)
where J is the matrix of first order partial derivatives.

AP term = P m - P (erm (AC) - P t e r m (DC)
(3.7.2)
AQ,erm = Qtfrm ~ Q,erm(AC) - Q term (DC)
(3.7.3)
and
P l m n (DC) =/(V t e r m , x)
(3.7.4)
Qterm(DC) =/(V term> X)
(3.7.5)
Applying the AC fast decoupled assumptions to all Jacobian elements

related to the AC-system equations, yields
AP/V
Ad
B'
DD AA7
AP term /V tern
AQ/V
AV
AQ erm /V tem
BH
AA!'
AV te
BB"
Ax
(3.7.6)
where all matrix elements are zero unless otherwise indicated. The matrices
[B'] and [B"] are the usual single-phase fast decoupled Jacobians and are
kept constant. The remaining matrices vary at each iteration in the solution
process.
80
The element indicated as B^ in eqn. 3.7.6 is a function of the system

variables and, therefore, varies at each iteration.
The use of an independent angle reference for the DC equations results
in
5Pterm(DC)/30term = 0
i.e. the diagonal Jacobian element for the real-power mismatch at the
converter-terminal busbar depends on the AC equations only and is,
therefore, the usual fast decoupled B' element.
In addition
which will help the subsequent decoupling of the equation.

To maintain the block-successive iteration sequence of the fast decoupled
AC load flow, it is necessary to decouple eqn. 3.7.6. Therefore, the Jacobian
submatrices must be examined in more detail. The Jacobian submatrices are
'term
{dP term (AC)/3V tenn } + - ^ - {dP term (DC)/3V term }

Following decoupled power flow practice
D D = 0 + - L - {5Pterm(DC)/<5Vterm}
* term
and since
P term (DC) = V(
3P term (DC)/dV lerm = 0
Therefore
DD = 0
Similarly
-^[MPter
"term
i-[5P u
Merm
The powerflowsolution 81
[AA"]=-[8AQterJdx]
"term
1
Mem
BB" = dR/dVteil
[A] = dR/dx
1
5Q,erm(AC)/5Vter
V ter
[aQUrm(DC)/avterm]
= B;;(AC)
In the above formulation the DC variables x are coupled to both the real
and reactive power AC mismatches. However, eqn. 3.7.6 may be separated
to enable a block-successive iteration scheme to be used.
The DC mismatches and variables can be appended to the two fast
decoupled AC equations in which case the following two equations result
AP/V
APtCrm/Vterm
R
AQ/V
AQ term /V ter ,
R
Ad
B'
AA'
A
Ax
(3.7.7)
"A?"
B"
BB"
AA"
AVterm
Ax
(3.7.8)
The iteration scheme, illustrated in Figure 3.5, is referred to as -PDC,

QDC-.
The algorithm may be further simplified by recognising the following
physical characteristics of the AC and DC systems:
the coupling between DC variables and the AC terminal voltage is

strong;
there is no coupling between DC mismatches and AC-system angles;
under all practical control strategies the DC power is well constrained
and this implies that the changes in DC variables x do not greatly affect
the real power mismatches at the terminals. This coupling, embodied in
matrix AA'of eqn. 3.7.7 can, therefore, be removed.
82

data input
triangulate B' and B"
converter busbars not used as pivots
calculate A P (AC system only)
IT
calculate Pienn(DC) and DC residuals ( R )
form DC Jacobian matrix

forward reduction of vector A P/V
solve reduced equation for P , ^ and Ax
back substitute for AO
calculate AQ (AC system only)

calculate Q ^ ^ D C ) and DC residuals (R)
form DC Jacobian matrix

forward reduction of vector AQ/V
solve reduced equation for Vterm and Ax
back substitute for AV
Figure 3.5
Flow chart for single-phase AC-DC load flow
83
These features justify the removal of the DC equations from eqn. 3.7.7 to
yield a -P,QDC- block successive iteration scheme represented by the
following two equations
(3.7.9)
[AP/V] = [B'][A0]
AQ/V
AQ
IV
term'
term
AV
B'
AA"
AVterm
BB
Ax
(3.7.10)
3.7.1 Multiconverter systems

Each additional converter adds a further five DC variables and a corresponding set of five equations.
As an example, consider the system shown in Figure 3.6. The top part
represents a point-to-point DC link (converters 1 and 2) and an aluminium
smelter (converter 3). The lower part consists of a three-terminal DC
interconnection (converters 4, 5 and 6) with converter 4 operating in the
rectifier mode and converters 5 and 6 in the inversion mode.
The structure of the reactive-power DC Jacobian for the unified method
is shown in eqn. 3.7.11
AQ/V
AV
AA';t
AQ t e r m i/V t e r m l
AV t e r m l
AV t e r m 2
AQterm2/ M e r m 2
AQ,e3/V ter m3
AA'33
AQ,erm4/Vter m4
AA44
^MOD
A Q t e r m 5 / V t e r m5
AA'5 5
AR2
AR3
AR 4
AR5
AR6
AV t e r m 4
AV t e r m 5
AV t e r m 6
AQ t e r m 6/V t e r m6
ARX
AV t e r m 3
Bn
AXl
K2
AX2
A
(30 x 30)
B33
n"
B55
B'e 6
AX3
AX 4
AX5
AX6
(3.7.11)
84
Figure 3.6
Multiterminal AC-DC system
where BÔD iS ^ e part of B'' which becomes modified. Only the diagonal
elements become modified by the presence of the converters.
Off-diagonal elements will be present in BÔD *f there is any AC connection between converter terminal busbars. All off-diagonal elements of BB"
and AA" are zero.
In this example, matrix A is block diagonal in 5 x 5 blocks with the
exception of the DC interconnection.
The DC-side configuration is used to derive the set of DC equations,
corresponding with the term R(3) of the point-to-point interconnection
(eqn. 3.6.11). For the test system of Figure 3.6 these are
- I<)3 R<(3 = 0
hi ~ hi = 0
(3.7.12)
3.7.2 Programming considerations

To retain the efficiency of the fast-decoupled load flow, the B' and B"
matrices are factorised only once before the iterative process begins.
85
The Jacobian elements related to the DC variables are nonconstant and

must be re-evaluated at each iteration. It is, therefore, necessary to separate
the constant and nonconstant parts of the equations for the solution routine.
Initially, the AC fast decoupled equations are formed with the DC link
ignored (except for the addition of the filter reactance at the appropriate
AC busbar). The corresponding reactive power mismatch equation for the
AC system is
v term J
(3.7.13)
where
is the mismatch calculated in the absence of the DC converter, and B" is the
constant AC fast decoupled Jacobian.
After triangulation down to but excluding the busbars to which DC
converters are attached, eqn. 3.7.13 becomes
where (AQ/V)" and (AQterm/Vterm)" signify that the left-hand side vector has
been processed and matrix B'" is the new matrix B" after triangulation.
The DC-converter equations may then be combined with eqn. 3.7.14 as
follows
(AQ/V)"
ermy
V
T
, AQ terro (DC)
V
ter
AV
0
B"'
AVtern
B;r + B;;(DC)
AA"
BB"
Ax
(3.7.15)
where
'ter
[5Q Urm (DC)/5V term ]
86
The unprocessed section, i.e.
AQ term Y , AQterm(DC)"
vten
Merir
B;;' + B;;(DC)
AA"
BB"
AV tern
(3.7.16)
may then be solved by any method suitable for nonsymmetric matrices.
The values of Ax and AVterm are obtained from this equation and AVterm
is then used in a back substitution process for the remaining AV to be
completed, i.e. eqn. 3.7.14 is solved for AV.
The most efficient technique for solving eqn. 3.7.16 depends on the
number of converters. For six converters or more the use of sparsity storage
and solution techniques is justified; otherwise, all elements should be stored.
The method suggested here is a modified form of Gaussian elimination
where all elements are stored but only nonzero elements processed.8
3.8 Convergence properties

The AEP standard 14-bus test system9 is used to test the convergence
properties of the AC-DC algorithm, with the AC transmission line between
busbars 5 and 4 replaced by an HVDC link.
Eight different control strategies are considered at the rectifier (m) and
inverter (n) ends. The corresponding specified DC constraints are:
1
2
3
4
5
6
7
Q
am P d m
am * dm
am * dm
a m Prfm
am ?dm
am Pdm
am h
vdm
7n
an
an
dn
dn
In
In
n
7n
In
' dn
dn
The DC-link data and specified controls for case 1 are given in Table 3.1.
Initial values for the DC variables x are assigned from estimates for the
DC power and DC voltage and assuming a power factor of 0.9 at the
converter-terminal busbar. The initial terminal busbar voltage is set at
1.0 p.u.

Table 3.1
87
Characteristics of DC link
AC busbar
DC voltage base
Transformer reactance
Commutation reactance
Filter admittance B*
DC link resistance
Converter 1
Converter 2
bus 5
100kV
0.1260
0.1260
0.4780
bus 4
100kV
0.0728
0.0728
0.6290
0.3340 ohms
Control parameters for

case 1
DC link power
Rectifier firing angle (deg)
Inverter extinction angle (deg)
Inverter DC voltage
58.6 MW
7
10
-128.87 kV
*filters are connected to the AC terminal busbar

Note: all reactances are in p.u. on a 100MVA base
This procedure gives adequate initial conditions in all practical cases as

good estimates of Pterm(DC) and Vd are normally obtainable.
With starting values for DC real and reactive powers within 50%, which
are available in all practical situations, all algorithms converge rapidly and
reliably.
In all cases, convergence (to a 0.1 3VTW/MVAR tolerance) is achieved after
4 P iterations followed by 3 QDC iterations.
To investigate the performance of the algorithm with weak AC systems,
two extra transmission lines are added between the converters and buses 4
and 5, their respective reactances being 0.3 and 0.4 p.u. The reactive power
compensation of the filters was adjusted to obtain similar DC operating
conditions to those previously achieved. Under the new configuration, cases
3 and 4 required no further iterations, whereas cases 1 and 2 only took an
extra QDC iteration to converge.
3.9 Modification of the power flow for use with the unit
connection
The use of direct-connected generators to HVDC converters is under
serious consideration.10 The two alternatives, termed unit and group
88
Figure 3.7
Unit-connected HVDC power station, in which generators and converters are integrated into single units, series-parallel and/or parallel
combination of units at the DC side is also allowed
connection and illustrated in Figures 3.7 and 3.8, respectively, are equivalent from the power-flow modelling viewpoint.
In the absence of filters, the voltage at the converter terminal is not
sinusoidal and cannot be used as the commutating voltage in the conventional formulation. Instead, the use of the generator internal e.m.f. behind
subtransient reactance (E") is suitable.
The fast-decoupled power-flow formulation described in previous sections
should be applicable by shifting the converter interface to a fictitious
terminal (the internal e.m.f. behind subtransient reactance) where the
waveform is assumed sinusoidal.
Although E" is neither accessible nor directly controllable and varies with
the load, its magnitude can be derived from the generator's subtransient
Y Y
Figure 3.8 Group-connected HVDC power station, in which machines and converters can be combined in groups via a transfer bus
89
xlc
Figure 3.9
Simplified phasor diagram of the unit-connected generator
phasor diagram for the nominal operating point and then kept constant (as
shown in Figure 3.9). This is done to satisfy the generator's conventional
power-flow specification (i.e. constant voltage) but at the fictitious internal
E" bus. The conventional power flow is then carried out on the assumption
of perfect filtering at that point.
However, the internal e.m.f. behind subtransient reactance is not directly
controllable. Instead, the generator excitation should be controlled to
provide the specified firing angle (amin).
Therefore, the commutating voltage (E") will not be shown in advance
and its magnitude and phase angle (/?)> as shown in Figure 3.9, must be
derived as part of the iterative solution.
The five-variable model of section 3.6 at the rectifier end is replaced by
a set of seven variables
(3.9.1)
Therefore, seven residual equations are needed to formulate the power flow
problem at the rectifier end. 11
The first two equations are common to the conventional model (with the
tap-ratio variable removed by making ar = 1), i.e.
Vd - k^'cosot - ~xcld = 0
n
where
3V2
(3.9.2)
90
Two more equations are derived from the generator subtransient phasor
diagram of Figure 3.9. That is
E - E"cos 0 - (x - x")\Ip\sin(fi + <j>) = 0
E" sin P-(x-
x")\Ip\ cos (p + <t>) = 0
(3.9.3)
(3.9.4)
To complete the set of equations, three variables must be specified, a typical

selection being
I, - 17 = 0
p
(3.9.5)
E- W = 0
(3.9.6)
cos a - cos amin = 0
(3.9.7)
The commutation reactance seen by a unit-connected converter is not

constant either, due to the pole-saliency effect of hydrogenerators. The
actual value of subtransient reactance to be used in eqns. 3.9.3 and 3.9.4 is
load and firing-angle dependent. The differences are, however, very small
and in general an approximate value of
(3.9.8)
is perfectly acceptable.
For greater accuracy, a departure from the conventional power-flow
model is needed. Alternatively, exhaustive dynamic-simulation studies of
the unit-connected rectifier station need to be made only once to derive
capability charts for a given station and these charts, stored in memory, can
be made available during the iterative solution.
3.9.1 An equivalent inverter model

In the absence of an AC system, there is no need for explicit representation
of the sending-end variables and the complete DC link can be represented
as an equivalent inverter. The only restriction from the rectifier end is a
check on the limiting capability of its Vd/Id characteristic.
In this case, the vector equation of independent variables for the unitconnected DC link is
x = \yd, Id, a, cos a, <, cosyf
(3.9.9)
which contains the five variables of the conventional (inverter) set and an
extra variable (cos a) representing the rectifier end of the link.
91
In line with conventional practice, the equivalent inverter will normally be

on extinction-angle control (yn) and the rectifier on current control, with
the specified current level derived from a constant-power setting (P^p) at the
inverter end.
Besides Psf and ym, a third control specification must be made to match
the six-variable formulation. It is suggested that am is used and the inverter
transformer tap position made free. This will provide the highest transmission voltage. Should the tap changer attempt to violate one of its limits,
this limit becomes the new control specification while freeing the value of a.
The vector of independent variables, x, which describes the state of the
DC link at the inverter end, can still be obtained by the application of the
Newton-Raphson algorithm. The set of residual equations, |R|, to be
satisfied upon convergence is
(3.9.10)
3
Vd - &iaVterm cos {n - y) - -xcld = 0
(3.9.11)
(3.9.12)
(3.9.13)
cos (n y) cos (7c ysp) = 0

cosa-cosamin =
(3.9.14)
(3.9.15)
The incorporation of the equivalent inverter into the fast-decoupled ACDC power-flow algorithm requires modification of the B", AA', AA" and BB"
submatrices of the Jacobian matrix in eqns. 3.7.7 and 3.7.8. These submatrices are greatly simplified in the equivalent inverter model as they contain
only a single element each. The new elements are
[AA'] = [<5Pterm(DC)/&]/Vterm
or
[AA']T =
[AA"] = [<5Qterm(DC)/<5x]/Vtern
92
or
[AA"] T =
_2a i B i V t e r m i
[BB"] = <5R/<5Vtern
or
[BB"]T =
3.10 Components-related capability

In conventional power systems the power-flow solutions are subject to
operating-capability constraints associated mainly with synchronous-generator buses.
A typical capability chart for a synchronous generator is shown in Figure
3.10.
_rotor heating limit

. stator heating limit
turbine maximum
power limit
turbine
zero
power
limit
Figure 3.10
real power
'practical
stability
limit
Typical capability chart for a synchronous generator
93
Each operating constraint is represented as a locus on the complex power

plane. The area enclosed by the loci represents the possible range of power
output available from the generator terminals. The chart serves to clarify the
relationships between operating constraints and helps to indicate the
proximity of each constraint while the generator is running.
The same approach is used here to take into account the operating
constraints associated with the DC link. Capability charts are described12
with reference to a simplified form of the New Zealand HVDC link.
However, the principles discussed here can be applied to any two-terminal
HVDC link circuit.
3.10.1 HVDC test system

The simplified circuit shown in Figure 3.11 has been used as a basis for the
development of the capability chart.
The AC network at the Benmore converter terminal consists of an
equivalent hydrogenerator feeding a 16kV busbar, which is connected to
the South Island 220 kV system via a 400 MVA transformer having a 3.3%
reactance. The power base for the entire AC-DC system has been chosen
to be 100 MVA. The voltage level of the 16kV busbar is maintained by the
voltage regulators of the hydrogenerators. These generators can deliver
540 MW or act as synchronous compensators to supply a maximum of
324MVAr.
A bank of harmonic filters absorbs the harmonic currents from the
converter and also supplies lOOMVAr of reactive power. In this model the
filters are located on the 16kV busbar. The behaviour of these filters is
considered to be ideal so that the voltage waveform on the 16 kV busbar can
be assumed to be sinusoidal.
A 750 MVA converter transformer connects the 16kV busbar to the
Benmore terminal of the HVDC link. This transformer is modelled by a
Benmore
16kV
Figure 3.11
Circuit of HVDC link
Haywards
110kV
94
reactance of 2% (on the system MVA base). Since, for the purposes of the
model, it is fed by a sinusoidal voltage source, the transformer reactance can
also be regarded as the commutating reactance of the AC-DC converter.
The converter is modelled as a three-phase 12-pulse bridge rated at 1.2 kA
DC-side current and 525 kV DC-side voltage. The converter station actually
consists of four six-pulse bridges in series.
An overhead transmission line and an undersea cable transmit the power
to an identical converter at Haywards. The large smoothing reactors at each
end of the line are assumed to maintain a constant DC-side current. The
total resistance of the converters, smoothing reactors, transmission line and
cable is 25.56 Q.
Another 750 MVA transformer connects the Haywards converter to the
HOkV Haywards busbar. As at Benmore, the converter transformer reactance of 2% is regarded as being the commutating reactance because a bank
of harmonic filters maintains a sinusoidal voltage on the HOkV busbar.
The voltage level of the HOkV busbar is maintained by the voltage
regulators on synchronous compensators which can provide up to 260 MVAr
of reactive power. This is supplemented by 110 MVAr from the harmonic
filters. The busbar feeds the North Island AC network via a transmission
line.
Similarly to the capability chart of a generator, which represents the
power available from the machine terminals, the most useful information to
be derived from an HVDC-link capability chart is the power received at the
inverter end. In the test system, this is the power available to the Haywards
110 kV busbar from the converter transformer.
3,10.2 Type of HVDC constraint loci

A
B
C
D
E
F
G
H
I
J
K
commutation overlap angle of 60 at Haywards

maximum r.m.s. current in the 750 MVA Haywards converter
transformer
maximum DC-side current of 1.2 kA flowing through Haywards
converter valves
minimum DC-side holding current of 0.1 kA flowing through
Haywards converter valves
maximum DC-side voltage of 525 kV during rectification at Haywards
maximum DC-side voltage of 525 kV during inversion at Haywards
minimum firing angle of 3 during rectification at Haywards
minimum extinction angle of 18 during inversion at Haywards
maximum r.m.s. current in the 750 MVA Benmore converter
transformer
maximum DC-side current of 1.2 kA flowing through Benmore
converter valves
minimum DC-side holding current of 0.1 kA flowing through
Benmore converter valves
L
M
N
O
P
Q
R
S
95
maximum voltage of 525 kV during rectification at Benmore

maximum DC-side voltage of 525 kV during inversion at Benmore
minimum firing angle of 3 during rectification at Benmore
minimum extinction angle of 18 during inversion at Benmore
power available to Benmore 16kV busbar from the 220 kV South
Island busbar
maximum current flow in the 400 MVA interconnection transformer at
Benmore
maximum reactive-power supply of 324 MVAr available from Benmore
generator
maximum reactive-power consumption of 378 MVAr drawn by
Benmore generator
(Benmore is normally rectifying and Haywards is normally inverting.)
3.103 Constraint equations

The constraint loci are derived from the basic converter equations (in per
unit). With reference to Figure 3.12, the following equations are used.
*
(3.10.1)
~~ M>C*D
(3.10.2)
(3.10.3)
(3.10.4)
VDC =
IDC = [aVp/(V2Xc)][cosa - cos (a +
(3.10.5)
The factor K in eqn. 3.10.3 is approximately 0.995 for 12-pulse converter

operation with a commutation angle fj. well below 60.
The loci of constant DC-side current Ipc are given by substituting eqn.
3.10.3 into 3.10.2, i.e.
P 2 + Q 2 = [VpflK(3V2M)IDC]2
(3.10.6)
P+jQ
ysu
A
Xc
Figure 3.12 AC-DC converter
zx
h
dc
dc
96
Eqn. 3.10.6 is used to draw the loci for maximum and minimum valve
currents. These are circles centred on the origin with radii of Vp#K(3^/2/
Îocmax and VpaK(3V2/7t)IDCmin.
The maximum r.m.s. converter transformer current It is equal to the
per-unit MVA rating of the transformer. If the effects of commutation are
ignored, the converter transformer current can be related to the DC-side
current in per unit by
The DC-side current from eqn. 3.10.7 is substituted into eqn. 3.10.6 to
produce a circular locus centred on the origin with radius VpaK(3^j2/n)It/
yj2 describing the maximum converter transformer current. If the effects of
commutation are considered, the locus would be an approximate circle with
a slightly larger radius. Consequently, there is a small margin of safety
inherent in the circular locus described here.
The loci representing maximum DC-side voltage VDCmax are obtained by
combining eqns. 3.10.1, 3.10.2 and 3.10.3 to give
Q = - |P| Vt(V,aK(3%/2/7r)^rDCmax)2 - 1]
(3-10.8)
This is the equation of two straight lines extending from the origin of the
complex power plane. The gradients of the lines are
V[(VpaK(3 V2/7t)/VDCmax)2 - 1]
Two parametric equations are used to describe the locus representing the
minimum firing angle, amin. The variable parameter used is the DC-side
current 1 ^ . The real power co-ordinate of the locus is given by substituting
eqn. 3.10.4 into eqn. 3.10.1 to give
P = - [(3V2/7t)Vpcos(amin)IDC - (3/7r)XcIjy
(3.10.9)
The reactive power co-ordinate is given by substituting eqn. 3.10.3 into eqn.
3.10.2 and using the real power from eqn. 3.10.9 to give
Q = - V[V> 2 K 2 (3 V2M) 2 I^ - P2]
(3.10.10)
The parametric equations are used to draw the locus by gradually increasing
the variable parameter IDC starting from zero. For each new value of IDC, a
corresponding locus-point co-ordinate is given by eqns. 3.10.9 and
3.10.10.
The locus representing minimum extinction angle, dmin, is also parametrically described and is obtained by changing the sign of the power flow in
97
eqn. 3.10.9 and replacing amin with <5min to give

P = (3 v /2/7i)V p cos(^ min ). I DC - (3/7r)X c I^
(3.10.11)
Eqn. 3.10.10 is then used to provide the reactive-power co-ordinate.

Parametric equations are again used to describe the locus of the 60
commutation angle, pi. The variable parameter used is the firing angle
a. Eqn. 3.10.5 is used to calculate a value of DC-side current for each value
of a
IDC =
[aVp/(V2Xc)][cosa - cos(a + TT/3)]
(3.10.12)
This value of current is then substituted into eqns. 3.10.1 and 3.10.4 to
give the real-power co-ordinate
P = - [(3^2/7c)aVpcosaIDC - (3/TC)XCI^C]
(3.10.13)
The reactive-power co-ordinate is obtained from eqn. 3.10.10.

The DC-link power-transfer mapping is obtained by considering the
power loss in the DC-line resistance R
P; = P , - l c R
(3.10.14)
Combining eqns. 3.10.2, 3.10.3 and 3.10.14 provides the point-to-point

mapping relating the real and reactive power at HVDC-link terminals
P ; = P r - (P 2 + Qr2)R/[(3V2/7T)VprflrK]2
2
Q, = - V[(Vpi/Vpr) (,-/flr) (P, +
Qr2)
(3.10.15)
2
- P ]
(3.10.16)
where subscripts r and i refer to the rectifier and inverter, respectively.

A computer-plotting routine becomes indispensable at this stage to
implement the power-transfer mapping. Each locus point on the rectifier
capability chart must be mapped onto a corresponding point on the inverter
chart using eqns. 3.10.15 and 3.10.16.
The locus of power flowing into the Benmore busbar can be found by
analysing the transformer which connects the South Island busbar to the
rectifier busbar. If the variables are regarded as complex values, the power
flowing into the rectifier busbar is
S = (Vs*Vr/Zf) - (V*Vr/Zf)
(3.10.17)
where Vs is the South Island AC-busbar voltage and Zf is the transformer

reactance. As the angle between the two voltages changes, this equation
98
describes a circular locus centred on V*/Z, with a radius given by

|VJ.|V r |/|Z,|.
If the South Island busbar is ignored and the transformer is operated at
its maximum current rating, Ifmax, the power flowing into the rectifier
busbar is defined by
(3.10.18)
= VrI?*
max
This equation represents a circular locus centred on the origin.
3.10.4 Loci of operating constraints

The operating-constraint loci of the Haywards converter are shown in
Figure 3.13. In theory, there is a locus (A) representing the commutation
angle limit of 60 dominating the complex plane. However, such overlap
only occurs during transient conditions and the corresponding locus has
been excluded from Figure 3.13.
The constraints involving current flow are all represented by arcs of
circles. The two ends of each arc correspond to firing and extinction angles
of 0. Locus B is the outermost arc and represents the maximum converter
transformer current. Just within locus B is locus C representing the maximum converter valve current. The innermost arc is locus D which represents
the minimum converter valve current.
The two constraints involving maximum DC-side voltage are both represented by straight lines extending from the origin. Locus E represents
rectifier operating at maximum DC-side voltage and locus F represents
inverter operating at maximum DC-side voltage.
The control-angle constraints are represented by cycloids extending from
the origin. The minimum firing angle during rectification is represented by
locus G, and the minimum extinction angle during inversion is represented
by locus H.
-1000
real power, MW
-1000 1 reactive power, MVAr
Figure 3.13
Magnified loci of Haywards converter
1000
99
This combination of converter constraints restricts the Haywards converter to operating within the area bounded by loci D-F-H-C-E of Figure
3.13. However, it is not sufficient to consider only the constraints of the
Haywards converter. The Benmore converter and Benmore AC system must
also affect the power available from the link. The operating constraints of
the Benmore circuit can easily be referred to the Haywards end of the link
by using the DC-link power-transfer mapping. This is a point-to-point
mapping which relates the complex power entering the link from the
Benmore 16kV busbar to the complex power leaving the link at the
Haywards HOkV busbar. The mapping is carried out using eqns. 3.10.15
and 3.10.16.
The constraint-locus equations of the Benmore converter can now be first
formulated in the same manner as those of the Haywards converter. This
produces a set of Benmore-converter loci which refer to the power entering
the link at the Benmore 16kV busbar. These loci are identical to the
Haywards-converter loci shown in Figure 3.13. The DC-link power-transfer
mapping is then used to transform these loci to produce a new set of
Benmore-converter constraint loci which are referred to the Haywards
110 kV busbar.
The transformed Benmore-converter loci are shown in Figure 3.14. All
the constraints involving current flow are still represented by arcs of circles.
However, the shapes of all loci have been altered by the transformation.
The complex power supplied from the South Island 220 kV busbar to the
Benmore 16kV busbar is dependent on the voltage angle between the
busbars. This power can be represented by a circular locus. Locus P in
Figure 3.15 is a portion of this circle. The power is restricted by the current
rating of the 400 MVA interconnecting transformer. The circular locus O
represents maximum current flow in the transformer. The intersection of
loci P and Q therefore represents the maximum possible power delivery to
Benmore. The arc of locus P that lies within locus Q represents the full
range of power delivery.
-1000
real power, MW
-1000L
reactive power, MVAr
Figure 3.14 Loci of Benmore converter
1000
100

500 r
-500L
Figure 3.15
Power to Benmore from South Island
reactive p<)wer, MVAr
50
Q
real power. MW
1000
Q
1000
5
-500
Figure 3.16
Power to DC link from Benmore 16kV busbar
If the Benmore generator is run as a synchronous compensator, the

generator reactive-power variation of 378 to 324 MVAr will augment the
power delivered from the South Island busbar. Taking into account the
100 MVAr contribution of the harmonic filters, the range of reactive-power
variation becomes 278 to 424 MVAr.
The reactive-power variation is added to each point on the power delivery
arc to produce the loci shown in Figure 3.16. The enclosed region in this
figure represents the power that may be delivered to the Benmore converter
from the Benmore 16kV busbar. The loci R and S represent the reactivepower limits of the Benmore generator. The two loci labelled Q represent
maximum current flow in the interconnecting transformer. The label Q has
been deliberately retained for these loci because it has the same significance
as the locus Q in Figure 3.15.
3.10.5 Complete capability chart

The complete capability chart of the HVDC link is constructed by combining all the constraint loci. The loci of Figure 3.17 are formed by superim-

real power, MW
-1000
101
1000
-1000L
Figure 3.17
Capability chart of DC link
posing Figure 3.13 onto Figure 3.14. This chart describes the power that
may be delivered to the North Island from the link assuming that an
infinitely strong 16kV busbar exists at Benmore. The Benmore currentconstraint loci overlay the corresponding Haywards current-constraint loci
because the two converters are identical.
Manipulation of the control angles at Benmore and Haywards allows the
operating point of the link to vary within the area bounded by loci
KD-L-H-CJ-O-E. When maximum power is being delivered to the North
Island, the operating point is positioned by the intersection of loci CJ and
H. This closely matches the practical operation point.
If the Benmore generators are run as synchronous compensators and the
power-generation capability of the South Island is considered, the loci of
Figure 3.15 must be incorporated into the capability chart. This is achieved
by applying the DC-link power-transfer mapping to the loci of Figure 3.15
and superimposing the transformed loci onto Figure 3.17.
real power, MW
-1000
1000
iliillll
y0
\
CJ
XX
H
_mnr>
Figure 3.18
Capability chart of combined HVDC link and South Island network
102
The resultant capability chart is shown in Figure 3.18. The region of

negative reactive power in Figure 3.16 cannot be transformed because the
converters must always consume reactive power.
The boundaries of the capability chart in Figure 3.18 are defined by the
loci KD-L-Q-R-Q-E. In this case, the maximum power transfer is seen to
be restricted by the current rating of the Benmore interconnecting transformer.
3.11 System-related HVDC capability

The short-term power-transfer capability of a DC link in response to a
change in current order specification comes under the quasisteady-state
condition defined in Section 3.1. In this case, the AC-system e.m.f., minimum commutation-margin angle, tap changer's position, automatic voltage
regulation and the values of shunt capacitors and reactors attached to the
converter are assumed not to have changed when calculating the new
operating points.
3.1 LI Converter powercurrent characteristics13

For a given AC-system configuration, such as that shown in Figure 3.19 and
an initial operating state, determined from a power-flow study, there will be
a unique VJld characteristic.
When considering the inverter power capability, it is assumed that the
rectifier imposes no limitation to the supply of DC current at rated DC
voltage. Each subsequent point following the ld change is calculated by the
same steady-state equations used in the power flow and the resulting
quasisteady-state characteristics give a good indication of the link dynamic
performance.
ESCR
SCR<
Pd
-GD-S
Xc
Id
Qc
UL
Figure 3.19 Simplified representation of an HVDC inverter
Ud
103
The following starting conditions are defined with reference to Figure

3.19
power) = 1 p.u.
UL(AC terminal voltage) = 1 p.u.
Ud(DC voltage) = 1 p.u.
Id(DC current) = 1 p.u.
If the inverter is operated at minimum commutation-margin angle y the

resulting characteristics will represent maximum obtainable power for the
system parameters being considered. This curve is termed the maximum
power curve MPC. Powers below the MPC curve can be obtained by
increasing a and y but power above the MPC can only be achieved if one or
more system parameters are changed, e.g. reduced system impedance,
increased system e.m.f., larger capacitor banks etc.
A similar MPC curve can be obtained for the rectifier at minimum
constant a.
Maximum power curves are plotted on Figure 3.20 for an inverter
connected to AC systems of three different strengths.
1.8
.^
1.6
1.4
1.2 m
SCR =1.5 (ESCR =0.96)
\
Xc = 0.15 pu.y= 18
2.0(1.46) N B , Qc = Qd = 0.54 Pdn at UL = 1.0 pu
^
\
DC Power
^
^
AC Volts
3.0(2.46ir\
^ 1.0
o
0.8
3.0
0.6
0.4
0.2
u
0.2 0.4 0.6 0.8
1.0 1.2
1.4
1.6
1.8 2.0
DC Current (pu)
Figure 3.20
Variation of AC voltage and DC power with DC current

CIGRE, 1992 Reproduced by permission
104
An MPC curve exhibits a maximum value termed the maximum available

power (MAP), the peak of the curve in Figure 3.20. An increase of current
beyond the MAP reduces the DC voltage to a greater extent than the
corresponding DC-current increase. Thus, the rate of change dPd/dld is only
positive for operation at DC currents smaller than IMAP, the current
corresponding to MAP.
For SCR = 3, the operating point A in Figure 3.20 is below MAP and the
current is smaller than IMAP = 1-4 p.u.
For SCR = 1.5, the operating point A is beyond MAP, corresponding to
IMAP = 0.8 p.u. of rated DC current and the value of dPd/dld is negative.
It may appear that there is another possible operating point for
SCR = 1.5 at the left of MAP, point B. However, inspection of Figure 3.20
will indicate that the voltage corresponding to point B for SCR = 1.5 is too
high to be utilised, as indicated by point B'.
When the rated values of Fd, Id and UL (all at 1.0 p.u.) correspond to the
maximum point of the Yd/Id curve for operation with minimum y, then the
corresponding short circuit ratios are termed critical ratios (CSCR, CESCR
and CQESCR), as defined in Appendix II.
In this example, the critical short-circuit ratio, CSCR, is equal to 2 and
the operating point A coincides with the MAP of the curve for SCR = 2.
3.11.2 Converter power limits

If the voltage at the sending end of the DC link is kept constant, the
receiving-end power can be plotted as a function of its AC-terminal voltage
for a given SCR value. Figure 3.21 shows two such curves. One of them,
corresponding to an SCR of 3, achieves the desired operating point A. The
second curve plotted illustrates the result of a sudden change of the SCR
from 3 to 2, following a change in the system configuration.
The new maximum power (MAP) of the curve is lower than the power
before the disturbance at point A. An attempt by the converter to draw an
increasing current, in order to maintain the original power, would result in
voltage collapse.
An important factor, that differs in the DC case as compared with the
pure AC configuration, is the controllability of the DC converter, which can
prevent voltage instability and enhance the performance of the AC system.
For these reasons, it is convenient to plot power against DC current, the
controlled quantity, rather than against the voltage. The power flowequations can still be used to determine the quasisteady-state power curves
and, thus, the power limits of the inverter (or rectifier).
An example of the power limitations of an inverter is shown in Figure
3.22 for a sudden change of SCR from 3 to 2. Prior to the change, if the
operating point A of the MPC curve is close to MAP-1, a relatively
moderate change in the AC-system voltage may result in a MAP-2 lower
than the rated power.
0.2
0.4
0.6
0.8
1.0
1.2
105
1.4
Figure 3.21 AC voltage versus DC-power variation

Xc = 0.15pu,Y=18
Qc = Qd = 0.54Pdn at U L = 1.0 pu
1.6
1.4
SCR = 2.0
MAP-1
1.2
B
Pd DC Power
UL - - AC Volts
1.0
o
o 0.8
MPC-1
SCR = 3.0
CL
0.6
MPC - 2 \
SCR = 2.0
0.4
0.2
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
DC Current (pu)
Figure 3.22 AC-DC systemlow SCR power and AC-voltage curves sudden
change of SCR from 3 to 2
106
The required power cannot be maintained but the current limit imposed
by the rectifier prevents the collapse of the system despite the demand for
higher power from the master controller.
The operating point B on MPC-2 will be in the region where dd/dld
is negative and stable operation would not be achieved in power-control
mode but stable operation continues in constant-current control mode.
If the DC system is used to control AC frequency or to provide system
damping, operation in the constant-current mode would not be acceptable
except for a very brief period of time; e.g. while the fault is being cleared.
3.11.3 DC link power-transfer capability

In the previous section, the inverter-power capability was calculated under
the assumption that the rectifier operation does not impose further limitations. However, the rectifier is subject to similar limitations to the inverter;
at increasing DC current the voltage will reduce and the amount of control
in hand depends on the value of a in normal operation.
In this section, an example is given which takes into account both ends of
a DC link, initially operating with the parameters shown in Table 3.2.
The effect of a five percent AC-voltage reduction at the rectifier end (case
1) or inverter end (case 2) has been calculated using the quasisteady-state
solution, i.e. after converter controls have settled but before tap changers or
capacitors have switched.
The results are shown in Table 3.3.
For case 1 the rectifier a reaches its minimum of 3 and the inverter takes
over control of current (and power) by increasing its y which increases its
reactive-power consumption, hence reduces its AC-busbar voltage.
Power is reduced by up to three per cent for case 1, otherwise power
control is maintained (except for a brief dip until the master control takes
effect).
The AC voltage on the remote busbar is only slightly depressed.
The results in Table 3.3 are to some extent a compromise because they
depend on the values used for the current limits (1.05 and 1.0 in the above
Table 3.2
ESCR
K
a
e.m.f.
Example data
Rectifier
Inverter
2.75
15%
12
1.05
1.00
14%
25
1.17

Table 3.3
107
Calculated solution after voltage depression
Case
Power
p.u.
AC voltage on the
remote busbar
1
2
.970
1
.990
1
udc
.970
1
ld
a
rectifier
7
inverter
1
1
3
12
26.7
18.4
case). For example, an increase in current limit by 0.05 p.u. for case 1 would
cause the power to be maintained at 1 p.u. (instead of 0.97 p.u.) but with a
larger reduction of remote AC voltage, by 3.2 per cent instead of one per
cent. In other cases an increase in current limit would actually reduce power,
with a further voltage depression, hence would be pointless.
In the final steady state, the capacitor switching and tap-changer control
would restore the specified conditions.
3.12 Summary
Using the basic converter steady-state formulation developed in Chapter 2
for power frequency and symmetrical operation, this Chapter has described
the incorporation of AC-DC converters, point-to-point DC links and
multi-invertor schemes in the conventional power-flow solution.
The overall convergence rate of the AC-DC algorithms depends on the
successful interaction of the two distinct parts. The AC-system equations are
solved using the well-behaved constant tangent fast-decoupled algorithm,
whereas the DC-system equations are solved using the more powerful, but
somewhat more erratic, full Newton-Raphson approach.
In general, those schemes which acknowledge the fact that the DC
variables are strongly related to the terminal voltage give the fastest and
most reliable performance.
In practice, converter operation has been considered down to an SCR of
3. A survey of existing schemes shows that, almost invariably, with systems
of very low SCR, some form of voltage control, often synchronous condensers, is an integral part of the converter installation. These schemes are,
therefore, often very strong in a power flow.
It may, therefore, be concluded that the sequential integration should
converge in all practical situations although the convergence may become
slow if the system is weak in a power-flow sense.
An alternative model has been proposed for the integration of unitconnected schemes.
Constraint loci have been developed to take into account the limited
capability of the converter plant. The question of system-related DC-link
108
power-transfer capability has also been considered, using the maximum

power curve (MFC) and maximum available power (MAP) concepts.
3.13 References
1 VAN NESS, J.E., and GRIFFIN, J.H.: 'Elimination methods for load-flow studies',
Trans. AIEE, 1961, PAS-80, pp. 299-304
2 STOTT, B., and ALSAC, O.: 'Fast decoupled load flow', IEEE Trans., 1974,
PAS-91, pp. 1955-1059
3 BROWN, H.E., CARTER, G.K., HAPP, H.H., and PERSON, C.E.: Tower flow
solution by impedance matrix iterative method', IEEE Trans.,, 1963, PAS-82, pp.
1-10
4 TINNEY, W.F., and HART, C.E.: Tower flow solution by Newtoh's method',
IEEE Trans., 1967, PAS-86, (11), pp. 1449-60
5 STOTT, B.: 'Decoupled Newton load flow', IEEE Trans., 1972, PAS-91, pp.
1955-1959
6 TINNEY, W.F., and WALKER, J.W.: 'Direct solutions of sparse network equations by optimally ordered triangular factorization', Proc. IEEE, 1967, 55, (11),
pp. 1801-1809
7 ARRILLAGA, J., ARNOLD, C.P., and HARKER, B.J.: 'Computer modelling of
electrical power systems' (John Wiley & Sons, 1983)
8 BODGER, B.P.: 'Fast decoupled ac and ac-dc load flows'. PhD thesis, University
of Canterbury, New Zealand, 1977
9 IEEE Computer Applications Sub-Committee Standard Test System, American
Electric Power Service Corporation, 1962
10 CAMPOS BARROS, J.P., ARRILLAGA, J., BOWLES, J., KANNGIESSER, K.W.,
and INGRAM, I.: 'Direct connection of generators to HVDC converters: main
characteristics and comparative advantages', ELECTRA, August 1993, 149, pp.
19_39
11 ARILLAGA, J., ARNOLD, C.P., CAMACHO, J.R., and SANKAR, S.: 'AC-DC
load flow with unit-connected generator-converter infeeds', IEEE Trans., May
1993, PS-8, pp. 701-6
12 DE SILVA, R., ARNOLD, C.P., and ARRILLAGA, J.: 'Capability charts for an
HVDC link', Proc. IEE, May 1987, 134, (3), pp. 181-6
13 CIGRE Working Group 14.07: 'Guide for planning dc links terminating at ac
systems locations having low short-circuit capacities. Part I: AC/DC interaction
phenomena'. Report 68, June 1992
Chapter 4
The harmonic solution
4.1 Introduction
Chapter 2 (section 2.8) has described a model of the AC-DC converter in
the harmonic domain which, for a given set of switching instants, converter
terminal voltage and DC-current waveforms, can calculate the returned
currents.
currents.
In practice, however, the terminal-voltage waveforms are part ot the
overall solution which involves not only converter variables, such as controller characteristics and firing-angle constraints, but also the system voltage
sources, current sources and impedances. Iterative techniques are thus
necessary to solve all these variables together to reach a final correct
solution.
. .
.
.
The simplest iterative scheme used to simulate harmonic interactions is
based on fixed-point iterations. At each iteration the latest values of the
distorted converter terminal voltages are used as the commutating voltages
for the converter solution.
The calculated DC-voltage waveform is then impressed upon the DC-side
impedance to derive the DC-current waveform, which in turn, together with
the switching instants and commutation process, provides the AC-side
current harmonic injections. The latter are then used to derive the ACvoltage harmonics in the frequency domain for the next iteration.
However, the fixed-point iteration solution diverges when the AC-system
harmonic impedance is large and the commutating reactance small. This
Chapter describes the use of the Newton method to model the interaction
between the converter and the AC and DC systems.
4.2 Basic AC-DC system

The description of the full Newton solution is made with reference to the
system shown in Figure 4.1, which is derived from the CIGRE benchmark
The AC system is represented by a three-phase Thevenin equivalent and
the DC system by a T-circuit equivalent and a DC source. This test system
110
J J
\
controller
YgD
uc
alpha order
\\ filter
VD
Figure 4.1
l'-a
Twelve-pulse AC-DC system model
is operationally difficult in that it consists of a weak AC system resonant at

the second harmonic, coupled via the converter to a fundamental resonance
on the DC side. Further details of the test system are given in Appendix III.
The variables to solve in Figure 4.1 are as follows:
V, the three-phase harmonics of the converter terminal voltage;

Vp, the voltage at the equivalent star-g/delta transformer primary, on
the converter side of the lumped commutation resistance; as shown in
Figure 4.1, the transformer and thyristor conduction resistances have
been lumped into an equivalent resistance between each bridge transformer and the AC terminal of the 12-pulse converter;
Vp, as above for the star-g/star transformer;
Is, 1^, the phase currents flowing into the secondary windings of the
star-g/star and star-g/delta transformers, respectively;
Ip, Ip, the primary phase currents for the two transformers;
Vdjk, Vd0, the DC-voltage harmonics and DC component, respectively;
ldk, Id0, the DC-current harmonics and DC component, respectively;
F a0 , the average delay-angle mismatch equation;
F^, F%, the firing-angle mismatches of the thyristors in the bridges
attached to the star-g"/delta and star-g/star transformers, respectively;
F^, F|j, the end-of-commutation mismatches in the bridges attached to
the star-g/delta and star-g-/star transformers, respectively.
4.3 Functional notation of the converter equations

Chapter 2 has described the equations interrelating the converter components in the steady state. A functional notation3 is now introduced to
describe these relationships expressed in terms of the variables shown in
111
Figure 4.1. The starting point will be a relationship between the converter
phase currents and the converter terminal voltage.
Given the primary phase currents, the terminal voltage can be found by
(4.3.1)
m * = [ Y j ^ - t i p + i & + [Ytc]k\yth]k)
where the square brackets denote a three-phase quantity. The AC-system

impedance, [Y^1,
is calculated by inverting the sum of the admittances
attached to the converter bus. These are the filter admittance, Yfilter, and the
Thevenin source admittance, Yfc. This equation will be represented over all
phases and harmonics by
(4.3.2)
V=f1(I%I$
The converter terminal voltage is then related to the equivalent transformer

primary voltages by the voltage drops through the equivalent commutating
resistances
= [V]* - [RS][Ip]*
(4.3.3)
= [V]k - [R ][I^
(4.3.4)
The commutating resistance matrix is assumed frequency independent and

diagonal but possibly unbalanced. The use of phase currents from the
previous iteration in eqns. 4.3.3 and 4.3.4 leads to the derivation of a
smaller nonlinear system in section 4.4. This is denoted by the primed
quantities in these equations. Using the function notation again, where
phase currents from the previous iteration are treated as parameters, not
variables
V* =/ 2 (V)
(4.3.5)
V? =/ 3 (V)
(4.3.6)
Chapter 2 has described the derivation of the DC voltage across a six-pulse

bridge attached to either a star or delta-connected AC source, with an
inductive source impedance. Using the transformer models of section 2.8.4,
the primary voltages, Vp and Vp, can be transformed into equivalent star or
delta-connected inductive sources on the secondary side. The DC voltages
across each group are then added to obtain the total DC voltage and the
constant forward voltage drop associated with each group is subtracted from
the total DC-side voltage direct component. The calculation of the DC
voltage is again represented in functional notation
* =/ 4 (vS,v i*. ef, ef, <t>l
ao =/ 5 (V, V?, ldk,flf,Of,
tf)
tf,
tf)
(4.3.7)
(4.3.8)
112
The DC voltage is functionally dependent on the switching angles since

these define the limits of the convolution analysis used to calculate the
DC-side voltage across each group. The DC-current harmonics are present
in the calculation of the voltage samples which are convolved with the
sampling functions. The DC voltage, when applied to the DC system, yields
the DC-side current. For example
I^VA
(4.3.9)
ho = (Vô - 4Vfwd - E^Y.o
(4.3.10)
where Vfwd is the constant forward drop through a stack.

Summarising for any topology
(4-3.12)
The DC current, switching angles and primary transformer voltage can be

used to calculate the transformer secondary-phase currents applying the
analysis of section 2.8.4 to each bridge
(4.3.13)
I = / 9 ( V , I * , I d0> Of, <t>f)
(4.3.14)
The transformer analysis of section 2.8.4 then describes how the primary
currents are obtained
IP=/IO(IS)
I?=/u(V?,l)
(4.3.15)
(4.3.16)
The following expressions, derived in Chapter 2 for the converter switching

angles, the power control and the average delay angle, are now written in
the form of mismatch equations, i.e.
K^fufyUa, 0?,$)
(4.3.17)
1%=fi3(yZ,hk,tf,4>?)
*Wi4fl.I>. 0?>o)
tf=fis(hk>ho,0?>*o)
F s =/ 16 (V d0( I d0 )
(4-3.18)
(4.3.19)
(4.3.20)
(4.3.21)
Fo=/i7(VdO.I,/o)
(4.3.22)
Seventeen functional relationships have been described, representing 2328
113
Table 4.1 Functional relationships between 12-pulse rectifier variables

Variable
V
vPD
Vdk
OK
c
CO CO
I*
's
Ip
Ip
F^
?
r
e
F?
Fs
Fa,
No. of vars
300
300
300
100
1
100
1
300
300
300
300
6
6
6
6
1
1
Function of
'i(lp,lp)
400 '
400s
f (\i
\#D 1 jk>
r\S r\O
"/ > "/
iS
Y^/
JD\
Wi )
'5V"P "P Viki @j > Oj i 0 / Y^/ )
f6(Vdk)
^(vdo)
fe(V% [dk, \6
f g (Vp, \dk, \0f0,
f10(\l)
fu (Vp, IQ)
f12(Vp, 1 ^ t?f,
'\4\'dk>
'd0' ^
f,5(\dk, l d 0 , t >?,
0f, 0,D)
0f)
oc0)
f^dO, ld0)
equations in as many unknowns, for a solution up to the 50th harmonic.

These relationships are summarised in Table 4.1.
4.4 Mismatch equations

There is a great deal of redundancy in the system of equations summarised
in Table 4.1. By a variety of substitutions of functions for variables, the
number of simultaneous equations and variables can be reduced to 426. For
example, taking fx and substituting functions for the variables Ip and IP
, ho, 01 <^)),
The new function,/18, is a composition of several functions which describe
114
how the DC current and terminal voltage, together with the switching
angles, are used to calculate the primary phase currents. The primary phase
currents are then injected into the AC-system impedance to yield the
terminal voltage. The relationship
V =/ 18 (V, I*, I, o , 9f, 4> $, 0?)
(4.4.2)
can be written as a mismatch equation suitable for use in Newton's method

Fv = V -/ 1 8 (V, Idk, I, o , Of, $?, 0?, $ )
(4.4.3)
The voltage-mismatch equation (4.4.3), when decomposed into phases,

harmonics and rectangular components, yields 300 real equations. The
DC-side current is expressed as
=f19(y,idk,o?,<p?,es,ti)
(4.4.4)
which leads to the following mismatch equation:

F M = la - / 1 9 (V, Iik, I, o , Of, tf, flf, 4>1)
(4.4.5)
Eqn. 4.4.5 yields 100 equations when decomposed into harmonic real and
imaginary components and is a composition of functions which describe the
calculation of the DC voltage and its application to the DC-system model,
to yield the DC current. A further 26 mismatch equations are obtained in a
similar manner, related to the converter controller and the switching
instants
=fzo<y,l, $.<!>?)
(4-4.6)
(4-4.7)
( 4 - 4 -8)
(4-4.9)
115
(4-4.10)
V?, Idk, 0?, 0?, $ , #
3(V),
I*, fl?,
=/22(V, I*, 1,0. 0?, 0?, ^
tf)
(4-4.11)
=/ 2 3 (V, I * , I d0 , 0?, 6, fa c/>f)
(4.4.12)
A smaller system of 426 simultaneous mismatch equations in 426 variables

has now been developed and is summarised in Table 4.2. The reduced set
of variables to be solved for consists of the AC-terminal voltage, the DC
current, the switching angles and the average delay angle.
It would be possible to solve for a different set of variables; however, those
chosen have the advantage of being less distorted. In particular, the AC-side
terminal voltage is less distorted than the phase currents and the DC-side
current is less distorted than the DC voltage. In fact, a fundamentalfrequency AC-DC power flow will give a very reasonable estimate of the
fundamental voltage component on the AC side, and the DC-current
component on the DC side.
The interaction of the converter with the AC system has been specified in
terms of terminal-voltage mismatch. This requires the injection of phase
currents into the AC-system impedance to obtain a voltage to be compared
with the estimated terminal voltage.
The AC-system interaction can also be expressed in terms of a current
mismatch. The estimated terminal voltage is applied to the AC-system
admittance to obtain phase currents that are compared with phase currents
calculated by the converter model using the estimated voltage
F rv irvi
rv irv 1 f (f (\r\ i
#s rfi\
f (f (\r\ T
PP
(4.4.13)
Table 4.2 Mismatches and variables for the 12-pulse rectifier
Variable
No. of vars
Fv
300
lld
100
Function of
V - f 18 (V, \dk, 1^, Of, (t>f, 9f, </>f)
^15('d/c' 'dO' ^/ '
y v . i* u fl?, 0, ^f, 0)
o)
y v , i * U e?, <>?, ^f, 4>)
116
Note that the current mismatch is still expressed in terms of the same
variables as the voltage mismatch. The current mismatch has the advantage
that it does not require the system admittance to be inverted, a possible
difficulty if it has a high condition number. The current mismatch is also
the preferred method of modelling the interaction with a purely inductive
AC system, such as the unit connection, as the system admittance will
decrease with increasing harmonic order. Only the voltage mismatch will be
implemented here, as the AC-system admittance is usually invertible, and
the AC-system impedance is typically much less than one per unit.
The DC-current mismatch, F w , defining the interaction with the DC
system, can also be written as a DC-voltage mismatch, F yd . This mismatch
is obtained by injecting the estimated DC current into the DC-system
impedance and comparing the resulting voltage with the calculated DC
voltage
,,, 0?, 6?, tf, $?)
(4.4.14)
The DC-voltage mismatch would be preferred in the unlikely instance of a

capacitive DC system. Only the current mismatch has been implemented in
this Chapter, as the DC-side admittance to harmonics is generally less than
one per unit owing to the presence of a smoothing reactor.
4.5 Newton's method

The mismatch equations and variables obtained in section 4.4 are a mixture
of real and complex valued. Newton's method is implemented here entirely
in terms of real-valued equations and variables. All complex quantities are,
therefore, converted into real form by taking the real and imaginary
components. A decomposition into real form is required in any case, since
the AC-voltage and DC-current mismatch equations are not differentiable
in complex form. Newton's method is implemented by first assembling the
variables to be solved for into a real vector X
X = [R{\}, /{V}, R{Id}, /{I,}, 6, 4>, a0, I d0 ] T
(4.5.1)
The mismatch equations are likewise assembled into a real vector

F(X) = [fl{Fv}, /{F v }, R{ld}, /{F H }, Te, F,, F.o> F W0 ] T
(4.5.2)
Given an initial estimate of the solution X, Newton's method is used for

finding the solution vector, X *, which causes the mismatch vector to be zero
F(X') = 0
(4.5.3)
117
The iterative process is defined by

F(X N )=J N Y
X(N +
i) =
(4.5.4)
-Y
(4.5.5)
with convergence deemed to have occurred when some norm of the residual
vector F(XN) is less than a preset tolerance. Newton's method is not
guaranteed to converge, but convergence is likely if the starting point is
close to the solution.
4.5.1 The Jacobian matrix

Central to Newton's method is the Jacobian matrix, J N , of partial derivatives. The base system under consideration contains 426 equations and the
Jacobian is 426 elements square, as it contains the partial derivative of every
mismatch function, with respect to every variable. This is illustrated for the
12-pulse converter functions and variables in Figure 4.2, for a system with
constant current control.
There are two methods for obtaining the Jacobian elements; numerical
partial differentiation and the evaluation of analytically-derived expressions
for the partial derivatives.
Numerical differentiation
Numerical calculation of the Jacobian has the advantage of ease of coding
but is quite slow. Each column of the Jacobian requires an evaluation of all
the mismatch equations and the resulting calculation is only an approximation to the partial derivative. The numerical Jacobian is obtained by
sequentially perturbing each element of X, and calculating the change in all
the mismatches
SiJ
AF,AX,.
F(* + A * ) - F ( x )
Ax
5bJ
Provided that AXj is small enough, this gives a good approximation to the
Jacobian.
The numerically-derived Jacobian for the test system has been plotted in
Figure 4.3, for a solution up to the 13th harmonic. This Jacobian was
calculated at the solution, and so represents a linearisation of the system of
equations in Table 4.2 around the converter operating point.
Referring to Figure 4.3, the elements of the Jacobian have been ordered
in blocks corresponding to the three phases of terminal voltage, the DC
current, the end-of-commutation angles, the firing angles and the average
delay angle. The blocks associated with interactions between the DC-current
harmonics and the AC-voltage harmonics comprise the AC-DC
118
--
co
^T
c2
a.*
S~2
s
fid
CO
-U
""!
J5
>
A.*
co*
fa
HH3
CO
s
s
-S 2
d fit
fit
c^ g
s-2
CO
v-'
s
c2
fa^ N J fa^
co
^ fit
NJ
CO
CO
*m*r
CO
CO
CO
fe
cS
co
1
J
^ t fa
od ^T
co
CO
c^
% &
CO
HH3
cS
fa
fa
N? k j
CO
co
s
co
CO
CO
CO
I co
'fid
c^
^ r ftj
co I co
rig 3|S :
'Is S'
S s"
9 fe*
CO
8 28 ^ Es* 83 V j
w
& r, fa
CO
h-< fit
c^
p P 2 P P ^ p 9 P fit
si s s" s*i s s
s" S S s"
co
8 2P JP 3 P 5
P JS
CO
CO
2
s
CO
co
CO
82
8 2p J
2P
SN
s" s"
8 28 2
119
Numerical Jacobian
mismatches
Figure 43
variables
Numerically-calculated Jacobian for the test system; 13 harmonics
partition, which is 104 elements square. All other parts of the Jacobian are
called the switching terms. Within the AC-DC partition, elements have
been arranged within each block in ascending harmonic order, with the real
and imaginary parts of each harmonic alternating. Each block in the
AC-DC partition is, therefore, 26 elements square in Figure 4.3 but 100
elements square for a solution to the 50th harmonic.
The Jacobian displays several important structural features:
A The test system contains a parallel resonance in the AC system at the
second harmonic. This leads to rows of large terms in the Jacobian
aligned with the second-harmonic terminal-voltage mismatch (the resonance terms).
B A change in harmonic k on one side of the converter affects harmonics
k + 1 and k 1 on the other side of the converter, causing the double
diagonal structures in the AC to DC and DC to AC blocks. These are
the three-port terms.4
C The end-of-commutation mismatch is very sensitive to harmonics in the
terminal voltage and DC current.
D The individual firing instants are sensitive to harmonics in the DC
current.
120
AC-DC
power system analysis
E The average delay-angle mismatch, since it relates to the average DC

current, is extremely sensitive to changes in the fundamental terminal
voltage. There is also some sensitivity to harmonics coupled to the
fundamental: i.e. the 11th and 13th harmonic on the AC side.
F As would be expected, there is strong coupling between the switching
angles and the switching mismatches.
G The Jacobian contains a strong diagonal, approximately equal to one in
the AC-DC partition. A large diagonal is often beneficial when the
linear system corresponding to the Jacobian is solved.
There is also, apparently, no dependence of the terminal-voltage mismatches on the end-of-commutation angles, even though there is a sensitivity to the firing angles. A change in 6 moves the entire commutation-current
curve, whereas a change in cj) moves only the end of commutation, consequently having a negligible effect. This seemingly anomalous result is due
to the formulation of the mismatch equations, and a different behaviour
would be observed with a different formulation of the mismatch equations.
The most useful feature of the Jacobian, however, is the large number of
small elements. Since the Jacobian is only an estimate of the behaviour of
the nonlinear system in response to small perturbations, it is acceptable to
Va
Vb
Id
100
120
20
40
60
80
100
I,
Figure 4.4
Sparsity structure of the Jacobian; 13 harmonics
120
Va
Vb
121
Vc
Id
120
r
20
40
60
80
100
120
Get
Figure 4.5
Scan structure of the sparse Jacobian; 13 harmonics
approximate elements in the Jacobian without affecting the convergence of

Newton's method. Small elements may, therefore, be approximated by zero,
making the Jacobian sparse. The sparsity structure of the Jacobian is
illustrated in Figure 4.4, and again the structural elements described above
are evident. For a solution up to the 50th harmonic, the Jacobian is typically
96% sparse. The sparsity structure of Figure 4.4 was obtained by scanning
through the Jacobian, analytically calculating selected elements, and retaining those elements in the AC-DC partition larger than 0.05, and switching
elements larger than 0.02.
In scanning the Jacobian, it is not necessary to calculate all elements. On
the AC-DC side of the converter an odd harmonic never couples to an even
harmonic, and for transfers across the converter an odd harmonic only
couples to an even harmonic. The total time to scan the AC-DC partition
can, therefore, be halved by scanning with the checkerboard pattern of
Figure 4.5.
Analytical derivation
The analytical method of calculating the Jacobian matrix requires considerable effort to implement but is exceptionally fast. Frequently, the amount of
computation required to calculate the analytic Jacobian is of the same order
122
as that required to calculate the complete set of mismatches just once. For
the converter system of 426 mismatch equations, the analytic derivation of
the Jacobian is 20 times faster than the numerical derivation. A full
derivation of the Jacobian elements can be found in Reference 5.
As an illustration of the analytical derivation, the elements corresponding
to the end-of-commutation mismatch equation are described here.
The current in the phase that is commutating off at the end of the
commutation should be zero, and is given by
= 0
(4.5.7)
for a commutation on the positive rail, where

(4.5.8)
(4.5.9)
jkcoLeIdk~Vekb
*-rilr
jkcoL*eb
t
(4.5.10)
The end-of-commutation mismatch equation is, therefore, a function of all

the variables except the average-delay angle a0. A commutation on the
negative rail is accounted for by substituting ldm in the above equations
for Idm.
With respect to AC phase-voltage
variations
Differentiating eqn. 4.5.7 with respect to an arbitrary voltage phase and

harmonic yields
dR{Vdm}
dR{V3m}
dR{V6m}
dI{Vdm}
dl{\dm}
(4.5.11)
and
dI{Vdm}
(4.5.12)
The partial derivatives in these equations are obtained from eqns. 4.5.8 to
4.5.10 as
(4.5.13)
(4.5.14)

aD
dI{V*m}
123
ejm
(4.5.15)
(4.5.16)
It is assumed in the above analysis that VJJ, corresponds to a phase-ending

conduction. Multiplying by 1 gives the required partial derivative for the
case that Ydm corresponds to a phase-beginning conduction.
With respect to direct current ripple variation
The analysis is similar to that for AC phase-voltage variation. Differentiating
eqn. 4.5.7 with respect to DC ripple yields
VIX
\Kdmi
UMX K
\ dml
urL L
\ dm}
(4.5.17)
)
and
The partial derivatives in these equations are obtained from eqns. 4.5.8 to
4.5.10 as
(4.5.19)
L. + L,
Jk0i
dl{ldm}
Le + I ,
T = v^h-
(4.5.20)
(4.5.21)
(4-5.22)
This analysis assumes that the commutation is on the positive rail. A similar
analysis holds for a commutation on the negative rail but with \dm
substituted into eqn. 4.5.7.
With respect to end-of-commutation variation

This partial derivative gives the effect on the residual commutating-off
current at the end of the commutation, if the end of commutation is moved.
124
It is obtained simply by differentiating eqn. 4.5.7 with respect to 0,. to yield

^
=l
(4-5.23)
J
respect to firing-instant variation

The DC offset to the commutation, D, is a function of the firing instant, and
so the effect of 8t on F^,- is through D. Differentiating the expression for D,
eqn. 4.5.8 gives the required partial derivative
4.5.2 Sequence-components model

A useful improvement to the performance of the converter model is
obtained by applying linear transformations to the vector of mismatches and
variables at each iteration of the Newton solution. The objective is to
improve the sparsity of the Jacobian matrix by transforming to a new system
of mismatches and variables which is more diagonal.
Balanced against the improved Jacobian sparsity of this method, is the
extra calculation overhead associated with calculating the transform at each
iteration. It is, therefore, not feasible to completely diagonalise the Jacobian
matrix by calculating its eigenvalues, primarily because the matrix of
eigenvectors which describes the diagonalising transform is the same size as
the Jacobian matrix, but full.
Applying either the sequence or dqO transforms to quantities on the AC
side of the converter yields a considerable improvement in Jacobian sparsity, with insignificant transformation overheads. Under either of these
transformations, the zero-sequence component is completely diagonalised,
unless the star-g/delta transformer is unbalanced. Additional frequencycoupling terms between the AC and DC sides are also removed in the
sequence transform, since a harmonic k on the DC side couples mainly to
harmonics k + 1 in positive sequence, and k 1 in negative sequence on the
AC side.
Application of the dqO transform is not warranted since, in the steady
state, the dqO transform is just a sequence transform followed by a rotation
of the positive sequence into the direct and quadrature axes, and a rotation
of the negative sequence into the conjugate of these axes. Unlike the
synchronous machine, where the direct and quadrature axes are aligned
with the rotor, there is no such preferred phase reference for the converter.
It would, therefore, be necessary to choose a direct axis at every harmonic,
which introduces the least coupling between the direct and quadrature axes
quantities and other variables.
125
Solving power-system elements in sequence components affords a common frame of reference within which to model the system as a whole. In
such a case, it is necessary to integrate machine models, the load flow,
HVDC, FACTS and transmission-system components into a single iterative
solution. In addition to being more computationally efficient, the sequencecomponents frame of reference leads to greater insight into the interactions
between component nonlinearities, communications interference and power
quality.
In the existing model, interaction with the DC system is specified by
summing the DC-voltage harmonics across each bridge and applying the
resulting voltage harmonics to the DC-system linear model. The resulting
harmonic current
I' = /* (V, I , 0?, $f, 0*, <$)
(4.5.25)
should be equal to the DC-current harmonic ripple, Idk. The DC-side

current mismatch equations can therefore be written
** = hk ~ lit
(4.5.26)
A similar equation is applied on the AC side, by summing the phase

currents from each bridge and injecting them into the AC system. This
yields voltage harmonics
Yk =/ 1 8 (V, ldk, 0?, (f)f, 8j, (j)J)
(4.5.27)
which should equal the estimated converter-terminal voltage harmonics, Vk.

The AC-side voltage-mismatch equations can therefore be written
F V k = Vk - Vk
(4.5.28)
A sequence-components solution is implemented by using the sequencetransform matrix, T, to interface complex phase-component calculations to
the Newton solution in sequence components. For complex phasors, the
complex 3 x 3 sequence transform is
1 a2
1
JV
a*
(4.5.29)
126
where a = eJ2n/3. Setting
v
V"
as the sequence components yields W = TV and V = T 1W. If a threephase quantity has been decomposed in real rectangular components, as in
the Jacobian matrix, a real components-sequence transform matrix, T, can
be constructed
W = TV
which in full is
"l
v? 0
VR" 1
V," 0
V
1
R
+ 0
v,
v?
Ra
-la
VR
0 Ra
-la
VR
la
Ra
la
Ra
vf
Ra
-la
Ra2
-la2
VR
la
Ra
la
Ra
(4.5.30)
VJ
where R, I denote the real and imaginary parts of a quantity. The existing
phase-components mismatch equations summarised in Table 4.2 can now
be written as sequence-component mismatches, in terms of the sequencecomponents terminal voltage, using the sequence-transform matrix. Sequence-component mismatches for the 12-pulse controlled rectifier are
listed in Table 4.3.
Table 4.3 Mismatches and variables for the 12-pulse rectifier with the terminal
voltage in sequence components
Eqn.
\dk
Y
F?
r
Functional notation
k(T-1w, i* ef, tf)

F 23 (T- 1 w, i ^ \M, ej,
127
128
Table 4.4 Convergence and performance of the converter model, (b) constant unified
Jacobian, (d) constant sequence components Jacobian
Test no.
CPU time (seconds)
Main iterations
1fc
11.1
10.9
14.8
15.0
12.3
11.5
6
5
11
12
7
6
0.1712
0.0851
0.2562
0.2973
0.1130
0.1056
Id
2b
2d
3b
3d
The Jacobian matrix for a sequence-components solution is readily

obtained from the phase-component partial derivatives by application of the
chain rule
=I-T
The resulting Jacobian matrix is plotted at the bottom of Figure 4.6, and
shows a greater degree of sparsity than that for the phase components,
shown in Figure 4.3. This is primarily due to the absence of any coupling
between the zero-sequence AC-voltage harmonics and any of the other
converter variables.
Convergence of the sequence-components solution is compared with that
of the constant Jacobian phase-components solution in Table 4.4, for three
of the test cases. They are very similar, indicating that the converter model
can be directly combined with other components (such as the synchronous
machine) modelled in sequence components.
4.6 Computer implementation

In this section the application of Newton's method to the case at hand is
described in detail. Several issues are addressed that have not yet been
discussed. Of particular importance is the method of determining a suitable
starting point for the Newton method, the updating of the Jacobian matrix,
129
calculate switching mismatches

Yes
update switching
system variables
update system variables
update secondary variables
(a)
Figure 4.7 Flow chart for the sparse Newton solution

a Main harmonic system
b Switching system
the sparse solution of the linear Jacobian system and the stopping criteria
for the iterative process. These points are illustrated in the flow diagram for
the solution (Figure 4.7), where it can be seen that a two-stage process is
employed to calculate the starting point. A first estimate of the converter is
obtained by using a classical analysis, followed by a Newton solution of the
switching system, with no harmonics. If the switching system converges, a
full harmonic solution follows, after which the results are printed to output
files.
130
4.6.1 Initialisation
An initial estimate for the converter delay angle is obtained from the
equation
V,o = |V ( f c l |cosa- I, o
n
n
(4.6.1)
ignoring voltage-magnitude drop through the AC-system impedance. The

DC voltage is estimated from the voltage drop through the DC system and
the DC source
(4.6.2)
d0
The average commutation angle is obtained from

V,o = |VJ[cos a + cos(a + /*)]
n
(4.6.3)
These angles are then used to assemble the individual firing and end-ofcommutation angles
6t = pi + a
(4.6.4)
& = 0 | + A
(4.6.5)
These calculations yield a very rough estimate of the converter switching

angles, which is subsequently improved substantially by a Newton solution
of the switching system.
4.6.2 The switching system

The purpose of the switching system is to solve the relationships between
the fundamental terminal voltage, the DC current and the switching angles
for both bridges. The switching system is thus a complete model of the
12-pulse converter in the presence of constant terminal voltage and DCcurrent harmonics, since the harmonic quantities appear as constant parameters. The mismatch equations and partial derivatives for the 12-pulse
switching system have all been derived previously. The set of equations to
be solved in the switching system (for constant-current control) are
FaO(0?,6l?,V,I<l) = O
131
(4.6.6)
A flow diagram for the switching system is shown in Figure 4.7b, and the
structure of the switching Jacobian can be seen in Figure 4.8. Those
elements corresponding to the partial derivatives of voltage mismatch with
respect to end-of-commutation angle have been set equal to zero, since they
are always insignificant.
'dO
5-
10-
520-
25--
(
30--
10
Figure 4.8
15
20
Variable number
25
30
Sparsity structure of the switching Jacobian matrix (power control)
132
The switching Jacobian is updated every iteration of the Newton method

and convergence has been found to be rapid and robust. The convergence
criteria for the switching system are
<0001
l < 0.001
2} < o.ooi
15ff!
0.001
|F*| < 5 x 10~8

| F ^ | < 5 x 10" 8
(4.6.7)
Convergence typically occurs in four to ten iterations, for starting terminal

voltages ranging from 0.3 to 7p.u. A case with a system impedance of
1.3 p.u. at the fundamental has been solved (which required 11 iterations).
A failure to converge in 20 iterations has so far always implied that the
system has no solution. An invalid solution (e.g. negative firing angle) has
never occurred, nor have multiple valid solutions been observed. If the
DC-voltage source is negative, the system will solve as for a currentcontrolled inverter.
The solution obtained with the switching system is an excellent starting
point for the full harmonic solution, as the switching angles are largely
determined by interactions at the fundamental frequency. Since the switching system is three phase, it includes the effect of any unbalance.
4.63 Harmonic solution

The system of harmonic-phasor and switching-angle equations is quite large
(426 elements square) and, at each iteration of Newton's method, a linear
system this size must be solved for the update vector: F(XN) = JNYN. This
step represents the bulk of the computation required in Newton's method,
and so techniques for speeding up the overall solution are concerned with
details of the Jacobian linear system and its solution method. The Jacobian
has been made sparse and it is essential that this sparsity is exploited in an
efficient manner.
Three types of sparse linear solver have been implemented. One of
these, the sparse symmetric bifactorisation6 method was found to be
unsuitable, as it requires the Jacobian to be diagonally row dominant and,
although the Jacobian has a large diagonal, it is not diagonally row
dominant. The method of Zollenkopf pivots for sparsity, not for numerical
stability, and does not yield the correct solution when applied to the
Jacobian system. The two other sparse solvers that have been implemented
133
are an asymmetric sparse bifactorisation which pivots for a compromise

between numerical stability and sparsity, and the iterative conjugate gradient method.7 The sparse bifactorisation employed is the y 12m solver from
the netlib.* Both methods have been found to be satisfactory but suited to
different types of solution algorithm.
Frequently, the Newton method can be improved by calculating the
Jacobian matrix only once, on the first iteration, and keeping it constant
throughout the solution. In this case, the sparse bifactorisation method is
fastest, as the bifactorisation need only be calculated once. On subsequent
iterations, the linear system is solved using the factorised Jacobian from the
first iteration. This method also avoids many of the indexing overheads
associated with sparse bifactorisation, since the sparsity structure is constant.
Holding the Jacobian constant leads to a larger number of faster iterations
to obtain the overall solution.
Another method is to update important parts of the Jacobian, holding the
bulk of the Jacobian constant. Convergence in the least number of iterations
has been obtained by holding the AC-DC partition constant and updating
the switching terms, since they can be recalculated quickly. In this case, the
Jacobian must be refactorised at each iteration, and the conjugate gradient
method is almost as fast. The conjugate gradient method typically requires
100 iterations to converge to an accuracy in the update that does not slow
the Newton solution. On the whole, sparse bifactorisation has been found
to be the most versatile.
4.6.4 Convergence tolerance

The basic requirement of convergence is that all of the mismatches F(X)f are
small enough. The mismatches, however, are of several types, entailing a
different convergence requirement for each type of mismatch. The convergence tolerances are listed below
< 0.001
<
Haoi
< 0.001
J ^ U 0.001
|Fgi|<5 x 10~ 8
|F^| < 5 x 10" 8
(4.6.8)
*accessible on the World Wide Web at http://netlib.att.com/netlib/yl2m/index.html
134
Note that convergence tolerance for the complex mismatches, F v , F,, F w is

expressed in terms of the magnitude of the mismatch. This means that the
error in an estimated value for a variable, for example V u , is smaller than
0.1% of its own length. The advantage of a relative mismatch of this type is
that it treats all harmonics equally. However, to prevent an attempt to
converge to an absolute error of zero for harmonics which are not present
(e.g. even harmonics in some cases), this convergence test is only applied to
harmonics that have a size larger than 10 ~~5 per unit. These convergence
tolerances can be made tighter to obtain more accurate solutions, if
necessary.
The tolerance set for the switching mismatches of 5 x 10" 8 corresponds
to 1.4 x 10~4 degrees at the 50th harmonic, or 0.1 nsec. This high tolerance
has been used in the calculation of the converter impedance and does not
require any extra iterations. Another type of convergence tolerance that can
be used, is to calculate a norm of the real mismatch vector; for example, the
1-norm
to a tolerance of |X|X < 10 5 is suitable for general purpose use. This type
of convergence test is fast and easy to apply but does not imply that all
harmonics have converged to a satisfactory accuracy.
4.7 Validation and performance

In this section the iterative model is verified against time-domain simulation
using the test system described in Appendix III. The steady-state solution
was obtained by running the EMTDC program, described in Chapter 6, for
one second with a time step of 20 /is, and then obtaining waveforms over
one cycle for subsequent comparison with the harmonic-domain solution.
The tests, designed to highlight any modelling and convergence deficiencies, are:
1 Base case with no harmonic sources in the AC system.
2 A 5% positive-sequence second harmonic is added to the Thevenin
voltage source in the AC system.
3 The leakage reactance of the phase b star-g/delta transformer is increased from 0.18 to 0.3 per unit.
4 A 0.1 per unit resistance is placed in series with the star-g/delta
transformer, and the secondary tap changer of that transformer is set to
1.1 p.u.
5 A 20% second-order harmonic is added to the Thevenin voltage source
in the AC system.
135
DC voltage, Test One
EMTDC
n. Harmonic Domain
1O
16
2O
2S
-4O
3O
35
3O
35
AO
3O
35
AO
3O
35
-*O
15
2O
25
harmonic order
Phase A current, Test One
53 O . 1 2
EMTDC
3*
O.1
O.O8
3 O.O6
Harmonic Domain
2) O.O4
CO
0.02
1O
15
2O
harmonic order
25
Phase A current, Test One
1O
Figure 4.9
15
2O
25
Comparison of time and harmonic-domain solutions for phase-currents

and DC-voltage spectra: base case
The results of test one, the base case, are shown in Figures 4.9 and 4.10. It
can be seen from the spectra that only the characteristic harmonics are
present and that there is a close match between the time and harmonicdomain solutions. The time-domain solution yielded small residual noncharacteristic harmonics in the spectra, which were suppressed in the phase
graphs. The DC-voltage waveform was generated from the harmonicdomain solution by plotting the Fourier series for each DC-voltage sample
during the appropriate interval, rather than inverse transforming the
136

Phase currents, Test One
0.002 0.004 0.006 0.008
0.01 0.012 0.014 0.016 0.018 0.02
time (sec)

600
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time (sec)
Figure 4.10 Comparison of time and harmonic-domain solutions for phase-current

and DC-voltage waveforms: base case
DC-voltage spectra. This eliminates Gibbs' phenomena associated with the

step changes in voltage but gives overly sharp voltage spikes. These are not
present in the time-domain solution owing to modelling of the snubber
circuits, which limits the dV/dt. The time-domain derived DC-voltage waveform is, therefore, more rounded. Clearly, if an accurate time-domain
waveform was required from the harmonic-domain solution, it would be
necessary to postprocess the waveshape using knowledge of the snubbercircuit time-domain response. The comparison of the waveshapes indicates
that all the switching angles are correct.
When a second-order harmonic voltage source is placed in the AC system
(test 2), a composite resonance is excited, resulting in noncharacteristic
harmonics and a high level of interaction between converter switching
angles and the AC-DC harmonics. Referring to Figures 4.11 and 4.12, odd
137
Phase currents, Test Two
0.002 0.004 0.006 0.008
0.01
0.012 0.014 0.016 0.018
0.02
time (sec)
DC voltage, Test Two
400
Figure 4.11
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018

time (sec)
0.02
Comparison of time and harmonic-domain solutions for phase-current

and DC-voltage waveforms: test 2
harmonics are present on the DC side and even harmonics on the AC side.
In particular, the fundamental resonance on the DC side is excited and
there is a large fundamental component in the DC voltage.
Unbalancing the star-g/delta transformer leakage reactance (test 3),
shown in Figures 4.13 and 4.14, causes the generation of many harmonics
and a large zero sequence on the AC primary side. This is due to sequence
transformation by the transformer, irrespective of its connection to a
converter. The zero sequence, also plotted in Figure 4.13, shows a close
agreement between the harmonic and time-domain solutions.
The main purpose of test 4 was to show that convergence is still acceptable
when the resistance present in the commutation circuit is not represented
in the Jacobian matrix. Also, setting the tap changer on one transformer but
not the other introduces six-pulse unbalance. It was necessary to increase
138

O.3
O.2
EMTDC
Harmonic Domain
O.1
15
2O
25
3O
35
4O
harmonic order
1O
15
2O
26
3O
35
4O
3O
35
4O
3O
35
4O
harmonic order
Phase A current, Test Two
SS O.12 -
3* 0.1 8 O.O8 . O.O6 fc O.O4

C
EMTDC
Harmonic Domain
O.O2
1O
15
2O
harmonic order
25
Phase A current, Test Two
nil! !L
25
Figure 4.12
Comparison of time and harmonic-domain solutions for phase-currents

and DC-voltage spectra: test 2
the AC-system fundamental source from 1.10976 to 1.20976 p.u. to enable

the current order to be satisfied. The results of test 4 (not displayed)
indicate the presence of six-pulse harmonics due to imbalance between the
two six-pulse groups.
Convergence of the sparse Newton solution is fast and robust, even if the
Jacobian matrix is held constant. A convenient measure of the convergence
is the 1-norm, the sum of the magnitudes of all the mismatches. At each
iteration, the 1-norm is reduced by an approximately constant convergence
factor, A, depending upon the difficulty of the system, and whether the
switching terms are updated. As indicated in Table 4.5, a smaller conver-
139
Phase currents, Test Three
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018

time (sec)
0.02
DC voltage, Test Three

600
400
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time (sec)
Zero sequence ac current, Test Three
I
-0.1
-0.2
Figure 4.13
0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
time (sec)

and DC-voltage waveforms: test 3
gence factor (and hence fewer iterations to convergence) is obtained if the

switching terms are updated (rows with subscript a). However, the constant
Jacobian method is always faster as each iteration takes approximately 0.6
seconds as opposed to 2.3 seconds. The constant Jacobian method is only
slower if it requires more than four times as many iterations to converge.
The only situation in which this has been observed is when the composite
resonance is excited by a very large (0.3 p.u.) second-harmonic source. The
140

EMTDC
ih Harmonic Domain
1O
15
2O
25
3O
35
3O
35
4O
harmonic order
15
2O
25
4O
harmonic order
Phase A current, Test Three
d
d.
0.12
EMTDC
0.1
Harmonic Domain
0.08
o.oe
O.O4
0.02
1O
15
2O
IH
25
m.
harmonic order
3O
Bl
35
til
...
4O
Phase A current, Test Three
HI w
f 1
III 111
m ijj
Hi
-100
-200'
Figure 4.14
1O
15
2O
25
3O
35
4O

and DC-voltage spectra: test 3
constant Jacobian method is, therefore, likely to be faster in any realistic

case.
It is evident from Table 4.5 that convergence is slowed by the low
harmonic-order composite resonance, and by the presence of a large
commutating resistance. The second-order harmonic composite resonant is
particularly difficult for the constant Jacobian method, as there is a higher
coupling between low-order harmonics and the switching angles. This is
evident in the convergence of test 5, which required 21 iterations. For more
realistic systems, convergence within eight iterations using a constant
141
Table 4.5 Convergence and performance of the solution, (a) updating switching
terms, (b) constant Jacobian
Test no.
CPU time
(seconds)
1a
1*>
2a
2b
3a
3b
4a
45
5a
5b
15.5
11.1
24.9
14.8
17.7
12.3
22.7
14.1
31.5
20.8
Switch
iterations
Main
iterations
-
6
6
6
6
6
9
9
6
6
6
8
11
5
7
7
9
11
21
X
0.1314
0.1712
0.1440
0.2562
0.1069
0.1130
0.2498
0.3129
0.3068
0.5382
Jacobian might be expected. The execution times listed in Table 4.5 are for
a Sun Sparcstation IPX.
4.8 Summary
This Chapter describes the use of iterative harmonic analysis for the
solution of static converters embedded in AC systems. The electric-circuit
constraints, control and switching functions are incorporated in a unified
way, and solved by the Newton method.
The steady-state interaction of a 12-pulse converter with weak AC and DC
systems has been analysed. In the test system, the AC side has been
represented by a three-phase frequency-dependent Thevenin equivalent,
and the DC system by a harmonic Norton equivalent.
The results of the Newton solution have been compared with a timedomain simulation of the same test system. The close agreement between
two such different methods of solution for the test system validates both of
them. Since the unified method converged quickly for such a difficult test
system, convergence is highly likely for any realistic system.
4.9 References
1 CALLAGHAN, CD., and ARRILLAGA, J.: 'Convergence criteria for iterative
harmonic analysis and its application to static converters'. ICHPS-IV, Budapest,
Hungary, October 1990, pp.38-43
142
2 SZECHTMAN, M., WEISS, T., and THIO, C.V.: 'First benchmark model for HVdc
control studies', Electra, April 1991, (135), pp. 55-75
3 SMITH, B.C., WATSON, N.R., WOOD, A.R., and ARRILLAGA, J.: 'A solution for
the steady state interaction of the ac/dc converter with weak ac and dc systems'.
Trans. IEEE, Paper 96-SM 447-3 PWRD
4 LARSON, E.V., BAKER, D.H., and McIVER, J.C.: 'Low order harmonic interaction on ac-dc systems', IEEE Trans. Power Deliv., January 1989, 4, (1), pp. 493-501
5 SMITH, B.C.: 'A harmonic domain model for the interaction of the HVDC
converter with ac and dc systems'. PhD thesis, University of Canterbury, New
Zealand, 1996
6 ZOLLENKOPF, K.: 'Bifactorisationbasic computational algorithm and programming techniques'. Conference on Large sets of sparse linear equations, Oxford,
England, 1970
7 PRESS, W., TEUKOLSY, S., VETTERLING, W., and FLANNERY, B.: 'Numerical
recipes in FORTRAN, the art of scientific computing' (Cambridge University
Press, 1992, 2nd edn.)
Chapter 5
Three-phase power and

harmonic flow
5.1 Introduction
The power-flow solution, described in Chapter 3, is based on the assumption of symmetrical and undistorted operation, the latter achieved by
perfect filtering at the converter terminals. In practice, there is always some
asymmetry in the AC system which, as described in Chapter 2, generates
noncharacteristic harmonics. These harmonics can excite resonances and
jeopardise the operation of control and protection equipment, as well as
involve extra power loss which should be accounted for in the power-flow
solution.
Thus, a three-phase fundamental and harmonic-frequencies solution of
the AC system and converter is required. The elements for such a solution
are the three-phase power flow,1'2 three-phase harmonic representation of
the AC system3 and the three-phase harmonic converter model. 4 " 6 In the
models proposed so far,7'8 the problem is partitioned into several modules
which are then solved separately. For example, a separate Newton-Raphson
module is used at each iteration to find the switching instants. Interaction
between these modules is then solved iteratively.
Convergence of the sequential method depends directly on the degree
of decoupling between modules; even a relatively small degree of coupling
can cause divergence. For a converter operating under conditions of
unbalance, there is no a priori justification as to why a sequential method should
work. This being the case, a Newton solution of all the unknowns is preferred.
In this Chapter, a three-phase power-flow and harmonic-converter model
are combined and solved using both the sequential and unified Newton's
methods.
5.2 The three-phase power flow

The structure, formulation and numerical solution of symmetrical power
flow have been described in Chapter 3 with reference to the fast-decoupled
method.9 This method has also been proposed for three-phase power-flow
analysis.1
144
The decoupled method is faster than a pure Newton solution by a

constant factor, independent of the network size. This is because the
network-admittance matrix is sparse, and consequently the number of
terms in the system Jacobian scales linearly with the network size. However, for medium-sized systems, and with the ready availability of fast
computers, the speed advantage of the decoupled method is not so
important.
The simplest algorithm for the incorporation of AC-DC converters in the
three-phase power flow is to iterate alternately between updates of the
power-flow and converter variables using two Jacobians. In the power flow,
the converter can be represented by a constant load, and in the converter
model the power flow can be represented by a voltage source behind the
system impedance at fundamental frequency. The advantage of this method
is that it is easy to implement, being compatible with any existing powerflow algorithm.
However, the AC-DC interactions degrade the convergence of the
sequential solution, if convergence is obtained at all.10 A combined power
and harmonic-flow solution is more likely to succeed with a unified
method.
Unification of the power-flow and converter equations imposes a different
set of requirements on the power-flow implementation than those which led
to the development of the fast-decoupled solution. First, the converter
model is necessarily in Cartesian co-ordinates. If the power-flow equations
are in polar co-ordinates, as required by the decoupling concept, nonlinear
polar transforms must be carried out at each iteration to interface with the
converter AC terminal (at fundamental frequency). This is likely to degrade
convergence substantially, as well as complicating the power-flow implementation. An additional factor to consider is that the converter equations
take longer to calculate than the solution of the prefactorised Jacobian
system. It is, therefore, desirable to reduce the number of convertermismatch equation evaluations by improving the convergence rate. This can
be achieved by using the full Jacobian matrix, with no decoupling. Taken
together, the above points indicate that a unified power-flow and harmonic
solution in Cartesian co-ordinates will be more efficient than one in polar
co-ordinates.
Although the converter solution is best implemented in sequence components, this does not impose a similar requirement on the power flow. The
sequence transform is fast and linear, and does not affect the convergence
of Newton's method. In fact, there are practical reasons why some of the
three-phase power-flow equations should be in phase components and other
equations in sequence components.
The object of a three-phase power flow is solving for zero and negativesequence voltages and currents. At generators, the internal e.m.f. is purely
of positive sequence, and thus the real power and voltage-magnitude
specifications refer to the power in the positive sequence and the magnitude
of the positive-sequence voltage, respectively.
145
At the slack bus a similar argument holds, except that in this case both
Cartesian components of the positive-sequence voltage are specified, rather
than just the magnitude. The presence of negative and zero-sequence
currents in the network is primarily due to sequence transformation in
transmission lines and unbalanced loads. Load buses are, therefore, best
specified by independent real and reactive powers in each phase.
5.2.1 Mismatch equations

The unknowns to solve for in the power flow are the voltages throughout
the network. Whether in sequence or phase components, there are six
unknowns at each three-phase bus, corresponding to the real and imaginary
parts of each phase or sequence of voltage. Six mismatch equations in these
variables are required at every bus to enable a solution. At load buses, these
equations are readily obtained by taking the Cartesian components of the
complex power mismatches in phase components
F = -#{vflli*, + sfl,}
F 2 i = 7{V I* F 3 i = R{V
I*
+ sw}
** =
F =
-/{v e ,i5 -
(5.2.1)
where the subscripts a, b, and c refer to the phase, I to the bus and V and I
are the nodal-phase voltage and current r.m.s. phasors. V and I are related
by the nodal admittance matrix for the system
I = YV
(5.2.2)
Consequently, It is a function of all the nodal-bus voltages to which bus I is

connected. Since the majority of buses are PQ, a solution in phase components for the voltage is slightly easier. At generator buses, the real power in
the positive sequence and the magnitude of the positive-sequence voltage
are specified, leading to the mismatch equations
F 2/ = | V S / - V + / V * ;
(5.2.3)
Four more equations are required. These are obtained with reference
to Figure 5.1, by writing KirchofFs current law for the negative and
zero-sequence currents at the generator terminal and taking Cartesian
146
positive sequence
V
negative sequence
zero sequence
sP
VI
Figure 5.1
Sequence components model of the synchronous generators
components
F 5 , = R{V0lY0l
F 6 i = /{V 0 ,Y 0 ,
(5.2.4)
At the slack bus, the positive-sequence voltage magnitude and phase are
specified, which is equivalent to specifying both the real and imaginary
parts
2(
= /{Vslack-V+i}
(5.2.5)
The mismatch equations are in phase components at PQ buses and

sequence components at all generator buses. Evaluation of the mismatch
equations proceeds as follows:
1 Calculate the nodal currents throughout the system by multiplying the
nodal voltage vector, V, by the system admittance matrix, Y. This is a
sparse calculation and the admittance matrix is stored as a list.
2 At PQ buses calculate the power mismatches using equation set 1.
3 At PV buses, sequence transform the nodal voltage, V,, and current, It,
and calculate mismatches using equation sets 3 and 4.
4 At the slack bus, proceed as for PV buses, but calculate mismatches using
eqns. 5.2.5 and 5.2.4.
5.2.2 The power-flow Jacobian

With some mismatch equations in sequence components, many of the
partial derivatives required for the Jacobian matrix are easier to derive
using real matrices, as opposed to complex numbers. For example, a
Three-phase power and harmonic flow 147

complex admittance
Y = YR +;Y,
(5.2.6)
can be written as a real tensor matrix

fY
Y = L,
Y,~l
(5-2-7)
In the following, the bar convention is used to indicate a real tensor matrix,
the subscripts R and I to refer to the real and imaginary parts of a complex
number and the subscripts / and m to refer to the number of a bus. A
submatrix of a larger matrix will be referred to by specifying the range of
row and column indices. For example, H = ](a:btC:d) is a matrix consisting of
rows a to b, and columns c to d of the matrix J.
The advantage in using real matrices as opposed to complex numbers in
this analysis, is that the conjugation operator is replaced by a left-multiplicative constant,
a*o[*]a
(5.2.8)
where
(5.2.9)
This makes conjugation differentiable in terms of real Cartesian components.
With nodal voltages in phase components, partial derivatives of mismatch
equations at PQ buses are easiest to derive. Assume that / is the bus number
of a PQ bus. There are six power mismatches at this bus, labelled Fkl, where
k = 1,2,3,4,5,6. The power mismatches at bus / depend on the nodal
voltages, in all three phases at every bus, m, to which it is attached (including
itself). Let Yb = YR +jYY be the branch admittance connecting phase p of
bus / to phase q of bus m. Then assuming that p " q, or that I ^m, the
following partial derivatives hold
3Vi
-R-K M - l
(5.2.10)
148
where VR = R{Vqm}, and V, = I{Vqm}. If p = q and I = m, the partial derivatives are
RYR
(2p-l)l _ y y
vV
(5.2.11)
V.Y.-IR
<5V,
IR = J R{I pl }andI I =/{I p( }.

The mismatch equations at generator buses are in sequence components,
and are derived by first forming partial derivatives of the nodal voltage and
current in sequence components, with respect to phase-component voltages
at all connected buses. Sequence-component nodal voltages at a generator
bus are obtained by multiplying by the sequence transform matrix
(5.2.12)
oi
1 a
Expanding eqn. 5.2.12 into real components, the sequence-components

matrix can be viewed as a matrix of partial derivatives
AV + 1 I
AV 0IR
AV,
on
AV_ IR
AV_,
2 0
0 2
2 0
0 2
2 0
0 2
-1
73
-73
-1
-1
-73
73"
"AV O I R
-1
AV a ( I
AVWR
IsO
AV + IR
AV W I
-1
73
-1
-73
AV C / R
-1
AVd,
-73
-1
(5.2.13)
or in a more compact form

AV
+,
AV,
0(
=T
(5.2.14)
For example, the fourth element in the first row corresponds to

(5.2.15)
dV,6(1
A similar equation may be written for the sequence components of

the nodal current at the generator bus. In this case, multiplying the second
row by 1 takes account of the conjugation operator applied to the positive
sequence
2
0
AI*n
1 2
AIOIR
~
6
0
Alon
2
AI-, R
0
AI-,,
AI*+iR
0
-2
0
2
0
2
-V3 - 1
-1
1
0
2
2
0
2
0
-1
-1
-V3
V3
-1
/Q
1
0
2
/SI
-l
AI a i R
AI a i l
AIWR
AIM,
(5.2.16)
AI C(R
AI,n
or in a more compact form
= T
(5.2.17)
AIC,
AI-,
T h e chain rule can now be applied to relate changes in the phase voltage at
other buses to the right-hand side of eqn. 5.2.16. If Ylm is the 3 x 3
admittance connecting bus / to bus m, given by
(5.2.18)
3>21
}>33
150
then the decomposition to Cartesian components is a 6 x 6 tensor

JllR
"Jill
Jl2R
-Jill
Jl3R
Jill
JllR
Jl2I
Jl2R
3*131
Jl3R
J21R
"Jill
J22R
-J22I
J23R
-J23I
Jill
J21R
J22I
J22R
J23I
J23R
(5.2.19)
Changes in the sequence-component nodal current due to changes in

phase-component voltages at bus m are, therefore
A[*]i+;
"AVflm"
(5.2.20)
= T'Y AV hm
AIOi
AV cm
Consider now the negative and zero-sequence mismatches at generator

buses
(5.2.21)
If / 7^ m, t h e n t h e last four rows of T ' Y a r e t h e partial derivatives
= T'Y( 3:6>1:6 )
(5.2.22)
If I = m, t h e negative a n d zero-sequence shunts m u s t b e a d d e d to T ' Y as

follows
= (
0
0
0
0
0
0
0
0
0
0
0
0
Y0R
-Y0I
Y0R
0
Yo,
0
Y-R
Y_,
Y_
(5.2.23)
(3:6,1:6)
The remaining partial derivatives to be calculated are those of F lf and F2l

at all generator buses. At PV buses
+ V + ,I*,}
(5.2.24)
This can be written in tensor form
F = iP+f + [tV, I f %mVm]

L
m=l
(5.2.25)
which is readily differentiate

/\r
"T
* / \(\ 1 fi'i
ri^/^
, riVx
' ' ^*
1
...
(5.4.26)
The voltage magnitude mismatch is
F2, = Vi-V + I VT,
(5.2.27)
F 2( = V s 2 p -[fV,T'V ( ] (1>1)
(5.2.28)
Converting to tensor form
the partial derivatives of which are
-Si 2 i-6 = I t 0 ' -' -'0> ]

^Vm '
*m
(5 2 29)
l[-V +l T[ 1:2 ,i;6) - Vt/T ( 1 : 2 1 : 6 ) ] ( U : 6 ) / = m
At the slack bus,
F 2( = /{V s l a c k -V + i }
(5.2.30)
F,(1:2) = Vslack - TV,
(5.2.31)
or, in real form
The partial derivatives with respect to terminal voltages are, therefore, just
^ < - = {f:V'OTO:O'O] _ v * f
\ * l (5.2.32)
152
input system data
create system admittance matrix

yes
calculate constant part of Jacobian
calculate nodal current injections
output results
solve sparse system for updates
Figure 5.2
Flow diagram for the three-phase power flow; FNIT is the number of
full Newton iterations, MAXIT is the maximum number of iterations
Partial derivatives of the power-flow mismatch equations have now been

derived in 6 x 6 blocks corresponding to the derivatives of the mismatches
at bus / to the Cartesian components of the phase voltages at bus m. These
6 x 6 blocks are stored directly into a sparse form of the Jacobian
*> = # "
(5-2.33)
In the above derivation of 6 x 6 Jacobian blocks, many redundant calculations have been specified which are not carried out in practice. For example,
in eqn. 5.2.26 only the first row of [V+lT'Ylm] is calculated.
5.2.3 Newton's method

Newton's method iteratively solves the nonlinear system of equations by
calculating updates at each iteration from
JAXN = F(XN)
(5.2.34)
so that X N + 1 = XN AXN. In practice, the linear system of eqn. 5.2.34 is

solved by LU decomposition, i.e. hy = F(XN) is solved for y, followed by a
solution of UAXN =y. The Jacobian sparsity is exploited by employing a
sparse bifactorisation for the solution of the linear system at each iteration.
Despite the use of sparsity, the bifactorisation still represents the bulk of the
calculation in each iteration. Overall convergence is greatly accelerated by
holding the Jacobian constant in the vicinity of the solution, where the
equation system is more linear. Typically, the Jacobian need only be
updated on the first two iterations. Subsequent iterations are then very fast,
since the LU decomposition of the Jacobian does not have to be recalculated.
A flow diagram for the power flow, Figure 5.2, includes a module which
checks for the presence of a zero-sequence block in the AC-system admittance matrix. This is discussed in detail in section 5.2.5. Convergence is
deemed to have occurred when the largest mismatch is smaller than a preset
tolerance.
5.2.4 Performance of the power flow

The test system to be solved is that of Figure 5.3. It is a nine-busbar network
corresponding to the lower part of the South Island of New Zealand. The
slack bus is at ROXBURGHO11, and unbalanced loads have been placed at
ROXBURGH220, INVERCARGO11 and TIWAI-220. Additional data
for the system is listed in Table 5.1. The transmission-line models are three
phase, and include mutual coupling between adjacent circuits. Initially, all
transformers are connected star-g/star-g.
154
ROXBURGH-011
(Slack)
ROXBURGH--220
P+JQ
INVERCARG011
P+JQ
INVERCARG220
TIWAI220
Figure 5.3
Test system for the three-phase power flow
The Jacobian for the test system is illustrated in Figure 5.4 by a meshplot
of the magnitude and a sparsity plot of the location of elements in the
matrix. Structural sparsity would be much greater for a realistically-sized
system and is clearly not symmetric. The largest elements are associated
with PQ mismatches at buses with the most connected lines, or the highest
AC-system admittance.
In order to assess the performance of the power flow, a variety of test cases
are solved. The base case is that described in Table 5.1, which contains
severely unbalanced loads. After converging in four iterations, the sum of
the magnitudes of all the mismatches, the 1-norm residual, was
2.2 x 10~ 10 p.u. This was for a solution in which the Jacobian was updated at
every iteration and with a starting point of 1 p.u. positive-sequence voltage at
every PQ bus. Updating the Jacobian only twice increased the number of
iterations to seven, with a final 1-norm residual of 3.9 x 10 ~7. The nodal
voltages at the solution are given in sequence components in Table 5.2. Note
that the voltage constraints at generator buses have been satisfied.

Table 5.1
Power flow constraints for the test system
Bus name
Bus type
INVERGARG220
PQ
INVERGAR011
PQ
ROXBURGH-220
PQ
MANAPOURI220
PQ
MANAPOURI014
PV
TIWAI-220
PQ
NTH-MAKAR220
PQ
INV-MAN--TMP
PQ
ROXBURGH-011
Slack
Constraints
Sa =
Sb =
Sc =
Sa =
Sb =
Sc =
Sa =
Sb =
Sc =
Sa =
0+y0MVA
0+y0MVA
0+y0MVA
50+y15MVA
45 + y'20 MVA
52+y18MVA
48+y20MVA
48+y12MVA
51.3 + y'28.3 MVA
0+y'0MVA
s c = o + yo MVA
P+ =500MW
|V+| = 1.05 p.u.
S a =170+y85MVA
Sb=160+y85MVA
Sc=170+y70MVA
Sa = 0+y0MVA
s b = o + yo MVA
s c = o + yo MVA
Sa = 0-fy0MVA
sb = o + yo MVA
Sc = 0+y0MVA
V + = 1.05^0p.u.
Placing a star-g/delta at the slack bus effectively shifts the angle reference
by 30, so that the starting voltages are offset by 30. In this case, convergence required 23 iterations to a final residual of 2.4 x 10" 6 if the Jacobian
was updated twice, five iterations for a full Newton solution and seven
iterations if updated four times. The large number of iterations required in
the first case was due to the large initial mismatch, implying that the
Jacobian calculated after the second iteration was significantly different to
that at the solution. A more versatile approach is to update the Jacobian at
every iteration until the largest mismatch is smaller than a preset tolerance,
at which point the Jacobian is held constant.
If all transformers are star-g/star-g, except for that at the INVERCARGO11 load bus, which is made star-g/delta, convergence is much more
difficult. To obtain convergence, it is necessary to start the voltage at
I <8
'
;.

!

:
:f
:
: :
:.
'.
99.
jf.tM9.M9.
t9.
..
Equation number
:_
Magnitude

Table 5.2
Solved nodal voltages in sequence components for test one
Bus and seq.
Seq. voltage (p.u.)
INVERCARG220
+ve
-ve
zero
INVERCARG011
+ve
ve
zero
ROXBURGH-220
+ve
ve
zero
MANAPOURI220
+ve
-ve
zero
MANAPOURI014
+ve
ve
zero
TIWAI-220
+ve
-ve
zero
NTH-MAKAR220
+ve
-ve
zero
INV-MAN--TMP
+ve
-ve
zero
ROXBURGH-011
+ve
-ve
zero
Phase ()
PQ
0.96285
0.01264
0.00043
-7.06927
-166.75234
38.40967
PQ
-7.98316
-165.72855
29.61255
0.95719
0.01287
0.00491
PQ
-1.78554
131.37500
-64.09227
0.99201
0.00284
0.00240
PQ
-1.63163
-156.61677
17.09903
1.02863
0.00362
0.00077
PV
1.02177
-156.61677
17.09903
1.05000
0.00181
0.00038
PQ
-8.10666
-166.20626
72.62988
0.95139
0.01441
0.00512
PQ
-6.65871
-166.16024
47.6068
0.96865
0.01212
0.00409
PQ
-6.65427
-166.78061
43.49631
0.96813
0.01197
0.00413
slack
1.00000
0.00142
0.00120
0.00000
131.37500
-64.09227
158
INVERCARGO11 with a phase angle of 30 and with a magnitude greater

than 1.05 p.u. Even if the Jacobian is updated every iteration, seven
iterations are required for convergence. The solution so obtained indicates
a very large zero-sequence voltage at the INVERCARGO11 bus only. It is
0.23397 p.u. Z_-110. If the starting voltage is increased to 1.1 p.u. at
INVERCARGO11, a different solution is obtained. The zero-sequence
voltage is 0.29267 p.u. Z_56.6 at the load bus, and the positive-sequence
voltages throughout the network are very similar. In fact, for a wide range
of starting conditions only these two solutions are obtained.
Even if all loads throughout the system are balanced, two highly distorted
solutions are found at INVERCARGO11; 0.096 L 61.74 p.u. and
0.11 L 83.96 p.u. of zero-sequence voltage. These two solutions are
intrinsic to the three-phase power flow, and are related to the magnitude of
the zero-sequence admittance at load buses. When a three-phase load is
connected to a transformer connection which blocks zero-sequence current,
there are two approximately antiphase solutions for the zero-sequence
voltage. If the zero-sequence admittance is large, one of the solutions for
the zero-sequence voltage becomes small and is usually converged towards,
since it is closer to the starting point of the Newton solution. In general, the
Newton algorithm will diverge if there are several zero-sequence-blocked
loads, as the multiple solutions lie too close together.
5.2.5 Zero-sequence blocking

A direct analytic solution for the zero-sequence voltage at a load bus is
possible for the simple system of Figure 5.5. This system consists of a
positive-sequence voltage source, an ideal star-g/star connected transformer
with no leakage reactance and a three-phase load. At the load bus, the
positive-sequence voltage (V+) is the same as the source, the zero-sequence
voltage (Vo) is unknown and the negative-sequence voltage (V_) is zero. The
zero-sequence current (Io) flowing into the load is zero, since there is no
path for it and the positive (I + ) and negative (I_) sequence currents are
unknown.
Three equations can be written for the specified loads on each phase
v.i* = s0
VCIC* = Sc
(5.2.35)
In sequence components these become
(v 0 +v + >a* + n ) = sa
(V0+a2V+)(a2I*
+al*) = Sb
(Vo + aV+)(aI* + a 2 l* ) = Sc
(5.2.36)
159
Figure 5.5 A three-phase load connected to the ungrounded side of an ideal

star-g/star transformer and, thence, to a positive-sequence voltage
source
When I_ and I + are eliminated from these equations, a quadratic in Vo and

V + is obtained
-S 0 V* + S_V+V0 - S+V2+ = 0
(5.2.37)
where
s 0 = s a + sb + sc
$+=Sa+a$b+a%
S_ = Sfl + a2Sb + aSc
(5.2.38)
This result confirms that there should be two solutions for the zero-sequence
voltage, if there is no path for zero-sequence current. To investigate this
effect further the test system of Figure 5.6 was solved using the three-phase
power flow.
In this system, the leakage reactance of the star-g/star transformer was set
to 0.01 p.u., and the leakage reactance of the star-g/star-g transformer was
varied from 10 p.u. to 0.01 p.u. in 100 steps. The load was held constant at
Sa = 49.5 -h/19.5 MVA, Sb = 50.5 + j"20.0 MVA and Sc = 50.5 + J20.5 MVA.
With the zero-sequence leakage reactance of the star-g/star-g transformer set to 10 p.u., two solutions to the power flow were obtained,
corresponding to approximately antiphase zero-sequence voltages at the
load bus. Starting from the larger of the zero-sequence voltage solutions, the
leakage reactance of the star-g/star g transformer was reduced in 100
steps down to 0.01 p.u., a more normal situation. The solution of each
intermediate power flow was used as the starting point for the next. The
outcome of this process is plotted as the top curve in Figure 5.7, which is a
plot of zero-sequence voltage magnitude as a function of leakage reactance.
The other two curves in Figure 5.7 show the effect of sliding the leakage
reactance down from 10 p.u. at the lower of the two zero-sequence voltagemagnitude solutions, and up from 0.01 p.u., starting with the zero-sequence
voltage set to zero. In the middle is a region of divergence. It is clear from
this diagram that there are always two solutions at load buses but that in
160

star-g/star
P+jQ
Figure 5.6 A test system with a variable path for zero-sequence current
normal circumstances one of the solutions is too far away from a purely
positive-sequence starting point to be converged towards. As the zerosequence impedance looking into the AC system from a load bus is
increased, the two solutions move closer together until they are of approximately equal magnitude. If the zero-sequence impedance is greater than
about 1 p.u., either solution can be converged towards, depending upon the
starting point.
In any real AC system there will always be a path for zero-sequence
Variation in Load-Flow Solutions
Figure 5.7
4
5
6
zero seq impedance (p.u.)
10
The effect of varying the zero-sequence system-impedance magnitude on

the two solutions for the zero-sequence voltage at a load bus
current. However, in constructing a simple power-flow model of an AC

system, such a path could easily be left out, causing divergence or unrealistic
solutions. This problem is best avoided by calculating the zero-sequence
impedance at every load bus prior to solution, using the system-admittance
matrix. If the zero-sequence impedance is greater than a preset maximum,
a warning is given and the location of the load concerned indicated. The
advantage of this approach is that it is automatic, and treats the zerosequence block like any other admittance matrix problem, e.g. cut sets and
unconnected buses.
5.3 Converter harmonic model

The converter model has been described in Chapter 4 with reference to the
simplified circuit of Figure 5.8; the relevant variables and mismatches are
listed in Table 4.2. The AC system is represented by a three-phase harmonic
Thevenin equivalent. A separate harmonic-analysis program is used to
calculate the harmonic impedances of transmission lines, loads, shunt and
machine components. The program then assembles the system-harmonic
admittance matrix and reduces it to the equivalent harmonic admittance at
the converter bus (Ytc in Figure 5.8). The DC system is also represented by
a Thevenin equivalent, prepared in a similar manner.
Each converter transformer may have a different leakage reactance and
resistance in each phase. Additional unbalance may exist between phases in
the filters owing to component tolerance, and between six-pulse groups
owing to tap setting or thyristor forward resistance. Snubbers are not
modelled, nor is skin effect or saturation in the converter transformers.
dc system
ac system
measured dc current
alpha order
Figure 5.8 Twelve-pulse converter model to be solved
162
The basic control types represented in the present model are constant
current, constant power and minimum gamma. In the case of constantcurrent control, the controller is assumed to be a simple PI type, with a
low-pass current-transducer characteristic. The twelve firing and end-ofcommutation instants are included as part of the Newton solution. Under
current control, the firing instants are, therefore, modulated by ripple in the
DC current. The end-of-commutation instants are modulated owing to
distortion in the terminal voltage and DC current.
The firing controller is assumed to be referenced to an ideal PLO
tracking the terminal voltage. This means that all firings will be perfectly
equispaced if there is no ripple in the DC current, despite any distortion in
the terminal voltage.
The AC-side terminal voltage is described in harmonic-sequence components. Sequence components are chosen as there is usually no interaction
between zero-sequence voltage harmonics and the converter, thereby improving sparsity in the Jacobian matrix.
5.4 Combined solution

5A.I Sequential method
A decoupled solution of the power flow and harmonic interaction is
implemented by iterating sequentially between power flow and harmonic
solutions, as illustrated by the flow chart in Figure 5.9. Interaction between
the two is mediated by the parameters that are used to interface them. It is,
therefore, possible to use any existing three-phase power flow in this
algorithm, including a fast decoupled power flow in polar co-ordinates.
Assuming that the converter is situated at a PQ bus, updates to the
specified complex power in each phase, consistent with the converter
solution, are all that is required to interface the converter to the power flow.
When the converter model is being solved, the powerflowis most effectively
represented by a Thevenin equivalent. The Thevenin impedance is the
three-phase system impedance at fundamental frequency, while the
Thevenin voltages are chosen so that the converter terminal voltages match
those calculated by the power flow.
Referring to Figure 5.10a, an initial solution for the power flow, with the
converter represented by its estimated power, Sc, yields the voltage at the
converter bus, Vc. The current drawn by this load in each phase is then
ICy = S*/V*, where y refers to one of the three phases. If this current is
drawn from the Thevenin equivalent to the powerflow(Figure 10b), then in
order to maintain Yc as the converter terminal voltage, the Thevenin source
must be
Vrt = Vc + IcZrt
(5.4.1)

start
evaluate converter mismatches
yes
converter
Newton
solution
yes
calculate and factorise
converter Jacobian
use Jacobian to update
converter variables
calculate converter
complex power
evaluate power flow mismatches
power
flow
Newton
solution
calculate and factorise
power flow Jacobian
use Jacobian to update
load flow variables
calculate power-flow
Thevenin-equivalent
internal voltage
Figure 5.9
Flow chart for the decoupled method of solution
163
164
Converter Representation
Sc-P+jQ
Linearising
Inductor
(a)
Load-flow Representation
Vth
Converter Equations
(b)
Converter Equations
Figure 5.10
a Representation of the converter model in the power flow

b Representation of the power flow in the converter model
c Combined power flow and converter model
If the converter model is now solved with this new Thevenin voltage, an
updated value of Vc and I c will be obtained. For the next solution of the
power flow (Figure 5.10a), the specified power at the converter bus must be
updated to be
SCy = VCvI*
(5.4.2)
Likewise, when the Thevenin voltage for the converter model is calculated,

the current shunted by the linearising reactor must be added to the current
shunted by the specified converter complex power
*>-%-%,
(5 4 3)
- -
The above-described linearisations are approximate and incomplete, since

only selected variations are linearised and held constant. A full linearisation
of the converter representation in the power flow would require the
calculation of 36 partial derivatives, these being the derivative of the real
and reactive powers in each phase with respect to the real and imaginary
components of the voltage in each phase. These linearisations could be
calculated either by solving 36 independently-specified perturbations of the
power flow and converter models, or by a Kron reduction of the Jacobian
matrix for each system down to a 6 x 6 matrix of partial derivatives. The
resulting matrix is a second-rank amplification tensor. Either method is
computationally intensive, being similar in complexity to the entire solution. The resulting algorithm can be shown to be equivalent to a full
(undecoupled) Newton solution but is slower and more difficult to implement.
In order to investigate convergence of the decoupled method, several
versions of the algorithm were implemented and applied to three different
test systems, of varying degrees of difficulty. The decoupled algorithms were
of the following types:
A With the linearising shunt present, the power flow and converter steps
were solved to complete convergence at each iteration of the decoupled
method. It was necessary to calculate and factorise the Jacobian matrix
once at the start of each of the power-flow and converter solutions.
B With the linearising shunt present, only one iteration of Newton's
method was applied in the power-flow and converter updates. Again, it
was necessary to recalculate the Jacobian matrix for both the power-flow
and converter model at each iteration of the decoupled method.
C As for method B, but with the linearising reactor absent.
The convergence criterion for both the power flow and converter model was
that the largest mismatch was less than a preset tolerance, typically
1 x l(T 5 p.u.
5.4.2 Unified Newton method

Unification of the power-flow and converter models into a single Newton
solution is simply a matter of collecting the converter and power-flow
mismatches into a single set, with a correspondingly larger Jacobian matrix.
Essentially, no new partial derivatives are required in the Jacobian matrix,
as the power flow is not a function of any of the variables from the converter
166
model except the AC-side voltage at fundamental frequency, and the

converter model is not a function of the voltages at other buses in the power
flow. The two models can be interfaced at fundamental frequency on the
converter AC side either by modifying the power mismatch at the converter
bus so that the specified power is a function of all the other converter
variables, or by modifying the converter voltage mismatch at fundamental
frequency so that the Thevenin voltage is a function of the voltages at other
buses in the power flow.
A conceptually simpler and easier mismatch to implement, is a current
mismatch at fundamental frequency in phase components on the converter
AC side. The current mismatch is simply a statement that the nodal current
flowing into the AC system at the converter terminal at fundamental
Fundamental
Phase Voltages
Sequence Harmonic Voltages
Figure 5.11
Structure of the combined power-flow and converter Jacobian for a

solution up to the 13th harmonics
frequency is equal to the current flowing out of the converter and any shunts
connected there
Ft/c = Ii + I 2 + I 3
(5-4.4)
where I x , I2,13 are defined in Figure 5.10c. The power-flow and converter
modules can readily supply these currents. The Jacobian for the unified
system is illustrated in Figure 5.11 as a plot of the structure. All other
elements in the Jacobian are small and have been discarded to improve
sparsity. The terms associated with current mismatch at fundamental frequency appear as horizontal and vertical bands coincident with the converter bus TIWAI
220. These terms, being in-phase components, contribute
approximately 300 elements to the Jacobian matrix over and above those
associated with the decoupled power flow and converter Jacobians. The
overheads associated with combining the two models into a single solution
are therefore minimal, especially since only one set of sparse storage vectors
need be defined and manipulated instead of two.
5.4.3 Test system

The same test system of Figure 5.3 is used for the combined solution. The
following cases are discussed:
test
test
test
and
1 all loads are balanced between phases;

2 loads in each phase unbalanced by up to ten per cent;
3loads unbalanced by up to ten per cent, with harmonic filters
power-factor compensation removed from the converter bus.
The voltage base for this test system was 220 kV, the power base was
100 MVA, the system impedance at the converter bus was 0.03 Z_70, the
converter power was approximately (610 +j'75)MVA and it was running
under constant-current control as a rectifier. The rectifier DC system and
filters corresponded to those at the rectifier end of the CIGRE benchmark
system. The rectifier delay angle was 15.
5.4.4 Convergence characteristics

Data pertaining to the performance of the different algorithms for the three
test systems is listed in Table 5.3. The unified method was substantially
faster in all cases but required approximately 350 more terms to be stored
in the single Jacobian matrix than the sum of the terms stored in the two
Jacobians of the decoupled methods.
Of the decoupled methods, a significant improvement in convergence is
obtained by the use of a linearising shunt. There is no advantage in
converging the separate power-flow and converter systems at each iteration
of the decoupled method (type A). The best decoupled method is, therefore,
168
Table 53
Performance of the decoupled versus full Newton solutions, 50 harmonics
Test no.
Type
NJR
TNIT
Time
J size
unified
dcp A
dcp B
dcpC
unified
dcp A
dcp B
dcpC
unified
dcp A
dcp B
dcpC
1
7
8
15
1
7
9
15
1
8
9
*
4
7
8
15
4
7
9
15
5
8
9
*
3.7
24.1
19.8
36.1
3.6
22.0
19.9
35.0
3.9
30.1
20.4
6971
6540
6542
6540
6991
6542
6542
6544
7355
7008
7008
*
NJR is the number of Jacobian updates and refactorisation required

TNIT is the total number of iterations to convergence
type B where only one iteration of Newton's method was applied to the
converter and power-flow systems at each decoupled iteration. With the
linearising shunt absent, the decoupled method failed to converge in test 3,
when no filters were present. For all cases recorded in Table 5.3, the starting
point was from a converged, fundamental-frequency only, solution.
Convergence of the decoupled method is entirely dependent on the size
of the partial derivatives of the quantities that are being iteratively solved
for: in this case, the Thevenin equivalent voltage, Vth, and the converter
equivalent power, Sc. In three phases, and real and imaginary parts, there
are twelve quantities involved. The decoupled method is a fixed-point
iteration for these quantities
[SC]N+1 = converter (\Vthf)
[\y
N+1
= powerflow ([SJ
N+1
(5.4.5)
)
(5.4.6)
For convergence to be guaranteed
(5.4.7)
The average size of these partial derivatives is 0.3779 and 0.4178, for the

Load Flow Convergence
10
Figure 5.12
Iteration
20
Converter Convergence
10
Iteration
20
Convergence characteristic of the decoupled and unified methods; (U)

the unified method, (B) decoupled with linearising shunt reactor, (C)
decoupled without linearising reactor
decoupled methods with and without the linearising inductance, respectively. Convergence of the decoupled methods is, therefore, uncertain at
best.
Even when the decoupled method does converge, a consequence of the
large coupling derivatives is that at each iteration the gains made by each
Newton subsystem are reduced. This leads to an oscillatory convergence
characteristic, illustrated in Figure 5.12. Figure 5.12 is a plot of the sum of
the magnitude of the mismatches (the 1-norm) of the power-flow and
converter systems at every step of the decoupled and unified methods.
Again, the advantage in using a linearising shunt is evident. A consequence
of the oscillatory convergence is that Jacobian matrices for the converter
and power flow must be recalculated at each iteration of the decoupled
method. This is the most significant contributing factor in the long execution time of the decoupled method, since the Jacobian must also be
refactored.
5 A3 Powerflow/converterinteraction
If interaction between the three-phase power flow and converter model is
not solved simultaneously, some degree of error must exist in the solution
obtained. For example, a common approach is to estimate the converter
170
power and hold it constant in a power flow, or to hold the Thevenin

equivalent voltage constant in the converter model, based upon the result
of a single-phase power flow.
At best, the approximated quantities are the same as those which would
be obtained from a full simultaneous solution but lacking any asymmetry
between phases. To assess the resulting error, a full solution of test 1 was
obtained using the decoupled method. Next, the Thevenin voltage source
was made equal to its positive-sequence component and the equivalent
converter power in each phase was set equal to the average over the three
phases. The power-flow and converter models were then reconverged.
The outcome of this process is that both the power-flow and converter
models have been solved in the absence of any unbalance arising from their
interface to the other system but with all other factors being equal. The
results give an estimate of the best possible solution that could be expected
if the effects of asymmetry are ignored. A comparison of the power-flow
solutions is listed in Table 5.4, and of the converter solutions in Figure 5.13.
The effect on the power flow of not modelling asymmetry due to the
converter is a moderate underestimate of negative sequence throughout the
system. Zero sequence throughout the system is substantially larger due to
the converter.
The converter solution for the positive-sequence 3rd harmonic and
negative-sequence fundamental are substantially larger if the power-flow
asymmetry is fully represented. There is also an impact on all odd triplens,
but the characteristic harmonics are not affected. Odd triplens are present
in both solutions owing to asymmetry in the Thevenin impedance of the AC
system at fundamental frequency, and all harmonics up to the 50th.
Positive Sequence Voltage Harmonics
10
15
20
25
30
35
Negative Sequence Voltage Harmonics
Be
Left - Exact
Right - Approximate
15
20
25
Harmonic Order
Figure 5.13
Effect of not modelling power-flow unbalance in the converter model

Table 5.4 Sequence-components nodal voltages throughout the test system calculated
by exact Newton iteration and an approximate constant-load representation of the
converter
Bus and seq.
INVERCARG220
+ve
-ve
zero
ROXBURGH-220
.f-ve
_ve
zero
MANAPOURI220
+ve
-ve
zero
MANAPOURI014
+ve
-ve
zero
TIWAi-220
+ve
ve
zero
NTH-MAKAR220
+ve
ve
zero
INV-MAN-TMP
+ve
-ve
zero
ROXBURGI+-011
+ve
_ ve
zero
Exact
Approx.
PQ
1.01217
0.00876
0.00780
1.01211
0.00705
0.00183
PQ
1.04252
0.00150
0.00044
1.04250
0.00077
0.00040
PQ
1.04470
0.00149
0.00036
1.04470
0.00192
0.00048
PV
1.05000
0.00096
0.00024
1.05000
0.00075
0.00018
Converter
1.00618
0.00803
0.00182
1.00625
0.01000
0.00989
PQ
1.01541
0.00673
0.00165
1.01547
0.00840
0.00741
PQ
1.01499
0.00655
0.00150
1.01504
0.00817
0.00720
Slack
1.05000
0.00075
0.00022
1.05000
0.00038
0.00020
5.5 Summary
To make it compatible with the converter-harmonic model of Chapter 2 and
thus enable an integrated Newton solution of the power flow and harmonic
interaction, the three-phase power flow described in this Chapter uses
172
Cartesian, rather than polar co-ordinates for the mismatches. The overall
solution is formulated in terms of phase components, which suits the P, Q
bus specifications; however, at generator buses, the mismatches are in
sequence components.
Interaction between the three-phase power flow and a three-phase harmonic converter model has been solved using both decoupled and full
Newton methods. The decoupled method displays good convergence if a
linearising shunt is present but is slow because the Jacobian matrices need
to be recalculated and factorised at every iteration. The full Newton method
(with constant Jacobian), is faster and displays more robust convergence.
The decoupled method is compatible with any existing three-phase power
flow, including a polar fast-decoupled type. A special but simple three-phase
power flow is required for the unified method as the converter model must
be framed in real variables. Overall, code for the unified method is shorter
and less complicated, as there is only one set of sparse storage, mismatch
evaluation, convergence checking, etc.
The effect of not modelling the power-flow/distortion interaction is a
moderate underestimate of unbalance in the power-flow solution, and a
large underestimate of distortion at low order noncharacteristic harmonics
in the converter model.
The AC-DC iterative algorithm can easily be extended into a generalpurpose model with the capability of including several AC systems, DC
systems, possibly with multiple terminals in each system, and all integrated
with the power-flow equations. This is essentially a software engineering
task, as each half pole contributes a block to the main diagonal of the system
Jacobian matrix, with diagonal matrices coupling to other harmonic sources
in the same system. The main task is to code the program so as to be
versatile, easy to use and modular.
5.6 References
1 ARRILLAGA, J., and HARKER, B.J.: 'Fast decoupled three-phase load flow',
Proc. IEE, August 1978, 125, (8), pp. 734-740
2 WASLEY, R.G., and SLASH, M.A.: 'Newton-Raphson algorithm for three phase
load flow', Proc. IEE, July 1974, 121, (7), p. 630
3 DENSEM, T.J., BODGER, P.S., and ARRILLAGA, J.: 'Three phase transmission
systems for harmonic penetration studies', IEEE Trans. Power Appar. Syst, February 1984, 103, (2), pp. 310-317
4 YACAMINI, R., and de OLIVEIRA, J.C.: 'Harmonics in multiple convertor
systems: a generalised approach', IEE Proc. B, March 1980, 127, (2), pp. 96-106
5 SMITH, B.C., WATSON, N.R., WOOD, A.R., and ARRILLAGA, J.: 'A Newton
solution for the steady state interaction of ac/dc systems', IEE Proc, Gener. Transm.
Distrib., March 1996, 143, (2), pp. 200-210
6 SMITH, B.C., WATSON, N.R., WOOD, A.R., and ARRILLAGA, J.: 'A solution
for the steady-state interaction of the ac/dc converter with weak ac and dc
systems'. ICHPS Conference, Las Vegas, October 1996

7 XU, W., MARTI, J.R., and DOMMEL, H.W.: 'A multiphase harmonic load flow
solution technique'. IEEE PES, Winter Meeting 90, WM 098-4, PWRS, 1990
8 VALCARCEL, M., and MAYORDOMO, J.G.: 'Harmonic power flow for unbalanced systems'. IEEE PES, Winter Meeting 93, WM 061-2 PWRD, 1993
9 STOTT, B., and ALSAC, D.: 'Fast decoupled load flow', IEEE Trans., 1974, PAS
93, pp. 859-867
10 CALLAGHAN, C, and ARRILLAGA, J.: 'A double iterative algorithm for the
analysis of power and harmonic flows at ac-dc terminals', Proc. IEE, November
1989, 136, (6), pp. 319-324
Chapter 6
6.1 Introduction
The simulation of electromagnetic transients associated with HVDC
schemes is carried out to assess the nature and likely impact of overvoltages
and currents, and their propagation throughout both the AC and DC
systems. Transient simulation is also performed for the purpose of control
design and evaluation. Transients can arise from control action, fault
conditions and lightning surges. The response to these events is initially
dominated by electrical resonances in lumped RLC components, and
travelling waves in distributed-parameter components such as overhead
lines and cables. Electromechanical transients, usually studied over longer
timescales, are considered in Chapters 7 and 8.
The complexity of the DC system and the AC-DC system interactions has
necessitated the development of special tools capable of adequate representations. The principal tools are DC simulators (transient network analyser,
TNA), and digital computers.
The DC simulator has long been the most powerful tool for the design of
HVDC systems. It is a versatile tool for simulating dynamic performance
under various conditions, ranging from fundamental frequency to high
frequency; however, it is costly and less flexible than digital simulation. With
the recent advances in digital computer technology, computers are playing
an increasingly important role. At present, both DC simulators and digital
computers are used, often in a complementary manner, in the design and
analysis of HVDC systems.1
The DC simulator is characterised by the scaled physical modelling of all
the power circuits associated with an HVDC system, although actual control
circuits can be used. Every component in the system is modelled either by
its scaled-down physical model or by an electronic model. Being a real-time
model, it is capable of simulating a large number of cases in a short time.
The DC simulator does not need any approximations or simplifications in
the bridge model, and can cover a wide range of frequencies of interest in
real-time operation. However, the extent of AC-network representation is
limited, and the frequency response of the simulator is limited to about that
of the switching surges and below. It is possible to minimise the AC-network
176
limitation by controlling the voltage sources to exactly match the voltage

swings as derived from large-system stability simulations by digital computer
programs,2 or more accurately by an extensive AC-system representation
through a digital simulator, with real-time capability interfaced to the
DC-simulator hardware.
Confidence in the accuracy and validity of the DC-simulator results has
grown due to the fact that all commercial HVDC systems in operation to
date have been designed with considerable support from simulator studies.
Digital simulation is based upon the fast solution of differential equations
describing the network to be solved. There are two formulations for these
equations: transient-converter simulation (TCS) based on state-space analysis,3 and the electromagnetic transient program,4 henceforth called the
EMTP method. The state-space method draws upon a large body of
research developed in control engineering and numerical analysis; it often
employs sophisticated algorithms for the numerical integration of differential equations. In the state-space method, the formulation and then solution
of the differential equations describing the system to be solved are independent tasks, permitting a flexible approach to either which can be tailored to
the system at hand. In contrast, the EMTP method combines the integration
method and problem formulation into a single algorithm, which has turned
out to be well suited to the requirements of power-system engineering.
Both methods work well for linear time-invariant networks, such as those
consisting of lumped RLC components. Distributed-parameter components
such as cables and transmission lines have been better modelled in the more
widely used EMTP method. Linear time-varying networks, in particular
power-electronic devices, have required that specific alterations be made to
both methods. On the whole, the state-variable method is more accurate and
efficient with regard to these devices; however their treatment in the EMTP
method has been very much improved. Hybrid approaches combining the
EMTP and state-space models for the time-invariant and power-electronic
parts of the network, respectively, have been implemented a number of
times. Nonlinear components, such as saturating magnetic paths, have been
modelled to a limited extent in the EMTP method. The EMTP method,
therefore, enjoys widespread use with good representation of all powersystem components associated with HVDC transmission. The state-space
approach offers slightly improved modelling of power-electronic components, and greater mathematical insight.
To overcome the bottleneck of simulation time, the need for a cost
effective real-time simulator is imperative. This demands high-speed processing power which is not achievable by present-day general-purpose
computers. Recent advances in digital-processor technology, coupled with
parallel-processing techniques and significant software development, have
led to the realisation of a fully digital power-system transients simulator
capable of real-time operation. The real-time digital simulator (RTDS),
177
developed by the Manitoba HVDC Research Centre,5 provides a very

powerful tool for the power-systems engineer since it combines the benefits
of traditional transient network analysers and software-based simulation
packages.
The RTDS employs the EMTP method of simulation, which is able to be
implemented in a parallel form. The very fast solution enables a large
number of studies to be performed not only on HVDC systems but also for
the testing of protective devices. Furthermore, by interfacing with a traditional HVDC simulator the digital simulator can represent an extensive AC
network resulting in an economic representation of a large system in hybrid
operation. Moreover, it can serve as an excellent tool for training personnel.
The duality between time and frequency-domain analysis means that any
software developed for electromagnetic-transient analysis can also be applied to harmonic analysis simply by simulating to the steady state and
taking the Fourier transform over one cycle. This further improves the
versatility of such software, and the long simulation time required to achieve
the steady state becomes less of a problem with the continuing advance in
computer technology. With very fast computers becoming widely available,
it is feasible to obtain many steady-state solutions as part of a harmonic
perturbation study. This permits the equivalent harmonic impedance of an
HVDC link or any other power-electronic device to be calculated accurately.6 Such information is invaluable in analysing harmonic interaction between the AC and DC systems, filters and other HVDC or FACTS (flexible
AC transmission system) devices.
6.2 The state-variable solution

A state-space AC-DC dynamic simulation program specifically developed
for HVDC transmission is TCS (transient converter simulation).7
In TCS the AC-system components are represented by equivalent circuits,
the parameters of which can be time and frequency dependent. The
variation in time may be due to generator dynamics following disturbances
(a subject discussed in Chapters 7 and 8) or to component characteristic
nonlinearities. The latter mainly occur at the converter transformers which,
following disturbances, may be subjected to overexcitation, DC magnetisation and in-rush effects, all of which must be represented in dynamic
simulation studies.8
The frequency response of power-system components affects the transient
response owing to the multitude of frequencies present in transient waveforms. Given the size and complexity of a power system, it is not practical
to model each of its components individually. Instead, TCS uses frequencydependent equivalent circuits, derived from the actual AC system being
represented. 910
178
6.2.1 Structure ofTCS

Data input is considerable for dynamic analysis representation, even for
simple systems. Single branch entries and manual formulation of network
differential equations is a tedious task in such cases. An automatic procedure
for data input is adopted in TCS and collation of the data into the full
network is left to the computer.
For each system component, a set of control parameters provides all the
information needed by the program to expand the given component data
and to convert it to a form required by the program. The component data
set contains the initial current information and other parameters relevant
to the particular component. For example, for the converter bridges this
includes the initial DC current, delay and extinction angles, time constants
for the firing-control system, the smoothing reactor, converter transformer
data, etc. Each component is then systematically expanded into its elementary RLC branches and assigned appropriate node numbers. Crossreferencing information is created relating the system busbars to those node
numbers. The node voltages and branch currents are initialised to their
specified instantaneous phase quantities of busbar voltages and line currents, respectively. If the component is a converter, bridge valves are set to
their conducting states from knowledge of the AC busbar voltages, the type
of converter transformer connection and the set initial delay angle.
The procedure described above, when repeated for all components,
generates the system matrices in compact form with their indexing information, assigns node numbers for branch lists and initialises relevant
variables in the system.
Once the system and controller data are assembled, the system is ready
to begin execution. In the data file, the excitation sources and control
constraints are entered followed by the fault specifications. The basic
program flow chart is shown in Figure 6.1. For a simulation run, the input
could be either from the data file or from a previous snapshot (stored at the
end of a run).
Simple control systems can be modelled by sequentially assembling the
modular building blocks available. Control-block primitives are provided
for basic arithmetic such as addition, multiplication and division, an integrator, a differentiator, pole-zero blocks, limiters, etc. The responsibility to
build a useful continuous control system is left to the user.
At each step of the integration process, converter bridge valves are tested
for extinction, voltage crossover and conditions for firing. If indicated,
changes in valve states are made and the control system is activated to adjust
the phase of firing. Moreover, when a valve switching occurs, the network
equations and the connection matrix are modified to represent the new
conditions.
During each conduction interval the circuit is solved by numerical
integration of the state-space model for the appropriate topology. Each

start
input data and establish
network equations
initialise network and controls
monitor and control

converter operation
modify equations and

produce output
update nodal inductance matrix
(L,ineqn. 6 2.15)
solve network equations and control

system blocks for one time step
advance time by h
modify equations according to
store all variables and state

information for post-processing
Figure 6.1
TCS flow chart
179
180
state-space model is of the form
x=f(t,x)
which for a linear time invariant system becomes
x = [A]x + [B]u
(6.2.1)
where x is the vector of state variables, [A] is the system matrix and u the
vector of sources. The system matrix can provide useful information about
the range of time constants and resonant frequencies in the system. A
summary of state-space analysis is provided in Appendix IV. Eqn. 6.2.1 is
solved by numerical integration or evaluation of an explicit closed-form
solution. Numerical-integration techniques are discussed in Appendix V.
6.2.2 Integration method

TCS employs an iterative implicit trapezoidal integration that has gained
wide acceptance owing to its good stability, accuracy and simplicity.3'11
Details of the method are given in Appendix V.
The integration method is illustrated by the following iterative procedures:
(i)
(ii)
For an initial estimate it is assumed tht xn+i = xn.

An estimate of xn + 1 based on the xn+1 estimate is obtained
(iii) The state variable derivative xn + 1 is then estimated from state equation
*n + l=/0n + M* w + l)
(iv)
(6.2.3)
Steps (ii) and (iii) are performed iteratively until convergence is

reached. Convergence is deemed to have occurred when all the state
variables satisfy
ÂBS{(xjn++\-xjn+1)}
(6.2.4)
where is the convergence tolerance. It is sometimes necessary to

specify an additional convergence constraint to ensure the state
181
variable derivatives have converged sufficiently, i.e.
j:i-*J+i)}
(6.2.5)
where d 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 the efficiency
of the simulation.
6.2.3 Choice of state variables

Although inductor current and capacitor charge are the most commonly
chosen state variables, it is better to use the flux linkage of the inductor (\j/)
and the capacitor charge (Q) as the state variables. Regardless of the
numerical integration algorithm used to solve the different equations, this
will reduce the error propagation caused by local truncation errors. 12 It is,
therefore, possible to use large step lengths without increasing the number
of iterations.13
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 recognise when
inductors and capacitors in a network are dependent or independent.
The use of capacitor charge or voltage as a state variable creates a problem
when a set of capacitors and voltage sources forms a closed loop. In this case,
the standard state-variable formulation fails, as one of the chosen state
variables is a linear combination of the others. This is a serious problem as
many power-system elements exhibit this property (e.g. the transmissionline model). There are several ways of overcoming this problem; TCS uses
the charge at a node rather than capacitor voltage as a state variable.
A dependent inductor is one with a current which is a linear combination
of the current in k other inductors and current sources in the system. This
is not always obvious owing to the presence of intervening networks. When
only inductive branches and current sources are connected to a radial node,
and if the initialisation of state variables was such that the sum of the current
at this radial node was nonzero, 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.8 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 variable rather than a state variable.12
However, each approach has some disadvantages. The phantom current
182
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 complicated and time consuming. An inductor
can still be dependent even if it is not connected directly to a radial node
consisting of inductive branches but has an intervening resistor/capacitor
network.
The identification of state variables can be achieved by developing a
node-branch incidence matrix where the branches are ordered in a particular pattern (e.g. current sources, inductors, voltage sources, capacitors,
resistors) and Gaussian elimination performed. The staircase columns
represent state variables.12 However, this is computationally expensive if it
is required frequently, such as in the case of converter switchings.
6.2.4 Forming the network equations

Although the state-space formulation can handle any topology, the automatic generation of the system matrices and state equations is a complex
and time-consuming process, which needs to be done every time switching
occurs. By restricting the allowable topologies and the application of tensor
analysis, the generation of system matrices and state equations is simpler
and computationally efficient. The main reason for the restriction of
topologies is computational efficiency. If a general matrix formulation and
elimination of node-branch incidence matrix was performed each time a
switching occurred (to determine the independent state variables), the
algorithm would be very slow but more general. There is therefore a
trade-off between generality and computational efficiency. The fundamental
branch representations used in TCS are illustrated in Figure 6.2.
By defining the capacitive node charge Q and inductive branch flux *F as
state variables
(6.2.6)
the vectors Va and Ilf representing the capacitive node voltage and inductive
branch currents, respectively, become dependent variables related to the
state variables by algebraic relations.
The resulting set of equations can be written as
state-related variables
(6.2.7)
(6.2.8)
183
h0
Figure 6.2 Fundamental branch representations in TCS
dependent variables
Tf __
TO
rjz
(6.2.9)
K{.V
(6.2.10)
(6.2.11)
and the state equations

(6.2.12)
(6.2.13)
where
(6.2.14)
'Xr
(6.2.15)
(6.2.16)
E, V and I represent the branch e.m.f. source voltage, node voltage and
branch current vectors, respectively.
R, L and C are the resistance, inductance and capacitance matrices,
respectively. The p operator in TCS represents a derivative w.r.t. electrical
angle rather than time.
The rest of the symbols and suffixes in the above equations represent the
following
a = node with at least one capacitive branch connected
/? = node with at least one resistive but no capacitive branches
connected
184
y = node with inductive but no resistive or capacitive branches

connected
r, lfc,s = resistive, inductive, capacitive or current source branch
The above definitions give the topological or branch node incidence
matrices, their general elements being
Kfpi = 1 if node i is the sending end of branch p
= 1 if node i is the receiving end of branch p
= 0 if node i is not connected to branch p
where K^y, K'^, K'y are null by definition.
Integrating for Q a and xl and substituting in the dependent equations,
the total system solution can be obtained.
Valve on/off states are represented by a short-circuit or an open-circuit
condition and the network topology is changed accordingly.
6.2.5 Valve switchings

When a converter valve satisfies the conditions for conduction, i.e. sufficient
forward voltage with an enabled firing-gate pulse, it will be switched to the
conduction state. If the valve-forward voltage criterion is not satisfied the
pulse is retained for a set period without upsetting the following valve.
Accurate prediction of valve extinctions is a difficult and time-consuming
task which can degrade the solution efficiency. Sufficient accuracy is
achieved by detecting extinctions after they have occurred, as indicated by
valve-current reversals; by linearly interpolating the step length to the
instant of current zero the actual turn-off instant is assessed as shown in
Figure 6.3. Only one valve may be extinguished per bridge at any one time,
and the earliest extinction over all the bridges is always chosen for the
interpolation process. By defining the current (I) in the outgoing valve at
the time of detection (t\ when the step length of the previous integration
step was h, the instant of extinction tx will be given by
tx = t~hz
(6.2.17)
where
All the state variables are then interpolated back to tx by

Vx = V , - z ( V t - V , _ )
(6.2.18)
185
(a)
Time
Figure 6.3 Interpolation of time upon valve-current reversal
The dependent state variables are then calculated at tx from the state
variables, and written to the output file. The next integration step will then
begin at tx with step length h as shown in Figure 6.3fe. This linear
approximation is sufficiently accurate over periods which are generally less
than one degree, and is computationally inexpensive. The effect of this
interpolation process is clearly demonstrated in a case with an extended
1 ms time step in Figure 6.5.
Upon switching any of the valves, a change in the topology has to be
reflected back into the main system network. This is achieved by modifying
the connection matrices. When the time to next firing is less than the
integration step length, the integration time step is reduced to the next
closest firing instant. Since it is not possible to integrate through discontinuities, the integration time must coincide with their occurrence. These
discontinuities must be detected accurately since they cause abrupt changes
in bridge-node voltages, and any errors in the instant of the topological
changes will cause inexact solutions.
Immediately following the switching, after the system matrices have been
reformed for the new topology, all variables are again written to the output
file for time tx. The output file therefore contains two sets of values for tx
immediately preceding and after the switching instant. The double solution
at the switching time assists in forming accurate waveshapes. This is
especially the case for the DC-side voltage which almost contains vertical
jump discontinuities at switching instants.
186
6.2.6 Effect of automatic time-step adjustments

It is important that the switching instants be identified correctly, first for
accurate simulations and, secondly, to avoid any numerical problems associated with such errors. This is a property of the algorithm rather than an
inherent feature of the basic formulation. Accurate converter simulation
requires the use of a very small time step, where the accuracy is only
achieved by correctly reproducing the appropriate discontinuities. A smaller
step length is not only needed for accurate switching but also for the
simulation of other nonlinearities, such as in the case of transformer
saturation, around the knee point, to avoid introducing hysteresis due to
overstepping. In the saturated region and the linear regions, a larger step
is acceptable.
On the other hand, state-variable programs, and TCS in particular, have
the facility to easily adapt to a variable step-length operation. The dynamic
location of a discontinuity will force the step length to change between the
maximum and minimum step sizes. The automatic step-length adjustment
built into the TCS program takes into account most of the influencing
factors for correct performance. As well as reducing the step length upon
the detection of a discontinuity, TCS also reduces the forthcoming step in
anticipation of events such as an incoming switch as decided by the firing
controller, the time for fault application, closing of a circuit breaker, etc.
To highlight the performance of the TCS program in this respect, a
comparison is made with an example quoted as a feature of the NETOMAC
program.14 The example refers to a test system consisting of an ideal AC
system (e.m.f. sources) feeding a six-pulse bridge converter (including the
converter transformer and smoothing reactor) terminated by a DC source.
N\
I
(a) SO /is time step
Figure 6.4 NETOMAC simulation responses

IEEE, 1986 Reproduced by permission
N\
(b) 1 ms time step
187
-o-o- valve 1 voltage; -*--*- valve 1 current
Time
Figure 6.5
TCS simulation with 1 ms time step (firing angle 25)
The firing angle was fixed at 25. Figure 6.4 shows the valve voltages and
currents for 50 fi$ and 1 ms (i.e. 1.0 and 21 degrees) time steps, respectively.
The system has achieved steady state even with steps as large as 20 times.
The progressive time steps are illustrated by the dots on the curves in
Figure 6Ab9 where interpolation to the instant of a valve-current reversal is
made and from which a half time-step integration is carried out. The next
step reverts back to the standard trapezoidal integration until another
discontinuity is encountered.
A similar case with an ideal AC system terminated with a DC source was
simulated using TCS. A maximum time step of 1 ms was used also in this
case. Steady-state waveforms of valve voltage and current derived with a
1 ms time step, shown in Figure 6.5, illustrate the high accuracy of TCS
both in detection of switching discontinuities and reproduction of the
50 /*s results. The time-step tracing points are indicated by dots on the
waveforms.
Further TCS waveforms are shown in Figure 6.6 giving the DC voltage,
valve voltage and valve current at 50 jus and 1 ms (although the actual step
was modulated under the program's control).
In the NETOMAC case, extra interpolation steps are included for the 12
switchings per cycle in the six-pulse bridge. For the 60 Hz system simulated
with a 1 ms time step, a total of 24 steps per cycle can be seen from the
Figure 6.46, where a minimum of 16 steps are required. The TCS cases
shown in Figure 6.6 have been simulated with a 50 Hz system. The 50 jus
case of Figure 6.6a has an average of 573 steps per cycle with the minimum
requirement of 400 steps. On the other hand, the 1 ms time step needed
only an average of 25 steps per cycle. The necessary sharp changes in
188

- - valve I voltage<pu)
Rectified dc vollage(pu)
valve 1 cunent(pu)
- valve I cunem(pu)
Rectified dc voliage(pu)
valve 1 voltage(pu)
rJVW
1
l\
ri
'
1
\
0.460
n/"i/
0.468
Figure 6.6
\
\
0.476
0.484
VV"
0.492
0.500
0.468
0.476
0.484
Time(s)
Time(s)
(a) 50 us time step
(b) 1 ms time step
0.492
0.500
Steady-state responses from TCS (firing angle of 25)
waveshape are derived directly from the valve voltages upon topological
changes.
When the TCS frequency was increased to 60 Hz, the 50 jus case used
fewer steps per cycle, as would be expected, resulting in 418 steps compared
to a minimum required of 333 steps per cycle. For the 1 ms case, an average
of 24 steps were required, as for the NETOMAC case.
The same system was run with a constant-current control of 1.225 p.u.,
and at 0.5 s a DC short circuit was applied. The simulation results with 50 fis
and 1 ms step lengths are shown in Figure 6.7. This indicates the ability of
the TCS to track the solution and treat waveforms accurately during
transient operations (even with such an unusually large time step).
0.48
dc fault application
x Rectified dc voitage(pu)
A dc current(pu)
0.60
0.52
0.54
0.56
Time(s)
(a) 1 ms time step
Figure 6.7
0.58
0.60
0.48
dc fault application
x Rectified dc voltage(pu)
A dc current(pu)
0.50
0.52
0.54
0.56
Time(s)
(b) 50 jus time step
Transient simulation with TCS; DC short circuit at 0.5 s
0.58
0.60
189
6.2.7 TCS converter control

Present day control schemes are variations of the phase-locked oscillator
method developed by Ainsworth.15 A modular control system16 is used
which includes blocks of logic, arithmetic and transfer functions, enabling
simulation of feedback type controllers and improving the descriptive
representation of the control system.
TCS converter units include a built-in valve-firing control mechanism.
The converter bridges are modelled as six-pulse units either in star-star or
star-delta transformer configurations.
Valve firing and switchings are handled individually on each six-pulse
unit of a converter. For 12-pulse operation both bridges are synchronised
and the firing controllers phase-locked loop is updated every 30 compared
to the 60 used for the six-pulse converter.
The present version of TCS mirrors one recently implemented in the
EMTDC program discussed in section 6.6.2.
The firing-control mechanism implemented is equally applicable to six or
12-pulse valve groups; in both cases the reference voltages are obtained
from the converter commutating bus voltages. When directly referencing to
the commutating bus voltages any distortion in the bus voltage may result
in a valve-firing instability. To avoid this particular problem, a three-phase
phase-locked oscillator (PLO) is used instead which attempts to synchronise
the oscillator through a phase-locked loop with the commutating busbar
voltages.
In the simplified diagram of the control system illustrated in Figure 6.8,
the firing controller block (NPLO) consists of the following functional units:
(i)
(ii)
(iii)
(iv)
(v)
a zero-crossing detector unit;

AC-system frequency measurement;
a phase-locked oscillator;
firing-pulse generator and synchronising mechanism;
angle (a and y) measurement unit, etc.
Three phase
input voltages
NPLO
"order
Converter
Firing
Control System
VALFIR
I
Valve States
Feedback
Control
System
orP?
off
' measured
c .(fc'
Figure 6.8 Phase-locked oscillator-basedfiring-controlmechanism
'POggO'^-
190
Zero crossover points are detected by the change of sign of the reference
voltages and multiple crossings are avoided by allowing a space before
making the next crossing. Distortion in line voltages can cause difficulties
in proper zero-crossing detection, and the voltages must be smoothed
before being passed to the zero-crossing detector.
The time between two consecutive zero crossings, of the positive to
negative (or negative to positive) going waveforms of the same phase, is
denned here as the half period time, T/2. The measured periods are
smoothed through a first-order real-pole lag function with a user-specified
time constant. From these half-period times the AC-system frequency is
estimated, every 60 (30) for a six (12) pulse bridge.
Normally, the ramp for firing of the particular valve (c(l).. .c(6)) starts
from the zero crossing points of the voltage waveform across the valve. After
T/6 time (T/12 for 12 pulse), the next ramp starts for the firing of the next
valve in sequence. So, a of 0 for the incoming valve corresponds to the end
of T/6 time from the previous ramp, and so on.
It is possible that during a fault or due to the presence of harmonics in
the voltage waveform, the firing does not start from the zero crossover
point, resulting in a synchronisation error, B2, as shown in Figure 6.9. This
error is used to update the phase-locked oscillator which, in turn, reduces
the synchronising error, approaching zero at steady-state conditions. The
synchronisation error is recalculated every 60 (every 30 for 12 pulse).
The firing-angle order (oeorder) is converted to a level to detect the firing
instant as a function of the measured AC frequency by
Figure 6.9
q order (rad)
/AC(P.U.)
Synchronising error in firing pulse

(Source: Figure 7.3 of Reference 29)
(6.2.19)
191
As soon as the ramp c(n) reaches the set level specified by T o , as shown in
Figure 6.9, valve n is fired and the firing pulse is maintained for 120. Upon
having sufficient forward voltage with the firing pulse enabled, the valve is
switched on and the firing angle recorded as the time interval from the last
voltage zero crossing detected for this valve.
At the beginning of each time step, the valves are checked for possible
extinctions. Upon detecting a current reversal, a valve is extinguished and
its extinction angle counter is reset. Subsequently, from the corresponding
zero-crossing instant, its extinction angle is measured, e.g. at valve 1 zero
crossing, y2 is measured, and so on. (Usually, the lowest gamma measured
for the converter is fed back to the extinction angle controller.) If the
voltage-zero crossover points do not fall on the time step boundaries, a
linear interpolation is used to derive them.
As illustrated in Figure 6.8, the NPLO block co-ordinates the valve-firing
mechanism, and VALFIR receives the firing pulses from NPLO and checks
the conditions for firing the valves. If the conditions are met, VALFIR
switches on the next incoming valve and measures the firing angle, otherwise it calculates the earliest time for next firing to adjust the step length.
Valve currents are checked for extinction in EXTNCT and interpolation of
all state variables is carried out. The valve's turn on time (Ton) is used to
calculate the firing angle and off time (Toff) is used for the extinction
angle.
To verify the operation of the firing controller implemented in the TCS
program, two cases are presented in Figures 6.10-6.12 using the same test
system as in section 6.2.6. A constant aorder of 15 was used in the case of
Figures 6.10 and 6.11, where a step change was applied at 0.5 s for five
cycles by increasing the aorder to 75.
- d c Current
Rectified dc voltage(pu)
Firing angle(rad)
Extinction angle(nid)
Rectified dc voltage
A dc Current
3.0
. . .
.
1.5
Firing angle measured
Extinction angle measured
1
i f. u j i u .u n . ' i 1 V
v-
V-J
iiiinii
0.0
0.450 0.475 0.500 0.525 0.550 0.575 0.600 0.625 0.650
i.i.iih
iilillii
-1.5
0.450 0.475 0.500 0.525 0.550 0.575 0.600 0.625 0.65
Time(s)
(a) TCS
niiiii
Time(s)
response
(b) EMTDC response
Figure 6.10 Constant aorder operation with a step change
192

Rectified dc voltage(pu)
0.45
0.50
0.55
0.60
0.65
Rectified dc voltage
0.70
-1.0 40.75 0.45
0.50
0.55
Time(s)
0.60
0.65
0.70
0.75
Time(s)
(a) TCS
response
(b) EMTDC response
Figure 6.11 Constant-current control with a step change; DC voltage
o dc Current(pu)
v Firing angle measured (rad)
Firing angle measured (rad)
~~ dc Current(pu)
2.0 T
0.45
0.50
0.55
0.60
'"
'
0.65
0.70
Time(s)
(a) TCS response
-0.5
0.75
0.45
0.50
0.55
0.60
0.65
0.70
0.75
Time(s)
(b) EMTDC response
Figure 6.12 Constant-current control with a step change; DC current and a
In the case of Figures 6.11-6.12, an initial constant-current control of

1.225 p.u. was set up and at 0.5 s the current order was reduced by 50% for
five cycles. The responses are compared with similar control implementations of EMTDC (with a 20 fis time step) as shown in the respective figures.
Complete agreement was found in all cases.
6.3 The EMTP method4

The EMTP method is an integrated approach to the problems of:
(i) forming the network differential equations;
(ii) collecting the equations into a coherent system to be solved;
(iii) numerical solution of the equations.
193
vm
Figure 6.13 Resistive branch
The method is based upon the discretisation of components, given a

predetermined time step, which are then combined in a solution for the
nodal voltage. Branch elements are represented by the relationship which
they maintain between branch current and nodal voltage.
63.1 Discretisation of system components

The simplest element is a resistor connected between nodes k and m, as
shown in Figure 6.13
(6.3.1)
or in nodal form as
1
ik
1"
R
(6.3.2)
L~R at nodes k and m.

where lk, Im are the current injections
Resistors are accurately represented in the EMTP formulation, apart from
the proviso that R should not be too small. A consequence of R being too
small is that 1/R in the system matrix will be large, resulting in poor
conditioning of the solution at every time step.
For an inductor
'
dikm(t)
dt
(6.3.3)
Integrating from the previous time step
- AO = i f
- \m(x)]dx
(6.3.4)
194
*- At
Figure 6.14 Trapezoidal integration of inductor current
as illustrated in Figure 6.14. The right-hand side of eqn. 6.3.4 is approximated by the shaded trapezoidal, yielding
ikm(t) - ikm(t - At) s i At|j-(V 4 (0 - Vm(0) + i(Vk(t - A/) - VJt - A/))]

or
2^,(0-v m (0]
2L'
(6.3.5)
The branch current, therefore, consists of two components, one related to
the nodal voltages at the current time step which are to be solved, and the
other independent of the current-time step voltages. The independent
component can be represented as a current source, and the dependent
component as a branch resistance, yielding the Norton equivalent, shown
Utm
2L
A/
Figure 6.15 Inductive branch
195
in Figure 6.15. For the inductor

At
(6.3.6)
where
IkmO At) = ikm(t At) H
tVk(^ At) '
is often referred to as the history term.

Eqn. 6.3.6 can be rewritten in nodal form as
2L
-2L
it
-2L
2L
A*
The analysis for a capacitor is similar

km
(6.3.8)
dt
and integrating from the previous time step yields
V k (0 ~ VM(0 - (Vk(t - At) - Vm(* - At)) = i f' ik
(6.3.9)
Applying the trapezoidal rule

'""
/ - At)]
(6.3.10)
2C
2C
-g \yma -At)-
yk(t -
= ^ [V*(0 - V m (0] + lkm(t - At)
- At)
(6.3.11)
where
Ifcm(* - At) = ^
[Vm(i - A/) - Vk(< - A/)] - ikjt - At)
(6.3.12)
196
AL
2C
Figure 6.16 Capacitive branch
The resulting Norton equivalent, shown in Figure 6.16, is again dependent

on the time step, A/. In nodal form, eqn. 6.3.11 becomes
M_
2C
2C
-Af
2C
Vk
2C
A detailed description of transmission-line modelling is deferred to section

6.7.3, where a multiphase frequency dependent model is described. Singlephase lossless lines are a useful introduction, however, and are described by
the differential wave equations
dv
di
"dx~ Jt
dx
dt
(6.3.14)
(6.3.15)
where L and C are the distributed parameter inductance and capacitance

per unit length and x is distance along the line from the ending end.
The solution to eqns. 6.3.14 and 6.3.15 is a superposition of two travelling
waves moving along the line in different directions
i=f1{x-si)-f2(x+st)
(6.3.16)
where z = ^/L/C and s is the wave speed. Consider a wave starting at node
m and moving towards node k, as shown in Figure 6.17.
Multiplying eqn. 6.3.16 by z and adding it to eqn. 6.3.17 eliminates the
negative travelling wave, i.e.
v(t) + zi(t) = 2zf(x - st)
(6.3.18)
The left-hand side of this equation is a voltage that depends only on x st.
197
km
x=0
x=d
k
Figure 6.17 Propagation of a wave on a transmission line
The same value of x st can be obtained at both x = 0, t = tx (at node m)

or x = d, t = t2 (at node A) by appropriate choice of tx and t2, i.e.
-st1 =d-st2
(6.3.19)
then substituting t2 t and ^ = T in eqn. 6.3.19

_d
s
(6.3.20)
Thus
v(t - t) 4- zi(t -
T)
= t/(0 + zt'(0
(6.3.21)
However, at tl9 the right-hand side of eqn. 6.3.21 corresponds to node k,

and the left-hand side to node m
vm(t - T) + zimk(t - T) =
(6.3.22)
(6.3.23)
where
(6.3.24)
Similarly
(6.3.25)
198
Figure 6.18
Transmission-line equivalent
where
(6.3.26)
The transmission line has therefore been replaced by two Norton equivalents (Figure 6.18).
In nodal form, eqns. 6.3.23 and 6.3.24 become
With all system components replaced by Norton equivalents, the EMTP

solution proceeds at every time step by solving a linear system for the nodal
voltages throughout the network
[I] = [Y][V] + [Ihist]
(6.3.28)
The linear system is constructed by combining the nodal equations for every
system component with due regard to the system topology.
Current sources are modelled merely as an additive term to the left-hand
side of eqn. 6.3.28. With voltage sources included, part of the right-hand
side of eqn. 6.3.28 is known, as opposed to the left-hand side. A partitioning
of Y is therefore required with voltage source nodes ordered last
(6.3.29)
where the history current terms have been included in the left-hand side,
VT are the specified voltage sources, and IN are the known Norton currentsource injections. This system is solved by first finding V from
[IN - BVT] = [A]V
(6.3.30)
and then the injection at voltage sources from

I = [C]V + [D]VT
(6.3.31)
199
Provided that neither the system topology nor the time step change, the
system conductance matrix [Y] is constant throughout the simulation. [Y] is
also sparse, since each node is connected to only a few others. The presence
of transmission lines may also cause [Y] to be block diagonal, affording a
breakdown of the system to be solved into smaller subsystems (section 6.4).
The linear system, represented by eqn. 6.3.28, is solved by Gaussian
elimination and LU decomposition.
A simpler alternative implementation of voltage sources is to assume that
they have a source impedance |zsource| > R ^ , and to model them as Norton
equivalents. This completely avoids the requirement of having to partition
the conductance matrix as in eqn. 6.3.29.
63.2 Error analysis

The trapezoidal integration method used, although stable, introduces errors
which generally get smaller as the step length is decreased. However, if the
step length is made too small, errors increase again due to poor conditioning of [Y]. The poor conditioning is caused by the presence of both
inductance, which contributes a term Af/2L to [Y], and capacitance, which
contributes 2C/At. As At decreases, these two terms differ widely in scale.
However, in practice, this type of error has not so far been a problem with
double precision computations. A possible solution, in any case, is to
precondition [Y] before factorising. More significant errors are introduced
owing to the approximate solution for inductance and capacitance, however.
The trapezoidal integration method introduces a parallel stray capacitance
across inductors and adds a series inductance to capacitors.17
The trapezoidal integration for an inductor yields the exact solution for
a short-circuited lossless stub transmission line (shown in Figure 6.19).
The travel time for the line is
T= j
(6.3.32)
If the length of the line is /, then L = L7 and

T = ly/(L'C)
rnr^
>
V
- r 'Â
L 1
(6.3.33)
onrn
J_ <t
s
m
Figure 6.19 Transmission-line stub solution for an inductor yielded by trapezoidal
integration
200
so that z = 2h/At, where L' and C are the line inductance and capacitance
per unit length. A wave leaving k at t At travels to 5 by t At/2.
From eqn. 6.3.22, in slightly different form
Vkm(t - At) + zikm(t -At) = zis(t - ^j
(6.3.34)
The wave travels from 5 at t At/2 to m at t

~zis (t - j ) = ykm(t) = zikm(t ~ A0
(6.3.35)
Substituting eqn. 6.3.35 into eqn. 6.3.34 yields

^jVkm(t-At) + ikm(t-At)^
(6.3.36)
The total parallel capacitance is

At2
C = Cl = 4L
(6.3.37)
For example, if At = 50 jus and L = 0.5 H, then the trapezoidal integration

introduces capacitance of 1.25 nF. Since the exact frequency-dependent
impedance of both inductor and stub line are known, the frequencydependent error in the trapezoidal integration solution for an inductor is
readily calculated. For the stub line, the impedance is
Y
sc
*
tan(a)Z(L'C'))
coV(L'C')_
1
coAt/2
ijteiL'
tan(coA*/2) jcoL
' ' '
Consequently, the ratio of the impedance of the stub line to the impedance
of the inductor is
(6.3.39)
which has been plotted in Figure 6.20 as a function of the Nyquist frequency
1/2A2 of the simulation.
An error analysis for the trapezoidal integration of a lumped capacitance
is the dual of that for an inductor. The trapezoidal integration yields the
exact solution for an open-circuit stub line of transit time x = A//2 with
C7 = C, z = AJ/2C so that L = A*2/4C. Figure 6.20 is, consequently, also a
plot of the ratio of the actual admittance of the capacitor to the admittance
of the open-circuited stub line calculated by the trapezoidal integration.
201
Trapezoidal integration error

^^_
>
_^_
._.
,j
"1
0.9
0.8
0.6
or
0.4
: Y;
0.3-
0.2-
0.1
Figure 6.20
633
0.1
0.2
0.3
0.4
0.5
0.6
0.7
: i\
0.8
0.9
Trapezoidal integration error
Switching discontinuities
T h e basic EMTP-type algorithm requires modification in order to accurately

a n d efficiently model the switching actions associated with HVDC, thyristors, FACTS devices, or any other piecewise linear circuit. T h e simplest
approach is to simulate normally until a switching is detected a n d then
u p d a t e t h e system topology and/or conductance matrix. A switch can b e
represented, as shown in Figure 6.21, by either a n on/off resistance o r off
resistance only.
T h e former m e t h o d cannot represent an ideal switch, since R ^ must b e
large enough n o t to decondition the system conductance matrix. In practice, this coincides well with real switching devices which are n o t ideal either.
T h e second m e t h o d avoids the use of small resistances b u t requires a m o r e
severe change in system topology since there is o n e fewer n o d e in t h e
conduction state than in t h e off-state.
For either representation, the system-conductance matrix must b e reformed a n d factorised after each change in conduction state. This considerably increases t h e computational requirements of the simulation in
202

On
(a)
(b)
Figure 6.21
Switch representation
proportion to the number of switching actions (recall that the efficiency of

the EMTP technique lies in the fact that the conductance matrix is held
constant to avoid refactorisation). Nevertheless, for HVDC and most FACTS
applications, the switching rate is only several kHz, so that the overall
simulation is still fast. As an example, the CIGRE benchmark system
(Appendix III) was simulated with all valves blocked to assess the processing
overheads associated with refactorising the conductance matrix. The
results, presented in Table 6.1, indicate that in this case the overheads are
modest.
The reason for the small difference in computation time is the ordering
of the system. Frequently switched elements (thyristors, etc.) are ordered
last. However, infrequently switching branches are also included (such as
fault branches and CBs) which increase the processing at every switching
even though they switch infrequently. The CIGRE system used as an
example here is representative since larger systems would most likely be
broken into several subsystems, so that the ratios of switchings to system size
would be likely to be smaller.
In virtually all cases switching action or other point discontinuities will not
fall exactly on a time step, yielding a substantial error in the simulation as
shown in Figure 6.22 for a diode-ending conduction.
Table 6.1
Unblocked
Blocked
Overheads associated with repeated conductance matrix refactorisation

Time step
Number of refactorisations
Simulation time
10 fis
50 fis
10 ^s
50 us
2570
2480
1
1
4 mm 41 s
1 min21 s
4 min 24 s
1 min 9 s
203
Figure 6.22 Apparent reverse current in a diode due to placement of simulation

time steps
At (A) the diode should be switched off, but the negative current is only
detected at (B) and the conductance matrix is reformed at (C). This
approach is inadequate for HVDC applications, since any error in the
thyristor switching instants yields noncharacteristic harmonics on the AC
and DC sides. This is due to the fact that the converter acts as a modulator
of the large DC-side current so that a small variation in switching instant
yields a substantial current on the AC side. This effect is important for SSR
studies, configurations involving GTOs and the calculation of uncharacteristic harmonics.
An effective solution of this problem is to apply a linear interpolation at
(B) to find all the nodal voltages at a point very close to (A). An exact
solution for the zero crossing is a nonlinear problem; however, given the
close proximity of points t and t + At, a linear interpolation between them
introduces no significant error.
In the above example, the diode model must include logic which detects
that a switching has occurred between t and t + At, and an estimate, t + xv
of when. For a diode, a linear interpolation on the forward current yields
T =
Mf(t)
if(t)-if(t + At)
(6.3.40)
The nodal voltages at t + t are then approximately

v(t
^ v(t) + [v(t
At
At) - v(t)]
(6.3.41)
and similarly for branch currents in the system. If simulation proceeds

normally from t + T with the new conductance matrix, subsequent solution points will be shifted by t. For convenience, an additional interpolation can be made after the first time step with the new conductance matrix
to bring the simulation back onto the original time sequence, yielding the
sequence of steps illustrated in Figure 6.23. Another option is to accept
204
new conductance matrix
Figure 6.23 A double interpolation scheme to find the switching instant and
resynchronise time steps
unequally-spaced points in the solution (which complicates Fourier analysis

of the resulting waveforms), or fit a cubic spline to the solution and re-index
to any desired time base. In the PSCAD/EMTDC package, the interpolation
scheme of Figure 6.23 is used but with two additional interpolations
introduced to eliminate voltage and current chatter, to be discussed in the
next section.
6.3 A Voltage and current chatter due to discontinuities

As illustrated by Figure 6.20, the trapezoidal integration has the effect of
increasing the apparent impedance of inductors at high frequencies approaching the Nyquist rate. A corollary is that large voltages will be
developed across inductors if high-frequency currents are injected into
them. This effect is manifested most notably with respect to switching
actions or point discontinuities which necessarily contain high frequency
components.
Figure 6.24 shows an inductor in series with a diode ending conduction.
Assuming that the inductor current falls steadily to zero at T, the voltage
will be constant at VL. At T the diode switches off and the inductor voltage
T-2At
T-At
T+At
T+2At
Figure 6.24 A circuit which will display voltage chatter after the diode switches off
205
(a)
T-2A/
T-Af
T+Af
T+2Ar
(b)
v
Figure 6.25 a Correct solution for voltage across the inductor
b Chatter voltage at the diode anode
will fall instantaneously to zero, since dijdt = 0. The inductor voltage is

shown in Figure 6.25#.
The trapezoidal integration equation for the inductor is
91
VL(0 = ^ ( t L ( 0 - iL(t - A0) - VL(* - At)
(6.3.42)
From (t + At) onwards

= *L(* - M) = 0
so
VL(0 = - VL(* - At)
(6.3.43)
The inductor voltage consequently oscillates between VL instead of

falling to zero (Figure 6.256). Note, however, that the average inductor
voltage is correct at zero, and that the chatter does not grow larger. A
number of approaches to removing this chatter have been proposed and
implemented:
1
2
3
4
post-processing averaging;
half time step interpolation;
critical damping adjustment method;
damping resistors.
206
i>
R
Figure 6.26 Damped inductor method for reducing chatter
Application of the trapezoidal rule to the parallel combination of inductance

and damping resistance in Figure 6.26 yields
If i has abruptly reduced to zero but v(t At) is nonzero, then

v(t)=
-Xv(t-At)
(6.3.45)
where
R - 2L/A/
~ R + 2L/A*
is the decay constant. If R = 2L/A/, X = 0 and there is no chatter. However,
the simulation would be inaccurate because of the relatively small values of
damping resistance. Larger values of damping resistance lead to more
accurate simulations but increasingly persistent chatter after discontinuities.
Setting R = 2L/A/ is called critical damping, and substituting into eqn.
6.3.44
i{t) = i(t -At) + ^v(t)
(6.3.46)
which is the backward or implicit Euler method.

In the critical damping method, backward Euler integration is used for
the first integrating step after a discontinuity. By using a time step of size
At/2, the conductance matrix for the Euler integration is the same as that
for the trapezoidal integration. The network solution at At/2 after the
discontinuity is then used to reinitialise the history terms at the discontinuity to obtain a chatter-free trapezoidal integration solution to A after the
discontinuity. The critical damping method has been implemented in the
EMTP program. A slightly more accurate method which uses only trapezoidal integration, illustrated in Figure 6.27, has been implemented in the
PSCAD/EMTDC program.
Figure 6.27
207
The double half time step interpolation method employed in PSCAD/

EMTDC to remove chatter
(1) - (2)
(2) -> (3)
(3) -> (4)
(4) - (5)
trapezoidal integration
interpolation back to discontinuity, between (1) and (2)
trapezoidal integration with new conductance matrix
interpolation between (3) and (4) of half time step to
remove chatter
(5) -> (6) trapezoidal integration to get past t + At
(6) (7) interpolation between (6) and (5) to resynchronise time
steps
(7) - (8) normal integration resumes
The half time-step interpolation method works because an interpolation

of this length is the same as averaging between two consecutive trapezoidal
integration steps. Since the chatter is an oscillation of 8 around the correct
solution, such an average must eliminate the error. To illustrate the effect
of not interpolating or removing chatter, the circuit of Figure 6.28 is
analysed using PSCAD/EMTDC.
This is a difficult circuit to simulate because of numerical chatter associated with the inductor, and because the diode must turn on almost at the
same time the GTO is switched off. Failure to do this yields large voltages
across the inductor. The results for three cases are presented in Figures 6.29
and 6.30 for the inductor voltage and current, respectively:
(i) with interpolation for switching and chatter removal;
(ii) with interpolation for chatter removal only;
(iii) with no interpolation.
For case (ii) large voltage spikes appear as the inductor is open circuited for
one time step. In case (iii) the voltage spike sets the level of a continuous
voltage chatter between 15 kV.
208

GTO
1.0 Q
VL
0.1 H
Diode
Figure 6.28 Basis offorced commutation in inductive circuits
With Chatter Removal,

With interpolation
With Chatter Removal.

Without Interpolation
200 ~i
100 -
o -100 -200 Without Chatter Removal.

0.00
0.01
0.02
0.03
Time(seconds)
Figure 6.29 Inductor voltage (VL)
0.04
209
With Chatter Removal,

With Interpolation
10 -|
5 0
-5
With Chatter Removal.

10 -i
''.A
-5
Without Chatter Removal,

10
5
0
-5
0.00
0.01
0.02
0.03
Time(seconds)
0.04
Figure 6.30 Inductor current (IL)
6.4 Subsystems
Since the system-conductance matrix is refactorised every time switching
occurs, fast simulation can only be achieved if the factorisation is as efficient
as possible. This entails the use of techniques which utilise the sparsity of
the conductance matrix. The conductance matrix for a system of w-nodes is
of size n x n and, if factored in full, requires of the order of w3 floating-point
operations (0(n3) flops). Only a small fraction of the n2 elements of the
conductance matrix are nonzero, however.
Nonzero elements are present on the diagonal, and wherever two electrical nodes are connected by a nonzero conductance. Typically, each
electrical node is connected to only two others, and therefore contributes
three elements to the corresponding conductance matrix row. As a rough
210
10'
10'
10*
tioa
6
10"
102
10'
/cubic fit
50
100
150
200
250
No. of nodes
300
350
400
450
500
Figure 6.31 Even with sparsity, flops count for conductance-matrix factorisation
increases with the cube of the number of electrical nodes
estimate, there are approximately 3n nonzero elements, yielding a fill factor

of only Sn/n2 = 3/w.
Sparsity methods process (and possibly store) in an indexed fashion only
the nonzero elements, thereby obtaining a speed advantage. In order to
retain sparsity, a compromise is made in the pivoting order between
numerical stability and the introduction of new elements into the partiallyfactored matrix. The number of flops required to complete the reduction
depends on the fill factor, the distribution of the elements in the matrix,
and the condition number. On the whole, the reduction still requires 0(n3)
flops, as illustrated in Figure 6.31. These results were obtained by constructing conductance-like matrices and forming a sparse LU decomposition in
Matlab. Twenty reductions were obtained for each matrix order, and the
flops count was averaged in Figure 6.31. In all cases, the fill factor was
approximately 4/n, so that in practice a greater speed advantage would be
obtained by the use of sparsity. The most important feature of Figure 6.31,
however, is the cubic growth in computation with number of nodes. If a
factorisation is required after each switching action, then the presence of
211
switching devices implies that only small or moderate-sized systems can be

simulated. There are a number of solutions to this problem:
(i)
(ii)
(iii)
(iv)
compensation method;
partial-matrix reduction;
Sherman-Morrison formula;
subsystems.
(i) and (ii) are very similar, and have both been implemented in the EMTP
program, but were eventually discarded in favour of full-matrix reduction.
In (i), the system matrix is not modified after each switching. The correct
solution is obtained by injecting compensating currents at the switch
terminals obtained from a reduced Thevenin equivalent of the network.
The partial matrix reduction method confines the matrix reduction to
that part containing switches, by ordering nodes with switches last and
partitioning
ara
(6.4.1)
so that
= [D-CA ^
(6.4.2)
[I - BVS] = AV
(6.4.3)
In this method, A is fully bifactorised, and used to calculate CA 1 B, by back

substitution with columns from B. After each switching, D is modified and
only [D CA" *B] need be refactored. This method has been implemented
in the EMTDC program.
The presence of transmission lines and cables in the system being
simulated introduces decoupling into the conductance matrix. This is
because the transmission-line model injects current at one terminal as a
function of the voltage at the other at previous time steps. In the present
time step there is no dependency on electrical conditions at distant terminals of the line, frequently leading to a block-diagonal conductance
matrix
0
(6.4.4)
0
0
Each decoupled block in this matrix is a subsystem, and can be solved at
212
each time step independently of all other subsystems. When simulating

HVDC systems, it can frequently be arranged that the subsystems containing
each end of the link are small, so that only a small conductance matrix need
be refactored after every switching. Even if the link is not terminated at
transmission lines or cables, a subsystem boundary can still be created by
introducing a one-time step delay at the commutating bus.
Wherever a part of the system appears to change little from one step to
the next, it can be replaced with a linear equivalent, either Norton or
Thevenin, with little impact on accuracy. Such is the case where a node
voltage is stabilised by a shunt capacitance to ground, and a branch current
is smoothed by an inductance. HVDC systems provide a good example of
this situation with regard to filter banks and commutating reactance,
respectively.
This technique was used extensively in earlier EMTDC versions. However,
with the extensive use of sparsity, and with frequently switched nodes
ordered last, there is no real limit on the size of the conductance matrix
of a subsystem, and this approximation is no longer needed. By contrast,
there is no approximation when a transmission line is interconnecting the
two parts of a network and two separate subsystems are automatically
formed.
Figure 6.32 illustrates coupled systems that are to be separated into
subsystems. Each subsystem in Figure 6.326 is represented in the other by
a linear equivalent. The Norton equivalent is constructed using information
from the previous time step, looking into subsystem 2 from bus A. The
shunt connected at A is considered part of 1. The Norton admittance is
YN = YA + [1/Z(YB + Y 2 )]/[l/Z + YB + Y J
(6.4.5)
The Norton current is

I N = IA(* - At) + VA(* - At)YA
(6.4.6)
The Thevenin impedance is

Z TH = 1/YB[Z + 1/(YX + YA)]/[1/YB + Z + 1/CY, + YA)]
(6.4.7)
and the voltage source is

VTH = VB(* - At) + ZTHIBA(* - At)
If YA is a capacitor bank, z is a series inductor, and YB is small, then
Y N YA
Z TH Z
and
and
I N = IBA(Z A*) (the inductor current)

VTH = VA(t At) (the capacitor voltage)
(6.4.8)
213
(a)
Figure 632
Separation of two coupled subsystems by means of linearised equivalent

sources
In many instances, a DC link may be subdivided into subsystems as in Figure

6.33.
Subsystems 2 and 3 now contain only a small number of nodes, so that
the refactorisation after each switching is fast. In practice, by the use of
sparsity, a moderately large number of nodes can still be present in a
subsystem with switching elements. As an example, Table 6.2 compares the
execution time for the CIGRE benchmark system as more nodes are added,
and with both bridges blocked (so there is only 1 factorisation of the
conductance matrix).
The execution time is scaling approximately linearly with the number of
nodes in this instance. The extra nodes were added by placing strings of
shunt resistors on the commutating buses. Such a system leads to a
structured conductance matrix with few interconnectors, that is efficiently
factored by the sparse routines. Subsystem sizes are generally much smaller
than 200 nodes, and in the RTDS are limited to 36 nodes.
The method of interfacing subsystems by controlled sources is also used
to interface subsystems with component models solved by another method,
e.g. state-variable analysis. Synchronous machine and SVC models are
framed in state variables in PSCAD/EMTDC and appear to their parent
subsystems as controlled sources. As for interfacing subsystems, best results
214

(a)
subsystem 1
'
_L
subsystem 2
subsystem 3
subsystem 4
Figure 6.33 A typical subsystem equivalent for a DC link
are obtained if the voltage and current at the point of connection are
stabilised, and if each component/model is represented in the other as a
linearised equivalent around the solution at the previous time step. In the
case of synchronous machines, a suitable linearising equivalent is the
subtransient reactance, which sould be connected in shunt with the machine
current injection.
Table 6.2 Factorisation overheads as a function of subsystem size for the CIGRE
benchmark
Time step
(A*)
Factorisations
Number of nodes
CPU time
(s)
40
40
40
40
40
40
9680
1
9680
1
9677
1
46
46
92
92
184
184
150
117
194
168
294
247
215
6.5 The EMTP program

The EMTP program is modularised, with models of various physical
components contained in functional modules. These modules include,
among others, transformers, transmission lines, switches, surge arresters,
control systems and electric machinery.19 In general, EMTP is designed to
solve any network which consists of interconnections of resistance,
capacitance, inductance, single and multiphase n circuits, distributed transmission lines and certain other elements.17
Models of transmission lines and transformers have improved substantially throughout the years. Accurate models were developed to model the
distortion of waveforms in transmission lines and saturation in transformers.
The program contains a model for a built-in synchronous machine
with its mechanical system as well as universal machine models20 allowing
studies for 12 different types of machine. The mechanical behaviour of the
machine is modelled as a lumped RLC electric network for the mass-shaft
system to which the electromagnetic machine torque appears as a current
source.
Additional features are included such as a power-flow calculator, a
Fourier-analysis program, a frequency-scan program to derive frequencydependent impedances of a network by current injections, etc.
Transient analysis of control systems (TAGS) is a subprogram added to
the EMTP21 for the simulation of HVDC converter controls. It was later
extended for wider applications of other control systems and for modelling
special devices or phenomena which cannot be modelled directly with the
existing network components in the basic EMTP.
TACS is used to model control systems with the use of transfer-function
blocks, signal sources, switches, metering and other special devices for
proper control-system representation. From the network solution TACS
accepts network voltage and current sources, node voltages, switch currents,
status of switches and certain internal variables, and the network solution
accepts output signals from TACS as controlled sources and as commands
to open or close TACS controlled switches.
A At step delay is inherent when the electric network and TACS are solved
sequentially. Recently, to improve the control-system flexibility and to ease
the task of modelling digital controls, an EMTP-TACS-FORTRAN interface has been proposed, which it is claimed introduces no interfacing
errors. 22
There are five basic types of switch available through TACS; namely, a
time-controlled switch, a gap switch, a diode switch, a thyristor switch and a
measuring switch. The diode switch closes with a positive forward voltage
and stops conducting if the current reduces to less than a specified current
margin. The thyristor switch serves as the basic building block of HVDC
converters. It is similar to the diode switch except for the firing pulse
required to implement the closing.
Next Page
216
Other switch types exist which are specifically designed for statistical
overvoltage studies. EMTP provides the means for multiple runs with some
parameter variations, for example, the time delay between switch openings
and closing can be systematically incremented or randomly varied with a
given distribution during these runs.
Switches are modelled in EMTP as short circuits when conducting, and as
resistances when not conducting. This means that the number of nodes in
the system varies with the switch positions during a simulation.
6.6 The NETOMAC program

Based on the EMTP technique, NETOMAC is a system of digital programs
designed to study the electrical-power system for electromechanical and
electromagnetic behaviour. This includes a wide range of studies in the time
and frequency domains. A complete model valid over a wide range of
frequencies is computationally uneconomical; instead, a set of programs
each valid for a frequency range of interest have been combined into what
is called the program system NETOMAC.23 Although not widely available,
these have been used by HVDC manufacturers for quite some time.24
Two different modes can be selected for the calculation of transient
behaviour involving HVDC systems.25 Dynamic simulation of converter
networks is carried out using the instantaneous mode. The stability mode
(r.m.s. value mode) permits transient stability studies by the use of a
fundamental frequency quasisteady-state model. Both these cases can use
the same data records containing information on three-phase AC networks
and closed-loop controls. A special feature of the program, when carrying
out studies on AC networks, is its ability to switch over between the
instantaneous and stability modes during the calculation.
As well as fixed frequency-voltage sources, synchronous and asynchronous
generators can be modelled using Park's differential equations with mainfield saturation represented. Voltage and turbine control systems can also
be used with each synchronous machine. Passive network elements such as
linear and nonlinear inductances and resistances, capacitances, lines as n
sections or distributed parameter models and transformers with saturation
included are available. A four-terminal network provides isolation between
the potentials of the high-voltage and low-voltage sides similar to the real
transformers and permits the implementation of phase displacement in
three-phase transformers.
Control systems for synchronous machine excitation or turbine control
can be modelled using a block-oriented simulation language. A limited
number of blocks are available in the block library to build up a closed-loop
or open-loop control system.
Snapshots can be stored at the end of each run and interpreted as the
initial conditions for a subsequent run.
Chapter 7
7.1 Introduction
Stable operation of the power-system network is dependent on the balance
of mechanical and electromagnetic forces keeping the generators in synchronism. A disturbance on the network may or may not result in the system
falling out of synchronism and electrically collapsing. The assessment of
such behaviour is the purpose of electromechanical-stability studies.
It is normally assumed that, prior to dynamic analysis, the system is
operating in the steady state and that a load-flow solution is available.
In line with the power-flow philosophy described in Chapter 2, the
network solution of the stability studies uses the nodal-matrix method.
Thus, for each network-loading component an injected current is calculated
by solving the relevant differential and algebraic equations. The nodal
voltages are then obtained from the current injections and matrix admittance. However, as the network voltages affect the loading components, an
iterative process is often required.
Two types of stability study are normally carried out. The subsequent
recovery from a sudden large disturbance is referred to as transient stability
and the solution is obtained in the time domain. The period under
investigation can vary from a fraction of a second, when first-swing stability
is being determined, to over ten seconds when multiple-swing stability must
be examined.
The term dynamic stability is used to describe the long-term response of
a system to small disturbances or badly-set automatic control. The problem
can be solved either in the time domain or in the frequency domain. In this
Chapter both transient stability and dynamic stability are considered together in the time domain, the latter as an extension of the former.
The main consideration of the electromechanical simulation is the rotorswing stability of synchronous machines. In this respect, as compared with
the rotor long-time constants, the AC and DC-transmission systems respond
rapidly to network and load changes. The time constants associated with the
network variables are extremely small and can be neglected without significant loss of accuracy. The synchronous machine stator time constants may
also be taken as zero. The relevant differential equations for these rapidly
changing variables are transformed into algebraic equations.
276
The DC link is assumed to maintain normal operation throughout the

disturbance. This approach is not valid for larger disturbances such as
converter faults, DC-line faults and AC faults close to the converter stations;
these disturbances can cause commutation failures and alter the normal
conduction sequence.
The conditions leading to commutation failure are also discussed in this
Chapter, to test their occurrence at every step of the stability study. When
such an event is predicted the quasisteady-state model is not applicable and
a more rigorous alternative is discussed in Chapter 8.
7.2 Dynamic model of the synchronous machine

7.2.1 Equations of motion
The accelerating torque of a synchronous machine is defined by the product
of the moment of inertia of the machine, Imi, and the acceleration, a
Tfl = IMI.a
(7.2.1)
In a generating machine, the driving torque is mechanical and the load

torque is electrical. The overall accelerating torque is then
Tfl = T m j - T e
(7.2.2)
Therefore, machine acceleration will result from an increase in mechanical

torque or a decrease in the electrical torque.
If 8 is defined as the rotor angle of the machine with respect to a
reference frame rotating at synchronous speed co0 then
6 = co0t + 5
(7.2.3)
6 = co0 + 5
(7.2.4)
0 = (D0t + 3
(7.2.5)
where S is the initial position of the rotor with respect to the synchronously
rotating reference frame.
If 6 is the acceleration of the rotor, then from eqn. 7.2.1
= lmicb
where co is the shaft angular velocity.
(7.2.6)
Electromechanical stability 277

Since power is the product of torque and speed, then from eqn. 7.2.6
P. = lJ&
(7-2.7)
Angular momentum, M, is the product of the moment of inertia and the

speed, Ico. If the machine rotor speed is assumed constant at synchronous
speed, a normal and accepted assumption for stability studies, then M is
constant.
The accelerating power is then
Pfl = MS
(7.2.8)
If the rotational power losses of the machine due to such effects as windage
and friction are ignored, then the accelerating power equals the difference
between the mechanical power and the electrical power. From eqn. 7.2.2,
for a generator
P a = Pm - P e
(7.2.9)
M = P M - P e
(7.2.10)
Eqn. 7.2.10 is known as the swing equation.

Another constant commonly referred to is H, the kinetic energy at rated
speed per the rated power of the machine. H can be denoted as the inertia
constant of the machine and is usually given in units of (MWs/MVA).
The angular momentum M can be defined in terms of H as follows
^ ?
co0
(7.2.11)
where Gr is the three-phase MVA rating of the machine and coo is the
synchronous speed. Inserting eqn. 7.2.11 into eqn. 7.2.10 gives
***
O)0
Pm-r.
(7.2.12)
or
The electrical power angle 8e is related to the angle S by

S e = N p8
(7.2.14)
278
where N p is the number of pole pairs of the machine. Eqn. 7.2.13 then
becomes
The mechanical motion of the synchronous machine can now be described

by rewriting the second-order differential equation (eqn. 7.2.15) into two
first-order differential equations
=|^[P
-Pe]
(7.2.16)
(7.2.17)
Se = oj-co0
These equations describe the rotational swing of a synchronous machine.

The swing curves of a network of machines will determine whether or not
the system is able to remain in synchronism following a disturbance.
7.2.2 Electrical equations

The electrical characteristic equations describing a three-phase synchronous
machine are commonly denned by a two-dimensional reference frame. This
is well documented in the power-system literature1'2 and involves the use of
Park's transformations to convert currents and flux linkages (and therefore
voltages) into two fictitious windings located on axes 90 degrees apart.3'4
These axes are fixed with respect to the rotor position. One axis coincides
with the centre of the magnetic poles of the rotor (the d axis) and the other
lies along the magnetic neutral axis (the q axis). Electrical quantities can
then be expressed in terms of d and <jr-axis parameters.
The synchronous machine's inertia prevents the flux linkages from
changing instantaneously; machine models can be denned for steady-state,
transient and subtransient conditions depending on the speed of the
external changes. The d~q phasor diagram for the transient machine model
is shown in Figure 7.1.
Each phasor quantity can be represented in terms of its d and ^-axis
components,
(7.2.18)
E = Ed+jEq
(7.2.19)
(7.2.20)
The algebraic equations for the synchronous machine transient model in
>
279
> q axis
IdXd
Figure 7.1 Phasor representation of a synchronous machine in the transient state
Figure 7.1 can be written as follows

(7.2.21)
(7.2.22)
where
E' = transient internal voltage
V = armature terminal voltage
Ra = AC armature resistance
X' = transient reactance
I = armature current
and subscripts d and q denote the d-axis and #-axis components of the
parameter, respectively.
These equations lead to the formulation of the transient differential
equations
dK
dt
dt
[Ef-(Xd-X'd)ld-E'll]
rd0
[ _
[(X,g - 3QI, - E'd]

LV
where
Ey = field-winding voltage
X = synchronous reactance
TQ = transient open-circuit time constant.
(7.2.23)
(7.2.24)
280
The development of subtransient equations is very similar to that for

transient equations and yields the differential equations:
K - Vd = RflId - x;'I,
E; - V, = X'X + R a I,
,
*-
(7.2.25)
(7.2.26)
rfO
-^r^
(7.2.27)
(7 2 28)
- -
where
E" = subtransient internal voltage
X" = subtransient reactance
T'o = subtransient open-circuit time constant.
Although the synchronous-machine model is commonly represented as a
voltage source behind a fixed impedance (either the transient or subtransient reactance and the armature resistance), it is actually modelled in
transient-stability programs as a current source in parallel with a fixed
admittance.
Advanced models can, to some extent, take into account saturation of the
machine. This is achieved in the TS program by modification of the current
injection used in representing the machine.5 The single-phase, fundamental-frequency nature of the stability programs will, however, always limit the
usefulness of this modelling.
7.2.3 Synchronous-machine controllers

When stability analysis involves simulation times longer than about one
second, any effects due to machine controllers such as speed governors and
automatic voltage regulators (AVR) must be incorporated. In such cases, it
is no longer a reasonable assumption to consider the mechanical power (Pm)
constant after the first transient stability swing. The variation in Pm must
then be taken into account as it can have a significant effect on the system.
The automatic voltage regulator can also have an appreciable effect on
transient stability by varying the field voltage to try to maintain the terminal
voltage constant. Speed and voltage-controller models are thus necessary in
a stability program.
Composite models are normally included in the stability program to
simulate the behaviour of standard controllers.6'7
Figures 7.2 and 7.3 show composite AVR and speed governor models of
general purpose. By defaulting time constants and gains to zero, unity or
large values, the composite models can accommodate simpler models, as
required.
Se-f(Ef)
Specifiec
voltage
1
Controlled busbar
voltage
0+TrP)
Ka(l+Tbp)
U+Tap)
fl.
+
Fornax
1
(Ke+Tep)
EF
max
^f max
Field
voltage
\
yv
aux
Other
Kx
U+Txp)
(1+TyP)
(1+Tzp)
1
U+Tdp)
Kfp
<1+Tfp)
Typel
signal
Figure 7.2 Composite automatic voltage-regulator model

t00
282

initial
valve/gate i Gs
position
R(J+T2P)
(1+TIP)
Machine T +
speed
I
tf=
I
O+T3p)
Boiler or
water power
G
\ Cmax
>(n) >Pgv
Power to
turbine
Ref. speed
Figure 7.3 Composite model of a speed governor and valve

pgv
Power delivered
through gate
(a)
+ Mechanical power
to generator
Power delivered
through valve
(b)
Figure 7.4
Mechanical power
to generator
PI, IP and LP
turbine power
Basic linear models of turbines
a Hydro turbine
b Thermal turbine
Basic linear models of hydro and thermal turbines are shown in Figure
7.4. The hydro turbine model includes the penstock which gives the
characteristic lead-lag response of this type of turbine.
A rigorous formulation of the equations relating to the above controllers
can be found in Reference 5.
7.2.4 Generator representation in the network

In section 7.2.2 the synchronous-machine equations are given in the form
of Thevenin voltages behind impedances. For a nodal solution of the
network, the machine Thevenin model is converted into its Norton equivalent. The network-current injections vector will then include the synchronous-machine Norton equivalent currents, and the machine's
admittance is added to the system admittance at the machine's busbar.
k
283
Network
imaginary axis
direct axis
quadrature axis
> Network real axis
Figure 7.5 Synchronous-machine and network-phase references

Wiley & Sons, 1983 Reproduced by permission
The synchronous-machine equations are written in a frame of reference
rotating with its rotor. To use the network reference, as illustrated in Figure
7.5, requires the following transformations
cos (5 sin 3
sin 8
cos S
V,
cos d
sin 5
(7.2.29)
(7.2.30)
sin 3 cos (5
These transformations are equally valid for the currents.

When saliency exists, the values of X^J and X^ used in eqns. 7.2.25 and
7.2.26 and/or X'd and X^ used in eqns. 7.2.23 and 7.2.24 are different.
Therefore, the Norton-shunt admittance will have a different value in
each axis and when transformed into the network frame of reference, will
have time-varying components. However, a constant admittance can be
used, if the injected current is modified to retain the accuracy of the Norton
equivalent.8 This approach can be justified by comparing the two circuits of
Figure 7.6 in which Yt is a time-varying admittance, whereas Yo is fixed. At
any time, t, the Norton equivalent of the machine is illustrated in Figure
7.6a, but the use of a fixed admittance results in the modified circuit of
Figure 7.6b.
The machine current is
I = Y((E" - V) = Y0(E" - V) + I adj
and hence
I adj = (Yt - Y0)(E" - V)
(7.2.31)
284

>
(a).
(b)
Figure 7.6 a Norton equivalent of the synchronous machine

b Modified equivalent circuit
Wiley & Sons, 1983 Reproduced by permission
The injected current into the network which includes Y o is given by
Iinj=Tunadj+Iadj
(7-2.32)
where
A suitable value for Y o is found by using the mean of direct and quadrature
admittances, i.e.
(7.2.33)
where
Thus, the unadjusted current components are
1
Ra
-x,
Ra
(7.2.34)
Although the adjusting current does not appear to be affected by rotor
285
position in the machine frame of reference, it is when considered in the

network frame. From eqns. 7.2.31, 7.2.25 and 7.2.26
i( X ;-x;')
2
(Ra + x;x;')
K - v.
(7.2.35)
and using the reference transformation

adjr
-sin 2^ cos 23
E;-vr
sin 23
K-vm
cos 23
(7.2.36)
The total nodal injected current is, therefore

(7.2.37)
7.3 Load representation

In power-flow analysis the loads are normally specified as P and Q.
A general load characteristic commonly used in stability studies9 is defined
by
= kp(Wv(f)pf
(7.3.1)
(7.3.2)
where kp and kq are constants dependent on the initial values of P and Q.

The values of pv, qv9 pf and qfzre related to the type of load. Static loads
are, to a practical extent, unaffected by frequency (i.e. pf=qf=O) and, with
constant impedance loads, the factors qv and pv square the voltage value.
Typical values for load-characteristic parameters are given in Table 7.1. 10
However, the load characteristics represented by eqns. 7.3.1 and 7.3.2 are
only valid for small voltage deviations from the nominal. If the voltage is
small, even small errors in magnitude and phase produce large errors in
current magnitude and phase. These can be overcome by using a constantimpedance characteristic to represent loads where the voltage is below some
predefined value (e.g. 0.8 p.u.).
To be suitable for representation in the overall solution method, loads
must be transformed into currents injected into the transmission network
from which the terminal voltages can be calculated. A Norton-equivalent
model of each load must, therefore, be created. In a similar way to that
adopted for synchronous machines, the Norton admittance may be included
directly in the network-admittance matrix.
286

Table 7.1 Typical characteristic load-equation parameters
Load type
pv
qv
pf
qf
Fluorescent lamp
Filament lamp
Heater
Induction motor (half load)
Induction motor (full load)
Reduction furnace
1.2
1.6
2.0
0.2
0.1
1.9
3.0
0
0
1.6
0.6
2.1
-1.0
0
0
1.5
2.8
-0.5
2.8
0
0
-0.3
1.8
0
A constant-impedance load is, therefore, totally included in the networkadmittance matrix and its injected current is zero. This representation is
extremely simple to implement, causes no computational problems and
improves the accuracy of the network solution by strengthening the diagonal elements in the admittance matrix.
Nonimpedance loads may be treated similarly. In this case, the steadystate values of voltage and complex power obtained from the load flow are
used to obtain a steady-state equivalent admittance (Yo) which is included
in the network admittance matrix [Y]. During the stability run, each load is
solved sequentially along with the generators, etc. to obtain a new admittance (Y)
Y
= jvji
<7-3-3)
The current injected into the network thus represents the deviation of the
load characteristic from an impedance characteristic
Iinj=(Y0-Y)V
(7.3.4)
7.4 The AC transmission network

With the positive-sequence, fundamental-frequency nature of stability programs, it is common to represent power-transmission lines and transformers
by lumped-equivalent PI circuit parameters in the same way as with
power-flow programs. The impedances of these network components are
converted to admittances and included in an overall network-admittance
matrix. Fundamental-frequency representation is deemed adequate since
the frequency variation in stability analysis is usually small.
The behaviour of the network is described by the matrix equation
(7.4.1)
287
where [Iinj] is the vector of injected currents into the network due to
generators, converters and loads, and [V] is the vector of nodal voltages.
Any network (including loads) components represented by impedances may
be directly included in the network-admittance matrix, with the injected
currents (if any) due to these components set to zero.
7.4.1 System faults and switching

In general, most power-system disturbances to be studied will be caused by
changes in the network. These changes will normally be caused by faults and
subsequent switching action, but occasionally the effect of branch or machine switching will be considered.
Although faults can occur anywhere in the system, it is much easier
computationally to apply a fault to a busbar. In this case, only the shunt
admittance at the busbar need be changed, i.e. a modification to the
relevant self admittance of the Y matrix. Faults on branches require the
construction of a dummy busbar at the fault location and suitable modification of the branch data unless the distance between the fault position and
the nearest busbar is small enough to be ignored.
The worst case is a three-phase zero-impedance fault and this involves
placing an infinite admittance in parallel with the existing shunt admittance.
In practice, a nonzero but sufficiently low fault impedance is used so that
the busbar voltage is effectively brought to zero. This is necessary to meet
the requirements of the numerical-solution method.
The application or removal of a fault at an existing busbar does not affect
the topology of the network, and where the solution method is based on
sparsity exploiting ordered elimination, the ordering remains unchanged
and only the factors required for the forward and backward substitution
need be modifed. Alternatively, these factors can remain constant, and
diakoptical techniques can be used to account for the network change.
Branch switching can easily be carried out by either modifying the
relevant mutual and self admittances of the Y matrix or using diakoptical
techniques. In either case, the topology of the network can remain unchanged as an open branch is merely one with zero admittance. Although
this does not fully exploit sparsity, in almost all cases the gain in computation time by not reordering exceeds the loss of retaining zero elements.
7.5 Static power conversion models

HVDC links, aluminium smelters, chlorine producing equipment etc., all
rely on conversion from AC to DC and, in the case of HVDC links, back to
AC again. HVDC modelling can be quite complex since, dynamically, the
HVDC converter operates rapidly and usually under constraint from
multiple-control mechanisms. When perturbations are small, however, the
HVDC response is predictable and very fast, and a modified steady-state
288
model is acceptable. Most electromechanical-transient stability programs

currently utilise this approach.
7.5.1 Single-converter loads

A single-converter load consists of a number of rectifier bridges connected
in series and/or parallel, each bridge being phase shifted relative to the
others. With these configurations, high pulse numbers can be achieved
resulting in minimal distortion of the supply voltage. Rectifier loads can,
therefore, be modelled as a single-equivalent bridge with a sinusoidal supply
voltage at the terminals but without representation of passive filters. This
model is shown in Figure 7.7.
Rectifier loads can use diode or thyristor elements. In some cases, diode
bridges are used with tap-changer and saturable-reactor control. The effect
of the saturable reactors on diode conduction is identical to delay-angle
control of a thyristor over a limited range of delay angles. Both these control
methods can be modelled using a controlled rectifier with suitable limits
imposed on the delay angle (a).11
The DC load of a rectifier operating under current control is represented
by the equations
:
(7.5.1)
*d**d + Moad
(7.5.2)
(A + Rd)
where A is the constant-current controller gain and lds is the normal

DC-current setting as shown in Figure 7.8.
Constant current cannot be maintained during a large disturbance as a
limit of delay angle will be reached. In this event, the rectifier-control
specification will become one of constant-delay angle and eqn. 7.5.2 becomes
Id =
[(3V2/7r)flVtermcosamin - Vload]
[R, + ( 3 X ]
(7.5.3)
Delay angle a
load
Figure 7.7 Rectifier-load equivalent circuit
289
Natural voltage
characteristic
5 ^ aVJerm cos a mfn ,
(slope = : f x c )
Constant current
control characteristic
(slope = -A)
Figure 7.8 Simple rectifier-control characteristic

Wiley 8c Sons, 1983 Reproduced by permission
On the assumption that the DC current can change instantaneously, the
behaviour of the rectifier load can be represented by eqns. 2.2.6, 2.2.8,
2.2.16, 2.2.19, 2.2.21, 3.6.6 and either 7.5.2 or 7.5.3.
For some types of rectifier load, this may be a valid assumption but the
DC load may well have an overall time constant which is significant with
respect to the fault-clearing time. In order to realistically examine the
effects which rectifiers have on the transient stability of the system, this time
constant must be taken into account. This requires a more complex model
to account for extended overlap angles, when low commutating voltages are
associated with large DC currents.
When the delay angle, a, reaches a limiting value, the dynamic response
of the DC current is given by
where Ld represents the equivalent inductance in the load circuit. Substituting for Yd using eqn. 2.2.8 gives
Vload
where Tdc = L/R,,.
(7.5.5)
290
The controller time constant may also be large enough to be considered.

However, in transient-stability studies where large disturbances are usually
being investigated, faults close to the rectifier load force the delay angle (a)
to minimum very quickly. Provided the rectifier load continues to operate,
the delay angle will remain at its minimum setting throughout the fault
period and well into the post-fault period until the terminal voltage
recovers. The controller will, therefore, not exert any significant control
over the DC-load current. Ignoring the controller time constant can,
therefore, be justified in most studies.
Abnormal modes of converter operation
Changes in the DC current and/or AC voltage of a converter can drive the
converter into abnormal modes of operation, where the normal steady-state
converter equations are no longer applicable. Reference 12 describes the
operating modes of a three-phase bridge converter. Detection of the
changing modes is carried out by calculation of the factor K defined as
(7.5.6)
Information on the firing angle of the converter and the factor K completely
defines the mode of operation of the converter. The different modes of
operation are shown in Figure 7.9, as also are the limits of the modes
defined by the factor K and the converter firing angle.
Mode A represents normal rectifier operation, and mode E represents
normal inverter operation.
In mode B, when a valve is fired, a commutation is still taking place on
the other pole of the bridge rectifier; therefore the commencement of
commutation is delayed until that commutation process is completed, or the
delay angle exceeds 30 degrees. The commutation angle decreases as the
delay increases, and a stable condition occurs when the commutation angle
is 60 degrees. The converter formula is now
(7 5>7)
where
K' = K + 73(1 - K 2 ) 1/2
R; = (3/rc)Xc + (3/2)Re
and Re is the transformer resistance.
The additional delay (al) before the valve conducts is
al = 6 0 - a - c o s " 1 (K)
(7.5.8)

BOUNDARIES:
- D
K - cos(60-a)
'
2 cos(a)
- E
- C
30
- D
2//3 cos(a-30)
- F
cos(a-60)
1//3 O cos(Yn)
- F
C
E :
90-Y
< o < 90 -
90 < a < 120

120
<a
K - cos(a) - cos(150-* 0 )
K - cos(a) " cos(a*60-^ 0 )
K - oas(a) * cos(YQ)
K - 1//3 (003(Y.) cosCa-30))
120
150
180
Firing Angle
Figure 7.9 Boundaries of abnormal modes of converter operation
In mode C, the situation is similar to that of mode B; however, there are

periods of short circuits on the AC and DC sides of the converter, caused
by two commutations overlapping. This is not a dangerous situation, since
the valve current cannot exceed the DC-line current, due to the unidirectional flow of current through the valves. The short circuits clear when one
of the commutations is complete. The converter equations for this type of
operation are
ju= 120 - cos - ^ S K - 1)
R
term ~" **c ld
(7.5.9)
(7.5.10)
where
R; = (9/TI)XC + (3/2)Re
The total delay before the valve fires is 30 degrees.
The situation in modes D and F is similar to that of mode C, where there
are periods of AC and DC short circuits, the difference being that there is
no additional delay. Mode D represents abnormal rectifier operation, and
mode F represents abnormal inverter operation. The converter equations
292
for this type of operation are

\i = s i n ' ^ S K - cos(a - 30)] - a + 60
V, = ^
nn
VACcos(a - 30) - RÎd
(7.5.11)
(7.5.12)
R" is the same as in eqn. 7.5.10.

For the converter in the B, C, D and F modes of operation, the DC-power
equation is
PDc = V,I, + (3/2)ReId2
(7.5.13)
From Fourier analysis of the current waveform of the converter in C, D and

F modes of operation, the following equation is obtained for the calculation
of terminal power factor angle, <j>
cos(2oO - cos(2(5')
(7 5 14)
where
5' = OL + ii + 30
a' = a - 30
A changing mode of operation can be considered as a discontinuity, and if
during one time step the mode of operation of a terminal changes, the
solution of eqn. 7.5.5 becomes invalid. The ideal method for implementing
mode changes is to find the exact instant when the mode of operation of
the terminal changes, and adjust the time step used so that for every
solution of eqn. 7.5.5, the terminal is in the same mode of operation over
the whole time step. This method is difficult to implement, since it requires
special logic to determine exactly when a mode change occurs, and then a
number of step length changes have to be made. Instead, when a mode
change is detected at the end of a DC time step, the DC currents are kept
constant, and the discontinuity equations presented in section 7.6.2 (i.e.
eqn. 7.6.6) are formed and solved. Thus, the DC-system variables are
continuous over the mode changing boundaries and any errors introduced
using this technique are small.
A mode change can also occur when solving the discontinuity equations:
in both cases, the value Vd changes. Inspection of eqns. 7.5.7, 7.5.10 and
7.5.12 shows that in the abnormal modes of operation, the value of Rc
changes; however, in eqn. 7.5.7 there is a nonlinear relationship between
DC voltage and current, not allowing a definite value to be determined for
Rc in mode B operation. The solution adopted is to leave Rc constant
293
regardless of the mode of operation of the converter, and to vary Vd to

account for the changing modes. In abnormal modes of operation, Yd is,
therefore, a function of AC voltage, firing angle and DC current.
Converter representation in the network
The nominal bus-shunt admittance (y0), derived from a preliminary power
flow, is included directly into the network-admittance matrix [Y]. The
injected current into the network in the initial steady state is, therefore,
zero. In general
j
(7-5.15)
where
J
" iv t e r n
and
(7 5 17)
- -
The static-load rectifier model does not depart greatly from an impedance
characteristic and is well behaved for low terminal voltages, the injected
current tending to zero as the voltage approaches zero. However, when the
rectifier model is modified to account for the dynamic behaviour of the DC
load its characteristic departs widely from that of an impedance. Immediately after a fault application, the voltage drops to a low value but the
injected-current magnitude does not change significantly. Similarly, on fault
clearing, the voltage recovers instantaneously to some higher value while the
current remains low.
When the load characteristic differs greatly from that of an impedance, a
sequential-solution technique can exhibit convergence problems,13 especially when the voltage is low. With small terminal voltages, the AC-current
magnitude of the rectifier load is related to the DC current, but the current
phase is greatly affected by the terminal voltage. Small voltage changes in
the complex plane can result in large variations of the voltage and currentphase angles.
Good convergence is achieved with a unified algorithm11 by reducing the
AC network, excluding the rectifier, to an equivalent Thevenin source
voltage and impedance as viewed from the primary side of the rectifier
transformer terminals. This equivalent of the system, along with the rectifier, can be described by a set of nonlinear simultaneous equations which
can be solved by a standard Newton-Raphson algorithm. The solution of the
294
reduced system yields the fundamental AC current at the rectifier terminals.

To obtain the network-equivalent impedance, it is only necessary to inject
1 p.u. current into the network at the rectifier terminals while all other nodal
injected currents are zero. With an injected current vector of this form, a
solution of the nodal network eqn. 7.4.1 gives the driving point and transfer
impedances in the resulting voltage vector
(7.5.18)
LoJ
and
% = 1 + JO
(7.5.19)
The equivalent circuit shown in Figure 7.10 can now be applied to find the
rectifier current (Ip) by using the Newton-Raphson technique.
The effect of the rectifier on the rest of the system can be determined by
superposition
[V] =
(7.5.20)
where
fVff] = [Y]- 1 [ir nj ]
(7.5.21)
and [Ifnj] are the injected currents resulting from all other generation and
loads in the system.
If the network remains constant, vector [Z] is also constant and thus only
needs re-evaluation on the occurrence of a discontinuity.
The equivalent system of Figure 7.10 contains seven variables (Vterm, lp,
d, \j/9 a, Vd and Id). With these variables, four independent equations can be
formed. They are eqn. 2.2.8 and
(7.5.22)
V,. I , - V3aV t e r m . I p cos(0 - >/0 = 0
(7.5.23)
Eqn. 7.5.22 is complex and represents two equations. Substituting for V,,
and I p using eqns. 2.2.16 and 7.5.1 reduces the number of variables to five.
A fifth equation is necessary and with constant-current control, i.e. with the
295
Dynamic
load element
"load
Figure 7.10 Equivalent system for Newton -Raphson solution
delay angle within its limits, this can be written as

(7.5.24)
Eqn. 2.2.8, suitably reorganised, and eqns. 7.5.22 to 7.5.24 represent
[T?(x)] = 0 of the Newton-Raphson process and
[X]T = [Er, 0,tfr,a,
(7.5.25)
When the delay angle reaches a specified lower limit (oemin), the control
specification, given by eqn. 7.5.24 changes to
=0
(7.5.26)
Eqn. 2.2.8 is no longer valid. The DC current (Id) is now governed by the
differential eqn. 7.5.5. If the trapezoidal method is being used, this equation
can be transformed into an algebraic form and eqn. 2.2.8 is replaced by
ld = ka. E r . cos a kb = 0
(7.5.27)
The variables ka and kb contain information from the beginning of the

integration step only and are thus constant during the iterative procedure
= h/(2+kc.h)
kb = (1 - 2kc.ka)\d(t) +
(7.5.28)
S /2
2 k/i V
^ n a.ka.yteim(t)cosoi(t) + ' ' 'oad
(7.5.29)
296
where
kc = ( ^ )
+ 1/Tdc
(7.5.30)
and represents the time at the beginning of the integration step and h is
the step length. Commutation angle \i is not explicitly included in the
formulation, and since these equations are for normal operation, the value
of k in eqn. 2.2.16 is close to unity and may be considered constant at each
step without loss of accuracy. On convergence, fi may be calculated and a
new k evaluated suitable for the next step.
7.5.2 DC links
For as long as the DC link is operating normally, i.e. without commutation
failures, the converter steady-state equations can be used to simulate its
behaviour at each step of the stability study.
The initial steady-state operating condition of the DC link is obtained
from the power-flow solution, with which the control type and mode,
current and margin settings will have been established.
At each iteration of the stability study the control mode must be reconsidered. If the link operates on current control this can be done by first
assuming mode 1 (i.e. with the rectifier on c.c. control), and by combining
eqns. 2.2.19 and 3.3.6 a DC current can be determined as
dmodel
_ I*, - [(3V2/ft)/a,..Vtermi. cosyJ/A,
[1 + (Rd -
(mXJ/A}
Assuming this current to be valid, then DC voltages at each end of the link
can be calculated using eqns. 2.2.8 and 2.2.17. The DC link is operating in
mode 2 (i.e. with the inverter on c.c. control) if
Vêx-V^Ô
(7.5.32)
The DC current for mode 2 operation is given by
[(1
For constant power control, under control mode 1, the DC current may be
determined from the quadratic equation
(7.5.34)
297
Table 7.2 Current setting for constant-power control from

quadratic equation
within = within the range I dmin to I dmax
outside = outside the range I dmin to I dmax
greater = greater than I dmax
less = less than I dmin
I,
within
outside
within
greater
greater
outside
within
within/
greater
less
greater
less
less
greater of \dl and \i2
less
where Pds is the setting at the electrical mid point of the DC system, i.e.
P * = (Ps, + P y 2
(7.5.35)
The correct value for Idmodei can then be found from Table 7.2. Control
mode 2 is determined using eqn. 7.5.33 and in this case the following
quadratic equation must be solved
r- *dmode2
^0**dmode2
* dmarg ~~ *ds ~
"
(7.5.36)
where
+
_V2
ar. Vtermr.cos armin
(7.5.37)
(7.5.38)
If the link is operating under constant-power control but with a current

margin then for control mode 2
-MLod.2 + (K - M*narg)Idmode2 + Mdmarg ~ *dsr = 0
(7.5.39)
It is possible for the DC link to be operating in control mode 2 despite

satisfying the inequality of eqn. 7.5.16. This occurs when the solution
indicates that the rectifier firing angle (ar) is less than the minimum value
(armin). In this case, the delay angle should be set to its minimum and a
solution in mode 2 is obtained.
298
It is also possible that when the link is operating close to the changeover
between modes, convergence problems will occur in which the control mode
changes at each iteration. This can easily be overcome by retaining mode 2
operation whenever detected for the remaining iterations in that particular
time step.
The value of the DC current obtained from the above equations can
then be used to calculate the inverter and rectifier DC voltages and also the
AC-line currents. The remaining variables for the rectifier and inverter ends
can be found from a unified iterative solution of the set of equations
containing these variables, such as shown in eqn. 7.5.25 for the case of a
rectifier on constant-current control.
The active and reactive power at both converters can then be calculated
with
(7.5.40)
(7.5.41)
(7.5.42)
r
(7.5.43)
t a n (/>j
DC power modulation
It has been shown in the previous section that under the constant-power
control mode, the DC link is not responsive to AC-system terminal conditions, i.e. the DC-power transfer can be controlled disregarding the actual
AC-voltage angles. Since, generally, the stability limit of an AC line is lower
than its thermal limit, the former can be increased in systems involving DC
links by proper use of fast-converter controllability.
The DC power can be modulated in response to AC-system variables to
increase system damping. Optimum performance can be achieved by
controlling the DC system so as to maximise the responses of the AC system
and DC line simultaneously following the variation of terminal conditions.
1 1
AC system AL DC system
controller
controller
ai
DC network
DC control modes
AC/DC interface
0)
Figure 7.11 AC-DC dynamic control structure

(i) AC-system controller
(ii) DC-system controller
(iii) AC-DC network
(iii)
299
The dynamic performance under DC-power modulation is best modelled

in three separate levels.14 These levels, illustrated in Figure 7.11, are (i) the
AC-system controller, (ii) the DC-system controller and (iii) the AC-DC
network:
(i)
The AC-system controller uses AC and/or DC-system information to derive the current and voltage-modulation signals. A block
diagram of the controller and AC-DC signal conditioner is shown in
Figure 7.12.
(ii) The DC-system controller receives the modulation signals AI and AE
and the steady-state specifications for power P o , current I o and voltage
Eo. Figure 7.13a illustrates the power-controller model, which develops the scheduled current setting; it is also shown that the current
order undergoes a gradual increase during restart, after a temporary
blocking of the DC link.
The rectifier current controller, Figure 7.13fc, includes signal limits
and rate limits, transducer-time constant, bandpass filtering and a
voltage-dependent current-order limit (VDCOL).
The inverter-current controller, Figure 7.13^, includes similar components plus a communications delay and the system-margin current
(ijFinally, the DC-voltage controller, including voltage restart dynamics, is illustrated in Figure 7.13^.
(iii) The DC current Id and voltage Ed derived in the DC-system controller
constitute the input signals for the AC-DC network model which
involves the steady-state solution of the DC system (neglecting the
DC-line dynamics which are included in the DC-system controller).
Here the actual AC and DC-system quantities are calculated, i.e.
control angles, DC current, voltage, active and reactive power. The
converter AC-system constraints are the open-circuit secondary voltages Ear and Eai.
AC/DC
signals
r3~^
^k i
AC system
controller
>
PTd
>
PTd
PTd
pTd
Figure 7.12 AC-system controller
-A1,
}*
A/;
-AEr
300

'Ol
1
1+pT
dr
Current order
restart dynamics
\
=' ^1 ^ o ,
(a)
Current rate
limit
I
02
Current
limit
hi
Regulator
b4
dr
D(P)
1+pT
OmaxI
(b)
Current rate
limit
Communication
delay
Current
limit
k>5
(c)
E
0J
02
Regulator
l
Voltage rate
limit
04
(d)
Figure 7.13 DC-system controller

a Power controller
b Rectifier-current controller
c Inverter-current controller
d DC-voltage controller
N(P)
D(P)
O7
N(P)
D(P)
301
DC link representation in the network
From the AC-network point of view each converter of a point-to-point or

multiterminal DC link behaves in a manner similar to that of the single
converter. Thus, a nominal bus-shunt admittance (y0) is calculated from the
initial load flow for both the rectifier and inverter ends, and injected
currents are used at each step in the solution to account for the change from
steady state calculated from eqn. 7.5.15. Note that the steady-state shunt
admittance at the inverter (yOi) will have a negative conductance value as
power is being supplied to the network.
The accuracy of the solution will be poor if a reasonable step length is not
chosen. There is no means of estimating the value of errors in the
trapezoidal method but the number of iterations required to converge at
each step increases more rapidly than the step length and thus provides a
very good indication of the errors. It is suggested16 that the step length
should be doubled if the number of iterations per step is less than three and
halved if the number of iterations per step exceeds 12.
7.6 AC-DC transient stability programs

Generally, the DC scheme interconnects two or more, otherwise independent, AC systems and the stability assessment is carried out for each of them
separately, taking into account the power constraints at the converter
terminal. If the DC link is part of a single (synchronous) AC system, the
converter constraints will apply to each of the nodes containing a converter
terminal.
The basic structure of a transient-stability program is given in Figure 7.14
and a description of a suitable converter subroutine and AC-DC interface
is shown in Figure 7.15.
The converter subroutine is based on the unified iterative solution
described in section 7.5.1. It is important to note that the hyperplanes of
the Newton solution are not linear and good initial estimates are essential
at every step in the procedure. A common problem in converter modelling
is that the solution converges to an unrealistic result of reactive-power
generation. It is, therefore, necessary to check against this condition at every
iteration. With integration step lengths of up to 25 ms, however, convergence is rapid.
The main functions indicated in the diagrams of Figures 7.14 and 7.15
are discussed next.
7.6.2 Trapezoidal integration

From the integration techniques described in Appendix V, the implicit
trapezoidal is the preferred method for the solution of the multimachine
302

Q
start
J)
read stability input data

from loadflow results
calculate machine initial
conditions
determine stability
step length
perform switching and
subsequent bifactorisation
if necessary
calculate currents injected
by each synchronous
machine (7.2.32)
calculate currents injected
by each converter (7.5.15)
calculate bus voltages from
bifactorised network (7.5.20)
determine machines nonintegrable variables
perform trapezoidal
integration to evaluate
variables
Figure 7.14 AC-DC transient-stability program
main program
303
converter subroutine
solve machines
fort!]*
solve [V] = [ Y]-1 [ l]m
without converter
yes
check for switching
obtain new
Thevenin impedance
vector
extract Thevenin
voltages from [V]
_L
calculate new initial

conditions and mode
I calculate mismatches
yes
convergence
calculate Jacobian for
given mode of
operation
>0
evaluate updates for

variables
iter =
iter+
yes
block converter
I update variables
test mode of control
and mode of operation
calculate converter
variables
iter >max
yes
test for commutation

failure (if inverting)
fail
superimpose
converter AC current
on[V]
error message
solve stability equations
Figure 7.15 Flow diagram of a unified-converter solution
transient-stability problem. The solution at the end of n + 1 steps is given by

A +1
"
(7.6.1)
The trapezoidal method is the most accurate L-stable finite-difference

method possible.
The method, as expressed by eqn. 7.6.1, is implicit and requires an
304

z(t)
y(t)
(1+pT)
Figure 7.16 Simple transfer function
iterative solution. However, the solution can be made direct by incorporating the differential equations into eqn. 7.6.1.
For example, consider the trivial transfer function shown in Figure 7.16.
The differential equation for this system is given by
py(t) = (G.z(0 ~y(t))/T
(7.6.2)
with the input variable being denoted by z to indicate that it may be either
integrable or nonintegrable.
The algebraic form of eqn. 7.6.2 has a solution at the end of the (n + l)th
step of
yn + 1 =cn+1
+mn + 1.zn+1
(7.6.3)
where
Cn+i
= (1 - 2bn+1)yn + bn+1.G.zn
m*+i=K+i-G
(7.6.4)
(7.6.5)
and
hH+l)
(7.6.6)
Provided that the step length h remains constant, it is unnecessary to

re-evaluate b or m at each step, i.e.
n+1
There is little to be gained by this, however, as it is a simple process and it

is often desirable to change h during a study.
7.6.3 Initial conditions

Like any other numerical-integration method, the implicit trapezoidal
requires adequate initial conditions at the start and immediately after the
occurrence of a disturbance.

The load-flow program provides initial conditions to start the stability
program. These are the terminal powers and AC voltages, DC voltages, DC
currents and firing angles. An initial solution of the DC system is required
which assumes that the power system is in steady-state operation prior to
the commencement and, thus, the DC-system inductances can be neglected.
This is achieved by modelling the converter as a voltage source, Vc, and a
resistor, Rc, i.e.
(7.6.8)
for the converter and
V ( t a -V ( O = R r t I r t
(7.6.9)
for the transmission line.

These equations can be put into the form
[VDC] = [YDC]"1[IDC]
(7.6.10)
where [VDC] is a vector of nodal voltages, [IDC] is a vector of injected currents

due to the converters and [YDC] is an admittance matrix.
The nodal voltages obtained on solution of eqn. 7.6.10 are used to
calculate the DC-network currents. The real and reactive powers are calculated and then compared with the values given in the input data to check
whether the load-flow data is sufficiently accurate to start the stability
program. The DC-network voltages and currents thus calculated are used to
start the dynamic solution.
Since it is assumed that any type of controller can be modelled, an
iterative algorithm is needed in which the controller is solved by considering
that it has been in a steady-state condition for a long period of time, and
that all time delays can be neglected. This algorithm is shown below:
(i)
(ii)
Determine the required firing angle from the load flow.

Solve the controller with settings higher, lower and equal to the initial
terminal DC current/power and, by comparison with the value calculated in (i), determine whether the setting should be increased or
decreased.
(iii) Change the setting by a set percentage, solve the controller equations
for the firing angle and calculate error when compared with firing
angle calculated in (i).
(iv) If the error is less than the tolerance, then exit, otherwise reduce the
percentage change in the setting and go to (iii).
With the iterative procedure used above, it may be impossible to determine
the settings of some controllers; these will have to be calculated manually
306
and fed as part op the input data of the TS program. However, for all normal
controller configurations tested, the algorithm correctly determined the
required current/power settings.
After a discontinuity, such as a fault application or switching on the AC
system, good starting conditions have to be determined to ensure that
the DC equations converge to the correct solution at the next time step.
The DC network contains only passive components, and the network can
experience an instantaneous drop in DC voltage. However, the DC-network
inductance prevents the network DC currents from changing instantly.
Immediately following a discontinuity, new DC voltages have to be determined for use as starting values in the DC system solution at the next time
step.
By summing the rate of change of DC current at every node in the DC
network, and setting the total change to zero, the following equation is
derived for the DC network at a discontinuity
[V] = [H]- 1 tC]
(7.6.11)
where
H" = 1/14 + 1/Ly

C< = (V*c - I X
7.6.4 Test of operating mode
The DC-network model is required for use in a transient-stability program,
where due to the disturbances modelled, the mode of operation of the
HVDC link can change. Terminals can, therefore, change from constantcurrent to constant-voltage operation, and with I and G depending on the
mode of operation selected, some method is required to check the DC
voltages to ensure that the correct mode of operation is chosen. The
following expressions can be written for the DC voltages at the rectifier and
inverter ends
dsr
V
dsi
^(V2V term cos(a min )/X c -I ds )

n/($Xc)
- 1/KC
= W^ V term cos Uo)/ X c ~ hs)

TC/(3X C ) - 1/KC
K )
,*, g , gx
' " '
If a rectifier is assumed to b e o p e r a t i n g in t h e voltage control, VC, m o d e
307
and the terminal voltage is less than Vdsr, the mode choice is incorrect. If
the rectifier is assumed to be in the constant-current control, CCC, mode
and the DC voltage is greater than Vdsr the mode choice is incorrect. For an
inverter, if the DC voltage is greater than Vdsi the mode choice is incorrect.
For any valid set of AC voltages, the following algorithm is used to
determine the DC voltages of the HVDC link:
(i)
Using the latest AC voltages, and other variables, as well as the

approximate controller functions, calculate the new current settings
and/or minimum firing/commutation-margin (y) angle.
(ii) For the latest values of AC voltages, current settings and OLmiJy angles
calculate Vdsr and Vdsi for all DC converters, using eqns. 7.6.12 and
7.6.13.
(iii) Assume a mode of operation for the DC link, and form the admittance
matrix and vector of injected currents.
(iv) Invert the admittance matrix and calculate the DC voltages.
(v) Check terminal voltages against Vdsr and Ydsi calculated in (ii) to
ensure that the correct mode of operation has been assumed; if not,
change mode of operation and repeat (iii), (iv) and (v) above.
With reference to multiterminal HVDC schemes, if it is considered possible

for more than one terminal to be in the VC mode at the same time, and
impossible for all terminals to be in the CCC mode at the same time, then
for a three-terminal HVDC link, there are seven possible modes of operation, and 63 possible modes of operation for a six-terminal HVDC link. If
steps (iii) to (v) above are performed by testing all modes of operation of
the link in order, the process will be very slow when modes of operation are
changing. The solution adopted is to ensure that all possible modes of
operation of a link with one terminal only in VC mode are tested; other
modes of operation are obtained by changing modes of operation of
terminals found in violation of the test in step (v) above. Through numerous
simulations, not one instance was found where the program failed to find
the correct mode of operation of the link; however, instances were found
where it was impossible to converge to a correct link solution regardless of
what link-operating mode was chosen.
7.6.5 Converter program interface

In the main TS program, the power-system differential equations and
network-admittance matrix are solved iteratively. As stated above, the AC
network is represented as a voltage source in the solution of the DC
network. The HVDC terminals are represented in the AC network as active
and reactive power loads/sources, and a Norton equivalent is calculated for
each terminal.
308
With the DC voltages known, the DC terminal currents are calculated,

for rectifier
I, = I*-j^
(7.6.14)
for inverter
h = - l
d s
(7.6.15)
For terminals in the voltage-control mode, the formulae are

for rectifier
r ... V2V term COS(q min )
1zVd
r ^7"
x;
sx:
for inverter
d
^7
^/.O.l I)
Fourier analysis of the AC-current waveform gives
The active and reactive powers can then be obtained, i.e.

P,,c = V(JIlJ
(7.6.19)
Qdc = Pdctan((t>)
(7.6.20)
The converters are represented in the AC network as an admittance,

calculated from the initial power consumption and AC voltage as follows
G . = PDC/VL
Bs = -Qoc/VLn
(7-6.21)
(7.6.22)
Any variation in the AC voltage and/or power consumption of the converters

is accounted for by injecting a current into the AC network, the
309
value of which is determined as follows

G = PdXerm-Gs
B=-Qdc/Vt
erm-Bs
(7.6.23)
(7.6.24)
I , = -V r e G + VimB
(7.6.25)
I6=-VimG-VreB
(7.6.26)
DC converters are represented as current sources of value lg +jlb and

admittances of value Gs + jBs in the AC-network solution, and the AC
network is represented as a voltage source of value Vterm in the DC-network
solution. The DC and AC networks are solved in a sequential manner until
a solution is converged upon. The HVDC controller functions are solved as
part of the DC-network equations.
7.6.6 Test for commutation failure

Region E of the graphs in Figure 7.9 represents normal steady-state inverter
operation, but the extended commutation area following a large disturbance
is likely to lead to commutation failure. In such a case, the quasisteady-state
solution is inappropriate.
A commutation failure occurs when an outgoing converter valve is not
given sufficient time to commutate and deionise before the commutating
voltage reverses. The line DC current is then shifted back from the
incoming valve to the outgoing valve. A commutation failure occurs owing
to either inadequate predictive control of the converter-extinction angle,
low converter AC voltage, high DC current or a combination of the above.
During a commutation failure, the valve-firing sequence changes and,
since valve firings are not explicitly represented in the quasisteady-state
model, the behaviour of the converter during a commutation failure is not
represented. As an example, the TS program cannot distinguish between
double and single commutation failures, or model the behaviour of the
converter as it recovers normal operation. The only method of modelling
converter behaviour during a commutation failure is to use a program
modelling valve firings in detail. In practice, when successive commutation
failures are detected on a bridge, the bridge is blocked and any further
analysis requires detailed dynamic simulation. The onset of commutation
failure is discussed next.
Commutation failure is an unlikely event under normal operational
changes, owing to the fast controllability and the nominal commutationmargin angle of the converter. However, electromechanical stability is
generally carried out following system disturbances, and particularly shortcircuit faults. These can cause substantial temporary voltage reductions of
310
the inverter-commutating voltages as well as increasing the link direct

current, the right combination for the occurrence of a commutation failure.
The disruption of the normal switching sequence following a commutation
failure leads to considerable waveform distortion of the voltage waveforms
at the converter terminals, which further encourages abnormal operation.
Under such conditions, the quasisteady-state solution is inapplicable and the
assessment of converter behaviour requires electromagnetic-transient simulation.
It is possible to assess the likelihood of a commutation failure based on
depressed but undistorted voltage waveforms. This is because it is not
waveform distortion that leads to the first commutation failure but rather
the commutation failure event which leads to the waveform distortion. This
approach is used in a recent CIGRE document15 to analyse the conditions
leading to the onset of commutation failure. Using the basic commutation
equations, the document calculates the infringement on the commutation
margin as a result of a given per unit, reduced commutating voltage, also
taking into account any possible DC-current increase.
This condition is expressed by the following equation for the case when
the prefault current corresponds to full load (ldF^)
AV = 1 -
Id Xcpu + (ldFL/ld)[cos(y0
+ </>)- cosy]
or
*cpu ^
AV 1 - ^
x
h cvu + cos(j0 + </>)- cosy
(7.6.27)
where
AV = inverter sudden commutation-voltage reduction required to produce the theoretical onset of commutation failure; it refers to the
phase voltage for single-phase faults, specifically on the faulted
phase
for symmetrical 30 reductions, the per unit AV is the same for
phase and line quantities
Id = DC current at the beginning of the commutation
Id = DC current at the end of commutation
Xcpw = commutation reactance (in per unit), i.e.
x
=X =
b
V2W.
FL
y = nominal commutation margin angle
311
y0 = critical commutation margin angle for valve turn off

<j) = phase shift due to a hypothetical voltage reduction in only one
phase; it can be shown15 that it is related to the voltage reduction
by the equation
The two dependent factors AV and <j> can be derived from an iterative
solution of eqns. 7.6.27 and 7.6.28.
Prior to control action by the converters, any DC current increase caused
by the inverter AC-voltage reduction (i.e. the factor iyi^) is critical for
commutation failures. The current increase depends on the values of the
smoothing reactor, commutation reactance, line parameters and DC-side
filters.
For the short period between fault initiation and the first commutation
failure, the DC-current rise is relatively small and very nearly linear.
Extensive EMTDC simulations have been carried out on two schemes with
very different DC-side and commutation parameters to assess the range of
variation of iyi d to be expected.
In the New Zealand scheme, that ratio varied from 1.1723 to 1.1457 for
7=18 (the nominal value in current use) to 1.238 for y = 24, the critical
commutation margin being y0 = 8.
Substitution of these values in eqn. 7.6.27 results in voltage reductions of
between 12 and 14% for the case of a symmetrical fault.
Due to the zero-crossing phase shift caused by the voltage unbalance, the
probability of commutation failure is higher in the case of a single phase-toground as compared with the symmetrical faults.
However, the difference between the 3$ and single phase is, to a large
extent, counteracted by the larger DC-current increase caused by the
symmetrical fault. The combination of voltage magnitude and phase shift
following a single-phase fault reduces the voltage area available to the
commutation; this effect is more pronounced in the case of the starconnected transformer as compared to the delta connection. It is, therefore,
expected that the converter group with the star connection may be more
prone to commutation failures. As the standard HVDC converter contains
star and delta-connected groups, there is no need to consider the delta
connection when assessing the onset of commutation failure.
Thus, at every step of the stability study, the inverter voltages are checked
against the critical values to decide whether the quasi steady state model is
still applicable. If these voltages are below the critical levels, indicating the
onset of commutation failure, the prediction of DC-link performance needs
transient simulation, as described in the next Chapter.
312
AC-DC
power system analysis
7.7 Test system and results16

The test system selected for these studies consists of a two-terminal
1000 MW monopolar HVDC link connected on the inverter side to a simple
AC system, and on the rectifier side to a more complex AC system
incorporating a number of generating buses.
The HVDC system is based on the CIGRE benchmark model described
in Appendix III. The inverter AC-system short-circuit ratio, including the
converter-terminal filters, is 2.5. The HVDC link is shown in Figure 7.17
along with the inverter-side AC-system representation.
The rectifier AC system is a simplified portion of the New Zealand South
Island system and is shown in Figure 7.18. The AC system contains four
generating buses in an interconnected loop composed of six buses in total.
Double-circuit transmission lines exist between Roxburgh and Clyde, and
Clyde and Twizel, and single-circuit lines connect the rest of the system.
Loads representing further system connections are present at both Twizel
and Roxburgh. Each generator has associated with it both an automatic
voltage regulator (AVR) and a speed governor. Details of the test-system
parameters are outlined in Appendix III.
The converter rectifier terminal also has a set of filters as per the CIGRE
benchmark model, and with these the short-circuit ratio is calculated to be
2.7.
The rectifier is specified under constant-current control and the inverter
under minimum-gamma control. Under a fault situation, the mode of these
terminals may be swapped if the rectifier DC voltage falls below that of the
inverter.
0.0365
0.5968
2.5
2.5
345kV2ll.42kV
1196 MVA
n HMH
0.0365
MW
2H.42kV23OkV
1196 MVA
26.0
To
Rectifier
ac
system
0.5968
0.7406
warn
15.04
167.2
Figure 7.17 Test system HVDC link and inverter AC system
|3.23
0.0606
0.935
TXvizcl
Clyde
Benmore
-HH*
Livingston
Aviemore
Figure 7.18 Test system rectifier AC system
Table 73
Rectifier AC-system steady-state busbar voltages
Bus
ivu
Aviemore
Benmore
Clyde
Roxburgh
Livingston
Twizel
1.015
1.000
1.043
1.042
1.024
1.015
1.338
0.000
6.479
6.338
2.415
1.564
Table 7.4 Steady-state DC conditions
VDC
"DC
a
PDC
QDC
Rectifier
Inverter
500 kV
2kA
17.65
1000MW
583MVAr
490 kV
2kA
15
980 MW
545 MVAr
313
314
Electromechanical-stability programs are somewhat limited in their simulation of abnormal operating conditions. As described in section 7.5,
an abnormal operating condition exists when the commutation angle
exceeds 60 degrees. If this occurs at the inverter terminal, the inverter is
simply bypassed. The rectifier terminal is kept under operation and
maintains the current to at least 30% of its nominal setting. After a set
period, if the problem causing abnormal operation has been removed, the
inverter is restarted and the power ramped back up to normal using current
control.
The test system in the steady state has the rectifier AC-system bus-voltage
magnitudes and angles as shown in Table 7.3. The corresponding DC
steady-state conditions are shown in Table 7.4.
Roxburgh
-Aviemore
Benmore
Clyde
1.10
1.05&
1.00
"*...-'
^-
0.90
(a)
0.0
1.0
2.0
3.0
4.0
5.0
Time (s)
(b)
Figure 7.19
TS rectifier AC-system bus voltages
6.0
7.0
8.0
9.0
10.0
315
7.7. / Minor disturbance

The disturbance consisted of switching out the shunt capacitor connected at
the rectifier bus. The rectifier AC generating bus-voltage magnitudes are
shown in Figure 7.19. The voltage drop is evident on all the generating
buses in the AC system, as is the eventual return to nominal steady-state
conditions through the action of the generator controllers.
The generator field voltages are shown in Figure 7.20a. The AVR limit at
Aviemore is evident, as is the controllers' attempt to maintain their set
bus-voltage magnitudes. The Benmore generator at the rectifier bus, as
expected, is the most affected by the disturbance and a new steadystate value is eventually approached. The relative machine angles in Figure
7.206 also show the effect of settling to this new steady-state condition at
Benmore.
Roxburgh
Aviemore
2.1 -
Field >/olta
2.5 -
l.o -
Benmore
Clyde
1.7-
4/
0.90.5-
(a)
-Roxburgh
Aviemore
Benmore
o.o
1.0
2.0
3.0
4.0
5.0
6.0
Time (s)
(b)
Figure 7.20 Rectifier AC-system generator parameters
7.0
8.0
9.0
10.0
316
7.7.2 Major disturbance

A three-phase fault was applied at the rectifier AC-system terminal and
cleared in three cycles.
The rectifier terminal was selected above the inverter as rectifier faults are
often worse for transient stability. An inverter fault still allows generation at
the rectifier to supply line losses and local load. A rectifier fault, however,
especially near generators, blocks both AC and DC power. This results in
greater transient acceleration of the generators.
The results of a transient stability program with a QSS representation of
the DC link are shown in Figures 7.21 to 7.24. At time t = 17 s the
three-phase fault is applied and then subsequently cleared three cycles later.
Figures 7.21^ and 7.21& show the generating bus-voltage magnitudes
Roxburgh
Aviemore
(a)
- Roxburgh
Aviemore
Benmore
55
35
I 15
PQ
0.0
1.0
2.0
3.0
4.0
5.0
Time (s)
(b)
Figure 7.21 Rectifier AC voltage
6.0
7.0
8.0
9.0
10.0
317
Rectifier current (pu)
-Rectifier voltage (pu)
7.00
6.00
jg 5.00
i l 4.00
>
3.00-
2.00
1.00-
0.00
(a)
Rectifier alpha
(b)
- Rectifier P
- Rectifier Q
1200
0.0
1.0
2.0
3.0
4.0
5.0
Time (s)
(c)
Figure 7.22
Rectifier DC variables
6.0
7.0
8.0
9.0
10.0
318

Inverter voltage (pu)
Inverter current (pu)
7.00
6.00
8
40
>
3.00
2.00
5.00
I
1.00
0.00
(a)
Inverter gamma
100
60
40
20
0
(b)
- Inverter P
Inverter Q
1200
10.0
Figure 7.23 Inverter DC variables
-Roxburgh
Benmore
- Aviemore
Clyde
319
-1.0
(a)
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
Time (s)
(b)
Figure 7.24 Rectifier AC-system machine response
and angles (angles relative to the reference bus Clyde). The effect at the
fault terminal is particularly evident as the voltage magnitude drops to zero
and the voltage angle rises nearly 50 degrees. The return to a stable voltage
angle can also be seen, with an initial oscillation due to the generator
controllers eventually decaying away.
Figures 7.22 and 7.23 display the effect of the fault on the DC variables
at each terminal of the link. At the instant of the fault, the rectifier terminal
is blocked and the inverter terminal bypassed. Once the fault is removed,
the rectifier terminal is unblocked and the DC current through the rectifier
ramped to 30% of its nominal value. The inverter terminal is then unbypassed and the DC real power ramped back up to its nominal value.
The machine response to the fault is shown in Figure 7.24. The field
voltages and machine angles vary and oscillate through the response of the
AVR but eventually settle back to steady-state values. The Aviemore
320
field-voltage response is somewhat different than that for the other generators owing to its larger forward regulator gain and lower upper-field
voltage limit.
7.8 Summary
This Chapter contains a concise description of the synchronous-generator
transient mechanical and electrical response to disturbances in the power
system. The modelling of other system components is also discussed with
emphasis on the AC-DC converters.
The structure of a basic transient-stability program is then described with
representation of the DC-link behaviour on the assumption that the link
maintains continuous controllability during the disturbance; however, a
check for the onset bf commutation failure has been included as part of the
solution. Prediction of such an event would indicate the limit of applicability
of the algorithm and the need to adopt the more detailed stability simulation discussed in Chapter 8.
7.9 References
1 KIMBARK, E.W.: 'Power system stabilityvol. Ill' (Synchronous Machines,
Dover Publications Inc., 1968)
2 CLARKE, E.: 'Circuit analysis of a-c power systemsvol. IF (John Wiley 8c Sons
Inc., New York, 1950)
3 PARK, R.H.: 'Two reaction theory of synchronous machinespart I: generalized
method of analysis', IEEE Trans., July 1929, 48, pp. 716-730
4 PARK, R.H.: 'Two reaction theory of synchronous machinespart IF, IEEE
Trans., June 1933, 52, pp. 352-355
5 ARRILLAGA, J., ARNOLD, C.P., and HARKER, B.J.: 'Computer modelling of
electrical power systems' (John Wiley 8c Sons, Chichester, England, 1983)
6 IEEE committee report: 'Computer representation of exciter systems', IEEE
Trans., June 1968, PAS-87, (6), pp. 1460-1464
7 IEEE committee report: 'Dynamic models for steam and hydro turbines in
power-system studies', IEEE Trans., 1973, PAS-92, (6), pp. 1904-1915
8 ARNOLD, C.P.: 'Solution of the multimachine power system stability problem'.
PhD thesis, UMIST, Manchester, UK, 1976
9 DANDENO, P.L., and KUNDUR, P.: 'A noniterative transient stability program
including the effects of variable load-voltage characteristics', IEEE Trans., 1973,
PAS-92, (5), pp. 1478-1484
10 BERG, G.L.: 'Power system load representation', Proc. IEE, March 1973, 120, (3),
pp. 344-348
11 ARNOLD, C.P., TURNER, K.S., and ARRILAGA, J.: 'Modelling rectifier loads
for a multi-machine transient stability programme', IEEE Trans., 1980, PAS-99,
pp. 78-85
12 GIESNER, D.B., and ARRILLAGA, J.: 'Operating modes of the three-phase
bridge converter', Int.]. Elect. Eng. Educ, 1970, 3, pp. 373-388
13 DOMMEL, H.W., and SATO, N.: 'Fast transient stability solutions', IEEE Trans.,
1972, PAS-91, (4), pp. 1643-1650
321
14 'Hierarchical structure'. IEEE working group on dynamic performance and

modelling of DC systems, 1980
15 CIGRE Working Group 14.02: 'Commutation failures in HVDC transmission
systems due to AC system faults', Electra, December 1996, (169), pp. 59-85
16 ANDERSON, G.W.J.: 'Hybrid simulation of AC-DC power systems'. PhD thesis,
Chapter 8
Electromechanical stability with

transient converter simulation
8.1 Introduction
The previous two chapters have described independently the electromagnetic and electromechanical behaviour of AC-DC power systems. In practice, however, the two are closely interrelated and the results obtained for
either of them independently will be in error.
It is, of course, possible to include the equations of motion of the
generators in the electromagnetic-transient program to provide more realistic solutions. However, considering the different time constants influencing the electromechanical and electromagnetic behaviour, such an approach
would be extremely inefficient. The electromagnetic-transient simulations
use steps of (typically) 50 //s, whereas the stability programmes use steps at
least 200 times larger.
To reduce the computational requirements the NETOMAC package1 has
two separate modes. An instantaneous mode is used to model components
in three-phase detail with small time steps in a similar way to the EMTP/
EMTDC programs.2 The alternative is a stability mode and uses r.m.s.
quantities at fundamental frequency only, with increased time-step lengths.
The program can switch between the two modes as required while running.
The HVDC converter is either modelled elementally by resistive, inductive
and capacitive components, or by quasisteady-state equations, depending on
the simulation mode. In either mode, however, the entire system must be
modelled in the same way. When it is necessary to run in instantaneous
mode, a system of any substantial size would still be very computationally
intensive.
A more efficient algorithm that combines the two programs described in
Chapters 6 and 7 is described here. 3 ' 4 It is a hybrid algorithm which takes
advantage of the computationally inexpensive dynamic representation of
the AC system in a stability program, and the accurate dynamic modelling
of the converter nonlinearities.
The slow dynamics of the AC system are sufficiently represented by the
stability program and, at the same time, the fast dynamic response of the
324
Detailed
analysis of dc
system (and possibly
some of ac system)
Figure 8.1
The hybrid concept
HVDC system is accurately represented by electromagnetic simulation.

A hybrid approach is also useful to study the impact of AC-system dynamics, particularly weak AC systems, on DC-system transient performance.
Disturbance response studies, control assessment and temporary overvoltage consequences are all typical examples for which a hybrid package is
suited.
The basic concept, shown in Figure 8.1, is not restricted to AC-DC
applications only. A particular part of an AC system may sometimes require
detailed three-phase modelling and this same hybrid approach can then be
used. Applications include the detailed analysis of synchronous or static
compensators, FACTS devices or the frequency-dependent effects of transmission lines.
Detailed modelling can also be applied to more than one independent
part of the complete system. For example, if an AC system contains two
HVDC links, then both links can be modelled independently in detail and
their behaviour included in one overall AC electromechanical-stability
program.
8.2 Description of the hybrid algorithm

The proposed hybrid algorithm utilises electromechanical simulation as the
steering program, and the electromagnetic-transients program is called as
a subroutine. The interfacing code is written in separate routines to
minimise the number of modifications and thus make it easily applicable to
any stability and dynamic-simulation programs. To make the description
more concise, the component programs are referred to as TS (for transient
stability) and EMTDC (for electromagnetic transient simulation). The combined hybrid algorithm is called TSE.
325
Stability Program
(a)
EMTDC Program
Stability Program
Stability
Program
(b)
Figure 8.2 Example of interfacing procedure
With reference to Figure 8.2a, initially, the TSE hybrid reads in the data
files and runs the entire network in the stability program, until electromechanical steady-state equilibrium is reached. The quasisteady-state
representation of the converter is perfectly adequate as no fault or disturbance has yet been applied. At a selectable point in time prior to a network
disturbance occurring, the TS network is split up into the two independent
and isolated systems, system 1 and system 2.
For the sake of clarity, system 1 is classified as the AC part of the system
modelled by the stability program TS, and system 2 is the part of the system
modelled in detail by EMTDC.
The snapshot data file is now used to initialise the EMTDC program used,
instead of the TS representation of system 2. The two programs are then
326
interfaced and the network disturbance can be applied. The system 2

representation in TS is isolated but kept up to date during the interfacing
at each TS time step to allow tracking between programs. The AC network
of system 1 modelled in the stability program also supplies interface data to
the system 2 network in TS as shown in Figure 8.2b.
While the disturbance effects abate, the quasisteady-state representation
of system 2 in TS and the EMTDC representation of system 2 are tracked.
If both of these system 2 models produce the same results within a
predefined tolerance and over a set period, the complete system can then
be reconnected and used by TS, and the EMTDC representation terminated. This allows better computational efficiency, particularly for long
simulation runs.
8.2.1 Individual program modifications

To enable EMTDC to be called as a subroutine from TS requires a small
number of changes to its structure. The EMTDC algorithm is split into three
distinct segments, an initialising segment, the main time loop and a
termination segment. This allows TS to call the main time loop for discrete
periods as required when interfacing. The EMTDC options, which are
normally available when beginning a simulation run, are moved to the
interface data file and read from there. The equivalent-circuit source values,
which TS updates periodically, are located in the user-accessible DSDYN file
of EMTDC (described in Chapter 6).
The TS program requires only minor modifications. The first is a call of
the interfacing routine during the TS main time loop as shown in Figure
8.3. The complete TS network is also split into system 1 and system 2 and
isolated at the interface points, but this is performed in separate code to TS.
The only other direct modification inside TS is the inclusion of the interface
current injections at each TS network solution.
8.2.2 Dataflow
Data for the detailed EMTDC model is entered into the program database
via the PSCAD graphics. Equivalent circuits are used at each interface point
to represent the rest of the system not included in the detailed model. This
system is then run until steady state is reached and a snapshot taken. The
snapshot holds all the relevant data for the components at that point in time
and can be used as the starting point when interfacing the detailed model
with the stability program.
The stability program is initialised conventionally through power-flow
results via a data file. An interface data file is also read by the TSE hybrid
and contains information such as the number and location of interface
buses, analysis options and timing information. The data flow diagram is
shown in Figure 8.4.
read stability input data

from loadflow results
calculate machine initial
conditions
determine stability
step length
J_
perform switching and subsequent bifactorisation

if necessary
yes
pass information
to detailed model
J_
solve EMTDC
extract information
from detailed solution
and include in stability
programme
solve stability equations
T = T + step length
i
output results
Figure 8.3 Modified TS steering routine
327
328
positive
sequence
data for
whole
system
Figure 8.4
output
Dataflow
8.3 TS/EMTDC interface

Hybrid simulation requires exchange of information between the two
separate programs. The information that must be transferred from one
program to the other must be sufficient to determine the power flow in or
out of the interface. Possible parameters to be used are the real power, P,
the reactive power, Q, the voltage, V, and the current, I, at the interface
(Figure 8.5). Phase-angle information is also required if separate phase
frames of reference are to be maintained.
An equivalent circuit representing the network modelled in the stability
program is used in EMTDC and vice versa. The equivalent circuits are as
Figure 8.5
Hybrid interface

V
329
Interface
System 1
System 2
Figure 8.6 Representative circuit
shown in Figure 8.6, where E t and Zt can represent the equivalent circuit
of system 1 and Ic and Z2 the equivalent circuit of system 2.
8.3.1 Equivalent impedances

The complexity of the equivalent-impedance representation varies considerably between the two programs.
In the TS program, Ic and Z2 represent the detailed part of the system
modelled by EMTDC. TS, being positive-sequence and fundamental-frequency based, is concerned only with the fundamental real and reactive
power inflow or outflow through the interface. The equivalent impedance,
Z2, then is arbitrary, since the current source Ic can be varied to provide the
correct power flow.
To avoid any possible numerical instability, a constant value of Z2,
estimated from the initial power-flow results, is used for the duration of the
simulation.
The EMTDC program represents system 1 by a Thevenin equivalent (E t
and Zj) as shown in Figure 8.6. The simplest Zx is an R L series
impedance, representing the fundamental-frequency equivalent of system 1.
It can be derived from the results of a power flow and a fault analysis at the
interface bus.
The power flow provides an initial current through the interface bus and
the initial interface-bus voltage. A fault analysis can easily determine the
fault current through the interface for a short-circuit fault to ground. If the
network requiring conversion to an equivalent circuit is represented by a
Thevenin source, E1? and Thevenin impedance, Zl9 as shown in Figure 8.7,
these values can thus be found as follows:
from the power flow circuit
(8.3.1)
330
Ei
(a)
(b)
Figure 8.7
Derivation of Thevenin equivalent circuit

a Power-flow circuit
b Fault circuit
and from the fault circuit

\ = IFZ,
(8.3.2)
Combining these two equations
IF-
(8.3.3)
Ex can then be found from either eqn. 8.3.1 or 8.3.2.

During a transient, the impedance of the synchronous machines in system
1 can change. The net effect on the fundamental power in or out of the
equivalent circuit, however, can be represented by varying the source Ex and
keeping Zt constant.
331
EMTDC is a point-on-wave-type program and, consequently, involves

frequencies other than the fundamental. A more advanced equivalent
impedance capable of representing different frequencies is needed. This
question is discussed in section 8.6.
8.3.2 Equivalent sources

Information from the EMTDC model representing system 2 (in Figure 8.6)
is used to modify the source of the equivalent circuit of system 2 in the
stability program. Similarly, data from TS is used to modify the source of
the equivalent circuit of system 1 in EMTDC. These equivalent sources are
normally updated at each TS step length (refer to section 8.5). From Figure
8.6, if both Z1 and Z2 are known, additional information is still necessary to
determine update values for the sources Ic and E1. This information can be
selected from the interface parameters of voltage V, current Il9 real power
P, reactive power Q and power factor angle $.
The interface voltage and current along with the phase angle between
them are used to interchange information between programs. Reasons for
the selection of these interface parameters are discussed in section 8.4.
833
Phase and sequence-data conversions
An efficient recursive curve-fitting algorithm is described in section 8.4 to

extract fundamental-frequency information from the discrete pointoriented waveforms produced by detailed programs as EMTDC.
Analysis of the discrete data from EMTDC is performed over a fundamental-period interval, but staggered to produce results at intervals less than a
fundamental period. This allows the greatest accuracy in deriving fundamental results from distorted waveforms.
The stability program requires only positive-sequence data, so data from
the three AC phases at the interface(s) is analysed and converted to positive
sequence by conventional means. The positive-sequence voltage, for
example, can be derived as follows
Vps = ^(V a + aV, + a2Vc)
(8.3.4)
where
Yps = positive sequence voltage
Vfl, Vfc, Vc = phase voltage
a = 120 degree forward rotation vector (i.e. a = 1 L120)
Positive-sequence data from the stability program is converted to three
332
phase through simple multiplication of the rotation vector. For the voltage
Va = Vps
(8.3.5)
% = a2Vps
(8.3.6)
Vc = aVps
(8.3.7)
Some transient-stability programs have the additional capability of

modelling negative and zero-sequence networks. If this option is utilised,
sequence data from a stability program can then be converted to three
phase through the following formulae
a
* ps '
2
\. =a V
o
ns '
zs
yo.o.oj
+ aV + V
ps
'
ns
V, = aVnv 4- d2Y
c
ps
'
zs
+V
ns '
zs
(8 3 Q\
\**J,KJ*%JJ
f8 S 101
\^ *-' y*j
where
Vns = negative-sequence voltage
Vzs = zero-sequence voltage
Similarly, any unbalance in system 2 can be accommodated in the transientstability program.
83 A Interface variables derivation

In Figure 8.6, Et and Zt represent the equivalent circuit of system 1
modelled in EMTDC, and Z2 and Ic represent the equivalent circuit of
system 2 modelled in the stability program. V is the interface voltage and
Ij the current through the interface which is assumed in the direction
shown.
From the detailed EMTDC simulation, the magnitude of the interface
voltage and current are measured, along with the phase angle between
them. This information is used to modify the equivalent circuit source (Ic)
of system 2 in TS. The updated Ic value can be derived as follows.
From Figure 8.6
E1=I1Z1 +V
(8.3.11)
V = I2Z2
(8.3.12)
h = h +!c
(8.3.13)
333
From eqns. 8.3.12 and 8.3.13

= 1^2 + 1,22
(8.3.14)
From eqn. 8.3.11

E1 = I1Z1Z_(0I1+0Z1)+VZ_0V
= I^cosCfl,! + 0Z1) +jl1Z1 sin(0n + 0Z1) + Vcos0 v +;Vsin0 v
(8.3.15)
and
0n = ey-(j)
(8.3.16)
where 0 is the displacement angle between the voltage and the current.
Thus, eqn. 8.3.15 can be written as
x
= IiZj cos(0 v + P) +jl1Zl sin(0 v + jff) + V c o s 0 v +7 V sin0 v

= 1 ^ (cos 0V cos j? sin 0V sin /?) 4- V cos 0V
+>[I 1 Z 1 (sin 0V cos p + cos 0V sin j3) + V sin 0V]
(8.3.17)
where
= 0Zi ~
If
Ei = E l r
then equating the real terms
E l r = ( 1 ^ ! cosj? + V)cos0v -f ( - 1 ^ ! sin jS)sin0v
(8.3.18)
where Zt is known and constant throughout the simulation.

From the EMTDC results, the values of V, I and <j) are also known and,
hence, so is /?. E t can be determined in the TS phase reference frame from
knowledge of Zt and the previous values of interface current and voltage
from TS, through the use of eqn. 8.3.11.
From eqn. 8.3.18, making
AÎ^cosiS + V
(8.3.19)
B= -I^sinjS
(8.3.20)
334
and remembering that
A cos 0V + B sin 6y = ^ A 2 + B 2 cos(0v \j/)
(8.3.21)
where
the voltage angle 6y in the TS phase reference frame can be calculated, i.e.
(8.3.23)
The equivalent current source, \c, can be calculated by rearranging eqn.

8.3.14
ic = ^ ueY - eZ2) - \lLen
(8.3.24)
where 9n is obtained from eqn. 8.3.16.

In a similar way, data from the transient-stability program simulation can
be used to calculate a new Thevenin souree-voltage magnitude for the
equivalent circuit of system 1 in the EMTDC program. Knowing the voltage
and current magnitude at the TS program interface and the phase difference between them, by a similar analysis the voltage angle in the EMTDC
phase reference frame is
(8 3 25)
- -
where I cr is the real part of I c and

V
C = CQS 0Z2 - It cos <f>
(8.3.26)
v
D = cos 9Z2 - lx sin <j)
(8.3.27)
(8.3.28)
(8.3.29)
335
Knowing the EMTDC voltage angle, 6y, allows calculation of the EMTDC
current angle, 6n, from eqn. 8.3.28. The magnitude value of Ex can then be
derived from eqn. 8.3.11.
8.4 EMTDC to TS data transfer

A significant difference between TS and EMTDC is that, in TS, sinusoidal
waveforms are assumed. However, during faults the EMTDC waveforms are
very distorted. There is still no consensus on a universal definition for power
when waveforms are nonsinusoidal.5 Reactive power, for example, has two
primary definitions. The Budeanu definition is accepted by the American
National Standards Institute (ANSI) and the Institute of Electrical and
Electronic Engineers (IEEE), although the International Electro-technical
Commission (IEC) uses a different definition proposed by Fryze.6
In an earlier hybrid program,7 the total r.m.s. real power was used as the
interfacing variable to transfer information. The total r.m.s. power is made
up of fundamental power (Ff) and harmonic power (P^), i.e.
rms
The r.m.s. power can be extracted directly from the waveforms and,
therefore, Fourier transform or curve fitting methods are not necessary, i.e.
<8A2)
This method greatly reduces the computing time necessary for the data
extraction from EMTDC.
The choice of r.m.s. power was made on the basis that harmonic power
flow will result only from in-phase components of harmonic voltage and
current. The assumption was made that if a system contains only a lowresistive component, then the harmonic power flow is not significant.
This, however, is not valid for every situation and, particularly, at the
inverter end of an HVDC link, the effect of the resistive component of the
network is not insignificant. Certain harmonic frequencies in a network may
also be parallel resonant or close to parallel resonance and exhibit more
resistance than reactance. The presence of a transient may excite the
resonant frequency and greatly affect the results. It is important, therefore,
to model.the fundamental power flow accurately.
Another factor to consider is that the direction of fundamental power may
not necessarily be the same as the direction of harmonic power. With an
HVDC link, although fundamental power is drawn into a rectifier, much of
the harmonic power flow will be in the reverse direction. The r.m.s. real
336

Instantaneous power
rms power over T
800 ::
600 -
f\, ; \?V V? V
'
0.20
0.28
400
200 -
ti ; ' y
0 -200 -400-600
-800 10000.00
0.04
0.08
0.12
0.16
0.24
0.32
0.36
0.40
Time (s)
Figure 8.8
Total r.m.s. power following a single-phase fault at the inverter end
power measured is the difference between these two powers and so not
entirely representative of the fundamental power load of the HVDC link.
Conversely, at the inverter end, the r.m.s. real power will include harmonic
power flow into the AC system. This may exaggerate the amount of true
fundamental power from the inverter.
Figure 8.8 shows an EMTDC-simulated result of a single-phase fault at the
inverter end of the CIGRE HVDC benchmark model (see Appendix III).
The instantaneous power is shown, along with the total r.m.s. power over
discrete one-cycle intervals. For hybrid interfacing purposes, data is transferred at discrete intervals equal to the stability program time step, and
these intervals are significantly larger than the interval between the
discrete output points constituting the instantaneous power. The r.m.s.
power is then taken over a fundamental period to represent its use in a
hybrid situation.
Figure 8.9 shows an analysis of the single-phase fault comparing fundamental-frequency power with the total r.m.s. power. The fundamentalfrequency power was derived using the curve-fitting method described in
the next section to extract both fundamental voltage and current. The
comparison shows that, particularly during the fault time, there exists a
significant amount of Fh or harmonic r.m.s. power in the total r.m.s. power.
When the three-phase network is unbalanced the fundamental-frequency

Fundamental power
337
Total rms power
900 800 -
700 -
'
'
0.28
0.32
600500
400
300
200
100
0-100
-2000.00
0.04
0.08
0.12
0.16
0.20
0.24
0.36
0.40
Time (s)
Figure 8.9 Fundamental power versus total r.m.s. power
real power consists of positive, negative and zero-sequence components, i.e.

*fns + * fzs
(8.4.3)
and the negative and zero-sequence powers cause additional power loss in
the network. Figure 8.10 shows the sequence components of the fundamental r.m.s. power shown in Figure 8.9.
Fundamental-frequency negative-sequence currents, in the presence of
damper windings, can produce a braking torque which will retard the
rotor.8 Damper windings, however, also serve to lower the negative-sequence impedance of a machine which, in turn, reduces the negativesequence voltage.9 Which of these two opposing effects is dominant depends on the resistance of the damper windings. High-resistance windings
cause the braking torque to be the significant effect. The braking power of
the negative-sequence current is
Pb = \ I^R,
I2 (r - r)
(8.4.4)
(8.4.5)
338

- Fundamental positive sequence power
Fundamental negative sequence power
Fundamental zero sequence power
000 -I
900
800 -
700
'
600
500
400
300 200 100
-100 -200 0.00
0.04
0.08
P-'
-1.
0.12
0.16
--,
0.20
0.24
0.28
0.32
0.36
0.40
Time(s)
Figure 8.10
Sequence components of fundamental-frequency r.m.s. power
where
lrn = negative-sequence rotor current
R,. = rotor resistance
Isn = negative-sequence stator (or armature) current
rn = negative-sequence resistance of the machine
rp = positive-sequence resistance of the machine
The negative-sequence resistance can be approximated from the rotor and
the armature resistance, i.e.
r ^(R s + |R r )
(8.4.6)
where Rs is the armature resistance.

Retardation of the rotor can also be caused by DC components in the
armature windings. Three-phase faults at or near machine terminals can
cause DC components of short-circuit armature current which can have a
definite braking effect on the machine.9 The braking power in this case is
P b = 2Is2(r - rp)
2
^ I (r __ r )
(8.4.7)
(8.4.8)

- Fundamental positive sequence power
339
Total rms power
Fundamental power
1000
-200
0.00
0.04
0.08
0.12
0.16
0.20
0.24
0.28
0.32
0.36
0.40
Time (s)
Figure 8.11
Comparison of total r.m.s. power, fundamental-frequency power and

fundamental-frequency positive-sequence power
where
I s = the effective very low-frequency value of the armature current
iDC = instantaneous DC component of armature current
The total r.m.s. power, then, is not always equivalent to either the
fundamental-frequency power or the fundamental-frequency positive-sequence power. A comparison of these three powers is shown in Figure 8.1 L
The difference between total r.m.s. power and positive-sequence power can
be seen to be highly significant during the fault.
The most appropriate power to transfer from EMTDC to TS is, then, the
fundamental-frequency positive-sequence power. This, however, requires
knowledge of both fundamental-frequency positive-sequence voltage and
fundamental-frequency positive-sequence current. These two variables contain all the relevant information and, hence, the use of any other power
variable to transfer information becomes unnecessary.
8.4.1 Data extraction from converter waveforms

At each step of the transient-stability program power-transfer information
needs to be derived from the distorted converter waveforms. This can be
achieved using the FFT which provides accurate information for the whole
340
(a) Discrete window of fundamental period length
(b) Discrete window of half fundamental period length
(c) Staggered discrete window of fundamental period length
(d) Continuous window with decay weighting
Figure 8.12 Analysis techniques
frequency spectrum. However, only the fundamental frequency is used in

the stability program and the simpler curve-fitting approach (CFA), described in Appendix VI, should provide sufficient accuracy.
Four CFA alternatives are shown in Figure 8.12. Two of them, a and b,
are discrete rectangular windows of full and half-fundamental period duration. Figure 8.12c uses a full period in an overlapping manner to produce
results twice per cycle. Finally, a squared-decay weighted window is used in
341
2.00
-5.00
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
Time (s)
Figure 8.13
Simulated fault waveform
Figure 8.126? to emphasise the more recent trends of the waveform; it is

effectively a nondiscrete or infinitely-sized moving window.
The accuracy of the CFA algorithm is compared with the results of an FFT
for the voltage waveform of Figure 8.13. This waveform, derived with the
EMTDC program, corresponds to a single-phase fault on the rectifier side
of the CIGRE benchmark model (described in Appendix III).
Figures 8.14a and 8.146 show the voltage magnitude of the various
techniques described above. The fixed full-period window using both the
CFA and the FFT gave practically identical results.
Figure 8.14a compares the half-period fixed window with the full period
and shows the greater amplitude response of the smaller window. A
complete Fourier analysis of the first cycle after the fault is removed,
however, shows considerable harmonic components including DC. As
shown previously, the DC component can cause the fundamental result from
a half-period fixed-window length to be significantly different than it
actually is.
The decay-weigh ted window (Figure 8.146) displayed an obvious delay in
its results, especially during fault recovery. A high initial peak was also
registered which again could have been partially due to the presence of DC.
The staggered full-period analysis displayed very good results for both
initial-value and amplitude-variation tracking. At both the fault application
and removal, the staggering process was reset so as not to apply an analysis
window over the discontinuity.
Phase results for the waveform are shown in Figure 8.15. The analyses
using the full and half-period discrete windows are very similar except for
342

CFA(T/2)
CFA (T) and FFT
ii
_____
(a)
- Staggered CFA
Weighted CFA (continuous)
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
Time (s)
(b)
Figure 8.14
Postfault magnitude analysis
an initially larger decrease in the half-period window results directly after

the fault removal. The phase response increased in both cases by about 20
degrees above normal after this initial decrease, before settling near to its
prefault value.
From the above analysis, when data is required at intervals of less than a
cycle, it is recommended that the CFA of a fundamental period length in
the staggered manner be used. However, if data is only required at one cycle
intervals, the CFA analysis of a nonstaggered fundamental period length is
the optimum choice.

CFA(T)
CFA(T/2)
343
Staggered CFA
Weighted CFA (continuous)
30
20
10
i
i
i
-10
!* "
i
i
i
-20
-30
-40
-60
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
Time(s)
Figure 8.15
Postfault phase analysis
8.5 Interaction protocol

The data from each program must be interchanged at appropriate points
during the hybrid simulation run. The timing of this data interchange
between the TS and EMTDC programs is important, particularly around
discontinuities caused by fault application and removal.
Figure 8.16 shows three possible alternatives and assumes, for simplicity,
that the TS step length equals the fundamental-period length (T).
In Figure 8.16a both programs interchange information at the TS time
step, every period. This method uses the previous time-step (t At) data to
update the equivalent circuit for the present time step.
Figure 8.166 shows the use of the previous time-step data when passing
information from TS to EMTDC but present time-step data when passing
information from EMTDC to TS. When the two programs are run concurrently TS passes its information to EMTDC. EMTDC then runs to t 4- At,
and the information gathered over this present time step is given back to
TS at time /. TS now runs to t + At, using equivalent-circuit information
derived for that particular time step.
The third option, in Figure 8.16^, represents the use of present time-step
information for both data paths. For example, TS is run to time t -\- At and
the information from this run used to update the TS equivalent in EMTDC
at time /. EMTDC is then also run to time t + At and, similarly, the
equivalent circuit of the EMTDC solution in TS is updated. This process is
then repeated over the same time step until the data passed between
344
EMTDC
EMTDC
EMTDC
Time
Figure 8.16
Interfacing methods
successive runs is effectively the same. This is consequently an iterative

procedure and, therefore, increases the computation time and complexity
of the interfacing.
The electromechanical AC-system solution exhibits a relatively slow dynamic response to any disturbances and the accuracy of previous time-step
information passed from TS to EMTDC is, therefore, an adequate assumption. On the other hand, the HVDC converter exhibits fast dynamic
behaviour. Present time-step information is, thus, more appropriate to pass
from EMTDC to TS when solving the TS network equation. Therefore, the
optimum method is that of Figure 8.16&. It is noniterative and computationally economical yet, at the same time, accounts adequately for the dynamics
of the respective components.
345
Stability
step length
EMTDC
Dynamic program
step length
Figure 8.17
Normal interaction protocol
The interfacing philosophy can be modified to a cross between the

methods of Figure 8.16a and b to cater for TS step lengths which are less
than a fundamental period. This is a more typical situation and the
information-exchange protocol for this case is shown in Figure 8.17. A
portion of the figure is sequentially numbered to show the order of
occurrence of the variable interchange. In the example, the stability step
length is exactly one half of a fundamental period.
Following the sequential numbering on Figure 8.17, at a particular point
in time, the EMTDC and TS programs are concurrent and the TS information from system 1 is passed to update the system 1 equivalent in EMTDC.
This is shown by the arrow marked 1. EMTDC is then called for a length of
half of a fundamental period (arrow 2) and the curve-fitted results over the
last full fundamental period processed and passed back to update the system
2 equivalent in TS (arrow 3). The information over this period is passed
back to TS at the mid point of the EMTDC analysis window which is half a
period behind the current EMTDC time. TS is then run to catch up to
EMTDC (arrow 4), and the new information over this simulation run used
to again update the system 1 equivalent in EMTDC (arrow 5). This protocol
continues until any discontinuity in the network occurs.
When a network change such as a fault application or removal occurs, the
interaction protocol is modified to that shown in Figure 8.18. The curvefitting analysis process is also modified to avoid applying an analysis window
over any point of discontinuity.
The sequential numbering in Figure 8.18 explains the flow of events. At
the fault time, the interface variables are passed from TS to the system 1
equivalent in EMTDC in the usual manner, as shown by the arrow marked
1. Neither system 1 nor system 2 has yet been solved with the network
change. The fault is now applied in EMTDC which is then run for a full
fundamental-period length past the fault application (arrow 2) and the
information obtained over this period passed back to TS (arrow 3). The fault
346

Disturbance
Applied
Stability
step length
EMTDC
Dynamic program
step length
Figure 8.18
Interaction protocol around a disturbance
is now also applied to the TS program which is then solved for a period until
it has again reached EMTDC's position in time (arrow 4). The normal
interaction protocol is then followed until any other discontinuity is
reached.
A full-period analysis after the fault has been applied is necessary to
extract accurately the fundamental-frequency component of the interface
variables. The mechanically-controlled nature of the AC system implies a
dynamically-slow response to any disturbance and so, for this reason, it is
considered acceptable to run EMTDC for a full period without updating the
system 1 equivalent circuit during this time.
8.6 Interface location

The original intention of the initial hybrid algorithm7 was to model the AC
and DC solutions separately. The point-of-interface location was consequently the converter bus terminal. The detailed DC-link model included
all equipment connected to the converter bus, such as the AC filters, and
every other AC component was modelled within the stability analysis. A
fundamental-frequency Thevenin's equivalent was used to represent the
stability program in the detailed solution and vice versa.
An alternative approach has been proposed10 where the interface location is extended out from the converter bus into the AC system. This
approach maintains that, particularly for weak AC systems, a fundamentalfrequency equivalent representing the AC system is not sufficiently adequate
at the converter terminals. In this case, the extent of the AC system to be
included in the DC system depends on phase imbalance and waveform
distortion.
347
Although the above concept has some advantages, it also suffers from
many disadvantages. The concept is proposed, in particular, for weak AC
systems. A weak AC system, however, is likely to have any major generation
capability far removed from the converter-terminal bus as local generation
serves to enhance system strength. If the generation is, indeed, far removed
out into the AC system, then the distance required for an interface location
to achieve considerably less phase imbalance and waveform distortion is also
likely to be significant.
The primary advantage of a hybrid solution is in accurately providing the
DC dynamic response to a transient-stability program, and in efficiently
representing the dynamic response of a considerably-sized AC system to the
DC solution. Extending the interface some distance into the AC system,
where the effects of a system disturbance are almost negligible, diminishes
the hybrid advantage. If a sizeable portion of the AC system requires
modelling in detail before interfacing to a transient-stability program can
occur, then one might question the use of a hybrid solution at all and
instead use the more conventional approach of a detailed solution with AC
equivalent circuits at the system cut-off points.
Another significant disadvantage in an extended interface is that AC
systems may well be heavily interconnected. The further into the system that
an interface is moved, the greater the number of interface locations
required. The hybrid interfacing complexity is thus increased and the
computational efficiency of the hybrid solution decreased. The requirement
for a detailed representation of a significant portion of the AC system serves
to decrease this efficiency, as does the increased amount of processing
required for variable extraction at each interface location.
The advantages of using the converter bus are:
the detailed system is kept to a minimum;

interfacing complexity is low;
computational expense is minimised;
converter-terminal equipment, such as filters, synchronous condensers,
SVCs, etc., can still be modelled in detail.
The major drawback of the detailed solution is in not seeing a true picture
of the AC system, since the equivalent circuit is fundamental-frequency
Figure 8.19
Test system
348
based. Waveform distortion and imbalance also make it difficult to extract

the fundamental-frequency information necessary to transfer to the stability
program.
The problem of waveform distortion for transfer of data from EMTDC to
TS is dependent on the accuracy of the technique for extraction of
interfacing variable information. If fundamental-frequency quantities can be
accurately measured under distorted conditions, then the problem is solved.
Section 8.4 has described an efficient curve-fitting algorithm to extract the
required information from distorted waveforms. It has been shown that
using the technique described, fundamental-frequency quantities can be
accurately measured.
Moreover, a simple fundamental-frequency equivalent circuit is insufficient to represent the correct impedance of the AC system at each frequency. Instead, this can be achieved by using a fully frequency-dependent
equivalent circuit of the AC system11 at the converter terminal instead of
just a fundamental-frequency equivalent. A frequency-dependent equivalent
avoids the need for modelling any significant portion of the AC system in
detail yet still provides an accurate picture of the system impedance across
its frequency spectra.
To show the effect of a frequency-dependent equivalent, the test system
of Figure 8.19 is used with an inverter-effective short-circuit ratio of 2. The
DC link is represented by the CIGRE benchmark model, described in
Appendix III.
A three-phase solid short-circuit fault is applied at the inverter terminal.
Three cases were investigated, the first being the entire system represented
by the detailed solution. The second and third cases were the hybrid
solution interfaced at the converter bus, one case with a fundamentalfrequency Thevenin representation of the stability program in the detailed
solution, and the other with a frequency-dependent equivalent.
The inverter-terminal voltage results for the three cases are shown in
Figure 8.20. The benchmark EMTDC case and the frequency-dependent
equivalent case are identical. The fundamental-frequency equivalent case
(Figure 8.206) shows more distortion and prolonged effects from the
disturbance than the benchmark EMTDC case.
The inverter AC-current results are shown in Figure 8.21. These, too,
show a significant difference between the fundamental-frequency equivalent
case (Figure 8.216) and the benchmark EMTDC solution. The overcurrents
during the fault are less for the fundamental-frequency equivalent case,
although the distortion is again more evident and prolonged. The case
using a frequency-dependent equivalent for the AC-system equivalent in
EMTDC is again identical to the benchmark EMTDC solution.
These results show the inadequacy of a fundamental-frequency equivalent
at the converter terminal. Additionally, the results show that using a
frequency-dependent equivalent is also entirely adequate for presenting the
correct effects of waveform distortion to the converter terminal.

-Phase A
PhaseB
PhaseC
(a) EMTDC
(b) TSE Hybrid - Fundamental frequency equivalent
5
o
-1.4
0.98
1.00
1.02
1.04
1.06
1.08
1.10
1.12
1.14
Time (s)
(c) TSE Hybrid - Frequency dependent equivalent
Figure 8.20
Inverter AC voltage
1.16
1.18
349
350

Phase B
-Phase A
PhaseC
(a) EMTDC
(b) TSE Hybrid - Fundamental frequency equivalent
-50.0
0.98
1.00
1.02
1.04
1.06
1.08
1.10
1.12
1.14
Time (s)
(c) TSE Hybrid - Frequency dependent equivalent
Figure 8.21
Inverter AC current
1.18
1.18
351
8.7 TSE hybrid algorithm

The hybrid-algorithm interface is shown in Figure 8.22. It contains the five
possible states summarised in Table 8.1. The default state is 0 for the first
time that the algorithm is called. This occurs once TS has reached steadystate electromechanical equilibrium and is ready for interfacing, prior to a
system disturbance.
Under the 0 state, EMTDC is called for a full fundamental period ahead
of TS and its interfacing variables extracted and converted to positive
sequence. The state is now set to 1.
Under the 1 state, TS is run for one half of a fundamental period and
the variables at its interface location are compared with those of EMTDC.
This is simply to check and ensure that the data in both files is correctly set
up. The TS network is then split into systems 1 and 2, and Nortonequivalent circuits set up in each system. The state is now set to 1,
representing normal interfacing conditions, and TS is solved until its time
is concurrent with EMTDC.
Under the normal interfacing of state 1, variable information is given to
EMTDC from TS to update its equivalent circuit. The equivalent circuit of
system 1 in the TS model of system 2 is updated in the same way. EMTDC
is then called for one half of a period and the information measured over
the last fundamental period of its simulation used to update the equivalent
circuit of system 2 in TS.
In the interface data file, a time can be specified as a minimum termination time for EMTDC, after any network disturbance is cleared. Once this
time is reached in the program, the interface variables of both system 2
models are compared at each TS time step. If they correlate within a
predefined tolerance over a set period, then EMTDC can be terminated and
the TS representation of system 2 once again takes over. During the period
when they are being checked, if at any time the variables are outside the
specified tolerance, the checking-period time is restarted.
When a network admittance change is about to occur, such as a fault being
applied or removed, the interfacing routine is entered with a state value of
2. At this point in time, TS is concurrent with EMTDC. The TS-interface
variables are passed to EMTDC which is then run for a full period from the
network discontinuity time. It is not accurate to apply the analysis window
over the network discontinuity, so the staggered window process must be
restarted. The variables from EMTDC are passed back to TS at the time of
the disturbance and the state is changed to 3.
State 3 allows TS to catch up to EMTDC before the normal interfacing
procedure of state 1 recommences.
352
compare variables of
EMTDC with TS
Ts=Ts+stepl
yes
isolate TS systems
land 2
set up equivalent
circuits
yes
state=l
Ts=0
Ts=Ts+stepl
call EMTDC for a
period T
convert EMTDC o/p
to positive sequence
state=-l
Ts=0
Ts=Ts+stepl
Figure 8.22
TSE hybrid algorithm

[step 1] TS program step length
[Ts] TS step length counter
[Tend] Minimum finish time for EMTDC

A
353
B
passTS system 1
variables to EMTDC
passTS system 1
variables to EMTDC
passTS system 1
variables to TS sys.2,
passTS system 1
variables to TS sys.2
call EMTDC for a

period of T
call EMTDC for a

period of T/2
convert EMTDC o/p

convert EMTDC o/p

set new equivalent

circuits in TS sys.l
Dass EMTDC variables

to TS sys.l
pass EMTDC
variables to TS sys.l
Ts=0
Ts=Ts+stepl
state=3
Ts=0
Ts=Ts+stepl
yes
extract TS system
2 variables
compare variables of
TS sys.2 and EMTDC
terminate EMTDC
reconnect TS
systems 1 and 2
Figure 8.22 {Continued)

[K] Time counter for EMTDC and TS comparison of variables
[Tc] Time over which comparison must be within tolerance
354

c
pass TS system 1
variables to EMTDC
pass TS system 1
variables to TS sys.2
call EMTDC for a
period of T/2
convert EMTDC o/p
pass EMTDC variables
toTSsys.l
state =1
Ts=0
Ts=Ts+stepl
Figure 8.22
Table 8.1
State
-1
0
1
2
3
(Continued)
Interfacing states
Event
isolate TS network ready to interface with EMTDC
initial time through interfacing routine
normal interaction between TS and EMTDC
system network change signalledrun EMTDC for a full period past
the fault and return information to TS
TS catching up with EMTDC after a system-network change
355
8,8 Test system and results

The test system shown in Figures 7.17 and 7.18 is also used here. As
explained in section 8.6, the high levels of current distortion produced by
the converter during the disturbance require a frequency-dependent model
of the AC system.11
Figure 8.23 shows the frequency-dependent equivalent circuit of the AC
system at the rectifier end of the test system and Figure 8.24 the corresponding harmonic impedances.
Figure 8.23 Frequency-dependent equivalent
Equivalent circuit
Actual system
25002000 -
1000-
1
i
1500-
5000-
100 200 300 400 500 600 700 800 900 1000 1100 1200
Frequency (Hz)
Figure 8.24 Frequency spectrum of the equivalent circuit in Figure 8.23
356
A three-phase short circuit is applied to the rectifier terminals of the link

at t = 1.7 s and cleared three cycles later.
8.8.1 Electromagnetic transient response

To highlight the differences between the TS and EMTDC programs when
used independently from each other, Figure 8.25 compares their results for
three-phase fault simulation.
On fault application, the AC voltage (graph a) drops to zero, the rectifier
is blocked and the converter DC current (graph b) collapses. The DC-line
voltage (graph c) is effected by the inductive and capacitive components of
the line.
When the fault is cleared (at t = 1.76 s), a dynamic overvoltage is set up
until the full power to the link is restored.
The rectifier is unblocked at t = 1.8 s and a few ms later the inverter takes
over conduction from its bypass valves. Following an initial current inrush
into the DC-line capacitance, the power setting is ramped up. At t = 1.97 s
the inverter is restarted and the link current dips, the current drop being
more pronounced, and maintained for a longer period, in the EMTDC
solution.
At the end of the power ramp, the AC current has returned to its nominal
value and after a small oscillatory settling of the controllers so do the DC
voltage and current.
8.8.2 TSE hybrid response

Clearly, the EMTDC and TS simulation results show substantial differences
during the fault and its subsequent recovery. In some situations the
differences may affect the transient stability of the AC system. Even if
stability is maintained, the recovery prediction is unrealistic when either the
electromagnetic-transient response of the DC system or the electromechanical modelling of the AC system are omitted.
This section describes the response of the TSE hybrid method which uses
the EMTDC and the TS solutions to model the HVDC link and AC system,
respectively.
Figure 8.26a displays the rectifier-terminal AC-voltage levels calculated
with the TS and TSE alternatives. The main difference between the two is
the transient overshoot evident in the TSE solution owing to the elemental
modelling of components in the EMTDC program.
Figures 8.266 and c show the AC-voltage response predicted by the
EMTDC and the TSE hybrid programs, respectively. The main difference
between them occurs in the period between fault removal (t = 1.76 s) and
t = 2.1 s, where the fixed-voltage source of the EMTDC model predicts a
higher voltage.

EMTDC only
357
TSonly
2.50
-2.60
(a)
- EMTDC only
TS only
(b)
-TSonly
EMTDC only
1.20
1.00
3
0.80
0.60
0.20
0.00
1.50
1.65
1.80
1.95
2.10
2.25
2.40
2.55
2.70
2.85
Time (s)
(c)
Figure 8.25
EMTDC versus TS response to a three-phase short circuit

a AC-voltage waveform
b DC-line voltage
c DC current
3.00
358

- TSE (TS variable)
TSonly
1.40
1.20 1.00
0.80
0.60
0.40
0.20
0.00
(a)
1 pu reference
- EMTDC only
-2.50
(b)
- T S E (EMTDC variable)
1 pu reference
2.50
1.50
1.50
Figure 8.26
1.65
1.80
1.95
2.10
2.25
Rectifier AC-voltage comparisons
2.40
2.55
2.70
2.85
3.00

EMTDConly
359
TSonty
TSE (EMTDC variable)
DCci
1 00
0.800.60 0.400.200.001.50
1.65
1.80
1.95
2.10
2.25
2.40
2.55
2.70
2.85
3.00
Time (s)
Figure 8.27 Rectifier-terminal DC-current comparison
-EMTDConly
TSonly
10.00
(a)
TSonly
- T S E (EMTDC variable)
10.00
-2.50
1.50
1.65
1.80
1.95
2.10
2.25
2.40
Time (s)
(b)
Figure 8.28 Rectifier-terminal DC-voltage comparisons
2.55
2.70
2.85
3.00
360

TS
-TSE
1200
1.50
Figure 8.29
1.65
1.80
1.95
2.10
2.25
2.40
2.55
2.70
2.85
3.00
Real and reactive power across interface
The rectifier DC currents, displayed for the three solutions in Figure 8.27,
show a very similar variation for the TSE and EMTDC solutions, except for
the region between t = 2.03 s and t = 2.14 s, but the difference with the
TS-only solution is very large. The corresponding DC-voltage responses are
shown in Figure 8.28.
Figure 8.29 compares the fundamental positive-sequence real and reactive powers across the converter interface for the TS and TSE solutions.
The main differences in real power occur during the link-power ramp.
The difference is almost a direct relation to the DC-current difference
between TS and TSE shown in Figure 8.26. The oscillation in DC voltage
and current as the rectifier terminal is deblocked is also evident.
As for the reactive power Q, prior to the fault, a small amount is flowing
into the system owing to a surplus MVAr at the converter terminal. The fault
reduces this power flow to zero. When the fault is removed and the AC
voltage overshoots in TSE, the reactive MVAr also overshoots in TSE and,

- Roxburgh
361
Avicmore
Benmore
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
Figure 8.30 Machine variablesTSE (TS variables)
since the DC link is shut down, a considerable amount of reactive power

flows into the system.
Finally, Figure 8.30 shows the machine-angle swings with respect to the
Clyde generator (see test system of Figure 7.18). These indicate that the
system is transiently stable.
8.9 Summary
The transient-stability (TS) algorithm described in Chapter 7 has been
modified to include a detailed simulation of the HVDC converter behaviour;
the latter is obtained with the EMTDC program. Much of Chapter 8
describes the interface of the TS and EMTDC programs, which has been
made sufficiently general for use with alternative versions of the TS and
EMTP solutions.
To demonstrate the advantages of the hybrid solution, two different
disturbances have been applied to a simple test system and the results
compared with those of the conventional electromechanical-stability and
electromagnetic-transient approaches.
First, a minor disturbance has been used to validate the hybrid technique
by showing that the response is practically identical to that of the electromechanical-stability program TS.
The major disturbance, a three-phase fault at the rectifier-converter
terminal, exemplifies the differences between the two types of solution. The
electromechanical solution shows the slower dynamic response of the AC
system, and the electromagnetic solution displays the fast dynamics of the
rapidly switched converter. In the hybrid method these responses are
362
combined, thus giving a more realistic overall picture of the entire system
performance. Electromechanical modelling of the AC system has been
improved by inclusion of the actual response of the fast-acting HVDC link,
and the link behaviour has been made more realistic through the inclusion
of the generator-controller response varying the AC-system voltage at the
converter terminal. Neither program on its own can provide a realistic
response of both AC and DC systems.
8.10 References
1 KULICKE, B.: 'Netomac digital program for simulating electromechanical and
electromagnetic transient phenomena in ac systems'. Siemens Aktienngesellschaft, E15/1722-101
2 WOODFORD, D.A.: Validation of digital simulation of dc links', IEEE Trans.,
September 1985, PAS-104, (9), pp. 2588-95
3 ANDERSON, G.W.J., ARNOLD, C.P., WATSON, N.R., and ARRILLAGA, J.: 'A
new hybrid ac-dc transient stability program'. Int. conf. on Power systems transients
(IPST), September 1995, pp. 535-540
4 ANDERSON, G.W.J.: 'Hybrid simulation of AC-DC power systems'. PhD thesis,
5 EGUILUZ, L.I., and ARRILLAGA, J.: 'Comparison of power definitions in the
presence of waveform distortion', Int.J. ofElectr. Eng. Educ, April 1995, 32, (2),
pp. 141-153
6 FRYZE, S.: 'Wirk-, blind- und scheinleistung in elektrischen stromkreisen mit
nichtsinusformigem verlauf von strom und spannung'. Elektrotech. Z., June 1932,
pp. 596-599
7 HEFFERNAN, M.D., TURNER, K.S., ARRILLAGA, J., and ARNOLD, C.P.:
'Computation of AC-DC system disturbancesparts I, II and III', IEEE Trans.,
November 1981, PAS-100, (11), pp. 4341-63
8 CLARKE, E.: 'Circuit analysis of ac power systemsvol. II' (John Wiley & Sons
Inc., New York, 1950)
9 KIMBARK, E.W.: 'Power system stabilityvol. Ill, synchronous machines'
(Dover Publications Inc., 1968)
10 REEVE, J., and ADAPA, R.: 'A new approach to dynamic analysis of AC networks
incorporating detailed modelling of DC systemsparts I and II', IEEE Trans.,
October 1988, PD-3, (4), pp. 2005-19
11 WATSON, N.R.: 'Frequency-dependent ac system equivalents for harmonic
studies and transient converter simulation'. PhD thesis, University of Canterbury,
New Zealand, 1987
Appendix I
Newton -Raphson method
I.I Basic algorithm

The generalised Newton-Raphson method is an iterative algorithm for
solving a set of simultaneous equations in an equal number of unknowns
/*(xw) = 0
for = 1 ^ N
m= l - N
'
At each iteration of the N - R method, the nonlinear problem is approximated by the linear-matrix equation. The linearising approximation can
best be visualised in the case of a single-variable problem.
In Figure I.I xp is an approximation to the solution, with error Axp at
iteration p. Then
f(xp + Axp) = 0
(1.2)
This equation can be expanded by Taylor's theorem

f{xp + Axp) = 0
(Axp)2
=/(*') + Axpf(xp) 4- ^-~r(xp)
+ .
(1.3)
If the initial estimate of the variable xp is near the solution value, Axp will
be relatively small and all terms of higher powers can be neglected. Hence
f(xp) + Axpf'(xp) = 0
(1.4)
or
a.5>
The new value of the variable is then obtained from
xp+1 =xp + Axp
(1.6)
364
^solution
Figure LI
Single-variable linear approximation
Eqn. 1.4 may be rewritten as
f(xp) = -JAx*
(1.7)
The method is readily extended to the set of N equations in N unknowns.

J becomes the square Jacobian matrix of first-order partial differentials of
the functions fk(xm). Elements of |J] are defined by
j km
3
dx
(1.8)
and represents the slopes of the tangent hyperplanes which approximate

the functions fk(xm) at each iteration point.
The Newton-Raphson algorithm will converge quadratically if the functions have continuous first derivatives in the neighbourhood of the solution,
the Jacobian matrix is nonsingular and the initial approximations of x are
close to the actual solutions. However, the method is sensitive to the
behaviours of the functions/k(%OT) and, hence, to their formulation; the more
linear they are, the more rapidly and reliably Newton's method converges.
Nonsmoothness, i.e. humps, in any one of the functions in the region of
interest, can cause convergence delays, total failure or misdirection to a
nonuseful solution.
1.2 Techniques to make the Newton-Raphson solution

more efficient
The efficient solution of eqn. 3.3.7 at each iteration is crucial to the success
of the N - R method. If conventional matrix techniques were to be used, the
storage (an2) and computing time (cm3) would be prohibitive for large
systems.
Newton-Raphson method
365
For most power-system networks the admittance matrix is relatively

sparse and, in the Newton-Raphson method of power flow, the Jacobian
matrix has this same sparsity.
The techniques which have been used to make the Newton-Raphson
competitive with other power-flow methods involve the solution of the
Jacobian-matrix equation and the preservation of the sparsity of the matrix
by ordered triangular factorisation.
Sparsity programming
In conventional matrix programming, double-subscript arrays are used for
the location of elements. With sparsity programming,1 only the nonzero
elements are stored, in one or more single vectors, plus integer vectors for
identification.
For the admittance matrix of order n the conventional storage requirements are n2 words, but by sparsity programming 66 + 3n words are
required, where b is the number of branches in the system. Typically,
b = 1.5n and the total storage is 12n words. For a large system (say, 500
buses) the ratio of storage requirements of conventional and sparse techniques is about 40:1.
To solve the Jacobian-matrix eqn. 3.3.7, represented here as
[AS] = U][AE]
(1.9)
for increments in voltage, the direct method is to find the inverse of (J] and
solve for [AE], i.e.
[AE] = (J]"1[AS]
(1.10)
In power systems [J] is usually sparse but [J]1 is a full matrix.

The method of triangular factorisation solves for the vector [AE] by
reducing [J] to an upper triangular matrix with a leading diagonal, and then
backsubstituting for [AE], i.e. reduce to
[AS'] = [U][AE]
and backsubstitute
The triangulation of the Jacobian is best done by rows. Those rows below
the one being operated on need not be entered until required. This means
366
that the maximum storage is that of the resultant upper triangle and
diagonal. The lower triangle can then be used to record operations.
The number of multiplications and additions to triangulate a full matrix
is |N 3 , compared to N 3 to find the inverse. With sparsity programming the
number of operations varies as a factor of N. If rows are normalised, N
further operations are saved.
Optimal ordering
In power-system power flow, the Jacobian matrix is usually diagonally
dominant which implies small round-off errors in computation. When a
sparse matrix is triangulated, nonzero terms are added in the upper
triangle. The number added is affected by the order of the row eliminations,
and total computation time increases with more terms.
The pivot element is selected to minimise the accumulation of nonzero
terms, and hence conserve sparsity, rather than minimising round-off error.
The diagonals are used as pivots.
Optimal ordering of row eliminations to conserve sparsity is a practical
impossibility due to the complexity of programming and time involved.
Instead, two types of semi-optimal scheme are commonly used, i.e. preordering2 and dynamic ordering.3
1.3 References
1 OGBUOBIRI, E.C., TINNEY, W.F., and WALKER, J.W.: 'Sparsity-directed decomposition for Gaussian elimination on matrices', IEEE Trans., 1970, PAS-89,
(1), pp. 141-50
2 STOTT, B., and HOBSON, E.: 'Solution of large power-system networks by
ordered elimination: a comparison of ordering schemes', Proc. IEE, 1971, 118, (1),
pp. 125-34
3 TINNEY, W.F., and WALKER, J.W.: 'Direct solutions of sparse network equations
by optimally ordered triangular factorization', Proc. IEEE, 1967, 55, (11), pp.
1801-1809
INDEX
Index Terms
Links
A
Abnormal modes
AC load flow
291
61
control equations
77
converter variables
72
mismatch equations
per unit system
73
80
75
146152
74
sequential method
162
unbalanced
146
154
170
73
78
293
21
62
152
287
293
unified (simultaneous) solution

instability
104
Additional delay
290
Admittance matrices
Analytic Jacobian
121
Asymmetry
170
Automatic voltage regulator
280
312
27
34
B
Back-to-back
269
Backward Euler method
206
Bergeron
224
This page has been reformatted by Knovel to provide easier navigation.
40
Index Terms
Links
C
Cable
218
228
241
23
34
135
170
Chatter
204
220
CIGRE
223
CCC, see Constant current control

CESCR
Characteristic harmonics
benchmark model
372
109
167
202
248
263
312
336
340
348
13
47
75
89
309
12
22
29
41
130
134
275
303
309
15
46
102
19
90
47
310
15
117
130
162
222
288
294
307
371
Commutating voltage
Commutation
angle
290
failures
margin (see also Extinction angle)
307
reactance
11
310
voltage, (see also Commutating

voltage)
Connection matrices
Constant current control
13
233
Index Terms
Links
Constant extinction angle control
78
191
222
Constant power control
15
46
223
17
63
68
86
117
121
128
133
145
154
167
180
293
301
297
Convergence
Convergence factor
139
Convergence tolerance
74
133
180
Converter control
78
189
215
252
258
223
289
23
38
45
77
162
222
288
307
23
46
77
96
162
297
12
46
119
characteristics
constant current
constant power
firing instant
124
power modulation
298
Converter transformer
13
17
24
36
72
94
230
connection
leakage reactance
magnetising characteristic
on-load tap changer
17
155
134
230
236
233
252
77
134
235
252
139
139
saturation
36
Index Terms
Links
Converter variables
18
71
75
Convolution
51
112
227
Core saturation instability
36
252
Cross-modulation
35
Current mismatch
116
166
Curve fitting
228
241
335
387
39
106
252
298
331
D
Damping
206
DC faults
264
DC ripple
10
123
69
84
163
Decoupled power flow
80
91
162
Delay angle
15
19
46
73
110
120
130
288
Direct axis
124
278
Dynamic stability
275
dq axes
278
Decoupled load flow (see also

Decoupled power flow)
E
Eigenvalue
Electromagnetic Transients Program
124
224
385
310
324
Chapter 6
Chapter 8
Chapter 6
EMTDC Program 4
Chapter 5
Index Terms
Links
EMTP, see Electromagnetic Transients

Program
Equidistant firing control (see also
Symmetrical firing)
Equivalent PI model
286
ESCR
367
Euler
51
Exciter
Extinction angle
206
385
265
47
78
125
177
258
324
61
79
68
88
F
FACTS
201
Fast decoupled AC load flow (see also

Fast decoupled AC power flow)
Fast decoupled AC power flow
145
162
Fast Fourier transform
339
341
36
87
94
32
41
46
77
87
110
120
190
288
12
46
120
124
190
FFT, see Fast Fourier transform

Filters
100
Firing angle
Firing delay, see Firing angle

Firing instant (see also Firing angle)
Fixed point iteration techniques
Fourier analysis
fundamental current
168
13
292
15
35
341
Index Terms
Links
Fourier series
23
43
135
Fourier transform
41
177
335
276
278
283
328
333
31
177
225
324
348
355
27
109
275
61
86
182
207
259
Frames of reference
Frequency dependent
Frequency domain simulation
G
Gaussian elimination
199
Gibbs phenomena
136
GTO
203
H
Harmonic
currents
3136
domain
41
flow
impedances
135140
145
37
40
254
355
interaction
162
169
phasors
41
45
54
sources
136
transfers
30
History term
195
206
217
220
228
Hydro turbine
282
Index Terms
Links
I
Inertia constant
277
Instability
36
189
252
Integration
176
207
301
181
184
191
292
295
304
345
180
207
234
303
185
203
11
16
78
A-stable
385
backward Euler method
206
implicit
301
numerical
385
Runge-Kutta methods
383
-stable
303
step length
trapezoidal method
Interpolation
Inversion
194
98
deionisation angle, see extinction
angle
extinction angle
Iterative methods
47
78
41
47
62
84
89
109
117
153
180
229
294
296
301
344
Index Terms
Links
J
Jacobian
303
numerical
117
analytical
121
Jacobian matrix
119
6568
79
82
91
117
153
166
K
Kron reduction
165
L
Lead-lag circuit model
282
Linearising shunt reactor
169
Load characteristics
285
293
39
252
M
Magnetising reactance
MAP
104
MATLAB
210
Matrix sparsity
287
Mismatch equations
4347
219
241
78
112
150
Mismatch functions
65
113
Modulation theory
26
252
84
Multiterminal
307
301
Index Terms
Links
N
NETOMAC
186
216
61
66
79
91
145
229
293
295
363
43
46
65
67
114
116
145
153
New Zealand link
34
93
Nodal formulation
196
198
Newton-Raphson
equations
67
flow diagram
66
Jacobian elements
79
Newtons method
312
Noncharacteristic harmonics (see also

Uncharacteristic harmonics)
Norton equivalent
631
194
212
233
238
282
285
307
351
Numerical Jacobian
117
Numerical stability
133
210
241
O
Ordering of sparse matrices
210
287
P
Parks transformation
278
Per unit for DC system
74
217
Index Terms
Phase locked oscillator
Links
44
189
222
PLO, see Phase locked oscillator

Power
angle
277
factor
15
292
331
flow
115
326
modulation
298
Q
QESCR
369
Quadrature axis
124
284
15
61
6671
85
95
101
298
308
335
16
78
R
Reactive power
360
Rectification
11
characteristics
77
controlled
16
delay angle
15
19
46
36
119
136
253
335
enforced delay, see Additional delay

Resonance
Runge-Kutta methods
RTDS
385
3
176
268
213
Index Terms
Links
S
Saliency
90
283
Sampling functions
41
51
Saturation stability factor
39
252
38
124
150
338
36
102106
259
186
221
35
94
Salient pole rotor, see Saliency

112
SCR, see Short circuit ratio

Sequence components
Short circuit ratio
146
367
Six pulse
9
259
Smoothing reactor
34
116
Sparsity
ordering
programming
Speed governor
287
368
61
365
280
312
133
210
Stability,
numerical
241
representation of plant in
network
301
State space analysis
381
Static VAR compensation
263
265
Step length
181
184
199
217
292
296
323
343
Stub line
Subharmonic
199
34
36
Index Terms
Subtransient reactance
Links
90
Swing curves
278
Switching system
130
Switching terms
119
214
133
Symmetrical components, see Sequence

components
Symmetrical firing
Synchronous machine
22
276
angular momentum
277
controllers
280
damper windings
337
inertia constant
277
models
214
negative sequence impedance
338
negative sequence power
338
Parks transformation
278
saliency
saturation
subtransient reactance
90
337
265
283
280
90
transient reactance
279
zero sequence power
338
Synchronous reactance
279
System damping
106
214
253
T
TACS (transient analysis of control
systems)
215
TCR (thyristor controlled reactor)
259
TCS (transient converter simulation)
177
298
Index Terms
Links
Three phase
faults
251
316
338
356
power flow
145
representation
141
Time domain simulation

Time step (see also Step length)
Transformer core saturation
Transformer model
152
Chapter 6
275
292
323
37
252
230
269
Transient converter simulation, see

TCS
Transient
model
278
reactance
279
stability
Transient stability program structure
216
275
301
324
361
302
327
334
352
Transmission line model
196
223
241
Transmission lines
196
211
286
224
286
Trapezoidal integration
301
302
Travelling wave
226
365
equivalent PI
TS, see Transient stability

Turbine
hydro
282
thermal
282
Twelve (12) pulse
2736
189
259
110
Index Terms
Links
U
UMEC (unified magnetic equivalent circuit)
Unbalance
230
3
17
19
34
43
111
124
157
160
332
338
203
extinction
184
191
switching
184
Valve group
189
220
Voltage mismatch
114
Uncharacteristic harmonics (see also

noncharacteristic harmonics)
V
Valve

Ac DC Power System Analysis Arrillaga, - Jos

Uploaded by

Copyright:

Available Formats

Ac DC Power System Analysis Arrillaga, - Jos

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ac DC Power System Analysis Arrillaga, - Jos

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

IEE Power and Energy Series

The Institution of Electrical Engineers

Published by: The Institution of Electrical Engineers, London,

British Library Cataloguing in Publication Data

Printed in England by Short Run Press Ltd., Exeter

has practically eliminated the need for physical simulators. Consequently,

The power flow solution

With respect to AC phase-voltage variations

Three-phase power and harmonic flow

Electromagnetic transient simulation

The PSCAD/EMTDC program

Electromechanical stability with transient converter simulation

The short-circuit ratio (SCR)

1.1 Basic AC-DC configuration

Three-phase bridge configuration

AC-DC power system analysis

12-pulse converter configuration

Figure 1.3 Basic HVDC interconnection

1.2 AC-DC simulation philosophy

Being under continuous electronic control, DC links are free from

1.3 Steady-state simulation

AC-DC power system analysis

tion between the fundamental and other frequencies. This interaction is

1.4 Fault analysis

1.5 Harmonic analysis

rithm which also involves the AC and DC systems. Important contributions

1.6 System stability

The AC-DC converter in steady state

AC-DC power system analysis

Moreover, in response to an applied harmonic distortion, sideband

2.2 Power frequencysymmetrical operation

The forward-voltage drop in a conducting valve is neglected so that

The AC-DC converter in steady state

(ii) The converter-transformer leakage reactances as viewed from the

Six-pulse converter waveforms

AC-DC power system analysis

Typical six-pulse rectifier operation

a Positive and negative direct voltages with respect to the transformer

2.2.1 Analysis of the commutation circuit1

The AC-DC converter in steady state

AC-DC power system analysis

Taking as a reference the voltage crossing between phases R and Y

where Vterm is the phase-to-phase r.m.s. voltage, referred to the secondary

and integrating from cot = a

The instantaneous expression for the commutating current is thus

and substituting the final condition, i.e. ic = I d at cot = a + fi yields

[cos a - cos(a + fi)]

2.2.2 Rectifier operation

The AC-DC converter in steady state

The r.m.s. magnitude of a rectangular current waveform (neglecting the

for a + / i < cot < + a

The fundamental component of the current waveform defined by eqns.

AC-DC power system analysis

2.2.3 Inverter operation

Six-pulse inverter operation