1196 IEEE Transactions on Power Delivery, Volume 3 , No.

3 , July 1988

FREQUENCY-DEPENDENT A.C. SYSTEM EQUIVALENTS FOR HARMONIC STUDIES AND TRANSIENT CONVERTOR SIMULATION

N.R. Watson J. Arrillaga, Sen.Mem.

University of Canterbury (New Zealand)

Abstract - A generalization of the impedance SYSTEM FREQUENCY RESPONSE AT A CONVERTOR TERMINAL

loci is first described capable of displaying the
complete frequency response of any given power system The frequency response of a power system at the
including multi-phase related effects such as phase point of harmonic injection can be derived either from
asymmetries and mutual couplings. This information is measurements or from frequency-domain studies, and the
used to derive frequency dependent equivalent circuits results are normally presented in the form of a single
suitable for integration in the time-domain solutions impedance-locus plot. However the exact frequency
of power system waveforms. Direct and optimization response must include network asymmetries and mutual
techniques are used in the derivation of the couplings, thus requiring a 3 x 3 matrix impedance for
equivalent circuits. Their effect on impedance each harmonic. A generalization of the impedance loci
frequency matching and computational efficiency are is used in this paper to display the enormous amount
compared. of information that results. A s an illustration, the
self and mutual impedance loci of the system of Figure
INTRODUCTION 1 (the lower part of the New Zealand network) observed
from the TIWAI BUS are shown in Figure 2 .
Harmonic studies are beginning to play an
important part in power system analysis. The harmonic
currents produced by non-linear components are 4 - I -L
normally derived on the assumption of a strong, i.e.
perfectly sinusoidal, voltage supply. These nominal
harmonic currents are then injected into the a.c.
network (assumed linear) to determine the levels of
voltage distortion.
However, when the non-linear load is a large
power convertor and the a.c. system harmonic TlWAl
impedances are large, the supply voltage waveforms are f
not perfectly sinusoidal and the derivation of the
injected
algorithm.
frequency
harmonic currents requires an iterative
Moreover, if
is
the injected harmonic
close to a parallel resonant the Fig. 1. Test System
P135HW 36 M V A i

algorithm often diverges. In such cases there is no

alternative to the use of transient convertor Because power systems are generally bilateral,
simulation (TCS) for the derivation of harmonic the (n,m) loci are seen to be identical to loci (m,n).
levels.[ll However the acceptability of TCS harmonic Moreover for the particular line symmetry of the test
studies depends on the existence of a.c. system system, two phases, A and C, are seen to be similar
equivalents responding accurately to power and while phase B differs substantially.
harmonic frequencies. Some dynamic models lack mutual coupling
These frequency-dependent equivalents are also capability and in such cases it is possible to reduce
required to analyse the behaviour of a.c./d.c. the full matrix information to three diagonal terms
interconnections following convertor or system which contain the mutual impedances implicitly. The
disturbances. The scaled-down simulators and TCS method is valid for systems with little phase current
programs currently used to investigate the dynamic asymmetry, as is normally the case with convertor
behaviour of h.v.d.c. convertors are by necessity very plant. This is achieved by post multiplying each of
limited in their frequency dependence representation. the 3 x 3 matrices by the matrix
An equivalent circuit consisting of tuned RLC
branches has been advocated by N.G. Hingorani [ 2 1 as a

ml
possible solution for HVDC studies and a three-phase
extension of the method has been proposed by the
authors at a recent conference.[3] The accuracy of
these equivalent circuits is restricted by the lack of
where a = -
+ j 5
mutual coupling representation and their 'matching'
capability is limited to one quantity, the harmonic
impedance, and to a few selected harmonic frequencies.
This paper presents a generalization of the and extracting the diagonal terms. This is important
above techniques, capable of matching the impedances because for some harmonics the mutual terms have as
of mutually coupled and asymmetrical systems for any much effect as the self terms. The resulting loci for
required continuous range of frequencies. the system of Figure 1 are displayed in Figure 3.
The frequency response is probably better
assessed with reference to impedance/frequency plots
87 WM 171-2 A paper recommended and approved rather than impedance loci. These are illustrated in
by the IEEE Transmission and Distribution Committee Figure 4 for the system of Figure 1.
of the IEEE Power Engineering Society for presenta-
tion at the IEEE/PES 1987 Winter Meeting, New Orleans, Implementation in Time Domain Studies
Louisiana, February 1 - 6, 1987. Manuscript submit-
ted August 28, 1986; made available for printing Generally time domain solutions will be required
November 17, 1986. to derive the voltage waveforms resulting from the
interaction between a static convertor and the supply
system. Thus the complex impedance-frequency
information contained in the loci matrices (such as

0885-8977/88/0700-1196$01.0001988 IEEE
 1197

X(RI X(R 1
1
t
350

zac

150

-50 -260
-5
I
5 Q

-250 -250 1

3=0 t ha
t bb
3=0 t hc

150

-60

-250 -250

350 t zch
350
zcc

150

-5 -50

-250 1 -250 .250

Fig. 2. Impedance Loci Matrix of Test System

375 375 37s

250 i za 250
zb
250

125 125

-125 -129

-250 -250

Fig. 3 . Diagonalized Matrix Impedance Loci

1198

(R ) (R)
600 - 600

500 500
zab zac
400

300

200

io0

0
50 450 m50 1250 50 450 a50 1250

5!LJ
moo

200

L O O 50 450 a50 1230

50 450 1250 50 450 a50 12ao Hz

600

500 SO0

400 400

300 300

200 200

L 00 100

0 0
SO 450 650 la50 50 450 850 1250 50 450 650 1250 HZ

Fig. 4 . Impedance/Frequency Matrix Derived from Figure 2.

Self circuits - +

that of Figure 2 ) must be converted into equivalent

circuits. Every element of the 3 x 3 matrix must be
synthesised into an equivalent circuit.
The equivalents matching the self terms are
represented explicitly as shunt branches. The mutual
terms are processed separately by using diakoptical
techniques; the phase currents are first impressed
upon the mutual circuits and the resulting voltages
are then incorporated in the overall circuit as
additional voltage sources per phase. Thus the
overall frequency dependent model of the a.c. system
to be used in the time domain solution is as shown in
Figure 5. Mutual circuits
Extension to Multiconvertor Systems

Let us now consider the case of a two-convertor

system as shown in Figure 6. This system needs two
points of observation, i.e. busbars A and B. The
convertor connected to busbar B will affect the
impedances of the a.c. system when observed from
busbar A and vice versa. This interdependence Equivalent circuits
increases the number of mutual terms, i.e. each phase
has a self and five mutuals.
An extension of the diakoptical technique to
this case requires six self and thirty mutual circuits
for a complete frequency-dependent model.
Although extension of the theory to more than
two convertors is straight forward, the number of
frequency dependent circuits required increases with
the square of the number of convertors and the
complexity of each frequency dependent circuit is Fig. 5. Structure of the Frequency Dependent Equivalent
dependent on the number loops of the frequency loci. Model
 1199

Therefore it becomes impractical to model accurately a

large number of convertors connected to the a.c.
system.

A.C. Use Hingorani's topoloqy Decide eauivalent

System and direct synthesis to circuit topology and
derive a multi-phase optimize the circuit
equivalent parameters

-. circuit
Is the equivalent
acceptable-\optimize 1

Fig. 6. Two-convertor System

EQUIVALENT CIRCUIT SYNTHESIS

L Design a correction
branch
4 I
Add another branch to
the equivalent circuit

An initial equivalent circuit topology is

selected to suit the requirements of the time-domain
algorithms, such as the use of node-type partitioning Fig. 7. Flow Diagram for the Direct and Optimization
and tensor matrix analysis.[41 These restrictions Algorithms
increase dramatically the solution efficiency in the
case of a.c.-d.c. simulation, which require regular peaks and troughs, to obtain the best fit between the
topological changes, by avoiding involving the whole equivalent and real system responses.
network during localised convertor valve switchings. The selection of small scaling factors results
Since the topology used by Hingorani,[2] i.e. in larger impedance levels, which are easier to
RLC shunt branches, is amenable to Efficient Transient compensate. The compensation is achieved by means of
Convertor Simulation [4] this form has been retained correcting RLC (filter) branches which drop the
as the starting point in the derivation of the impedance levels in the regions of large discrepancy.
equivalent circuit. From inspection the region of largest deviation from
The basic matching philosophy consists of the required response is found and three matching
selecting the values of R, L and C that give the peaks points are selected in this region. The formulation
and troughs at the correct frequencies and the Q of Appendix B is then used to fit the correcting RLC
(quality factor) of each branch such as to cause the branch to these three points and thus obtain the
equivalent circuit response between these frequencies values of RLC required. This correction filter is
to approximate the actual system response. The added to the equivalent circuit to form a new
analytical part of the algorithm is described in equivalent and the process repeated if necessary to
Appendix A. find a new region of discrepancy and another
An acceptable error between the actual system correction filter.
and the synthesised circuit depends mainly on the This process continues until the required
following factors: accuracy is obtained, though the latter may be limited
(i) The type of study. For steady state harmonic by the increasing computing requirements. The CPU
penetration studies minimal error at the harmonic time used in a.c./d.c. dynamic simulation is related
frequencies is required while the error at to the size of time steps of the numerical solution,
intermediate frequencies are unimportant. This is which itself depends on the time constants of the
clearly unacceptable for transient convertor equivalent circuit. In each case the acceptability of
simulations as some intermediate frequencies could be a particular equivalent circuit and the required CPU
excited. time, are assessed by means of a short-duration
(ii) The feasibility of altering automatically the dynamic simulation study.
equivalent circuits to get the error down to a
prescribed accuracy. Optimization Method
(iii) The availability of an economically viable
alternative to model the actual system explicitly. Optimization methods [5][6] use either the
This relates to the amount of CPU time required to Search or Gradient approaches. The latter being more
provide the simulation and accuracy required. difficult to implement due to the need for derivative
The basic steps of the equivalent circuit information.
synthesis are illustrated in the flow diagram of A gradient optimization method based on a
Figure 7. This diagram includes two alternative special implementation of the
paths, i.e. the direct and optimization methods, Broyden-Fletcher-Goldfarb-Shanno (BSGF) technique has
which are described in the next two sections. been used to minimize the difference(sca1ar quantity
' 8 ' ) between the actual system and equivalent circuit
Direct Method responses. Its form is:

The mathematical formulation used to make an e = -

h{g(x,Al), g(x,A2) ---- 9(xrA,))
initial assessment of the RLC values to give the peaks
and troughs at the correct frequencies is described in where g is termed the objective function
Appendix A . Scaling of circuit parameters is then A1---Am are the values of the independent
carried out, without altering the position of the variables at the m sample points (which are complex
1200
functions of frequency).
and h is the Error Criterion Function
Among the various types of Error Criterion, the 688
=
Least Squares method has been found to be the most
suitable in our problem. Its general form is: 588

minimize e = sum{ (wi.g(x)

- )2} 408

where wl,---wm are weighting factors to emphasize 388 (a)

qi(x) in regions of importance.
280
The optimization method is more flexible in the
selection of equivalent circuit topology: the only 188
limitation in this respect is the possibility of
evaluating its frequency response. However, the 8 f
circuit topology must be decided before the
optimization process.
The CPU time requirements of the multi-variable
optimization techniques is several orders of magnitude
greater than those of the direct approach but they can
be substantially reduced if the circuit derived by the
direct method is used as a first approximation in the
optimization approach.

EFFECTIVENESS OF THE ALGORITHM

Figure 8(a) displays the frequency response of

one of the mutual elements of the 'actual' test system
of Figure 1, together with that of Hingorani's 121
harmonic equivalent. Although the results are in good
agreement at low frequencies the second and third
peaks (at 845 and 876 Hz) show substantial disparity.
The use of a correction filter in the region of
largest discrepancy ( 8 6 2 to 906 Hz) modifies the
688
harmonic response as shown in Figure 8(b). The new 1
branch reduces considerably the error in the regions
508
where the impedances of the equivalent circuit were
too high, but no marked improvement is noticeable
488
where the equivalent circuit impedances are too low.
As explained earlier these are best corrected by a 388 (C)
combination of scaling factor and subsequent use of
correction filters. 200
The region of greatest discrepancy of the new
equivalent circuit is now between 750-810 HZ. The 188
addition of a second correction filter in this region
produces a far better match as shown in Figure 8(c). 8 f
With reference to multi-variable optimization 58 258 458 658 OS0 1858 1258
techniques, the effectiveness of the synthesised
equivalent circuit is very dependent on the choice of
error criteria. If the optimization criterion is the
minimization of the maximum deviation, the response, Fig. 8. Direct Synthesis of a Mutual Element of the
illustrated in Figure 9 ( a ) , is unsatisfactory, i.e. a Impedance/Frequency Matrix
large discrepancy is introduced in the first peak. In Actual system response (continuous line)
this case it is necessary to use more elaborate Initial equivalent (dotted line (a))
weighting functions which are a function of the With one additional correction filter
discrepancy as well as the frequency. On the other (dotted line (b))
hand the use of the least squares as the error With two additional correction filters
criterion, illustrated in Figure 9(b), shows a marked (dotted line (c))
improvement. Although the sum of the errors squared
is minimized in this case, there are larger In order to illustrate the effectiveness of the
discrepancies at some individual frequencies when frequency-matched equivalents described above, the
compared with the direct solution. test system of Figure 1 was modified to include an
Thus an interactive approach based on the direct h.v.d.c. convertor at the TIWAI Bus. Figure 10 shows
method enhanced by the use of correction filters the d.c. voltage waveform, derived from Transient
generally has less error over most of the frequency Convertor Simulation, when a short-circuit occurs on
range as compared to an optimization derived the d.c. line. The continuous line shows the response
equivalent. However, each correction filter increases with a frequency-matched a.c. system equivalent and
the complexity of the equivalent circuit and hence the the dotted line that of a Thevenin equivalent derived
computational burden when carrying out time domain from the short-circuit capacity of the a.c. system.
studies. On the other hand the CPU time required for There is a significant difference between the two
a multi-variable optimization study is several orders responses which shows the importance of using accurate
of magnitude greater than the direct approach. Hence frequency-dependent equivalents for the a.c. system.
if a relatively small number of time domain The time domain simulation algorithm includes
simulations are required the direct approach is accurate modelling of the non-linear characteristics
preferable while for many, the optimization technique of the convertor plant, such as convertor transformer
is probably best as a comparable accuracy equivalent saturation, surge arresters, etc.
can be obtained with fewer circuit elements.
 1201

individual a.c. system busbars. It does, however,

800 z
provide accurate assessment of the voltage and current
500
waveforms at the convertor terminals, which is
essential in Transient Convertor Simulation. This
400
information can also be used in harmonic penetration
studies and filter design for the accurate derivation
300 (a) of harmonic levels throughout the system. For this
purpose the steady state harmonic currents finally
200
obtained from the Time Domain Simulation must be
injected into the actual network (represented in the
100
frequency domain) to derive the voltage distortion
levels at remote buses. We have already tried this
0 f technique with promising preliminary results.
50 250 450 850 BS0 1090 1250

800 2
REFERENCES

500
1 Kitchin, R.H. "New method for digital computer
evaluation of convertor harmonics in power
400 systems using state variable analysis", Proc.
IEE, Pt. C, Vol. 128, No. 4, pp. 196-207.
300 (b) Hingorani, N.G. L Burbery, M.F. "Simulation of
A.C. system impedance in HVDC system studies",
200 IEEE Trans. Vol. PAS-89, No. 5/6, May/June 1970,
pp. 820-826.
100 Watson, N.R., Arrillaga, J. L Joosten, A.P.B.
"A.C. system equivalents for the dynamic
0 f simulation of HVDC convertors", IEE Conference
50 250 450 056 D56 1050 1250 Publication 255 on AC and DC Power Transmission,
Sept. 1985, London, pp. 366-371.
Fig. 9. Optimized Mutual Element of the Impedance/ Arrillaga, J., Arnold, C.P. L Harker, B.J.
"requency Matrix "Computer Modelling of Electrical Power
Actual system response (continuous line) Systems", J. Wiley (1983).
Using maximum deviation criterion (dotted Adby, P.R. and Dempster, M.A.H. "Introduction
line (a)) to Optimization Methods", Chapman and Hall
Using least squares criterion (dotted line (1974).
(b)) Fletcher, R. Practical Methods of Optimization,
Vol. 1 Unconstrained Optimization, John Wiley L
Sons (1980). Vol. 2 Constrained Optimization,
John Wiley L Sons (1981).

APPENDICES

A. Derivation of Equivalent Circuit Parameters

The basic equivalent circuit contains a number

of single-tuned parallel branches as shown in Figure 5
and the first step is to obtain values for the
inductors and capacitors of the equivalent circuit
that will produce the same minima and maxima resonance
frequencies. The resistances of the branches are
ignored as they have very little influence on the
magnitude of the admittance at frequencies that are
sufficiently higher or lower than the series resonant
frequency of the branch.
The admittance of an n branch network is:

Fig. 10. Dynamic Simulation of the d.c. Voltage at the

Convertor Terminals following a Line Short-
Circuit
(i) Frequency-dependent equivalent Y(S
(ii) Thevenin equivalent
I
CONCLUSIONS 1=l
whe e :
Direct synthesis and multivariable optimization
techniques have been described for the derivation of U: =I=1,2...etc. and s = j w
three-phase frequency dependent a.c. system LIcI
equivalents. These equivalents model the frequency For an LC network the impedance is zero at minima and
dependence of the self and mutual impedance terms and infinite at maxima frequencies.
are amenable for implementation into time-domain Impedance minima occur when:
programs. Both methods have proved capable of
achieving high levels of matching accuracy, but the
optimization techniques require much greater computer (2)
requirements.
It should be noted that the equivalent circuit
does not give explicit harmonic information at the Impedance maxima will occur when the numerator
1202
of equation (1) is zero, i.e.

s ; 2-
k=l Lk J#k
fi (s2+w3) = 0 (3)
As Lk appears in the denominator of equation
(ll), to increase the admittance contribution of each
Let F1, F2, ...
Fn be the frequencies at which the branch, so as to reduce the impedance, the inductors
admittance is zero, then equation ( 3 ) can be written are divided by the scaling factor. In order to keep
as: the same resonance frequencies the capacitor values
n-1 are multiplied by the scaling factor.
S (S2+ME) where Mk = 2TFk (4)
k=l B. Derivation of Correction Circuit Parameters

By equating the coefficients of s in (3) and ( 4 ) (12)

the following equations result:

For s2n-1 (13)

(5)

For s2n-3

n n n-1
1 1 w2= 1 ME (6)
k=l Lk J=k k=l

For s2n-5

For s1
. .
(8)

Thus n equations are formed which are then Substituting equation (17) into equation (15)
solved for the n unknown inductances (L1,L2,...L,). gives;
The appropriate capacitor values are found using:

ut = e
k k
k=1,2. ..n (9)

where Wk are the required minima frequencies.

The admittance of the tuned network at any
particular frequency is given by:

Hence

Branch k is in series resonance when u; = 1/(ck

Lk). The corresponding admittance term for the kth
branch is 1/Rk. This term is significantly larger
than the contribution from the branches not in series
resonance and hence dominates the admittance. The
resistance that dominates at each minima is inserted
in the corresponding branch. This yields maxima point
impedances for the equivalent circuit which are much Given three points, equation (19) is solved to
higher than those of the actual system. The problem obtain the inductance of the required correction
can be corrected by the scaling of all inductor and branch. Equation (17) is solved next to obtain the
capacitor values. A common scaling factor is used for capacitance value. Lastly equation (12) is solved for
all inductors and capacitors' so that the resonance the required resistance, this completing the
frequencies remain unchanged. The appropriate scaling determining of the correction filter parameters.
factor is obtained by dividing the calculated
impedance of the network at a particular frequency by
the actual systems impedance. Equation (10) with R
ignored yields:
 1203

Discussion minimal error at the harmonic frequencies is required ... . But if one is
using the simulation to determine the harmonic frequencies to begin with
and the simulation has any errors built in, it appears one is in a dilemma or
D. J. Melvold (Department of Water and Power, Los Angeles, CA): The may have a false sense of security. As the paper stresses, it is important to
industry should be grateful for the authors continued development work on use accurate frequency-dependent equivalents for the ac system. The
analysis techniques. However, the difficulty usually encountered in
inability at present to secure accurate field data on real systems is the major
specification of the ac filtering for a new HVdc system is in the description stumbling block to be overcome.
of the ac system harmonic impedance and the risk the utility constantly must
assume due to inherent inaccuracy in current methods of deriving this Manuscript received February 23, 1987.
impedance. The industry appears to be making strides toward development
of more powerful tools for calculating harmonic impedance but with limited
use or questionable practical value. The reason I say this is because of the N. R. Watson and J. Arrillaga: The comments made by Mr. Melvold do
need to make field measurements, the least of which is not that in many not seem to require a reply. All we can do is agree with his dilemma and
cases the ac system to which the dc system is to be connected may not as yet perhaps add that our proposed model can represent to any specified level of
exist. accuracy any known linear system configuration.
The paper states that for steady state harmonic penetration studies Manuscript received May 6 , 1987

INFORMATION FOR AUTHORS

In January 1986 IEEE TRANSACTIONS ON POWER APPARATUS A N D SYSTEMS was replaced with three new TRANS-
ACTIONS which are sponsored by the Power Engineering Society. The new publications will continue to be devoted to all
aspects of electric power generation, transmission, distribution, and utilization. Original contributions of lasting value
to the profession are sought for publication in the appropriate TRANSACTIONS according to the subject matter as
follows.
IEEE TRANSACTIONS ON ENERGY CONVERSION:Research, development, design, application, construction, installa-
tion and operation of electric power generating facilities (along with their conventional, nuclear, or renewable sources)
for the safe, reliable and economic generation, conversion and control of electrical energy for general industrial, com-
mercial, public and domestic consumption.
IEEE TRANSACTIONS ON POWER DELIVERY: Research, development, design, application, construction, installation
and operation of apparatus, equipment, structures, materials and systems for the safe, reliable and economic delivery
and control of electrical energy for general industrial, commercial, public and domestic consumption.
IEEE TRANSACTIONS ON POWER SYSTEMS: Requirements, planning, analysis, reliability, operation and economics
electrical generating, transmission and distribution systems for general industrial, commercial, public and domestic
consumption.
In general, papers are preprinted and presented at a technical meeting of the Institute before being published in the
appropriate Transactions; discussions by qualified specialists are sought and published along with a closure by the
author. All contributions are reviewed by an appropriate technical committee of the IEEE Power Engineering Society
as listed on the inside back cover.
Following review, a decision is made as to publication. All papers accepted for publication will be published in full in
one of the TRANSACTIONS and a one page summary will be published in the IEEE POWERENGINEERING REVIEW.
COPYRIGHT: I t is the policy of the IEEE to own the copyright to the technical contributions it publishes on behalf
of the interests of the IEEE, its authors, and their employers, and to facilitate the appropriate reuse of this material by
others. To comply with the U S . Copyright Law. authors are required to sign an IEEE copyright form before publica-
tion. This form, returns to authors and their employers full rights to reuse their material for their own purposes. Authors
must submit a signed copy of this form with their manuscripts.
The original manuscript, typed on IEEE mats in accordance with the I E E E Power Engineering Society Publicntion
Guide f o r P o w r Engineers, November 1986, should be sent for review processing to the following address.
IEEE Headquarters
Society Special Services
345 East 47 Street
New York, NY 10017
I E E E Power Engineering Society Publication Guide f o r Power Engineers, November 1986, available on request
from the Society Special Services Department, at the above address, provides a guide to approved style and usage.
Detailed instructions, copyright forms, and standard mats for the preparation and handling of manuscripts can be
obtained from the IEEE PES Special Activities. Manuscripts should be limited to seven pages including illustrations.
Other editorial correspondence should be directed to the Editor.
Harold Gold
1037 North Primrose
Rialto, CA 92376