, , , ,

I ,

"

174

IEEE " a x i o n s on Power Systems, vo1.6, No.1, February 1991 A M U L T I P H A S E H A R M O N I C LOAD F L O W S O L U T I O N T E C H N I Q U E

Wenyuan Xu Student M.IEEE Jose R. Marti Member, IEEE Hermann W. Dommel Fellow, IEEE

Department of Electrical Engineering University of British Columbia Vancouver, B.C. Canada, V6T 1W5

ABSTRACT: The operation of nonlinear devices under unbalanced conditions may cause harmonic problems in power systems. A multiphase harmonic load flow solution technique for analyzing such problems is described in this paper. The harmonic load flows are obtained from iterations between the Norton equivalent circuits of the nonlinear elements and the linear network solutions at harmonic frequencies. Harmonics generated by static compensators with thyristor-controlled reactors under unbalanced conditions are used to illustrate the method. The inclusion of the control characteristics of the static compensator and comparisons with field test results are described in a companion paper. KEYWORDS: network unbalances, harmonics, harmonic load flow, multiphase solution, static compensator.

In this paper, a multiphase harmonic load flow (MHLF) technique is described which solves the network at fundamental and harmonic frequencies in ,the presence of nonlinear elements and unbalances. It is based on the harmonic iteration scheme which has been used earlier to compute harmonics from HVDC converters, transformer saturation and thyristor-controlled reactors [7, 8, 91. The harmonics caused by static compensators with thyristor-controlled reactors under balanced and unbalanced conditions are used to illustrate the method. This technique was primarily developed as an improved initialiaation procedure for EMTP simulations, but it is also a useful tool by itself for multiphase harmonic load flow analysis with unbalanced conditions. The multiphase harmonic load flow technique is simple in concept. Besides static compensators, other harmonic-producing nonlinearities with or without control specifications can be included. In order to keep the explanation of the method simple, only static compensator operation with known conduction angles is considered in this paper. The inclusion of the control characteristics of the static compensator and case study results as well as field test comparisons are presented in a companion paper [lo].

The voltage and current waveforms in power systems are frequently distorted by harmonics. This distortion can cause various problems, ranging from capacitor failure to communications interference [l]. The propagation of these harmonics through the network must be assessed so that harmonic counter-measures can be properly designed. Harmonic load flow programs are an important tool for such harmonic propagation studies. The work on harmonic load flow solution techniques was pioneered by Heydt et al [l].Their well-known HARMFLO program can be used to analyze harmonic load flows caused by rectifiers, HVDC converters, and other nonlinear devices [2,3]. Similar techniques are described 51. in [4, There are also many harmonic problems related to the unbalanced operation of power systems. With unbalanced conditions, the generation and propagation of harmonics are more complicated. For example, extra so-called non-characteristic harmonics may be produced. It is important to evaluate the effects of these harmonics since no filters are generally installed to alleviate them. In view of the fact that there is always some degree of unbalance and that the harmonics are more sensitive to unbalances than the fundamental frequency component, unbalanced harmonic analysis has received more attention recently. Unbalance effects have been induded in the three-phase frequency scan technique proposed by Densem, Bodger and Arrillaga [5] and in the computation of noncharacteristic harmonics from rectifiers [6].

2. PRINCIPLE OF HARMONIC ITERATION

To explain the principle of harmonic iteration, a static compensator with thyristor-controlled reactors (TCR) will be used as the source of harmonics. This device is essentially a reactor in series with anti-parallel thyristor valves, as shown in Figure 1. The valves conduct on alternate half-cycles of the supply frequency, for durations which depend on their firing angles a,thereby creating adjustable reactive power generation or consumption [ll]. The operating range goes from no conduction at a = 180" to full conduction at a = 90". The duration of conduction is defined by the conduction angle D = 2(180 - a). If Q is less than 180, harmonic currents are generated, aa shown in Figure 1.

2.1 Harmonic Norton Equivalent Circuit of Thyristor-Controlled Reactor

To compute the harmonic currents with the harmonic iteration scheme, an equivalent linear model of the TCR shall be derived first. In general, the voltages appearing across this element will be distorted with harmonics,
n

90 W ! 098-4 FWRS A paper recommended and approved by t h e IEEE Power System Engineering Committee of the IEEX Power Engineering Society f o r preaentation a t the IEEE/PES 1990 Winter Meeting, Atlanta, Georgia, February 4 - 8, 1990. Manuscript aubmitted August 31, 1989; made a v a i l a b l e f o r p r i n t i n g November 17, 1989.

i(t) =
h= 1

(Vh(COs(hWt

$h).

(1)

With the valve fired at tf, the current through the reactor during one half of a cycle is determined by

0885-8950/91/02004174$01.00 (D 1991 IEEE

175

the Norton equivalent circuit are easily found from Eq. ( 7 ) , after I[, has been obtained from the Fourier analysis of Eq. (3). This relationship is symbolically expressed as

4
Figure 1: Thyristor-controlled reactor and waveforms.
or
n

Note that there is no coupling among the equivalent circuits of the different harmonic frequencies. They are also independent of any network unbalances.
2.2

Harmonically-Decoupled Network Solutions

With the nonlinear TCR represented as a Norton equivalent circuit a t harmonic frequencies, the node voltages of the entire network can easily be found by solving a system of multiphase node equations
[Yh-net~oork][Vh-nehuork]

i ( t )=

CIv,,((hwL)-l[sin(hwt

+ (bh)

.+hwtf

+ dhj],
(3)

= [Ih-network],

h=1 ,*

n 7

(9)

t f 5 t Ite; h=l 0 < t < t f and t, < t < T / 2 ; 0,

where t f is the instant of firing, t, that of extinction, L the inductance of the TCR, and T the period at fundamental frequency. For the second half cycle, the current will reverse in sign (Figure 1). To obtain the current phasors as a function of the voltage phasors, Fourier analysis is required. Since the closed-form Fourier analysis of Eq. (3) is complicated due to discontinuous conduction, the distorted current is generated point-by-point from the given voltage with Eq. (3) and then analyzed with discrete Fourier analysis. This produces the harmonic content expressed by
n

at each frequency. The nonlinear TCR effects are represented in these equations as currents in the vector [Ih-nehuork]. Once the node voltages have been obtained, improved values for the equivalent current sources can then be calculated from Eq. ( 8 ) , which in turn is used to compute improved voltages. This is the process of harmonic iteration. It is continued until the changes in the equivalent current sources are sufficiently small. This iterative process constitutes the basic idea of the multiphase harmonic load flow technique. To include the load flow constraint options of HARMFLO and sinular programs, Eq. (9) at h = 1 must be modified into multiphase load flow equations, as described next.

iharmonce(t) =
h= I

(Ihlcos(hwt

oh),

(4)

which was used as a current source model in [9] to represent the harmonic effects of the TCR. It is better to model the TCR as a Norton equivalent circuit, however. It can be shown that the equivalent inductance of a TCR for a purely sinusoidal voltage at fundamental frequency is [Ill: Le, = nL(u - sina)-1
(5)

3. MULTIPHASE FUNDAMENTAL FREQUENCY LOAD FLOW SOLUTIQNS The multiphase load flow solution must be able to handle unbalanced conditions. Unbalanced load flow analysis was first introduced by El-Abiad and Tarsi two decades ago 1121. Since then, much progress has been made [13, 141. An excellent summary of the state-of-the-art of three-phase load flow analysis can be found in [13]. 3.1 Modelling of Network Components In contrast to most existing techniques, the load flow constraints for each power system component are expressed as branch equations here, instead of constraints on node quantities. ' Since branches can be connected in any way by the user, this provides greater flexibility. 3.1.1 Three-phase Synchronous Machines The response of a synchronous machine is different for positive, negative, or zero sequence current injections. This must be taken into account in unbalanced load flow studies. Reference [12] has developed such a model (Figure 3(a)), with the branch equations (10) = [Ygl([Vkl - [Vml - [El) where
[V,] = [Vk-a V k - b Vk-=IT, voltages on side k, [V,] = [V,,-, Vm-b Vm_,.IT, voltages on side m, [ l k m ] = [Ikm-@ 1km-b Ikm-eIT, currents from side k to side ni, [E] = [Ep .'Ep uE,IT, internal voltages,
[Yglmutual

This equivalent inductance represents the TCR very well at fundamental frequency, and reasonably well at other frequencies. The differences between the current absorbed in jhwL,, and the actual current from Eq. (4) become the parallel current sources in the Norton equivalent circuit representation of Figure 2, with

Y L e q = (jhwLJ'
1h-q

= (jhwLeq)-'Vh - I h

(6) (7)

where

v h = IVh(e36hand I h = IIhleJeh.

For given voltages at the TCR teriuinal, current sources of

Figure 2: TCR equivalent model for multiphase harmonic load flow analysis.

= (yo - w 3 ,

,
176

'I,

Subscripts p, n, and o indicate positive, negative, and zero sequence components, respectively. For the negative sequence reactance, jw(L&' L : ) / 2 can be used. The zero sequence reactance is jwL,. For the resistance, the armature resistance can be used, though this is not quite correct because the negative sequence resistance can Le an order of magnitude larger.

G([Ikm], [ v k ] ,[ v m ] )= Fspcerfied

(15)

This equation and Eq. (10) for branch currents jointly define the three-phase machine model. 3.1.2 Multiphase Loads Loads are usually represented as constant power consumption in single-phase (positive sequence) load flow programs. For unbalanced load flow analysis, the different response to positive, ncgative and zero sequence voltages and currents must be modelled, in addition to the power constraints. In view of the variety of load characteristics, four basic types of load models are proposed. Type 1. Constant impedance load with known [Z] or [Y] branchmatrix representation.

a
m
9

k
0

I
0

k
0

m
0 0

Type 2. Load with constant active and reactive power specified as a single-phase branch (Figure 3(b)). It is defined as:

k
0

m
9

k
0

This load corresponds to the traditional PQ representation. However, it is defined as a branch between two nodes, rather than from node t o ground as in single-phase load flow programs. This allows large flexibility in the type of connection, as for example phase-to-phase loads in delta systems. Type 3. Static load (Figure 3(c)). In this type of load, it is assumed that the positive and negative sequence impedances are equal and that. the ratio of the positive to zero sequence impedance is available. In effect, it assumes that the load impedances are balanced for the three phases. The total active and reactive power is specified, but the positive and zero sequence impedances are not explicitly known. Using symmetrical components, the branch equations for this type of load in phase quantities can be expressed as

(a) three-phase synchronous machine. (b) individual branch load. (c) three-phase static load. (d) voltage source. (e) current source. Figure 3: Branch models for multiphase harmonic load flow studies. The machine internal voltage E p is unknown and must be adjusted to satisfy the machine load flow constraints for terminal voltage and power output. Slack machine. The specified constraints are the magnitude and the phase angle of the positive sequence voltage at the machine terminals.
- [ v m l ) = V#peeifiedr

(11)

where y is an unknown admittance and [ K ]is a known constant symmetric matrix determined from the positive to zero sequence impedance ratio. The unknown admittance must be adjusted to fulfill Eq. (18). Type 4. Rotating machine load with unequal negative and positive sequence impedances, as in the case of induction motors. For this load, it is assumed that both the negative and zero sequence admittances are known. The positive sequence admittance is not known and is to be determined from the three-phase active and reactive power consumption. With symmetrical components, this type of load can be represented as
Ikm-p Ikm-n Ikm-o

where

[TI = (1/3)[1 a

Q.'].

Machines with these constraints correspond to the slack nodes in conventional load flow studies. Under unbalanced conditions, the negative and zero sequence voltages will be nonzero because the network sees the machine as admittances Y, and Yoin negative and zero sequence representations.

PV machine. The specified constraints are the three-phase active power output and the magnitude of the positive sequence voltage at the machine terminals.

=G

( v k - p

-v m - p ) -v m - n ) -vm-0)

(19) (20) (21)

Yn(Vk-n

=K(Vk-0

where superscript

denotes conjugate transposed.

PQ machine. The specified constraints are the three-phase active and the three-phase reactive power output.

In stead of using Ypas an unknown, it is better to model the load as an internal voltage E , behind the negative sequence admittance Y,. E,, then becomes the unknown variable and Eq. (19) can be rewritten as
Ikm-p

= y n ( v k - p -V

m-p

-Ep),

(22)

_ I .

177

Transforniing Eqs. (20), (21) and (22) into the phase domain and including the built-in load flow constraints, this load model can be defined as:
[Ikm]

Collecting all the related equations tbgether, the multiphase load flow problem can be formulated as:

= [%I([vkl

- [Ikm]H([Vk]

-iVm1 - [vml) = ( p + jQ)speezf*ed.

(231 (24)

where

[E] = [E, u'E, uE,]',

P;lmutual

[Yglsclf = (K

(ro ro)/3, + 2ro)/3,

-

Note that with the introduction of E,, this type of load has the same structure as the P Q synchronous machine. It can then be siinply treated as a PQ machine with negative power geiiFrat ion By specifying the load flow cpnstraints at the branch level, the loads can be arbitrary connected between nodes or from node to ground. Loads can also be connected to the same node.
3.1.3 Voltage and current. sources

The general form of these equations can be written as

and

Voltage and current sources are again represented as branches. The voltage source (Figure 3(d)) is defined as
Vk - vm = Espeeifted.

[Iv] is the vector of currents from voltage sources, ( 1 ~ is 4 the vector of single-phase P Q load currents,

[Ifif] is the vector of machine currents,

[IL3]

(25)

is the vector of static load currents,

A current source between two nodes (Figure 3(e)) is defined as currents leaving two nodes,
I k = Ispeeified,
&n

[Ep]is the vector of machine internal voltages,

[y] is the vector of static load parameter y.

Equation (34) is a set of nonlinear algebraic equations, which

must be solved iteratively. Experience has shown that the Newton-

-Ispceified*

3.1.4 Other Network Components Overhead transmission lines, underground cables, transformers, react,ors and capacitors can all be modelled as coupled .rr circuits. Details are well-documented in references [13] and [15]. There are no load flow constraints associated with these components. 3.2 Formula.tion and Solution of the Load Flow Equations With all network components described at the branch level in the form of Norton equivalent circuits, it becomes easy to write the nodal equations for the entire network. The branch adnlittance niatrix of each component enters the larger network adiiiittance matrix according to well-known building rules [15], while the current sources between sides k and m enter as currents with a positive sign on side k, and with a negative sign on side m. With load flow constraints, these currents in [L]are unknown, and must be iteratively adjusted. As a result, the network equation is formed
as

Raphson method is probably the best method for conventional load flow studies [16]. It has also been chosen for the solution of Eq. (34). Rectangular coordinates are used here to separate the complex variables and equations into real form. Besides its simplicity, the rectangular representation has other advantages. For example, if there are no PV and PQ constraints, the Jacobian matrix becomes constant. The solution is then equivalent to the direct solution of the linear problem [Y][V] = [I]. With the Newton-Raphson method, the system of linear equations [J,][Axt] = - [ A F ( z c l ) ] (35) is solved in each iteration step, and the variables are then updated with [.;+I] = [zi] [Azi],

where

i is the iteration number,

[J;]the Jacobian matrix, [ A F ( z ; ) ] the residual vector.

With Eqs. (27) t,o (33), Eq. (35) becomes

[YPI
whcre

+ [Is]t [Iu] = 0,

(26)

[Y] is the network node admittance matrix constructed from the branch admittance matrices without load flow constraints,
[VI is the node voltage vector,

[Is] is the vector of current sources leaving each node,

1o;itl

[I,,] is the vector of unknown currents (associated with flow constraints) leaving each node.

178

The procedure for obtaining the submatrices in this Jacobian matrix is the same as in conventional load flow techniques. Once the Jacobian matrix is obtained, Eq. (35) is solved by Gauss elimination with sparsity techniques. The largest component in the residual vector is used to test for convergence. 3.3 Initialization z , ]for the iterations is more comChoosing an initial guess [ plicated in the multiphase case. The traditional initialization technique, which uses 1.0 per-unit node voltage magnitudes with respective 120" phase shifts among phases a, b, and c, becomes unreliable if there are phase shifting effects through wye-delta transformer connections. Since the convergence of the Newton-Raphson method is sensitive to the initial guess, a special initialization procedure is used before entering the iteration loop. The procedure is based on the observation that the load flow equations become linear if there are no PV or PQ constraints. To approximate the network this way, the components with PQ and PV constraints are modified as follows:
1. Machines with PV and P Q constraints and rotating machine loads are represent,ed as admittance matrices of very small ruagnitude. This approximates open-circuit conditions.
2. Other loads are represented as known admittances y whose

machines act as frequency converters. For example, a negative sequence current at fundamental frequency induces second harmonics in the rotor circuits, which in turn induce third harmonic voltages in the stator. This can be taken into account with the method suggested by Semlyen, Eggleston, a.nd Arrillaga [18].
2. Loads. The behaviour of loads under the combined effects of unbalanced and harmonic conditions is usually not well known. Using the load modelling techniques of reference [19], a multiphase load can be modelled as a combination of lumped R,L,C elements. If the test data needed for this representation is not available, the recommendations of reference [17] can be used. This reference suggests that the harmonic characteristics of a load can be modelled
as
Zlwd-h

= (Ra

jx#)//jXp,

(39)

where

Using these approximations, the single-phase P Q load is replaced and the static load is represented as by &,,,d-h,

values are determined from

where T h is the user-supplied positive to zero sequence impedance ratio of the load at the given frequency. These sequence parameters are then transformed into phase quantities.
3. Transnussion lines. Transmission lines are represented as exact multiphase A equivalent circuits calculated at the considered harmonic frequency [15].

for the single-phase P Q loads, and

for three-phase static loads. The voltage magnitude (VI is estimated to be equal to the user-supplied rated voltage of the load. With these approximations, the load flow solution becomes linear. Rather than writing a separate algorithm for this initialization, the normal Newton-Raphson algorithm is used with zero initial values, and the linear estimate [CL] is obtained in one iterat,ion. This becomes the starting point for the following load flow iterations.

4.2 Solution Techniaue With the various system components represented at each harmonic frequency, the problem formulation is the same as that of Eq. (34). However, since the machines and loads are represented as constant impedances at harmonic frequencies, the problem becomes linear and iterations are not required.

5. GENERAL FORMULATION OF THE MULTIPHASE HARMONIC LOAD FLOW TECHNIQUE

Based on Sections 3 and 4, a general purpose program was written for the solution of the network with load flow constraints at fundamental and harmonic frequencies. The three solution stages, initialization, load flow, and frequency scan are all done by the Newton-Raphson solution module, with or without iterations. The harmonic iteration of Section 2 for static compensators was added by using the Norton equivalent circuits of Figure 2. The general form of the multiphase harmonic load flow (MHLF) technique can then be described with Figure 4. The initialization is very simple: At harmonic frequencies, the equivalent currents Ih--eq representing the effects of nonlinear elements are set to zero. For the fundamental frequency load flow solut,ion, the PV and PQ components are modified into known Y matrices. With these simplifications, the approximate node voltages without harmonic distortion ( h = 1) are obtained in one iteration. Because the voltage harmonics are relatively small compared with the fundamental frequency components, using these node voltages as initial conditions is quite reliable. As shown in the flow chart, the MHLF technique consists of two basic parts. The first part is the construction of harmonic

4. MULTIPHASE HARMONIC SOLUTIONS

To solve the network at the harmonic frequencies, it is first necessary to define how the machines and loads respond to harmonics.
4.1 Harmonic Response of Network Components 1. Machines. As a first approximation, it is assumed that machines do not produce harmonics. For harmonic frequencies, they can t,hen be modelled as known admittance matrices, as suggested in reference [17]:

where h is the harmonic order, and XI-, and XI-,, are the negative and zero sequence reactances of the machine at fundamental frequency, respectively. The internal voltage is zero for h, > 1. A similar model is used for rotating machine loads.
For more accurat,e representations, it must be realized that

179

Norton equivalent circuits for the nonlinear elements (a TCR is used as an example). The second part performs the network solutions at the fundamental and harmonic frequencies. These two parts are interfaced through the process of harmonic iteration.

Thus the size of the Jacobian matrix is reduced considerably and the computational burden is only linearly proportional to the total number of harmonics. Thirdly, the technique has good convergence behaviour [lo].

I inuut data 1
I iteration-0 I
set initial TCR model by letting
0

6. PRACTICAL CONSIDERATIONS FOR STATIC COMPENSATOR ANALYSIS

Call "WORK-SOLUTION

h>n hsn o r harmonic h setupcomponent models f

Call NETWORK-SOLUTION

A practical static compensator, such as the one shown in Figure 5 [ll], is more complicated than the simple TCR unit of Figure 1. First of all, there is the delta connection of the TCR's. This connection is used to filter the zero sequence harmonics (e.g. 3rd, 9th). With the multiphase representation and the Norton equivalent circuits, the modelling of delta connection is straightforward, and the circulation of the zero sequence harmonic currents is automatically taken into account. When the operating conditions are unbalanced, the non-characteristic harmonics in the delta connection are automatically obtained.

compute new TCR Norton equivalent

TCR

8@IIQ 8
T
T TT TT T
. . .

TSC

filters

filters

TSC

"CR

(a) Static compensator stxucturc (single-phase diagram)

NETWOFUC-SOLUTION is a subroutine whose function is to form F ( [ x l ) = 0 equation and to solve it by Newton-Raphson method. Figure4: How chart of the MHLF technique. (A TCR is used as an example of nonlinear element.) The main characteristics of the MHLF technique can be summarized as follows:

(b) T m d e l t a connection
1. The MHLF technique is a multiphase program. It can be used either for single-phase or for three-phase harmonic load flow studies. Unbalanced operating conditions can be considered in the study. 2. Harmonics from other nonlinear elements can be analyzed with the technique. The linear equivalent circuit models for nonlinear elements not included internally in the program can be supplied by the user externally in the form of a subroutine. The MHLF program has been structured to make such interfaces easy.

Figure 5: '"he structure of a practical static compensator.

3. The MHLF technique is computationally efficient. First of all, its initialization is simple and reliable. Secondly, because the Norton equivalent circuits are harmonically-decoupled, the network solutions are performed one frequency at a time.

The second complication is the three-winding transformer connection. The phase shifting effects of the connection are critical for the cancellation of the 5th and 7th harmonics generated by the TCR's at the two secondary windings. With the multiphase modelling of transformers, the phase shift is automatically included [15]. Since the voltage ratio can also be included in the transformer model, the load flow analysis can be performed in either per-unit or physical quantities.

180

7. CONCLUSIONS
A multiphase harmonic load flow (MHLF) technique has been developed for the harmonic analysis of static compensators and other nonlinear devices under balanced or unbalanced conditions. This technique incorporates the harmonic iteration scheme into the multiphase framework. The MHLF technique consists of two major parts. The first part constructs harmonic Norton equivalent circuits for the nonlinear elements. The second part performs linear network solutions at fundamental and harmonic frequencies. User-supplied modules for particular nonlinear elements can also be easily interfaced with the program. The paper describes the development of the Norton equivalent circuit for the TCR with a known conduction angle as an example of nonlinear elements modelling and the extension of this modelling to multiphase network solutions at fundamental and harmonic frequencies. The inclusion of the control characteristics of static compensators into the MHLF solution and the results of cas? studies are described in a companion paper [lo].

W. Xu, J.R. Marti and H.W. Dommel, Harmonic Analysis of Systems with Static Compensators, Paper submitted for IEEE PES Winter Meeting 1990.

T.J. Miller, Ed. Reactive Power Control in Electric Systems,

New York: John Wiley&Sons Inc., 1982. A.H. El-Abiad and D.C. Tarsi, Load Flow Solution of Untransposed EHV Networks, Proc. of 5th Power Industry Computer Applications Conference, Pittsburgh, pp. 377-384, 1967.

J. Arrillaga, C.P. Arnold and B.J. Harker, Computer Modeling of Electrical Power Systems, New Zealand: John Wiley&Sons, 1983.

N.A. Wortman, D.L. Allen and L.L. Grigsby, Techniques for the Steady State Representation of Unbalanced Power Systems, IEEE Trans. on Power Apparatus and Systems, vol. PAS-104, pp. 2805-2824, Oct. 1985. H.W. Dommel, Electromagnetic Tmnsients Program Reference Manual (EMTP Theory Book), Bonneville Power Administration, Aug. 1986. G.W. Stagg and A.H. El-Abiad, Computer Methods in Power System Analysis, New York: McGraw-Hill Book Company, 1968. CIGRE-Working Group 36-05, Harmonics, Characteristic Parameters, Methods of Study, Estimates of Existing Values in the Network, Electra, no. 77, pp. 35-54, July 1981. A. Semlyen, J.F. Eggleston and J. Arrillaga, Admittance Matrix Model of a Synchronous Machine for Harmonic Analysis, IEEE Trans. on Power Systems vol. PS-2, pp. 833-840, Nov. 1987. A.S. Morched and P. Kundur, Identification and Modeling of Load Characteristics at High Frequencies, IEEE Trans. on Power Systems, vol. PS-2, pp. 153-160, Feb. 1987.

8. ACKNOWLEDGEMENTS
The financial assistance of the System Engineering Division of B.C. Hydro and Power Authority is gratefully acknowledged. The authors are also indebted to TransAlta Utilities Corporation, Calgary, Alberta and to B.C. Hydro and Power Authority, Vancouver, B.C. for providing data and field test measurements. 9. REFERENCES D. Xia and G.T. Heydt, Harmonic Power Flow Studies, Part I - Formulation and Solution, Part I1 - Implementation and Practical Application, IEEE Trans. on Power Apparatus and Systems, vol. PAS-101, pp. 1257-1270, June 1982.

W. Song, G.T. Heydt and W.M. Grady, The Integration of HVDC Subsystems into the Harmonic Power Flow Algorithm, IEEE Trans. on Power Apparatus and Systems, vol. PAS-103, pp. 1953-1961, Aug. 1984.
W.M. Grady and G.T. Heydt, Prediction of Power System Harmonics Due to Gaseous Discharge Lighting. IEEE Trans. on Power Apparatus and Systems, vol. PAS-104, pp. 554-561, March 1985.
A. Semlyen, E. Acha, and J. Arrillaga, Harmonic Norton Equivalent for the Magnetizing Branch of a Transformer, IEE Proceedings, Vo1.134, Part C, No.2, pp. 162-169, March 1987. T.J. Densem, P.S. Bodger and J. Arrillaga, Three Phase Transmission System Modelling for Harmonic Penetration Studies, IEEE Trans. on Power Apparatus and Systems, vol. PAS-103, pp. 310-317, Feb. 1984. D. Xia, Z. Shen and Q. Liao, Solution of Non-characteristic Harmonics Caused by Multiple Factors in HVDC Transmission System, Proc. of the Third International Conference on Harmonics in Power Systems, Nashville, IN, pp. 222-228, Oct. 1988. R. Yacamini and J.C. de Oliveira Harmonics in Multiple Convertor Systems: a Generalized Approach, IEE Proceedings, vol. 127, Part B, No. 2, pp. 96-106, March 1980.
wenyuan X u (St.M85) was born in China in 1962. He received
a B.Eng. degree from Xian Jiaotong University, Xian, China in

1982 and a M.Sc. degree from the University of Saskatchewan, Saskatoon, Canada in 1985. At present, he is a Ph.D. candidate at the University of British Columbia.
Jose R. Marti (M71) was born in Spain in 1948. He received a M.E. degree from Rensselaer Polytechnic Institute in 1974 and a Ph.D. degree in Electrical Engineering from the University of British Columbia in 1981. From 1970 to 1972 he worked for industry. In 1974-77 and 1981-84 he taught power system analysis at Central University of Venezuela. Since 1984 he has been with the University of British Columbia. Hermann W. Dommel was born in Germany in 1933. He received the Dip1.-Ing. and Dr.-Ing. degrees in electrical engineering from the Technical University, Munich, Germany, in 1959 and 1962 respectively. From 1959 to 1966 he was with the Technical University, Munich, and from 1966 to 1973 with Bonneville Power Administration, Portland, Oregon. Since July 1973 he has been with the University of British Columbia in Vancouver, Canada. Dr. Dommel is a Fellow of IEEE and a registered professional engineer in British Columbia, Canada.

H.W. Dommel, A. Yan and W. Shi, Harmonics from Transformer Saturation, IEEE Trans. on Power Delivery, vol. PWRD-1, pp. 209-215, April 1986. W. Xu and H.W. Dommel, Comput,a.tion of Steady-%ate Harmonics of Static Var Compensat,ors, Proc. of the Third International Conference on Harmonics in Power Systems, Nashville, IN, pp. 239-245, Oct. 1988.

181

Discussion
Adam Semlyen (University of Toronto): I would like to commend the authors for their novel approach in solving the multiphase load flow problem. Using a harmonic domain Norton equivalent for the representation of nonlinear elements has made it possible to include them in a Newton-Raphson solution of the whole system. The approach permits to solve even such difficult problems as encountered with a set of star connected nonlinear branches with isolated neutral. In the approach of the paper, the Norton equivalent circuits are considered to be harmonically decoupled. Consequently, the load flow solution has been performed sequentially, for one harmonic at a time, a significant saving compared to a fully coupled approach. In general, however, the linearized representation of a nonlinear element in the harmonic domainAmay couple the individual harmonics with each other. This appears clearly from equation (17a) of reference [A]

dependent set of harmonic current sources, or better still, as a harmonic Norton equivalent [A]. Then, the three phase TCR model is assembled. In this paper the authors are presenting a model, which represents each TCR branch as a voltage dependent set of harmonic current sources paralleled by a linear admittance. The linear admittance is based on fundamental frequency voltage and current considerations. This is amenable to a Norton equivalent at each harmonic frequency; as opposed to the harmonic Norton equivalent obtained in the harmonic domain. The former does not exhibit cross-couplings between harmonics, and standard factorization techniques will apply when solving the network. The latter exhibits cross-couplings between harmonics and more specialized inversion techniques will be required [B]. The authors have moved from the current source representation to the Norton equivalent representation of the TCR. Presumably, because of better convergence characteristics. Have the authors found this improvement to be significant? If so, then, the principle presented in this paper should also be applicable to other non-linear components, i.e. magnetic nonlinearities and electric arcs. The rationale behind this surmise is given below. Harmonic domain linearization lends itself to a Newton-type iterative solution, where the Jacobian matrix contains all the harmonics and the cross-couplings between harmonics. The Jacobian matrix corresponds t o the admittance matrix of the harmonic Norton equivalent. During full conduction state, the TCR behaves linearly, and the admittance matrix of the harmonic Norton equivalent becomes diagonal, i.e. no crosscouplings between harmonics exist. In this condition, the Norton equivalent presented by the authors and the harmonic domain Norton equivalent will coincide. Clearly, this is not the case at other conduction states, however, a close numerical correspondance will exist if the cross-couplings between harmonics are neglected and the Jacobian matrix is evaluated during the first iteration only and then, kept constant until convergence is achieved. The Jacobian matrix is diagonally dominant and the harmonic admittance of the entire network is even more diagonally dominant. Thus, it is quite likely that the Norton equivalent representation of this paper will exhibit very good convergence characteristics even in cases of harmonic voltage magnification. It has been found that for some non-linearities, such as iron cores, the number of iterations taken to reach convergence is the same, whether the Jacobian is evaluated at each iterative step or it remains constant after having been evaluated during the first iteration. For some other non-linearities, such as laminated iron cores, this is not the case. Nevertheless, the effects of neglecting cross-couplings between harmonics should be investigated for the case of iron cores, particularly, multi-legged transformers. The authors must be congratulated for a most valuable paper.

i = Yhrmv+iN which corresponds to a harmonic domain Norton equivalent with i and v being vectors of the harmonic components of currents and voltages and i N the Norton current vector. The matrix Y,-, in the above equation (the Norton admittance matrix) is generally f u l l . I wonder, therefore, what are the assumptions or procedures which permit to use the decoupled approach of the paper without loss of accuracy in the solution? Is the TCR a harmonically truly decoupled device? In order to illustrate that the coupling between harmonics can be significant, consider the following extremely simple example. Let i =w+$
be the equation of a nonlinear reactor and

w=l+f1cosOt (b) the input (flux linkage) of base frequency only. Direct substitution yields 3 1 i= -y:)~osot+ -y1:cos3ot 4 4 With ~1 =w1 be +Awl eqn.(c) becomes i = i h e +Ai where 3 1 i h e =(VI + -4~ : ) ~ ~ c o s o ~ 4 + - - w : ~ , c o s ~ ~ ~ (e) and
9 Ai = (1 + -w?)brAwIcosor
4

3 + -w: 4

~eA~1cos030t

(0

With

w1he = 1 eqn.(f) becomes

Ai = 3.258y11coswr+0.75Ay11coso30t (9 This equation shows that the same base frequency incremental input Ay1 produces a third harmonic output of 23% (0.75D.25) of the base frequency output. The resulting coupling is thus not insignificant,
[A] A. Semlyen and N. Rajakovic, "Harmonic Domain Modeling of Lamiited Iron Core", IEEE Trans. on Power Delivery, Vo1.4, N0.1, January 1989,p ~382-390. .

Manuscript received February 5, 1990.

E. Acha (OCEPS Group, University of Durham, U.K.): I

would like t o congratulate the authors for this timely and most interesting paper. In particular, I would like to address the newly developed TCR model, which is simple and yet comprehensive. TCRs are nonlinear components and their harmonic interaction with the power network is reached by iteration. This is especially true for cases in which harmonic voltage magnifications occur due to resonant conditions. One possible modelling approach is to represent each TCR branch as a voltage-

[A] L.J. Bohman and R.H. Lasseter.fHarmonic Interac-

182

tions in Thyristor Controlled Reactor Circuits, IEEE TransJuly actions on Power Delivery, Vol. 4, No. 3, pp. 1919-1926, 1989.

[B]A. Medina, J. Arrillaga and E. Acha.(Sparsity-Oriented Hybrid Formulation of Linear Multiports and its Application to Harmonic Analysis, To be presented at the IEEE PES Winter Meeting, Atlanta, GA, Feb 4-8,1990.
Manuscript received February 20, 1990.

Wenyuan Xu, Jose Marti and Hermann W. Dommel: We would like to thank all the discussers for their interest in the paper and for their valuable comments. We hope that the following comments will help to clarify some of the raised issues. Professor Semlyen: Professor Semlyen is correct in pointing out the harmonic coupling nature of nonlinear devices. The harmonic Norton equivalent curcuit for the TCR presented in the paper includes the coupling effects in the form of

The system is an iteratively adjusted harmonic current source solved for one harmonic frequency at a time. The solution is then compared with the characteristic of each nonlinear element and corrections are added to the elements harmonic current source. D r .Acha: We observed significant improvement of the convergence rate by moving from a current source model to a Norton equivalent model. The improvement is due to a more accurate estimation of the fundamental frequency voltage at the TCR terminal. We completely agree with D r . Achas harmonic-domain linearization analysis. According to our numerous test runs,it is very likely that convergence can be further improved if the diagonal elements of the harmonic-domain Jacobian m a t r i x are as the equivalent admittances in the harmonic iteration scheme with nonlinear inductors has been presented in reference (I). In general, we have found that the convergence rate is inversely affectedby the degrees of saturation and network harmonic voltage resonances. Reference

W. Xu, A Multiphase Harmonic Load Flow Technique, Ph.D.

Dissertation, University of British Columbia, Vancouver, B.C., Canada, February. 1990. Manuscript received April 12,1990.