Vous êtes sur la page 1sur 9

IEEE Transactions on Power Systems, Vol. 9, No.

1, February 1994 59
Dynamic Ward Equivalents for Transient Stability Analysis

Thomas L. Baldwin Lamine Mili Arun G. Phadke


Student Member Senior Member Fellow

Department of Electrical Engineeeng


Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061, U.S.A.

Abstract - In an effort to reduce the computing time of analysis when the system is reduced throu h the Ward
transient stability assessment, the paper presents a dynamic equivalencing method. Recently, Pai et al. f9] initiated a
equivalent which results fiom the elimination of the load buses method based on the assumption that the complex ratios of the
povided with voltage-dependent loads. The elimination is voltage phasors at the generator buses to those at the load
performed through a new version of the Ward equivalencing buses are constant. A different approach was proposed by
method. In this approach, the equivalent current injections are Vittal, et al. [lo]. Here, a transient energy function for a
expressed in terms of the retained bus angles and a sensitivity reduced system is built after the elimination of the load buses
provided with constant current and constant power loads. The
matrix E. The non-linearity of the load flow model is reduction employs a Ward equivalencing method in which the
accounted for through piecewise linear approximations by equivalent current injections are updated at each integration
updating the E matrix whenever the operating point moves step of the path-dependent term of the energy function. Each
beyond the validity of the linearization. The paper also derives step involves a single iteration of the Newton-Raphson
the expressions of the incremental changes in the generator procedure on the unreduced system, which makes the method
electric power and the transient energy function for the time consuming.
reduced system. The approach has been tested on several The paper presents a fast dynamic Ward equivalent that
systems with different sizes and characteristics. is not only able t o handle constant P Q loads [ll],but also can
eliminate any load buses provided with voltage-dependent
Key words: Network reduction; Complex loads; Dynamic loads represented by a polynomial function. It takes
equivalents; Transient stability analysis; Energy function. advantage of the ability of the PMU’s to measure the voltage
and current phasors at the buses where they are located. The
cornerstone of the technique is a correction formula that allows
1 - INTRODUCTION us to update the equivalent current injections at the retained
buses without explicitly carrying out a load flow calculation.
I t can also be used to update the generator electric powers and
The advent of Phasor Measurement Units (PMU’s) [1,2] the energy function of the system. This formula is derived
which are able t o track the dynamics of an electric power from a sensitivity analysis performed on both the power flow
system has prompted the development of closed4oop control of and the load models. It involves a sensitivity matrix E which
power systems during an emergency state [2-5]. As part of this is updated whenever the machine angles move beyond the
effort, work has begun to address the problem of incorporating range of validity of the linearization. Simulation results
PMU’s in power system state estimation as well as provide showed that, in the case where a single machine is swinging in
data for transient stability analysis. To this end, an algorithm
that finds the minimal PMU placement for system the system, the E matrix needs to be updated only five times
observability has been developed [5, 61. to cover the whole range of angular excursion. Further savings
The need to quickly perform transient stability in computing time can be obtained by calculating beforehand
assessment for on-line control requires the reduction of the some of these matrices and storing them in memory.
system through the elimination of the load buses. The Throughout the paper, it is assumed without loss of generality
conventional approach consists of modeling the loads by that the retained buses are the internal generator nodes
constant impedances and eliminating them through a Ward whereas the eliminated buses are voltage-dependent load
equivalencing technique. A major drawback of this approach is buses.
the unreliability of the analysis due to the oversimplification of The paper is organized as follows. Section 2 reviews
the load models. To overcome this weakness, Bergen and Hill voltagedependent load models. Section 3 derives the dynamic
[7] have advocated the use of complex load models while Ward equivalent based on sensitivity analysis. Section 4 gives
preserving the structure of the system. However, this the expressions of both the generator electric power and an
methodology comes at the expense of large computing times, energy function for the reduced system. Section 5 describes
which precludes its use in a real-time environment. All this some simulation results performed on the New-England
calls for an approach which performs both the reduction of the 10-machine system.
size of the system and retains the effects of complex loads.
I
Podmore and Germond 81 were the first to propose a
method that accounts for non- inear loads in transient stability
Notation: In this paper we will denote a matrix and a vector
quantity b y a bar below it, and a phasor quantity b y a bar above
at.
93 WM 244-4 PWRS A paper recommended and approved
by the IEEE Power System Engineering Committee of 2 - VOLTAGE-DEPENDENT LOADS
the IEEE Power Engineering Society for presentation
at the IEEE/PES 1993 Winter Meeting, Columbus, OH,
January 31 - February 5, 1993. Manuscript submitted This section begins by presenting the polynomial form of
August 27, 1992; made available for printing the voltagedependent load model which is used in the
December 28, 1992. development of the dynamic Ward equivalent. It is followed
by a simulation study that demonstrates the unreliability of
transient stability analyses when constant PQ loads are
approximated by constant impedance loads.
0885-8950/94/$04.00 0 1993 IEEE
60

2.1 VoltageDependent Load Modeling power-angle curves of generator #3 for the unreduced and the
reduced system depart from each other. The difference
Consider a load bus, say the jth bus, which is provided becomes significant for large machine angles, which invalidates
with voltage-dependent powers, P . and Q , represented as a the stability margin assessment. Both curves result from the
(J 4 application of a sequence of angular shifts of 10' in magnitude
polynomial function of the voltage magnitude of that bus, Vlj , to the angle of generator #3 while maintaining the angles of
namely the other machines at the base case values.

and

Here ap a are coefficients which express the proportion of


i' Qi
various loads lumped together in the real and reactive power
injections, and bp b . are the exponential coefficients that
i' PI
describe the load characteristics with respect to the load bus
voltage magnitude [U].For instance, setting the bpi and bQi
coefficients to zero in (1) and (2) result in constant power
loads whereas setting them to one or two yields constant
current or constant impedance loads, respectively. Their usual
range is between 0 and 2.
2.2 Simulation Studies

Let us illustrate the effect of approximating the constant


power loads by constant impedance loads in transient stability 0.0 0.2 0.q 0.6 0.8 1.0 1.2 1.9 1.6 1.8 2.0
analysis through simulations performed on the New England
10-machine system [13]. Its one-line diagram is depicted in Time (Sec)
Figure 1.

Unreduced system with constant PQ loads.


A Reduced system with constant impedance loads.

F'ignre 2. Machine angles of generator #1 vs. time.

Figure 1. One-line diagram of the 10-machine system.

The system is provided with constant PQ loads. The inertia


constants of the generators range from 30 to 42 seconds except
for machine #1 which has an inertia constant of 500 seconds.
A three-phase fault is simulated at bus 4. In these Anqular Shifts i n Generator # 3
simulations, the swing equations are numerically integrated by
means of a RungeKutta method. Figure 2 displays the rotor
angle curves for machine #l. We observe that the rotor angles
for the reduced system obtained after approximating the Unreduced system with constant PQ loads.
constant PQ loads by constant impedance loads greatly depart A Reduced system with constant impedance loads.
from those of the ori 'nal system. The discrepancy exceeds 50"
at the end of the Erst swing, which lasts approximately 2
seconds. This discrepancy is unacceptable from a transient Figure 3. Power-angle curves of generator #3.
stability viewpoint as seen in Figure 3. Indeed, the
61
3 - DYNAMIC WARD EQUIVALENTS The linearization of (9) allows us t o express the equivalent
current increments as a function of the machine angle
We will now derive the dynamic Ward equivalent which increments A 6 and a sensitivity matrix
-g
z,
namely
is attached to the internal nodes of the generators. It is
assumed that the load power injections are described by (1)
and (2). It is also assumed that the generators are represented -I? .
A -gf q = E A & (13)
by the classical model with a constant e.rn.f. in series with a
transient reactance. To this end, we have to linearize the phasors contained in
3.1 The Static Ward Equivalent A E , AYf and A& and relate them to A-4, . The linearization
-g
is carried out in 3 steps. First, it is applied to the load flow
For the original system, the bus current phasors are model, which leads to the expression of the load voltage
related to the bus voltage phasors and the bus admittance magnitudes A V and phase angles At9 in terms of A 6 .
4 4 -I3
matrix through Then, the linearization is extended to the load current
magnitudes AI and phase angles A$fj, which enables us to
4
relate them t o A 6 . Finally, it is applied to the voltage and
-g
(3) current phasors.

where the subscript L denotes the load buses to be eliminated 3.2.1 Linearizing the Load Flow Model
and the subscript g denotes the generator buses that are
retained. When the system is reduced to the internal generator By linearizing the load flow model about the base case,
nodes, the current-voltage relationship (3) reduces to the real and reactive power increments can be expressed in
terms of the voltage magnitude increments, the phase angle
increments and the Jacobian matrix J . In a matrix form, we
have
where

denotes the equivalent current injection vector and

- p y-'p
qg- 9-p
-gg -gf -U -fg (6)

denotes the equivalent bus admittance matrix [14].


By assuming that A E = 0, we use sensitivity analysis to solve
3.2 Sensitivity Analysis of the Equivalent Current Injection g
for Alf and AYt in terms of Ad only. To begin with, we infer
-g
The Ward equivalent of a system with voltage-dependent from (14) that
loads is only accurate at the operating point at which it has
been computed. However, as the operating point moves away
from the base case, this model does not represent the system
properly, for the equivalent currents at the generator buses are
no longer valid. To overcome this difficulty, we may resort to
sensitivity analysis in order to find appropriate correction
formulae for the equivalent current injections, which are Then we perform a first-order Taylor series expansion of the
updated through voltage4ependent load model described by (1) and (2) to get

Here the subscript "0" refers to a quantity associated with the


base case operating point and ATq denotes the incremental where
a
changes in the equivalent current injections resulting from the
shift in the generator angles. Its expression is obtained by
substituting
and
Ig = Xg& + YglITf (8) , dQ b .-I
Q f j = A = C aQ i bQ i Vf j Q 1
into (5) and subtracting the base case quantities. This yields dVfj i=l

A T = TggAEg + Xgf AYf - Ygfxi AIf , (9) In a matrix form, we have


where
A E = E -E , i = l , ...,n I (10)
gi gi gOi
62

Finally, we substitute (19)into (15) to obtain 3.2.3 Lineariging the Voltage and Current Phanors

At this point, we intend to express the incremental


phasor vectors A E , A Y f , and A& in terms of the magnitude
1
and phase angle increments of the load voltages and currents
given by (21) and (29). A first order approximation of

yields
This relation allows us to calculate the load bus voltage
phasors as a function of the generator angles.
673
3.2.2 Linearizing the Load Current Injections .= AEgi + 673 A 6gi = e
jdgi
(AEgi + jEgi A b g > .
g1 aEgi a 'gi
Now, we linearize the load current magnitude I and (33)
4
phase angle $ at a given load bus, say the jth bus, which
4 Under the assumption that AE = 0
gi
, (33) becomes
express in phasor form as
A Egl. = j ejS"' Egi A6gi, i = 1,- -- 1 n. (34)
Similarly, we have

Here $ is the power factor angle at bus jgiven by


fj

j$f.
A f(J . = e J (AIfj + j I .fj. A $ f j) , j = l , - . . , m .
(36)
The derivations are carried out as follows. First, we perform a
first-order Taylor series expansion of I about the base case, These expressions can be put into a diagonal matrix form as
4
which yields
Aa
E = E A 6-g (37)

where

Then, we expand the current angle to get


A$ fj = A 0 fj - A $ f j = A 0 fj - 4 'f j AVfj
Here & E,c,n, and E are diagonal matrices given by
7 (27)
where
II = diag (j j6'i Egi, , E= diag (ej okJ )
-

iZ = diag (j ejek Vfj) ,


- J D= diag (ej$tJ )
The latter is the rate of change of the load's power factor with
respect to the voltage magnitude at bus j. The relation (25)
and (27) expressed for j = 1,. -,m can be put into the matrix - j$t
E=diag(je JI) . (40)
form 4
Substituting (21) into (38) and (29) into (39), yields

where and
Xi= diag ( Y.
fJ
) , 9;= diag ( 4'( j ) . (30)

Finally, we substitute (21) into (29) t o obtain


3.2.4 Expression of the Equivalent Cnrrent Injections
r _ I ?

It is apparent that using (37), (41), and (42 into (9)


allows us to lump all the sensitivity matrices toget er into a
single matrix W.This leads to
h
63

Aa
Ieq=EAjg , (13) In (47) , AI:: and A$eq denote the magnitudes and phase
gi
where angles of the equivalent current increments given by (13).
-
w=v K+y
- G .(43) 4.2 Building a Transient Energy Function
In the base case, the expression of the system transient
energy function in the center of inertia (coi) reference frame is
Therefore, the equivalent current injections can be updated given by [I51
through

3.3 Updating the Sensitivity Factors n-1 n


For transient stability analysis, it is important that the
- x fi cij(cos Zgoij -cos ZSgoij)
,
i = l j=i+l
equivalent current increments be fairly accurate. However, as
the system deviates from the base case, the accuracy of the
increments deteriorates when they are calculated with a n-1 n
ii + ij

constant sensitivity matrix E. This difficulty may be Z..d(Z


11 g0i
+ 8gOj) . (49)
overcome by updating g whenever the linearization errors 6i + 6j
become too large. The updating of can be readily performed
by using (21) to calculate the new load bus voltage phasors. Here C.. = E E BY; , D.. = E E G??, Xi = bi - bcoi
This updating is carried out as soon as one of the machine U gOi POj 1J gOi POj IJ
angular deviations exceeds the range of the validity of the and G.1 = w.1 - wcoi . As the system moves away from the base
linearization. Simulation results (see Section 5.2) showed that
only five updatings are needed to cover the whole range of the case, the equivalent energy function is updated through
angular excursion when a single machine is swinging in the
system. Additional computing time savings may be achieved Veq(& = V:($,&) + AVeq($,&) .
by calculating beforehand some of these matrices and storing (50)
them in memory.
When using the dynamic Ward equivalent, the incremental
energy function, AVeq(&&) , is formed from the incremental
- DYNAmc EQUrVALENT IN STABILITY electric power given by (47). The latter is related to the
entries of the matrix through (13). Fortunately, this matrix
This section derives the expressions of both the generator is not affected by a change of the angle reference frame,
electric power and the energy function for the reduced system. because it only involves the bus admittance matrix and the
Jacobian matrix. Indeed, the former is independent of the
4.1 Expression of the Generator Electric Power an les and the latter has entries that are function of the angle
differences only. Therefore, the expression of the incremental
The swing equation expressed with respect to a transient energy is readily obtained by integrating the sum of
synchronous rotating reference frame is given as
the product of APeq and
e1
ii, yielding

where P . is the generator electric power obtained from a load


e1
AVeq@,&) = 1' 4

tS
i=l
EgoiAI;: COS ( ZgOi - A(:) iidt
flow solution. When the system is reduced to the internal
generator nodes, the equivalent electric powers using the (51)
classical machine model are found at the base case through
An appropriate change of variables in the limits of integration
yields

6i
where y"q
I]
L 19 1; are the entries of the equivalent bus Egoi AI;: COS (ZgOi - A(:) dzi .
admittance matrix Pq, expressed as in ( 6 ) . When using the
' 7% (52)
dynamic Ward equivalent, the equivalent current increments
are transformed into equivalent electric power increments, Note that the incremental equivalent current phasor,
which express as
A{: = AI:: L A g e s , is a function of all the generator angles
gi
6 which makes AVeq(Z) path dependent. However, an
-g '
approximation of the incremental transient energy is obtained
The latter are added to the base case electric power to get by assuming a linear trajectory between the post-fault stable
equilibrium oint 1
and $ as it is widely accepted (see for
instance [15]f
5 - SIMULATION RESULTS reported in Table 1. We observe that, when the E matrix is
updated at 2" , 75" , 92' , 105' , and 120' , the largest power
mismatch of generator #3 does not exceed 3 pu out of
The accuracy of the dynamic Ward equivalent has been maximum generator power of 35 pu, yielding a maximum error
evaluated through simulations performed on several test of 8.5% .
systems having a number of machines that ranges from 7 to 29. Table 2 reports the results for the New-England system
The tests have involved both multiple-swing-bus power flow provided with voltage-dependent loads that are expressed in
calculations and transient stability analysis for a wide range of terms of the voltage magnitudes raised to the power 3/2. The
operating conditions using different load models. This section -
reports only simulations carried out on the New-England -
W matrix has been updated at the same generator angles as for
system shown in Figure 1. This system is moderately to the constant P Q load case. Again the maximum error in
generator power does not exceed 2.6 pu, namely 8.5 % .
heavily loaded.
5.1 Numerical Integration of the Swing Equations
The simulations that have been performed on the
New-England system with constant PQ loads and its reduced m
version based on the constant impedance load approximation #
(see Figure 2 in Section 2.2) have been repeated while using I
the dynamic Ward equivalent. Figure 4 displays the rotor 0
0

angle curves for machine #1 when a three-phase fault is


applied to bus 4 for both the unreduced and the reduced
systems. We observe that the rotor angle for the reduced
system remains very close to that of the original system. The
discrepancy is only of 1' after 1 second. It does not exceed 10' .J
m
after 2 seconds. In these simulations, the E matrix is updated
whenever the angle deviation of a machine attains 10'. At the
end of the first swing, which lasts about 2 seconds, it has been
updated 10 times. Note that the more often the E matrix is
updated, the smaller is the discrepancy between the two
curves.

- 30
-160 -120 -ao -qo o qo EO 120

Flnqular Shifts in Generator # 3

o Unreduced system with constant P Q loads.


A Reduced system provided with a dynamic equivalent.

Figure 5. Power-angle curves of generator #3.

Table 1. Largest power mismatch for constant and updated


-
-
w for systems with constant PQ loading.
~~

Angular Gen. Mismatch Error Mismatch Error


shift Power const. w const. E updated updated jy
(de4 ( P d (PUT (%I (PU) (%I
0.0 0.2 0.11 0.6 0.8 1.0 1.2 1.q 1.6 1.8 2.0 10' 0.5 -0.05 -9.8 -0.05 -9.8
20' -1.4 -0.23 16.5 -0.23 16.5
Time (Sec) 30' 22.7 0.69 3.0 0.69 3.0
40" 27.1 1.62 6.0 1.62 6.0
50' 30.7 3.13 10.2 1.56 5.1
Unreduced system with constant PQ loads. 60' 33.4 5.32 15.9 -0.05 0.2
o Reduced system provided with a dynamic equivalent. 70" 35.0 8.30 23.7 2.97 8.5
80' 35.5 12.13 34.1 2.37 6.7
90' 34.8 16.87 48.5 0.74 2.1
Figare 4. Machine angles of generator #1 vs. time.

5.2 MultiplAwing-Bus Power Flow Analysis


6 - CONCLUSIONS
The accuracy of the dynamic Ward equivalent has been
further assessed through multipleawing-bus power flow
calculations performed on the New-England system with A modified version of the Ward-equivalencing method
constant PQ loads. Figure 5 displays the power-angle curves has been developed. It is suitable for transient stability
for generator #3 for the unreduced and reduced systems. They analysis using data provided by PMU's. The technique is able
result from the same angular shifts of 10' in magnitude as to eliminate any load buses to which are connected voltage-
those shown in Figure 3. The largest power mismatches are dependent loads that are modeled as a polynomial function of

.,. . . ._* -
. . , . .. . . .
65

Table 2. Largest power mismatch for constant and updated R. Podinore and A. Germond, Development of Dynamic
-
W for systems with I V I loading. Equivalents for Transient Stability Studies, Final Report,
EPRI EL-456. 1977.
Mismatch Error M.A. Pai, K. R. Padiyar, and C. Radhakrishna,
Angular Gen. Mismatch Error "Transient Stability Analysis of Multimachine AC/DC
shift Power const. w const. @
! updated updated Power System Via Energy Function Method", IEEE
(deg) (PU) (P4- (PU) (%I Transactions on Power Amaratzls and Svstems, Vol. 100.
-0.13 -1.2 No. 12, Dec. 1981,pp. 5027'4035.
10' 11.3 -0.13 -1.2 V. Vital, N. Bhatia, A.A. Fouad, G.A. Maria, and H.M.
20" 13.8 -0.41 -3.0 -0.41 -3.0 Zein El-Din. !IncorDoration of Non-linear Load Models in
30' 22.9 1.05 4.6 1.05 4.6 the Transient Energy Function Method", ZEEE
40" 27.3 2.22 8.1 2.22 8.1
4.02 13.0 2.63 8.5 Transactions on Power systems, Vol. 4, No. 3, Aug. 1989,
50" 30.9 pp. 1031-1036.
60' 33.5 6.57 19.6 0.52 1.6
70- 35.1 9.94 28.3 2.28 6.5 T. L. Baldwin, L. Mili, and A. G. Phadke, "Ward-type
14.22 40.1 -1.38 -3.9 Equivalents for Transient Stability Analysis",
80' 35.5 Proceedings of the IFAC International Symposium on
goo 34.7 19.45 56.1 -0.36 -1.0
Control of Power Plants and Power Systems, March 9-11,
1992,Munich, Germany, pp. 251-255.
P. L. Dandeno and P. Kundur, "A noniterative transient
stability program including the effects of variable
the voltage magnitudes. The cornerstone of the method is a load-voltage characteristics , IEEE Transactions on
correction formula which can be used to update the equivalent Power Apparatus and Systems, Vol. 92, No. 5, May 1973,
generator currents as well as the generator electric power and pp. 1478-1484.
the system energy function. Simulation results showed that M. A. Pai, Energy Function Analysis for Power System
the method gives very close results to those obtained from the Stability, Kluwer Academic Publishers, Boston, Mass.,
unreduced system. 1989.
Future research include the modification of the method t o F. F. Wu and A. Monticelli, "Critical Review of External
accommodate for two axes machine models with flux decay and Network Modelling for On-line Security Analysis",
excitation control. The intent is to overcome the inaccuracies Electric Power €4 Energy Systems, Vo. 5, No. 4, October
inherent t o the classical model of the generators. They also
include the development of analytical criteria for selecting the 1983,pp. 222-235.
T. Athey, R. Podmore, S. Virmani, "A Practical Method
machine angles a t which the E matrix has to be updated. for the Direct Analysis of Transient Stability", ZEEE
Another issue that needs to be addressed is the extension of the Transaddons on Power Apparatus and Systems, Vol. 98,
method to handle dynamic load models which are function of NO. 2, March/April 1979,pp. 573-584.
the frequency and/or the voltage derivatives.

ACKNOWLEDGEMENTS 8 - BIOGRAPHY

The support of NSF under grant number ECS-9257204 is


gratefully acknowledged. Thomas L. Baldwin (S '86) received the B.S. and M.S. degrees
in Electrical Engineering from Clemson University in 1987 and
1989, respectively. He is completing his Ph.D. degree in
Electrical Engineering a t Virginia Tech. His research interests
7 - REFERENCES include harmonic distortion, voltage collapse, transient
stability, and state estimation.
A. G. Phadke, J. S. Thorp, and M. G. Adamiak, "A New Lamine Mili (S '83, M '87, SM '91) received the B.S. degree
Measurement Technique for Tracking Voltage Phasors, from the Swiss Federal Institute of Technolo y, Lausanne, in
Local System Frequency, and Rate of Change of 1976, the M.S. degree from the University of Tunis in 1983,
Frequency", IEEE Transactions on Power Apparatus and and the Ph.D. degree from the University of Likge, Belgium, in
Systems, Vol. 102, May 1983,pp. 1025-1038. 1987. He is presently an Assistant Professor of Electrical
J. S. Thorp, A. G. Phadke, S. H. Horowitz, and M. M. Engineering at Virginia Tech. Dr. Mili is the recipient of the
Begovic, "Some Application of Phasor Measurement to NSF Young Investigator award (1992). His research interests
Adaptive Protection", ZEEE Transactions on Power include state estimation, transient stability, voltage collapse,
Systems, Vol. 2, May 1988,pp. 791-798. and power system control.
F. C. Schweppe, "Power System '2000': Hierarchical
Control Strategies", ZEEE Spectrum, July 1978, Arun G. Phadke (F'80) is a Professor of Electrical Engineering
pp. 42-47. at Virginia Tech. He was a System Engineer with Allis
T. E. Dy Liacco, "Real-Time Computer Control of Power Chalmers Company (Milwaukee, Wisconsin) from 1963 - 67.
Systems", Proceedings of the IEEE, Vol. 62, No. 7, There he participated in the design and development of the
July 1975,pp. 884-891. AC/DC Simulator Laboratory at the University of Wisconsin,
L. Mili, T. Baldwin, and R. Adapa, "Phasor Madison, from 1967 - 69. He then joined AEP in 1969. His
Measurement Placement for Voltage Stability Anal sis of reponsabilities at AEP as a Consulting Engineer included the
Power System", Proceedings of the 29th ZEEE ConJrence management of the Substation Computer Project. He was a
on Decision and Control, Dec. 9-7, 1990, Honolulu, HI, Visiting Professor at Virginia Tech during 1978 - 79 on leave
pp. 3033-3038. from absence from AEP. He joined Viginia Tech permanently
T. L. Baldwin, L. Mili, M. B. Boisen, Jr., and R. Adapa, in January 1982. Dr. Phadke is a Convener of CIGRE
"Power System Observability with Minimal Phasor Working Group 34.02on Computer Systems for Substations.
Measurement Placement", Paper No. 92 SM 583-5 PWRS
presented at the IEEE PES Summer Meeting, July 12-16,
1992,Seattle, Washington.
A.R. Bergen and D.J. Hill, "A Structure Preserving
Model for Power System Stability Analysis", IEEE
Transactions on Power Apparatus and Systems, Vol. 100,
No. 1, Jan. 1981,pp. 25-35.
Discussion The main computational burden when updating the
matrix is entailed by the factorization of the matrix G in
(21). To save computing times, we may take advantage of the
M. A. Pai (Dept. of Electrical Engineering, University of Illinois,
Urbana, IL): The paper is a very useful contribution to the fact that only few elements of E-1 need to be updated. These
transient stability analysis area by proposing an improvement to are typically some of the diagonal elements of Jzz and J,, that
existing methods for including voltage dependent loads. It is not represent the sensitivity of the load powers situated in the
a one shot approximation but is updated as the system evolves. vicinity of the accelerating machines. The other fast varying
The sensitivity matrix W essentially is built from the lineariza- elements are the entries of the E‘ and Q’ matrices, especially
when the coefficients api , bpi , aQi , bpi in (17) and (18) are
tion of the algebraic equations of the augmented Y matrix (i.e.,
including transient reactances). Current injections and hence large. In comparison, the other elements of GW1undergo small
Ai‘‘ at the internal modes of generators are updated through variations, and hence can be maintained constant at the base
the W matrix. Eq. (51) of the paper also expresses the incremen- case. As a result, sparsity techniques such as partial matrix
tal change in the energy function. The results of Fig. 4 compared
to those of Fig. 2 are improved considerably. Ih
factorization A and sparse vector methods [B] can be used.
In addition, i t e X/R ratios are relatively large, we can hold
constant or even neglect the PV and & e terms of the Jacobian
In Sec. 5.1 the authors state that the W matrix is updated matrix, that is, set J2, and J4z to zero in (22).
whenever the angle deviation of a machine attains lo”. This
would put a considerable computational burden if applied in
real time. Apparently one requires updating less frequently Now we will respond to each discusser separately.
while computing generator powers. Do they have any results for Dr. M.A . Pai:
voltage variations on the system with this approximation perhaps
by backsolving (3)? It may be useful in voltage stability studies. We thank Dr. Pai for bringing to our attention the fact
Extension to higher order models may not be difficult if they that maintaining a good accuracy of both the electric power
take a circuit theoretic approach to the “phasor” equations of and the energy function does not require to update the
detailed models since the internal voltage is a current dependent matrix as often as for the angles. Indeed, as shown in Figure
voltage source [A]. A, the range of accuracy of the power angle curve computed
Reference
with the dynamic equivalent is of 40’ in angular shift when W
is assessed at 0‘ and maintained constant as the machine
angles change. It degrades rapidly beyond that range. On the
[A] P. W. Sauer and M. A. Pai, “Dynamic modeling and simula-
tion of multimachine power system dynamics,” Control and other hand, when is assessed at go’, the range of accuracy is
Dynamic Systems, Advances In Theory and Applications (C. only of 15’ in angular shift . Hence the need to updated at
T. Leondes (Ed.)), vol. 43, Academic Press, San Diego, CA, 3 other angles, namely at 75’, 105’, and 120’. The resulting
1991. curve is displayed in Figure 5. A simpler scheme consists of
updating the matrix whenever the angular shift of any
machine exceeds 30’ in magnitude. Table A gives the errors in
S. Casper and C. 0. Nwankpa (ECE Department, Drexel Uni- the electric power of enerator #3 when applying such a
versity, Philadelphia, PA): scheme. In that case, & has been recalculated when the angle
of that generator is at 5.7’, 35.7’ , and 65.7” Thanks to these
The authors should be commended for their interesting paper recalculations, the power mismatch increases roughly linearly
on developing dynamic Ward equivalents. Linearization is per- (instead of quadratically) as the machine angles increase. It
reaches a maximum value of 7.6% at 95.7’, which is still
formed on the load flow model, load current injections and the acceptable. Similar results have been obtained for the system
voltage and current phasors followed by combining all these energy function.
into the equivalent current injection expression corresponding We agree with Dr. Pai that the dynamic Ward
equivalent can be applied to voltage stability studies as well.
to the new Ward sensitivity matrix. The authors justify the Indeed, relationship (21) allows us to calculate the voltage
need to update the Ward matrix when the machine angle de- phasors at all load buses as the machine angles evolve with
time. This opens new avenues for real time applications of the
viates by more than a set angle from the base case; the Ward method.
matrix is updated for this new base case. The discussors will
appreciate authors’ comments on the following.

Since this approach is geared towards on-line modeling, how


does the updating procedure affect the overall computation Table A. Largest power mismatches for the generator #3 of
simulation time? The test on the 38-bus system was based on the New England system with constant PQ loading.
a three phase fault at a particular bus in the system. How Angular Original Reduced Reduced Reduced
does ths choice of the disturbed bus affect the Ward matrix shift System System System System
(ded PG3 (PU) PG3 (PU) APG3 (pu) APG3 (%)
and results? In other words, are results obtained generic for
all buses in the system? 5.7 6.59 6.59 0.00 0.00
15.7 11.02 11.07 0.05 0.47
Manuscript received March 1, 1993. 25.7 15.19 15.44 0.25 1.67
35.7 18.94 19.65 0.71 3.76
45.7 22.13 22.87 0.74 3.34
55.7 24.66 25.60 0.93 3.77
Thomas L. Baldwin, Lamine Mili, Arun G. Phadke: We thank 65.7 26.38 27.73 1.35 5.12
the discussers for their interest in the paper. First, we will 75.7 27.22 28.63 1.41 5.18
comment on the computational burden of the method, an issue 85.7 27.10 28.77 1.67 6.16
raised by all the discussers, and then a separate response to 95.7 25.96 27.93 1.97 7.59
each of them will be given.
Drs. S. Caspw and C. 0. Nwankpa
P)
# Yes, the matrix does not depend on the fault location,
L
0
which means that the method is generic for all buses in the
0
rp
system. Good results have been obtained in various test
systems with different characteristics and fault locations. Due
k to space limitations, only a few of them have been reported
QJ
U and commented in the paper.
r)
m
I
0,
REFERENCES
0
a [A] S. M. Chan and V. Brandwajn, "Partial Matrix
Factorization", IEEE Transactions on Power Systems,
Vol. PWRS-1, No. 1, Feb. 1986,pp. 193-200.
[B] W.F. Tinney, V. Brandwajn, and S. M. Chan, "Sparse
Vector Methods", IEEE Transactions on Power
Apparatus and Systems, Vol PAS-104, No. 2, Feb. 1985,
pp. 295-301.
0 Manuscrip received April 2.1993.
Qnqular Shifts I n Generator # 3

o Unreduced system with constant P Q loads.


A Reduced system provided with a dynamic equivalent.

Figure A. Power-angle curves of generator #3.

Vous aimerez peut-être aussi