Sei sulla pagina 1di 13

IEEE Transactions on Power Systems, Vol. 8, No.

3, August 1993 1159

Voltage Stability Analysis Using Static and Dynamic Approaches


G.K. Morison B. Gao, Member IEEE P. Kundur, Fellow IEEE
Power System Planning Division, Ontario Hydro
Ontario, Canada
Abstract - This paper discusses voltage stability analysis of power Although many static approaches to voltage stability analysis have
systems using static and dynamic techniques. Using a small test been proposed [2-81, none has clearly demonstrated results
system, results of time domain simul0tion.s are presented to clarify consistent with time domain studies. In fact, it has been suggested
the phenomenon of voltage instability and to better understand [9] that results from static based techniques can not be correlated
modelling requirements. The same system is then analyzed using to results from full dynamic simulations. It is clear that there is a
a static approach in which modal analysis is performed wing need to bridge the gap between the two approaches and to
system conditions, or snapshots, which approximate different demonstrate the validity of static assessment.
stages along the time domain trajectory. The results obtained using
the dynamic and static methodr are compared and shown to be A small test system is used in this paper to show how voltage
consistent. instability can occur and to clearly resolve the situations in which
the modelling of loads, Under Load Tap Changers (ULTCs), and
Key Words - Voltage stability - Dynamic - Static - Modelling - generator Maximum Excitation Limiters (MXLs) are impactive on
Snapshots - Modal Analysis system voltage stability. First, results of time domain simulation
are presented for three system loadings to show various conditions
of voltage stability and to illustrate the key factors influencing
1. INTRODUCTION system performance. Next, static analysis is performed on the
system for the three load levels. The static approach used captures
In planning and operating todays stressed power systems, the system conditions, or snapshots, which approximate stages along
ability to maintain voltage stability has become a growing concern. the time domain trajectory. For each snapshot, modal analysis [8]
Although the term voltage stability is being extensively used in the is performed to evaluate voltage stability. Finally, the results of the
literature and in industry, general agreement on exactly what dynamic approach and the static approach are compared. It is
constitutes voltage stability does not yet exist. This is largely shown that the results obtained using the two approaches are
because distinctions between phenomena which represent different consistent, provided appropriate models are used.
forms of instability are often subtle. Also, the physical aspects of
voltage instability may not be well understood due to the complex 2. ANALYTICAL METHODS
nature of the phenomenon and the variety of ways in which it can
manifest itself in power systems. Recent literature has helped to 2.1 Dynamic Analysis
some extent in developing a common understanding of the
fundamentals of voltage stability, and with this, definitions, Features and modelliig capabilities of conventional transient
descriptions of the phenomenon, and methods of analysis are stability time domain simulation programs have been greatly
slowly evolving [l]. enhanced over recent years to make them suitable for the
assessment of long-term and voltage stability problems [lo],[ 111.
Time domain simulations, in which appropriate modelling is In this paper, the Extended Transient/Midterm Stability .hogram
included, capture the events, and their chronology, leading to (ETMSP) [101 is used for dynapic analysis. This program has the
voltage instability. Of all the analysis methods available, this modelling capability to account for dynamics which are important
approach still provides the most accurate replication of the actual in voltage stability analysis including those associated with voltage
dynamics of voltage instability. However, time domain simulations dependent loads, ULTCs, and generator MXLs.
are time consuming in terms of CPU and engineering required for
analysis of results, and they do not readily provide sensitivity 2.2 static Analysis
information or the degree of stability. These limitations generally
render the dynamic approach less suitable for the bulk of system The static analysis approach presented in this paper is a two step
studies in which the examination of a wide range of system procedure :
conditions and a large number of contingencies is required. For
this type of work, static analysis techniques, which require much
Step 1 :Establish a system condition, or snapshot, which closely
less CPU, can be used to provide much insight into the approximates a point along the time domain trajectory. This can be
voltage/reactive power problem. achieved by solving a set of system steady state algebraic
equations with appropriate models for controls and limits,
92 SM 590-0 PWRS A paper recommended and approved including those associated with generator MXLs and ULTCs. The
by the IEEE Power System Engineering Committee of equations required include network, steady state generator, and
the IEEE Power Engineering Society for presentation load power voltage characteristic equations. By judicious choice of
at the IEEE/PES 1992 Summer Meeting, Seattle, WA,
July 12-16, 1992. Manuscript submitted February 3, controls and limits invoked, and proper specification of active
1992; made available for printing May 1, 1992. power dispatch, snapshots which represent various stages along the
time domain trajectory can be obtained. For small systems,
snapshots can be obtained by manually manipulating powerflow
controls and parameters to obtain the desired operating conditions.
For large practical systems however, a snapshot program with ,
appropriate modelling would be required which could
automatically make system condition adjustments in order to obtain
0885-8950/93$03.00 0 1992 IEEE
1160
approximations of various time frames. Using the static method described above, postdisturbance voltage
stability can be assessed by performing modal analysis at several
Step 2 : Determine the system voltage stability at the selected snapshots representing different time frames in the time domain
snapshot(s). This is accomplished using the modal analysis simulation.
technique, details of which can be found in [8]. A brief description
of the technique follows. 3. STUDY OF A TEST SYSTEM
The linearized steady state system power voltage equations, 3.1 Test SystemDescription
including the effect of detailed device modelling, are given by,
In the study of large systems, device interactions are very
complicated and, as a result, the basic factors influencing voltage
stability are often difficult to isolate and understand. For this
reason, the small system shown in Fig. 1 is studied. It is based on
the system described in [12], with loads and generator outputs
Let AP = 0, then modified so that voltage instability may occur following the loss
AQ = [ J ~ ~ - J ~ =~ J,+W
J ~ ~ ~ A V of one of the branches. Complete data of the system is provided in
the Appendix. This system is simple yet realistic, and the study
results can be easily tracked and interpreted. Use of this simple
JR is called the reduced Jacobian matrix of the system. Let, system helps us to focus on specific factors influencing voltage
JR = U11 (3) stability.

and,
(4)

From (2) and (4). we have


AV = U - ~ ~ A Q

or,
AV = E F Q
I
Fig. 1 : Test System
6
where is the i* column right eigenvector and qi the i* row left
eigenvector of JR. 3.2 Dynamic Analysis

The i* modal reactive power variation is, System Modelliiq


AQmi = K& (7)
The disturbance considered is the loss, without a fault, of one of
the branches between bus 6 and bus 7. Time domain simulations
The corresponding i* modal voltage variation is were conducted for 300 seconds following the disturbance.

b o d at bus 11 is modelled by its natural voltage dependent


characteristic which is assumed to be 50% constant impedance and
50% constant current for both active and reactive power. The
action of the ULTC transformer supplying this load is modelled in
It is seen that, when the reactive power variation is along the
detail. Other system loads are modelled as constant MVA for both
direction of 4,the corresponding voltage variation is along the
active and reactive power, without ULTC dynamics.
same direction and the magnitude is amplified by a factor which
is equal to the magnitude of the inverse of the i* eigenvalue. The
system is voltage stable if all the eigenvalues of J, are positive,
Transformer ULTC actions are represented with time delays and
deadband. Time delays for ULTC operation are assumed to be 30
and voltage unstable if at least one of the eigenvalues is negative.
seconds for the first tap movement and 5 seconds for subsequent
The smaller the magnitude of the eigenvalue, the closer the
tap movements. The deadband is assumed to be i l %PA. of the
Corresponding modal voltage is to being voltage unstable. If the
controlled bus voltage. Tap range is k16 steps, and the step size is
eigenvalue is zero the system is on the verge of voltage instability.
98% (=0.00625 pa).
Although for convenience we refer to modal analysis as a sruric
MXL modelling is included for generator 3 only, as the other units
technique, it is more precisely, as pointed out by Dr. C.Condordia
never reach excitation limits in the cases examined. The block
[13], a gradient technique. For each steady state, the method
diagram of the MXL model is shown in Fig. 2. The function of a
examines the response of voltage to incremental changes in power,
MXL is to keep the field current under its continuous limit, as
and is therefore fundamentally based on slope or gradienf
shown in Fig. 3. If the field current exceeds the high setting
concepts.
(GdmaXz),the current is ramped down without a time delay to its
continuous limit. If the field current exceeds the continuous limit
1161

(Ifdmaxl) but is below the high setting, the current is ramped operation and the time at which the MXL acts. Once the ULTC
down to its continuous limit after a time delay dependent on the starts to operate, voltage at the load bus will gradually be brought
level of field current and the MXL settings. The MXL parameters up toward its reference value, and load power increases, provided
are as follows : the tap ratio does not reach its upper limit. Restoration of load bus
voltage and load power may increase the reactive demand on
Ifdmaxl = 3.02 pu (or 1.05xFull Load Current) generators sufficiently to raise generator field currents above their
Ifdmaxz = 4.60 pu (or 1.6xFull Load Current) continuous limits. If field currents exceed their high set limits.
ILM = 3.85 PU K, = 0.248 currents will be ramped down without a time delay, and time
K, = 12.6 frame 3, as described below, begins. If however field currents are
below their high set limits.current ramping will be delayed for a
The physical configuration of MxLs may vary from machine to period determined by the current level and MXL setting. During
machine, depending on exciter type and manufacturer. The model this period, generator terminal voltage will remain relatively
used in this paper is based on the MXL installed at Ontario constant, and load bus voltages continue to move towards their
Hydro's Bruce 'A' nuclear units. reference values in response to ULTC actions. For the w e under
study, voltage instability does not occur during this time frame.

Time frame 3 starts when the MXL initiates the reduction of field
current. The limiting of generator field current causes the generator
terminal voltage to be reduced. This in turn causes the load bus
voltages to go down, initiating further ULTC operations. Voltage
0
'1fd instability is most likely to occur during this time frame.

Fig. 2 : MXL Model For all the three load levels, time frame 1 runs between time 0 and
about 30 seconds, and time frame 2 starts at about 30 seconds, as
can be seen from Fig. 4.

At Load Level 1, the system never enters time frame 3 because


field currents of all machines stay below their continuous limits.
As can be seen from Fig. 4, ULTC operation restores the voltage
at bus 11 to close to its reference value at about 40 seconds, and
system voltage stability is maintained.

For the case of Load Level 2, time frame 2 runs between 30


seconds and about 180 seconds. During this time frame, ULTC
operation restores the voltage at bus 11 close to its reference value.
Fig. 3 : MXL Characteristic Time frame 3 begins when the MXL of machine 3 starts to bring
its field current down at about 180 seconds. The actuation of the
Simulation Results MXL triggers the following chain of events:

Three cases with different pre-disturbance system load levels are a. field current of machine 3 is reduced thereby reducing
considered. Total system load for each load level is : the machine terminal voltage
b. voltage at bus 11 drops below its reference value
Load Level 1 = 6655 MW and 1986 MVAR C. ULTC operates in an attempt to bring the voltage at bus
Load Level 2 = 6755 MW and 2016 WAR 11 back
Load Level 3 = 6805 M W and 2031 MVAR d. voltage and load power at bus 11 increases due to ULTC
operation
Fig. 4 to Fig. 7 show the time responses of the voltage at bus 11, e. field current of machine 3 is increased due to AVR
the field current of machine 3, the terminal voltage of machine 3, operation in response to the increase in load power at
and the tap ratio of transformer 10-11 for the cases with the three bus 11
different load levels. f. MXL reduces the field current of machine 3 and drops
the terminal voltage of machine 3
Referring to Fig. 4. three distinct time frames can be easily g. voltage at bus 11 is reduced due to the decrease in the
recognized: terminal voltage of machine 3
h. ULTC operates again
Time frame I is the period between the instant when the
disturbance occurs and the time of the first ULTC tap movement. ULTC operation continues, with bus 11 voltage being reduced,
This is the time frame normally considered in transient stability rather than increased, in response to each tap movement, indicating
simulation. At the end of this time frame, voltages at load buses the system has become voltage unstable. The voltage at bus 11
are lower than their pre-disturbance values. If load is not falls progressively until the ULTC reaches its upper limit at about
represented as constant MVA. load power also decreases because 260 seconds. After the ULTC reaches its limit, the system is
of the voltage dependency of the natural load characteristics. voltage stable as seen from Fig. 4.
Generally, load bus voltages will remain stable at their low values
until ULTCs operate. For the case of Load Level 3, time frame 2 runs between 30
seconds and about 50 seconds. As can be seen from Fig. 4, voltage
Time frame 2 is the period between the time of the first ULTC at bus 11 increases continuously during this time frame as the
1162

-
1.1-
MXL of generator 3 starts to reduce field current, as can be Seen
from Fig. 5. Limiting of the field current of generator 3 causes the Load Level 1
similar chain of events as in the case of Load Level 2. and the 5
voltage at bus 11 falls progressively until the tap reaches its upper

Load Level 1
I

Load Level 2

Load Level 3

m 0.85

Fig. 7 : Tap Ratio for Transformer 10-11


Fig. 4 : Voltage at Bus 11

::I &ad Level 3

l1.- 4

Fig. 8 Speed of Machine 3 (Load Level 3)


Fig.5 : Field Current of Machine 3
1163

Simulation Results

Table 1 shows the modal analysis results for different snapshots


corresponding to various time frames in the dynamic simulations
for the case of Load Level 1. All the eigenvalues are positive,
indicating that the system is voltage stable at all the three
snapshots. This agrees with the results obtained using the dynamic
approach.

Table 1 : Three Weakest Modes At Different Snapshots


(Load Level 1)

Eigenvalue

I 1 1 1 1 I
Snapshots
2 I 3

1
11.4 7.9 7.7
I 151.0 141.7 140.9
456.4 424.1 4228

Snapshot 1 : Pre-disturbance
Equivalent to time = 0
Snapshot 2 : Post-disturbance. before ULTC operation.
Fig. 9 : Voltage at Bus 11 Equivalent to time frame 1
Snapshot 3 : Post-disturbance, after ULTC operation.
(Load Level 3, Tap Ratio Unlimited)
Equivalent to time frame 2

Table 2 shows the modal analysis results for the case of Load
Level 2. All the eigenvalues are positive for snapshots 1-3
(corresponding to time frame 1 and time frame 2), indicating that
the system is voltage stable for these conditions. The system

*.-I
becomes voltage unstable at snapshot 4, (corresponding to a point
in time frame 3 before the ULTC reaches its limit), as indicated by
#.a## the negative eigenvalue. The system is voltage stable at snapshot
5, (all positive eigenvalues), because the ULTC has reached its
limit. This agrees with the results obtained using the dynamic
approach.

Table 2 Three Weakest Modes at Different Snapshots


(Load Level 2)

Snapshots
Eigenvalue ' 1 1 2 1 3 1 4 1 5
I 1. I 10.1 I 6.5 I 5.8 I -10.7 I 5.6 I
146.1 136.1 133.4 70.6 69.1
li? 448.3 416.1 412.1 100.9 120.1

Snapshot 1 : Re-disturbance
Equivalent to time = 0
Fig. 10 : Speed of Machine 3 Snapshot 2 : Post-disturbance
(Load Level 3, Tap Ratio Unlimited) Before ULTC operation
Equivalent to time frame 1
Snapshot 3 : Pw-disturbance
After ULTC operation
Before MXL action
3 3 static Analysip Equivalent to time frame 2
Snapshot 4 : Post-disturbance
System Modelling After MXL action
Before ULTC reaches limit
The same models used for the time domain simulation are used to Equivalent to the start of time frame 3
Snapshot 5 : Post-disnubance
obtain different snapshots following the disturbance and to form
After MXL action
the steady state system Jacobian matrix at each snapshot for the After ULTC reaches limit
evaluation of system voltage stability. Equivalent to the end of time frame 3
1164

Table 3 shows the modal analysis results for the case of Load maintain terminal voltage control, and remote machine 2 and
Level 3. The system is voltage stable for snapshots 1 to 3. machine 1 participate more heavily than machine 3.
(corresponding to time frame 1 and time frame 2), because all the
eigenvalues are positive. The system becomes voltage unstable for Table 4 : Participation Factors
snapshot 4, (corresponding to a point in time frame 3 before the (Snapshot 3 Load Level 2)
ULTC reaches its limit), as indicated by the negative eigenvalue.
The system is voltage stable at snapshot 5, (corresponding to a
point in time frame 3 after the ULTC has reached its limit), as
indicated by all the eigenvalues being positive. Again the static
results agree with the dynamic simulation results.

Table 3 : Three Weakest Modes at Different Snapshots


(Load Level 3)

Snapshots
Table 5 : Participation Factors
Eigenvalue (snapshot 4 Load Level 2)
1 2 3 4 5

11 9.3 5.7 4.7 -113 4.7

h 143.3 132.9 129.1 67.1 65.2

113 1443.8 1411.7 I 406.2 I 99.3 I 113.8

Snapshot 1 : he-disturbance
Equivalent to time = 0
Snapshot 2 : Post-disturbance
Before ULTC operation
Equivalent to time frame 1
4. CONCLUSIONS
Snapshot 3 : Post-disturbance
After ULTC operation Results of dynamic analysis using time domain simulations have
Before MXL action been presented to demonstrate voltage instability and to clarify the
Equivalent to time frame 2 influence of ULTCs. generator MXLs, and voltage dependent
Snapshot 4 : Post-disturbance loads. The same system was analyzed using a static approach in
After MXL action which modal analysis is performed for system conditions, or
Before ULTC reaches limit snapshots, which approximate points along the time domain
Equivalent to the start of time frame 3 trajectory. The results obtained using the two approaches are
Snapshot 5 : Post-disturbance shown to be consistent in indicating system voltage stability.
After MXL action
After ULTC reaches limit
Equivalent to the end of time frame 3 The static approach is shown to have a number of practical
advantages over the dynamic approach:

In addition to computing eigenvalues, the modal analysis program. 1) It is computationally less intensive.
using the eigenvectors of the reduced Jacobian matrix, computes 2) The sign of the eigenvalues computed in modal analysis
the participation factors of each bus, branch and generator to each clearly indicate if the system is stable or unstable at the given
mode [SI.Bus participations (PB) determine the areas most prone operating condition.
to voltage instability. Branch participations (eL)and generator 3) The participation factors clearly defme areas prone to voltage
participations (PG) indicate, for each mode, the branches and instability and indicate elements which are important in the
generators which have the largest impact on the voltage stability instability phenomenon. This provides insight into the
of that mode. The participation factors provide important mechanism of instability.
information regarding the mechanism of voltage instability.
Although time domain simulations are necessary for detailed
For the case of Load Level 2, Table 4 and Table 5 show the bus, dynamic analysis and studies requiringcoordinationof controls and
branch and generator participations for the weakest mode at protections, the above advantages make static analysis a suitable
snapshot 3 and snapshot 4, respectively (refer to Table 2 for the choice for the bulk of system studies where the examination of
descriptions of snapshot 3 and snapshot 4). From the eigenvalues system performance under a large number of conditions.
shown, we see that the system is voltage stable at snapshot 3 and
voltage unstable at snapshot 4. 5. REFERENCES

Bus participations indicate that the weak area consists of bus 11 IEEE Special Publication 90TH0358-2-PWR, "Voltage
and bus 10, which agrees with the dynamic simulation results Stability of Power Systems : Concepts, Analytical Tools. and
shown in Fig. 4. At both snapshot 3 and snapshot 4, line 9-10 is Industry Experience". Prepared by the Working Group on
the branch with the highest participation factor to the critical Voltage Stability, 1990.
mode, indicating that this branch is most critical in maitaining the
voltage stability of the critical mode. As for generator N. Flatabo, R. Ognedal. and T. Carlsen, "Voltage Stability
participations, local machine 3 participates the most at snapshot 3 Conditions in a Power Transmission System Calculated by
because its field current is not limited at this snapshot. Once the Sensitivity Methods", IEEE Trans. on Power Systems, Vol. 5 ,
field current of machine 3 is limited at snapshot 4, it is unable to No. 4. November 1990.
1165

P. Kessel, and H. Glavitch, "Estimating the Voltage Stability Shunts


of a Power System", IEEE Trans. on Power Delivery, Vol. Bus MVAR
PWRD-1, NO. 3, pp. 346-354. July 1986. 7 763
8 6 0 0
9 1710
R.J. Thomas and A. Tmnuchit, "Voltage Instabilities in hds
Electric Power Networks", Proc. Eighteenth Southeast Bus P O QWW
Symposium on System Theory, pp. 359-363, 1986. 8 3271 1015 (Load Level 1)
3320 1030 (Load Level 2)
Y. Tamura, H. Mori, and S. Iwamoto, "Relationship Between 3335 1035 (Load Level 3)
Voltage Instability and Multiple Load Flow Solutions in 11 3384 971 (Load Level 1)
3435 985 (Load Level 2)
Electric Power Systems", IEEE Trans. Vol. PAS-102, No. 5 , 3460 993 (Load Level 3)
pp. 1115-1125, May 1983. Load Flow Generator Data
Bus P o Vku)
R.A. Schlueter, et. al., "Voltage Stability and Security 1 3981 0.9800 (Laad Level 1)
Assessment", EPRI Final Repoxt EL-5967 on Project 1999-8, 4094 0.9800 (Load Level 2)
4152 0.9800 (Load Level 3)
May 1988.
2 1736 0.9646 (Load Level 1)
1736 0.9646 (Load Level 2)
C. Lemaitre, J.P. Tesseron, Y.Harmand, and Y.S.Zhao, "An 1736 0.9646 (Load Level 3)
Introduction of the Risk of Voltage Profile Instability for 3 1154 1.04oO (toad Level 1)
Real-Time Control Applications", IEEE Summer Meeting 1154 1.04oO (Load Level 2)
1989, Paper 89 SM 713-9 PWRS. 1154 1.04oO (Load Level3)
Mochine Parametem
Machine 1 : Infinite Bus
B. Gao. G.K. Morisoo, and P. Kundur, "Voltage Stability Machine 2 : H = 2.09. MVA Base = 2200 MVA
Evaluation Using Modal Analysis", 1991 IEEE PES Summer Machine 3 : H = 2.33. MVA Base = 1400 MVA
Meeting, Paper 91 SM 420-0 PWRS, July 28-August 1,1991, The following stator and rotor parameters apply to both machine 2 and
San Diego, California. machine 3 :
R, = 0.0046 X, = 2.07 Xq = 1.99
X - 0.155
M.K. Pal, Discussion to Above Reference [SI.
! .-
X = 0.215
X', = 0.28
X", = 0.215
X,' = 0.49

[lo] P. Kundur, G.J. Rogers. D.Y. Wong, M.G. Lauby, "A T', = 4.10 T = 0.56
,
'
Comprehensive Power System Stability Analysis Computer
Programs Package", Proceedings of the Power Plant & Power T", = 0.033 T", = 0.062
System Training Modelling & Simulation Conference, Miami Exciters
Beach, Florida, April 17-19. 1991. Both machine 2 and machine 3 have thyristor exciter with a gain of 400
and the sensing circuit time constant of 0.02 second.
[ll] M. Stubbe, A. Bihain, T. Deuse, J.C. Baader, "STAG A New
Unified Software Program for the Study of the Dynamic
Behaviour of Electrical Power Systems", IEEE Trans. on
Power Systems, Vol. 4, No. 1, February 1989.

(121 C.W. Taylor, "Voltage Stability : Load Characteristics and Kip Morison joined the Power System Planning Division of
R&D Needs", Paper Presented at EPRVCRIEPI Joint Ontario Hydro in 1981 and received his M.A.Sc. degree from the
Symposium on Power System Dynamics and Stability, University of Toronto in 1985. He currently works as a system
Monterey, California, May8-10, 1990. studies engineer in the Analytical Methods & Specialized System
Studies Department, and is involved io special system stability and
[13] C. Concordia. Comments on Above Reference [SI, Private conhol studies.
Communication.
Baofu Gao received his M.A.Sc. degree from the University of
APPENDIX Toronto in 1986. In 1990 he joined Ontario Hydro where he now
TEST mSTEM DATA works in the Analytical Methods & Specialized Studies
Department in the Power System Planning Division. He is also
Trausmissim Lmes (R, X & B in pu on 100 MVA Base) studying part-time at the Department of Electrical Engineering,
R X B University of Toronto for his Ph.D. degree.
5-6 o.oo00 o.Oo40 o.oo00
6-7 0.0015 0.0288 1.1730 Prabhashankar Kundur received the M.A.Sc.and Ph.D. degrees
9-10 0.0010 0.0030 O.oo00 from the University of Toronto, Canada io 1965 and 1967
Trausfonnels (R & X in pu on 100 MVA Base) respectively. He taught at Mysore and Bangalore Universities
R x Ratio during 1967-1969. In 1969, he joined Ontario Hydro where he is
1-5 O.oo00 0.0020 0.8857 cunently Manager of the Analytical Methods & Specialized
26 O.oo00 0.0045 0.8857
3-7 O.oo00 0.0125 0.9024 Studies Department in the Power System Planning Division. He
7-8 O.oo00 0.0030 1.0664 also holds the position of adjunct Professor at the University of
7-9 O.oo00 0.0026 1.0800 Toronto. Dr. Kundur was elected a Fellow of the IEEE in 1985
10-11 O.oo00 0.0010 0.9750 (Load Level 1) and is a member of several IEEE working groups and task forces.
0.9938 (Load Level 2) He is also a member of C i p Task Force 38-02-10 on Modelling
1.oo00 (Load Level 3) of Voltage Collapse Including Dynamic Phenomena.
1166

Discussion [A] M. K. Pal, "Voltage Stability Conditions Considering


Load Characteristics,'' IEEE Trans. on Power Systems,
M. K. Pal: This paper presents results of a voltage stability Vo1.7, No. 1, pp.243-249, Feb. 1992.
analysis using static and dynamic techniques. The paper firmly [B] M. K. Pal, Discussion of "An Investigation of Voltage
suggests that a static method is a valid approach to voltage Instability Problems," by N. Yorino, et al., 91 W M 202-2
stability analysis irrespective of the dynamic response PWRS, IEEEIPES winter meeting, February, 1991.
characteristics of the loads. The authors respond to the
comments made on their earlier work [9]. It appears that the
comments were largely misinterpreted. Indeed, provided the M. G. Lauby (EPRI, Palo Alto, CA): The authors are congratu-
load characteristic is favorable, a static analysis is all that is
required. It has been clearly demonstrated in [A] that when the lated for their paper comparing static (gradient) and dynamic
composite load exhibits slow dynamics, both small and large techniques to analyze voltage stability of the power system. The
disturbance voltage stability are assured by the existence of a authors have prepared a paper with rich engineering judgment
stable equilibrium point in the post-disturbance system, which and solid engineering mathematics. By understanding the char-
can be determined by static means. This is not true if the bulk acteristics of voltage stability, the authors are able to create new
of the load exhibits fast dynamics. There can be considerable approaches in the analysis and evaluation of this phenomena.
error in determining voltage stability by static means in such The static (gradient) technique closely approximates points
situations. For such loads a detailed dynamic analysis is
required. Simply reflecting the loads' static characteristics and along the time domain trajectory by solving a set of steady state
control status in the power flow model is not enough. equations with models for controls and limits. For very long
Consideration of load dynamics is also essential to correctly simulations (e.g. twenty minutes) in large systems (200 to 500
explain the system response following certain corrective actions, machines), what is the accuracy of the approximations? Is it
e.g. capacitor switching under voltage collapse conditions, possible for the static simulations to diverge from the actual
although an actual dynamic analysis may not be necessary. trajectory?
In the authors' example system, voltage stability is obviously
determined by the characteristic of the load at bus 11. The load The power system may also be unstable (e.g. transient, or
at bus 11 is a static load, rendered dynamic by ULTC action, long-term). The static technique can not identify this condition.
whose response speed is slow compared to that of the generator However, as with Small Signal analysis, the user of the static
control system. In this system, therefore, voltage stability technique must initially assume that the system is stable and
determined by static and dynamic means would be identical. The dynamic analysis must be performed to determine transient
authors' analysis supports the conclusion of [A]. (Actually, the stability. What are some of the system design issues that need to
snapshot approach is unnecessary. All that is needed is to ensure
that the post-disturbance equilibrium is possible and that it is be considered when building future power systems resilient to
stable, taking into account of the equipment limits.) voltage stability?
The agreement between static and dynamic analysis as Voltage stability appears to be a localized phenomena. That
presented in the paper is due to the specific load characteristics is, a local area is affected first, finally dragging the system down.
chosen. Any generalized conclusion based on this particular This characteristic might be used when calculating the steady
example can do great harm, if interpreted as being valid in state equations, especially for large systems. Have the authors
every situation irrespective of the load dynamic response taken advantage of this aspect in their analysis? Would this
characteristic. Had the authors repeated the analysis by
interchanging the loads at buses 8 and 11, the conclusion might approach be more efficient or will there be unacceptable accu-
have been different. racy? Are there short-comings to using the localized approach?
Note that the basic premise on which this paper as well the What about using this approach for on-line contingency selec-
authors' previous paper [8] is based, viz., positive eigenvalues tion of voltage stability?
of the power flow Jacobian (or the reduced Jacobian) indicates The authors have prepared an exciting paper. By using the
voltage stability, and that the voltage collapse point is static techniques described, the user is not forced to run thou-
determined by the singularity of the Jacobian, has no theoretical
basis, as has been clearly demonstrated in [B]. It is not difficult sands of transient stability cases for long periods after the
to show that, depending on the load response characteristics, the disturbances to detect voltage stability. Their comments on the
voltage stability limit can be well below the singularity point of above would be very much appreciated.
the Jacobian. Note also that the modal analysis as used by the Manuscript received July 24, 1992.
authors is indeed a static technique. The response of voltage to
incremental changes in power as indicated, for example, by
equation (6), is valid only as a criterion for the purpose of static
analysis. Although equation (6) shows that this response can Adam Semlyen (University of Toronto): I would like to congratulate
change sign with changes in the sign of the eigenvalues, in the authors for their interesting and useful paper. It presents both static
actual operation, except under specialized control, this sign and dynamic approaches for voltage stability analysis in the context of
change does not occur. Note that when an eigenvalue is zero, their consistency. The following remarks pertain to the problem of
the right hand side of equation (6) is infinity, which, of course, modal analysis presented in the paper.
will not happen in real life.
A note about the use of constant MVA static load model in a The matrix JR is almost symmetrical so that its eigenvalues are
dynamic analysis: A constant MVA load is not a static load. It probably always real and the left and right eigenvector matrices are
requires a valid dynamic representation in a dynamic simulation. close to orthogonal. Therefore, the method of modal analysis has, in a
The use of a static model for such loads can lead to results that practical sense, most of the advantages of Singular Value
are physically impossible. Even when the results look normal, Decomposition to be discussed below. Consequently, the remarks that
there can be considerable error when compared with the results follow are primarily of conceptual significance.
obtained from a model that represents their true response
characteristics. In order to refer to a single mode, equation (6) of the paper is rewritten
In general, the authors' method will determine the below for i= 1 (A, =A,,,,.<<h; this separation of the first mode reflects
loadability limit without regard to stability. In specific cases, the fact that the next one, even if excited, will have a much smaller
such as the one presented in the paper, this limit will also be the output - otherwise, the procedure applies for a group of modes):
voltage stability limit. The loadability limit depends on the load
level as well as the distribution of loads throughout the system,
which is rarely known precisely. As such, the determination of
the exact limit based on an assumed load distribution is not very
meaningful. Conventional power flow can get fairly close to the
limit very easily. The product 7,AQ is a scalar k which becomes an input to eqn.(a). It is
1167

thus clear that AV is proportional to i,,as stated. This is the direction Herminio J.C.P. Pinto & Nelson
I Martins (CEPEL - Centro de
of the modal output. Pesquisas de Energia Eletrica, Rio de Janeiro, FU, Brazil): The
The direction of the modal input AQ is uniquely defined if it is to authors are to be congratulated on this fine paper, which reflects their
maximize the input scalar k for a constant magnitude of I AQl . The large experience in this field. We share the authors' views on the
corresponding direction is then up along the steepest slope, or benefits gained from the coordinated use of both linearized and non-
gradient, in the parlance of the paper. This means that AQ should have linear system analysis to the voltage stability problem.
the direction q l T which, consequently, is the direction of the modal
input. It can be checked that the direcrion Il for AQ, while it has the Our voltage stability work is presently very much related to that of [8].
merit that it does not excite another mode, gives a smaller output IAVI We have produced a simple, though mathematically rigorous, dynamic
- often significantly smaller (the strongly unsymmetrical matrix [ y system model for the voltage stability methodology proposed by the
authors in [8] This dynamic model provides an answer to the
may serve as an example) - for the same magnitude of 1 AQl . The criticisms made in [9] and [13] on the methodology of [8], regarding
direction q l r is therefore the best choice for identifying the most the dynamic nature of a stability problem.
efficient controls to prevent voltage instability.
Of course, this definition of a modal input is at variance with the We will be soon reporting on our work, but feel that a discussion to
fundamental concept of an eigenvector which, as an input to a matrix, this paper is appropriated at this time.
gives an output of the same direction. This is also the approach
adopted in the paper. It seems that the requirement of optimal input The authors carry out modal analysis on the system equations:
conflicts with that of the independence of the individual modes! The
clarification of this problem by the authors would be most welcome. SI + a-'= 0 (A)
It would appear that conceptually more satisfactory results could be
obtained by using Singular Value Decomposition (SVD) for the where JR is the reduced Jacobian matrix defined in section 2 of the
analysis of input-output type of problems as the one examined in the paper. Equation (A) describes the dynamics of the system model
paper in question. With conventional notations, we have represented in the block diagram below:
y=AX (b)
where
A = UZVT (C)
Accordingly (see my discussion to Ref. [A]), the output modes are
given by the columns of the ortho-normal matrix U while the input
modes are defined by the columns of the ortho-normal matrix V. The
individual modes are orthogonal to each other and thus completely
independent. Moreover, since multiplications with ortho-normal
matrices preserve the Euclidean norm, the magnitude of the output
vector y is directly related to the magnitude of the input vector x of
mode i by the correspondent singular value ai. It may be useful to note
that U, V, and E are related to the eigenset of the matrix ArA as Figure A. Simple Dynamic Model Structure foI
follows: U and V are the matrices of its left and right eigenvectors, and Voltage Stability Analysis
E is the matrix of the square roots of its eigenvalues which are all
positive. If A is symmetrical, then its singular values U are equal to its Please note some analogy to the classical multimachine electro-
eigenvalues A. mechanical stability problem, described by equations.
The use of SVD has the following advantages:
It gives full modal information for outputs, inputs, and their ratio. s21 + K s = 0
The singular values represent precisely the ratio of the magnitudes where K S is the matrix of synchronizing torque coefficients
of output to input for any given mode.
The modes can be ranked by the associated singular values and are Let us now describe the dynamic behavior of our simple load model,
fully decoupled. considering the case of a single load bus:
The transfer matrix A does not have to be square, so that the effect
of a reduced number of input variables can be investigated.
The largest singular values and their associated singular vectors
can be directly calculated by simple iterative procedures.
If the largest singular values are exactly or nearly equal, an
invariant subspace for best control vectors can be determined. I 1
Inputs and outputs can be easily interchanged since U and V are I I
orthogonal matrices.
L
Because of the above listed features of the SVD, the reduction of the
original Jacobian matrix in eqn.(l) to the reduced Jacobian of eqn.(2) I

may not be necessary.


Figure B. Dynamic Model for a Single Load Bus
[A] P.-A. Lof, G. Andersson, and D.J. Hill, "Voltage Stability
Indices for Stressed Power Systems", IEEE paper no. 92 WM
101-6 PWRS, presented at the 1992 IEEE/PES Winter Meeting in
The gain k, in Figure B, is a real number equal to the derivative of the
New York, NY. QxV curve at the operating point around which the system has been
linearized. The load considered has a constant active power
Manuscript received J u l y 27, 1992. characteristic, i.e.,it is independent of the bus voltage magnitude.
1168

The closed loop transfer function of the system of Figure B is: Quasi-dynamicmethods without modeling of fast dynamics
is often effective--full dynamic simulation can be used to
confirm results.
The authors model about half of the load (both real and re-
active parts) a s static, constant power. This is often consid-
The closed loop system pole is therefore equal to - 1 / k , and the closed ered a poor approximation of induction motor load in
loop transfer hnction has zero gain at steady state (washout block) dynamic simulation. (Adjustable speed drives and power
Our linear dynamic load model has therefore the characteristc of electronic loads may, however, approach this model.) Could
automatically bringing the bus voltage deviation back to zero following t h e authors discuss t h e effect of realistic dynamic loads in
a step change in the load MVAr demand This is therefore a simplified both the time domain simulation and t h e modal analysis?
model of an OLTC action Note also that a system which is voltage
Figure 3 show a linear decrease in field current. I think this
stable will have eigenvalues in the left half plane, which is coherent
is only true for step changes into the integrator of Figure 2.
with modal analysis
As a side issue, can the authors confirm that, for load level
Many other first order dynamic models may also be proposed for the 2 or 3, and with tap changer a t limit, adding load will de-
load, all of which have different dynamic performances but provide crease load power, and inserting a capacitor bank will raise
equivalent information regarding the voltage stability condition of the voltage and load power? For a similar system (but with in-
system. duction motor rather t h a n static constant power loads), ref-
erence A demonstrates these effects.
Another point worth mentioning is the benefit of using transfer The interpretation of voltage stability for discontinuous tap
function residues or controllability factors [A] rather than participation changing with limits requires discussion. Modal analysis
factors to produce the various ranking lists needed (critical generators, applied a t a point in time between t a p changes (e.g., snap-
critical lines, critical loads, etc.). shot 4 for load level 2 ) should, mathematically, be stable
just as snapshot 5 is stable. How does modal analysis treat
We suggest that the terminology used by the authors (static approach the discontinuous tap changing? Regarding t a p changer
to voltage stability) be changed to miall signal voltage stability The limits, can we say that modal analysis shows periods of
suggested terminology will then be in harmony with modal analysis and voltage decay (along with generatiodnetwork participa-
with the well established small signal electromechanical stability tions) of a stable nonlinear dynamic system?
analysis [B]

Reference:
Martins, N. and Lima, L.T.G., Determination of Suitable A. C. W. Taylor, A Conceptual Analysis of Voltage Stability as
Locations for Power System Stabilizers and Static VAr Related to Load Characteristics, published by the North
Compensators for Damping Electromechanical Oscillations in American Electric Reliability Council in Survey of the Volt-
Large Scale Power Systems, IEEE Trans. on Power Systems,, age Collapse Phenomenon, August 1991.
Vol. PWRS-5, pp 1455-1469, November 1990.
Manuscript received August 10, 1992.
Martins, N. and Lima, L.T.G., Eigenvalue and Frequency
Domain Analysis of Small-Signal Electromechanical Stability
Problems, IEEE Symposium on Application of Eigenanalysis
and Frequency Domain Methods for System Dynamic M. Y. Vaiman and V. A. Faybisovich ( V & R Energy System
Performance, publication 90 TH 0292-3 PWR, pp 17-33, 1989 Research, Los Angeles, CA): We have some remarks to a very
interesting article by G. K. Morison, B. Gao, P. Kundur.
Manuscript received August 10, 1992. Remark 1. This paper compares voltage stability analysis of
power system using static and dynamic technique. For small
example authors show that the results of comparison are consis-
Carson W.Taylor, Bonneville Power Administration, Port- tent. However, the authors dont describe the conditions when
land Oregon: The authors demonstrate some important as- results obtained by using the dynamic and static methods are
pects of voltage stability. similar. Also the authors dont describe the conditions when the
A power system is a dynamic system with many complexi- results, that were obtained by using these approaches are not
ties (e.g., nonlinearities including discontinuous equip- consistent.
Only in the cases when violation of stability is a periodic one,
ment). Simulation, particularly static voltage stability
simulation, requires significant approximations. The au- the results obtained by the dynamic and static methods are
consistent. Indeed, all eigen-values of the matrix of the lin-
thors demonstrate that linearized sensitivity type static
earized steady state system power voltage equations are real
methods can, however, be useful in voltage stability analy-
ones (not complex) and change of their signs indicates the
sis. Weak points in the generatiodtransmission system are
stability violation. If the stability violation is oscillatory one we
identified.
must use only dynamic approach [l].
The authors give several reasons for more accurate time Remark 2. To use steady state system power voltage equations
domain simulation. Since effective (i.e., cost-effective) volt- for voltage stability analysis is better to utilize another tech-
age stability solutions often involve various control meth- nique. In monographs [2, 31 is shown an interesting property of
ods, longer-term dynamic simulation is no doubt the most stressed power system comprising the fact that the surface of
important tool for system design and demonstration. We potential function of its conservative model has not a conven-
are using the ETMSP 3.0 program for design of undervolt- tional shape of potential well but of a trough(groove) - shaped
age load shedding and automatic capacitor bank switching. one. Such a trough goes from the minimum of potential energy
1169

[31 M. Y. Vaiman, The Application of Lyapunov Function in


the Analysis of Stability of Electrical Systems Using Com-
I I puter Simulation, Published by University Education,
I
, I Moscow, 1992 (in Russian).
[41 M. Y. Vaiman, A. Z. Krupenin, Choice the Dangerous
Direction while Stressing Regime of Multi-machine Electri-
cal Systems, 10-th Scientific Conference Power System
Simulation, Kaunas, October 15-17, 1991 (in Russian).

Manuscript received August 11, 1992.

G. K. Morison, B. Gao, and P. Kundur: We would like to thank


all discussers for their interesting comments on the paper. We
address each discussion below.

Mr. Carson Taylor: The first question by Mr. Taylor concerns


the effect of induction motor load on the results of voltage
stability evaluation using both time domain simulation and modal
fi-y. / analysis. Because the main purpose of our paper was not the
investigation of the effect of load model on system voltage
stability, constant MVA load was used for one of the two loads
just to allow us to better focus on the effects of other devices
such as ULTC and generator MXL. Constant MVA is indeed a
very poor way of modelling induction motors. This is particularly
I
true at low motor terminal voltages because constant MVA
model does not capture the sharp increase in motor reactive
power consumption at low terminal voltage. Both our time
domain simulation program and modal analysis program use
appropriate detailed induction motor models. For time domain
simulation, the model captures the actual sharp increase in
motor reactive power consumption as the motor speed decreases
due to low terminal voltage. For modal analysis, effect of induc-
tion motors on system voltage stability is taken into account by
including the linearized induction motor model in forming the
\
, system Jacobian matrix. The sharp increase in motor reactive
\ power consumption at low motor terminal voltage is reflected by
a high negative sensitivity of motor reactive to the change in
motor terminal voltage.
Comments regarding Figure 3 are correct.
It is true that, for load level 2 or 3 with tap changer at limit,
(the stable equilibrium position of the model) to the criteria1 inserting a capacitor bank will raise voltage and load power.
saddle (one of the unstable equilibrium positions situated upon Modal analysis results at snapshot 5 for both load levels have
the boundary of the stability region). Fig. 1 illustrates the sur- also confirmed this point. However, the statement that under the
face of potential function of stressed system. Fig. 2 illustrates same conditions load power decreases as load admittance in-
the surface of potential function of non-stressed power system. creases has to be taken with caution, because the change in load
Fig. 1 and Fig. 2 are taken from [31. Comparing Fig. 1 and Fig. 2 power depends not only on the magnitude of the added admit-
one can deduce that stressed system has a trough shape form tance but also the power factor of the added load admittance.
and non-stressed system has a well shape form. ULTC model used in modal analysis assumes that the tap
In [4] this result is generalized on the steady state system position is a continuous linear function of bus voltage. While in
power voltage equations. While stressing regime the minimum reality taps always move in discrete steps, only a continuous
and criterial saddle are approached one to another. When they model can be included in forming the Jacobian matrix. Because
are merged the system is on the verge of voltage stability. the ULTC operation is assumed to be continuous, taps should
This approach allows to determine stability margin and condi- be assumed adjustable between two consecutive tap movements
tions when the system is on the verge of voltage instability. for snapshot 4.
This approach is clearly physical, it is simple for computer
simulation and there is no necessity to linearize steady state
system power voltage equations. Mr. Mark Lauby: The quasi dynamic, or snapshot approach, is
intended to be a practical compromise between the detail of
References time domain simulation, and the relative simplicity of so called
static approaches. The snapshot method embodies many of the
[ 11 V. A. Venikov, Transient Electromechanical Processes in advantages of these methods (such as comprehensive informa-
the Electrical Systems, Moscow, Published by University tion regarding mechanism of instability), while using approxima-
Education, 1985 (in Russian). tion to eliminate disadvantages (such as high CPU require-
[2] M. Y. Vaiman, Analysis of Systems Stable with Finite ments). Approximation means that snapshots will be unable to
Disturbances, Moscow, Science Press, 1981 (in Russian). exactly duplicate the time domain trajectory, and as systems get
1170
larger, and simulated time longer, these approximations are approximated by Equation (a) in Prof. Semlyens discus-
expected to result in increasing deviation from the true trajec- sion.
tory. However, with the proper accounting of discrete and con- AQ is along the direction of a right singular vector 6,,AV
tinuous controls, and of device timers, etc., we fully expect that is along the direction of the corresponding left singular
the technique can, as far as system voltage stability is concerned, uector p L ,and I AV I = I AQ I / U , , with U, the ith singu-
(1) determine whether the system can reach a final stable lar value of the matrix.
operating condition and (2) provide sufficient information at Each of these modal definitions has its advantages and disad-
intermediate points between the occurrence of the disturbance vantages as discussed below:
and final system steady state or system collapse.
It is quite possible that a system becomes transiently unstable

Because Equation (a) is not exact, q does not provide the
most sensitive direction of AQ which will produce the largest
before reaching the steady state determined by static analysis output AV. Also AQ along the direction of a left eigenvector
methods. In this regard, one still requires dynamic simulation to produces AV which is a linear combination of all the right (or
determine this. By design, the snapshot approach is intended to left) eigenvectors. Therefore, left eigenvectors have little use in
focus on the problems related to voltage stability. This decou- identifying the direction of AQ which will result in the largest
pling of phenomena for purposes of analysis allows the applica- AV and decoupling of different modes. The left eigenvectors,
tion of specialized and efficient methods tailored to assess however, can be used to determine the shortest distance of an
specific aspects of stability. As far as planning against system operating point to the hypersurface of saddle node bifurcations
voltage instability is concerned, our opinion is that the snapshot [I].
approach is adequate. The right singular vector 6, corresponding to the minimum
Though voltage stability can be considered a local phe- singular value m1 provides the direction of AQ which results in
nomenon in that a local area is affected first which may drag the largest AV. Therefore the direction of 6, is along the
down other parts of the system, the elements contributing to, steepest slope. However, because all the singular values are
and participating in, the instability, may be quite remote from positive, the SVD approach can not identify whether a mode is
the local area. For this reason, localized solution methods has stable (with positive eigenvalue) or unstable (with negative
not been considered in determining the network conditions. eigenvalue).
However, selective modal analysis has been considered, in which Because the reduced Jacobian matrix JR is very close to being
modal analysis is performed on a reduced portion of the system symmetric, our approach of using AQ along the direction of the
after the operating condition has been determined. By perform- right eigenvectors as modal inputs has practically all the advan-
ing modal analysis on local parts of the Jacobian, most mode tages of the SVD approach.
information is preserved and CPU time requirements could be
drastically reduced.
For on-line contingency selection, it is possible to envision the Mr. Pinto and Dr. Martins have proposed an arbitrary but
use of the snapshot approach coupled with modal analysis, useful dynamic load model. The development of this simple
provided CPU requirements can be controlled sufficiently. The dynamic load model has made it possible to relate the eigenval-
use of reduced system models and selective modal analysis ues of JR to the poles of the closed loop transfer function of the
would likely make this possible. However, it is expected that, as dynamic system.
with transient stability, on-line analysis would have to be supple- The idea of using transfer function residues rather than the
mented with detailed off-line studies. participation factors is definitely worth considering. Our modal
analysis program computes the right and left eigenvectors corre-
sponding to the small eigenvalues. It would be very easy to add
the calculation of the transfer function residues to the existing
Professor Semlyen: We find Prof. Semlyens discussions very program and compare the results of transfer function residues
thought provoking. It is becoming clear that most of us have no with those of participation factors. However, we promote the use
difficulty in accepting the essence of the voltage stability analysis of participation factors because of their strong physical meaning
technique based on the modal analysis of the reduced Jacobian which may be more practical than the more mathematical con-
matrix. However, Prof. Semlyen has pointed out that there need cept of residues.
to be further clarifications as to how a mode of voltage stability
should be best defined.
The modal analysis technique as applied in this paper exam- Mr. M. Y. Vaiman and Mr. V. A. Faybisovich: Our answer to the
ines the relationship between the input vector AQ and the first question by Mr. Vaimain and Mr. Faybisovich is the follow-
output vector AV. In our definition AQ for each mode is along
the direction of a right eigenvector [,, and the corresponding AV
is AQ/A,. If A, > 0, AV and AQ are along the same direction,
.
ing:
Voltage instability and small signal oscillatory instability are

and the system is voltage stable because the increase in reactive


power injection at a group of buses causes the voltage magni-
. two different types of power system stability problems.
Our modelling and analytical techniques were designed to
study system voltage stability and therefore can not be directly
tudes of the same group of buses to increase. If A, < 0, AV and applied to study small signal stability problems.
AQ are along the opposite directions, and the system is voltage
unstable because the increase in reactive power injection at a We are sorry that, because we do not fully understand the
group of buses causes the voltage magnitudes at the same group authors second question, and because the references are not
of buses to decrease. Therefore, mode of voltage stability de- readily available, we can not provide additional comments.
fined based on the right eigenvectors of JR makes it possible to
use the sign of the eigenvalue to determine the stability of each
Dr. M. K. Pal: Regarding Dr. Pals comment regarding the
mode.
inadequacy of reflecting dynamic characteristics into the power-
Prof. Semlyen has suggested the following two alternative
flow model when attempting to capture fast dynamics, we have
definitions of the mode of voltage stability:
the following comment. Fast dynamics which result in such
(1) AQ is along the direction of a left eigenvector q , AV is things as sudden increases in reactive power consumption are
1171

indicated using modal analysis by high Q / V sensitivities. Provid- pute using a particular analytical technique reflect actual system
ing the snapshots, on which modal analysis is performed, are problems and the solutions we come up with are meaningful.
obtained frequently enough so that these sensitivities are re- Dr. Pals last comment is that exact limits obtained based on a
flected in subsequent operating points, then the method will assumed load distribution is not very meaningful. Dr. Pal clearly
indeed capture fast dynamics. overlooks the fact that virtually all power system transmission
Regarding the comment on the generalization of the findings, and generation planning, as well as operating, is performed
there is no reason to believe that the results will not be consis- using assumed load distributions and growth rates based on load
tent for different types of load, providing the correct load forecasts. In this regard, a stability margin obtained using load
dynamic characteristics are used in obtaining the snapshots (as forecast information may in fact be the most practical measure
in dynamic simulation) and incorporated into the Jacobian used available to planners and operators.
for modal analysis. Dr. Pal again raises the comment that he
believes the singularity of the Jacobian is not a suitable indicator References
of voltage instability, particularly when different load character-
istics are present. This issue has been addressed many times [l] I. Dobson and Liming Lu, Using an Iterative Methods to
Compute a Closest Saddle Node Bifurcation in the Load
before, including in the closure of [2]. It is essential to under-
Parameter Space of an Electric Power system, Paper pre-
stand that, unlike the conventional powerflow Jacobian, lin-
sented at Voltage Stability and Security NSF Workshop,
earized load characteristics are included in the reduced Jacobian
used for modal analysis. Indeed if this were not the case, the Deep Creek Lake, MD, August, 1991.
singularity would not correctly correspond to instability. Inclu- [2] B. Gao, G. K. Morison, P. Kundur, Voltage Stability
Evaluation Using Modal Analysis, Paper 91 SM 420-0
sion of linearized dynamic characteristic in the Jacobian is
PWRS presented at the 1991 IEEE/PES 1991 Summer
provided to ensure that singularity corresponds to instability, for
that particular load type. It is important to keep in mind the Meeting, San Diego, California, July 28, 1991.
practical aspects of the problem; whether the results we com- Manuscript received October 1, 1992.

Potrebbero piacerti anche