POWERENG 2007, April 12-14, 2007, Setubal, Portugal

of Power System Voltage

Stability Using Modal Analysis

Determination

Chandrabhan Sharma, Senior Member, IEEE, and Marcus G. Ganness

Abstract-- Voltage instability within the power system has severe

consequences including a wide-spread low voltage profile and

system blackout. The development and use of accurate methods
to predict incipient voltage instability is crucial in preventing
such voltage collapse situations. A comparison of the common
steady-state methods for voltage stability analysis was
undertaken, along with a brief examination into the dynamic
approach. Special attention is given to the modal analysis
technique presented in [2], which provides both a relative
proximity of the system to instability, as well identifies the key
contributing factors to instability. A PC-based MATLAB
prototype application was developed to assist power system
operators in the assessment and analysis of the voltage stability of
a power network, at any given operating point, using the modal
analysis technique, in addition to useful features such as power
flow analysis and traditional V-PNV-Q analysis. Using multiple
practical test cases on the IEEE 30 bus system the modal analysis
technique was shown to provide an accurate estimate of the
system proximity to the voltage stability limit, and to consistently
and correctly predict the critical buses and the weakest branches
in the transmission system. These results clearly demonstrate the
suitability of modal analysis technique as applied to voltage
stability analysis, and the need for closer analysis, possible
refinement to the method and incorporation into more
commercial analysis packages.

Index Terms-voltage stability, voltage instability, voltage

collapse, steady state voltage stability analysis, modal analysis, QV curve.
I. INTRODUCTION

The increasing number of power system blackouts in many

countries in recent years, is a major source of concern. Power
engineers are very interested in preventing blackouts and
ensuring that a constant and reliable electricity supply is
available to all customers. Incipient voltage instability, which
may result from continues load growth or system
contingencies, is essentially a local phenomenon. However,
sequences of events accompanying voltage instability may
have disastrous effects, including a resultant low-voltage
profile in a significant area of the power network, known as
the voltage collapse phenomenon. Severe instances of voltage
collapse, including the August 2003 blackout in North Eastern U.S.A and Canada, have highlighted the importance
of constantly maintaining an acceptable level of voltage
stability. The design and analysis of accurate methods to
evaluate the voltage stability of a power system and predict
incipient voltage instability, are therefore of special interest in
the field of power system protection and planning.
1-4244-0895-4/07/$20.00 C 2007 IEEE

381

The voltage stability of a power system refers to its

ability to properly maintain steady, acceptable voltage levels
at all buses in the network at all times, even after being
subjected to a disturbance or contingency. A power system
may enter a condition of voltage instability when the system is
subjected to a steady increase in load demand or a change in
operating conditions, or a disturbance (loss of generation in an
area, loss of major transformer or major transmission line).
This causes an increased demand in reactive power. Voltage
instability is characterized by gradually decreasing voltage
levels at one or more nodes in the power system. Both static
and dynamic approaches are used to analyze the problem of
voltage stability. Dynamic analysis provides the most accurate
indication of the time responses of the system [1]. Dynamic
analysis is therefore extremely usefulfor fast voltage collapse
situations, following large disturbances such as loss of
generation and system faults, when specific information
concerning the complex sequence of events leading to
instability, is required. Dynamic simulations however, fail to
provide information such as the sensitivity or degree of
stability. More importantly, dynamic simulations are
extremely time consuming in terms of CPU and engineering
resources required for the computation and analysis of the
several differential and algebraic equations needed for
quantification of the phenomenon.
In most cases, the system dynamics affecting voltage
stability are usually quite slow and much of the problem can
be effectively analysed using the static approaches that
examine the viability of a specific operating point of the power
system. Static analysis methods, in addition to providing
information such as sensitivity or degree of stability, involve
the computation of only algebraic equations and are much
more efficient and faster than dynamic approaches. The static
analysis approach is more attractive than the dynamic
method, and well suited to voltage stability analysis ofpower
systems over a wide range of system conditions [1], [2].
Many static approaches for voltage stability analysis are
presented in the literature, most utilizing a variation of the
same common principle. Some popular static methods are
summarized and contrasted in this document. The modal
analysis of the load flow Jacobian presented by Gao, Morisson
and Kundur in 1992, is of particular interest. This method, in
addition to providing an accurate estimate of the system
proximity to instability using the system eigenvalues,
identifies the elements of the power system contributing the
most towards incipient voltage instability (critical load buses,
branches and generators). A prototype package for predicting
incipient voltage stability is developed incorporating modal

analysis as well as traditional V-P, V-Q analysis features. This

is used to investigate the effectiveness of the modal analysis
technique by an application to the IEEE 30 bus standard
system.
II. STATIC ANALYIS METHODS

Static analysis methods of voltage stability utilize a

'snapshot' of the system at a point in the time domain
trajectory [3] and determines the overall stability of the system
and/or the proximity and margin to unstable operation at this
specific operating point. Equation 1 represents the linearized
model of the power system at such an operating point.

FaP aO]

aoav FO1A(1)
LAP1]
LAQ] 8Q Q LAV]
Lao avi

where AP, AQ are mismatch active and reactive power vectors

A V, A0 are unknown voltage magnitude and angle correction

vectors

eP aP

beforehand the buses that are critical. For large power

networks this can be time-consuming and confusing. V-P
curves also give no useful insight into the cause of the voltage
stability problem in the network [6]
B. Q- V curves
Many power system planners still utilize Q-V curves in
analyzing performance criteria [5]. Such curves indicate the
sensitivity and variation of bus voltages with respect to
reactive power injections and absorptions. The Q-V curve at a
test bus is generated by placing a variable reactive power
source with infinite limits at the bus. Successive power flows
are performed for different scheduled values of bus voltage,
and the required reactive power injection is measured. Q-V
curves can therefore give the reactive power load margin at a
bus from a stable operating point to the point of voltage
instability. Such curves give no insight however into the
causes of the voltage stability, such as participating nodes and
branches. As with V-P curves, Q-V curves at several buses
need to be drawn to obtain a system-wide perspective of the
stability problem and decipher which buses in the system have
the smallest stability margins, since there is no way of
identifying beforehand which buses are critical. Although
computation can be easily implemented on software,
processing time can be quite large for practical networks with
several hundred buses.

00 oV is the Jacobian matrix of partial derivatives.

aQ 8Q

ao av]

Static methods usually involve determining power margins till

non-convergence of the power flow, determining the distance
of the power flow Jacobian matrix to singularity (bifurcation
point), or some form of spectral analysis of this Jacobian
matrix. [4]
A. V-P curves
The V-P curve for a bus is obtained by increasing the total
load at the bus or area and performing successive power flow
calculations, whilst maintaining a constant power factor. The
voltage at the bus is plotted as a function of the total active
power load, using conventional power-flow programs until the
bifurcation point or 'nose' point is reached at critical loading.
The points above the 'nose' correspond to voltage stable
operation. Continuation power-flow programs can be used to
obtain solutions below the bifurcation point, but this is not
usually necessary [5]. Generating V-P curves via automation
of power flow simulations can be implemented relatively
easily. As the size of the power network increases however,
the processing time required to generate V-P curve increases,
due to the increase in processing time to perform a single
power flow solution. For large networks, processing time can
be relatively large. More importantly, V-P curves at several
buses must be generated before a system-wide perspective of
voltage stability emerges, since there is no way of identifying

C. Minimum singular value

A voltage stability index proposed in [7] using the minimum
singular value of the power flow Jacobian matrix, can be used
to indicate the relative distance between the studied operating
point and the steady state voltage limit Singular value
decomposition is used to determine the rank of a matrix which
is equal to the number of its singular values[8]. In static
voltage collapse analysis, it is used to monitor the smallest
singular value up to the point it becomes zero[7].
The minimum singular value of the power flow Jacobian is a
relative measure of the proximity of the system to the stability
limit [8]. The method still does not indicate specific causes of
voltage instability such as highly loaded transmission lines
and generators reaching reactive limits. Although the
minimum singular value provides a measure of the relative
proximity of the system to instability, it cannot provide an
absolute or linear measure. This is simply due to the fact that
the approach of the system from a stable operating point to
bifurcation is non-linear. Hence practical monitoring of
voltage stability using this method, must be used in
conjunction with V-P, Q-V, or VPQ curves.

D. Q-VModal analysis
Q-V modal analysis method is proposed in [2] as
follows and uses the reduced Jacobian matrix JR to analyze
the system and provide both a relative proximity of the system
to voltage instability, as well as the mechanism or key
contributingfactors to instability.
382

is stressed, the value of A becomes smaller and the modal

voltage becomes weaker. If the magnitude of Ai is equal to
zero, the corresponding modal voltage collapses since it
undergoes infinite changes for reactive power changes. A
system is therefore defined as voltage stable if all the
eigenvalues of JR are positive. The bifurcation or voltage
stable limit is reached when at least one eigenvalue reaches
zero; that is, one or more modal voltage collapses. If any of
the eigenvalues are negative, the system is unstable. The
magnitude of the eigenvalues provides a relative measure of
the proximity of the system to instability. The critical modes
(associated with minimum eigenvalues) are of major
importance in the voltage stability analysis.
The left and right eigenvectors corresponding to the
critical modes in the system can provide information
concerning the mechanism of voltage instability, by
identifying the elements participating in these modes. The bus
participation factor measuring the participation of the kth bus
to the ith mode can be given as

If active power P is kept constant in equation 1 then

JPOJPv[AO]

(2)

JQO JQV _LAV-

LAQ-

AQ = JRAV

(3)

where JR is the reduced Jacobian system matrix and

=Lv -JQ JPO-1JPV]

JR

The modes of the power network can be defined by the

eigenvalues and eigenvectors of JR
Assume

JR =A7

where
; is the right eigenvector matrix of JR
C7 is the left eigenvector matrix of JR
A is the diagonal eigenvalue matrix of JR

R- =A1

(7)
77ik
Bus participation factors corresponding to the critical modes
can predict areas or nodes in the power system susceptible to
voltage instability. Buses with large participation factors to
the critical mode correspond to the most critical system buses.
Pki

(4)

Incremental changes in reactive power and voltage are related

by equation 3. Substituting equation 4, then
AV AA1AQ

For a change in reactive power injection

or

'AQ

AV = E

Ai

voltage variation is
from equation 2 is

(5)

AOmi =-Jp

where
2i is the ith

eigenvalue
is the ith column right eigenvector of JR

Therefore, the ith modal reactive power variation is

with

,ji

=1

the jth element of Xi

The ith modal voltage variation can therefore be written

AVmi

Ai

AQmi

(8)

max i
Branches with large participation factors to the critical modes
consume the most reactive power for incremental changes in
reactive load, and lead to problems concerning voltage
stability. [2]

where k i is a normalization factor such that

ji

A Vmi

AQI

AQmi = ki,i

AVmi, and the ith modal angle variation,

Assume AQimai = MAXj (AQji)

Then the participation factor of branch lj to mode i can be
defined as
(9)
AQjj

-I

AQmj, the resulting

With AV,AO known, the linearized reactive loss variation

across a transmission branch lj, AQO,i can be obtained.

X
and ii is the ith row left eigenvector of JR.
Ai di, and 7i define the ithmode of the system.

ki

ki

III. PERFORMING COMPLETE VOLTAGE STABILITY ASSESMENTS

It is useful to establish and facilitate a systematic and

(6)

From equation 6, the stability of a mode i with respect to

reactive power changes is defined by the modal
eigenvalue Ai. Large values of 2Ai suggest small changes in
the modal voltage for reactive power changes. As the system

complete method for voltage stability analysis and prediction

of incipient voltage instability in the power network. As seen
previously V-P/Q-V curves cannot be used alone in large
practical networks since there is no way to precisely determine
beforehand, for many different operating points and
contingencies, exactly which nodes and areas are susceptible

383

to instability (i.e. at what buses curves need to be drawn). the reactance of the highest participating branch was
Methods such as the minimum singular value do not readily decreased and V-Q curves regenerated at the critical buses.
provide information into the main factors contributing to This process was repeated, decreasing a medium
voltage instability. The modal analysis technique however can participating branch and the lowest participating branch by
be seen as extremely useful since it provides a relative the same percentage. The increases in reactive stability
proximity of the system to voltage instability, as well as the margins at the critical buses for each of the three cases were
key contributing factors to instability such as the weakest or compared, in order to quantify the effect of the relevant
critical buses and transmission branches. A simple systematic branch to system instability.
technique for initial assessment of voltage stability A. Test Case 1
incorporating modal analysis as well as traditional V-P/QV/VPQ curves can be conducted as follows:
TABLE I
1.
Perform a power flow computation to
THREE LOWEST SYSTEM EIGEENVALUES FOR TEST CASE 1 (CRTICIAL
FTC-TFNVT ATTF= 1 Q6AV
establish conditions at the specific operating
point and generate system Jacobian matrix
2.

Perform modal analysis to obtain a relative

measure of the system distance to
bifurcation, identify the weakest system
buses, and identify the weakest system
branches

3.

Generate V-P/Q-V/VPQ curves at weak

nodes identified in modal analysis to further
investigate specific stability margins.

4.

Implement stability solutions such as sizing

and placement of reactive compensation
devices at critical buses identified, and
enhancement of weakest transmission
branches identified.

TABLE 2

COMPARISON OF 8 MOST CRITICAL BSUES IDENTIFIED BY MODAL ANALYSIS

AND V-0 STABILITY MARGINS (TEST CASE 1)

30
19
18

0.060
0.057
0.056

18
29

3.63
2.83
3.05

19

A prototype application on the MIATLAB platform called

the Power System Stability Analyzer (PSSA) was developed
to perform such analyses.
IV. EVALUATING THE MODAL ANALYSIS TECHNIQUE

In order to demonstrate the usefulness of modal

analysis, three separate test cases, representing three different
stable operating points of the IEEE 30 bus test system were
used. The results are given in Tables 1 to 13 below.
For each operating point, the eigenvalues of the
reduced Jacobian matrix were first generated to obtain the
relative proximity of the system to instability. The bus
participation factors were then generated for the critical
mode to predict the critical buses in the system (buses with the
smallest margins to instability). In order to verify that the
buses predicted as critical were correct, V-Q curves were
generated to observe and compare the reactive stability
margins at each system bus. (V-Q curves are chosen since
reactive power transfer is closely associated to voltage
instability).
Next the branch participation factors for the critical
mode were generated to predict the critical transmission lines
in the system (line with largest contributions to voltage
instability). In order to verify that these results were correct,
384

TABLE 3
EIGHT HIGHEST PARTICIPATING BRANCHES (TEST CASE 1)

*-rh

35

11
13
17
34
27
14

p
1.00
037

0.30
0.27
0.20
0.15
0.11

29

0.11
TABLE 4

COMPARING THE EFFECTS OF BRANCHES WITH VARYING PARTICIPATION

FACTORS ON SYSTEM STABILITY (TEST CASE 1)

B. Test Case II

C. Test Case III

TABLE 9
THREE LOWEST SYSTEM EIGEENVALUES FOR TEST CASE 3 (CRTICIAL
EIGENVLAUE= 0.442)

TABLE 5
THREE LOWEST SYSTEM EIGEENVALUES FOR TEST CASE 2 (CRTICIAL
EIGENVLAUE= 1.161)

1.161

3898
8.499

Mode

13
19
20

(TP,T tAFP

TABLE 11
EIGHT HIGHEST PARTICIPATING BRANCHES (TEST CASE 3)

TABLE 7

EIGHT HIGHEST PARTICIPATING BRANCHES (TEST CASE 2)

Br* h
35
34
11

1.000
0.402
0.393

0.327

29

0.268
0.168
.
0120

14

Pawricipo

13

17
27

1.388

2.578

TABLE 10
COMPARISON OF 8 MOST CRITICAL BSUES IDENTIFIED BY MODAL ANALYSIS
ANT) V-() TAlITJITV MARPCANT (TFV'T CA(.F 3)

TABLE 6
COMPARISON OF 8 MOST CRITICAL BSUES IDENTIFIED BY MODAL ANALYSIS
Awn V n RT ATT TTV X4AR TNTQ

Eigenvalu
0.442

Branth
34
33
35

40

30

0.280

29
32

27

aripai
1.000
0.775
0.412

0.298
0.262

0.248

0.202
0.194

TABLE 12
COMPARING THE EFFECTS OF BRANCHES WITH VARYING PARTICIPATION
FACTORS ON SYSTEM STABILITY (TEST CASE 3)

TABLE 8
COMPARING THE EFFECTS OF BRANCHES WITH VARYING PARTICIPATION
FACTORS ON SYSTEM STABILITY (TEST CASE 1)

385

u .uz

25
24
23

0.21
0.2S
1) 1)

0.35

so)

0.03
0.05
0.14
Al 19

V. DISCUSSION

For the first test case, the three lowest system

eigenvalues are displayed in table 1 and are above zero,
confirming that the system is stable and the bifurcation point
has not yet been reached. The minimum or critical eigenvalue
is 1.963 and the corresponding mode is 8. Bus participation
factors to this critical mode are generated to predict the critical
buses in the system. The buses with highest participation
factors were expected to be the most critical buses or buses
closest to instability. Table 2 verifies that the buses with the
highest participation factors have the lowest reactive stability
margins in the system and therefore correctly correspond to
critical system buses. These include bus 26, 25, 24 and 23.
Moreover, the table shows that participation factors predict
the 5 most critical system buses in the exact same order they
are ranked by the actual V-Q curve reactive stability margins.
The 8 highest branch participation factors are displayed in
table 3, showing line 35 as having the highest participation
factor. As shown in table 4, slightly reducing the reactance of
branch 35, increases the stability margin at the 5 most critical
buses by an average of 0. 14p.u.The same percentage reduction
in reactance however, of a lower participating branch such as
branch 34 increases the stability margins at the buses by a
much less average of 0.07p.u. The same reduction in
reactance for the lowest participating branch 22, gives no
increase in stability margin whatsoever. The results verify that
branches identified with higher participation factors are
branches with higher contributions to system voltage
instability.
For the second test case all eigenvalues are still above
zero, confirming that the system is still stable. The critical
eigenvalue however, has now decreased to 1.161. Once again
as shown in table 6, the buses with the highest participation
factors have the lowest reactive stability margins in the system
and correctly correspond to critical system buses. In addition,
the participation factors predict the 8 most critical buses in
the exact same order they are ranked by the actual V-Q curve
reactive stability margins. Branch 34 is identified as having
the largest participation factor and table 8 shows that slightly
reducing the reactance of branch 34, increases the stability
margin at the 5 most critical buses by an average of
0.18p.u.The same reduction in reactance however, of a lower
participating branch such as branch 17 increases the stability
margins at the buses by a much less average of 0.08p.u and a
reduction in reactance for the lowest participating branch 22,

gives no increase in stability margin whatsoever. Once again,

the results verify that the branches identified with the highest
participation factors represent the branches contributing the
most to system voltage instability. The overall pattern is
continued in the third test case, with the bus participation
factors and branch participation factors reliably predicting
the weak buses and transmission branches.
At the first test operating point, all eigenvalues are
above zero and the minimum eigenvalue is 1.963. At the
second operating point, all the eigenvalues are still above zero
however, the minimum eigenvalue is reduced from 1.963 to
1.161. At the third operating point the minimum eigenvalue is
even closer to zero at 0.442. This suggests that the third test
case operating point is closest to the voltage stability limit,
followed by the second operating point, with the first being
the furthest or most stable test case operating point. This is
verified by observing the reactive stability margins at the
critical buses for each operating point (tables 2, 6 and 11). At
the first operating point, the reactive load margins to
instability for the 3 most critical system buses are 1.44 p.u,
2.66 p.u. and 2.83 p.u. At the second operating point, the
reactive stability margins for the 3 most critical buses are
lowered to 0.62 p.u, 0.86 p.u. and 1.12 p.u. For the third
operating point the margins are drastically reduced to 0.05 p.u,
0.12 p.u. and 0.21p.u. It is clear that the third test case
operating point is the closest to the voltage stability limit,
followed by the second and then the first test case, exactly as
predicted by the magnitude of the eigenvalues.
VI. CONCLUSION

The paper re-examines the problem of voltage

instability in power systems and the need for effective analysis
techniques to detect incipient cases of instability and prevent
possible escalation into wide-spread voltage collapse.
Common steady -sate analysis methods are reviewed
including modal analysis. The modal analysis feature was
tested on the IEEE 30-bus test system. The technique was
demonstrated as an extremely useful method for steady state
voltage stability analysis, providing a relative measure of the
system to the stability limit, and correctly predicting the
critical buses and weak transmission branches in the power
network. The paper therefore further reinforces previous
works such as [] which have successfully applied the modal
analysis method to test networks. The method can be
incorporated into user-friendly diagnostic tools which can be
applied to practical power systems and utilized by power
system operators and planners to assess voltage stability at
various operating conditions, as well as determine suitable
remedial actions for instability and collapse.

386

VII. REFERENCES
[1] Gao, B., Kundur, P., Morisson, G.K., "Towards the Development of a
Systematic Approach for Voltage Stability Assessment of Large-Scale
Power Systems", IEEE Transactions on Power Systems, vol. 11, no. 3,
pp 1314-1319, (August 1996). (3)

[2]
[3]

[4]
[5]

[6]

[7]

[8]

Gao, B., Kundur, P., Morisson, G.K, " Voltage Stability Evaluation
Using Modal Analysis", IEEE Transactions on Power Systems, vol. 7,
no. 4, pp. 1423-1543, (November 1992). (15)
Kundur,P., "Power System Stability and Control", Mc-Graw Hill, NewYork, 1994, pp 960-971.(1)
Gubina, F., Strmcnik, B., "Voltage Collapse Proximity Index
Determination using Voltage Phasors Approach", IEEE Transactions
on Power Systems, vol. 10, no. 2, pp 788-792, (May 1995).(6)
Alvarado, F., Dobson, I., Greene, S., "Contingency Rankingfor Voltage
Collapse via sensitivities from a single nose curve", Submission to IEEE
for possible publication, 1997.(9)
Balamourougan, V., Sachdev, M.S., Sidhu, T.S., "Technique for online
prediction of voltage collapse", IEEE Proceedings Generation,
Transmission and Distribution, vol. 151, no. 4, pp. 453-460, (July
2004)(10)
Anderson, C., Hill, D.J., Lof, P.A., Smed, T., "Fast Calculation of a
Voltage Stability Index", IEEE Transactions on Power Systems, vol. 7,
no. 1, pp. 54-64, (February 1992).(14)
Daniel, J.W., Noble, B., "Applied Linear Algebra", Prentice-Hall, NewYork, 1998 (13)

VIII. BIOGRAPHIES
Chandrabhan Sharma
Dr Sharma is a senior lecturer with the Faculty of Engineering, The
University of West Indies, Trinidad & Tobago. He is the Head of the Centre
for Energy Studies. He is a member of the Board of Directors of the local
Electric Utility and also a member of the Board of Directors of the largest
bank in the country. Prior to joining the Academic staff at the University, he
was attached to the petrochemical industry in Trinidad. His interests are in the
area of power system operations and control.
Marcus Ganness
Marcus Ganness is a final year student undergraduate student at the
University of The West Indies. He is expected to graduate in May 2007.

387