Вы находитесь на странице: 1из 8

www.ietdl.

org

Published in IET Generation, Transmission & Distribution


Received on 12th February 2009
Revised on 5th November 2009
doi: 10.1049/iet-gtd.2009.0071

ISSN 1751-8687

Voltage stability analysis based on probabilistic


power flow and maximum entropy
J.F. Zhang1 C.T. Tse1 W. Wang2 C.Y. Chung1
1
Department of Electrical Engineering, The Hong Kong Polytechnic University, Hong Kong, People’s Republic of China
2
Department of Electrical Engineering, Zhengzhou University, Zhengzhou, Henan, People’s Republic of China
E-mail: eejfzhang@gmail.com

Abstract: Methods of determination of voltage stability margin index had been well established. This study adopts
a new method to determine the probabilistic distribution of margin index taking into account the random
variations of bus loads. First, the probabilistic technique and the Jacobian method are combined to determine
the probabilistic characteristics of stability margins and nodal voltages at the maximum load points. Then,
according to these probabilistic characteristics, maximum entropy method is adopted to determine the
probabilistic distribution of stability margin. Last, the proposed method is investigated on two test systems
with random active and reactive loads. Monte Carlo simulations are used as a reference solution to evaluate
the accuracy of the proposed method.

1 Introduction of combined cumulants and Gram – Charlier series has been


used in probabilistic power flow to determine the major
There is an increasing concern about modern power system investment on improving transmission system inadequacy
voltage stability particularly after some blackouts because of [3]. Probabilistic eigenvalue analysis considering multi-
voltage instability or collapse in different countries [1, 2]. operating conditions has been used to analyse and design
Several methods based on different models for voltage robust power system stabiliser (PSS) to improve the angular
stability analysis have been presented. Static voltage stability stability of power system [4, 5]. The results of the
analysis based on power flow attracts researchers for its probabilistic load flow have been used to assess voltage
simple and fast calculation. It regards the maximum load instability [6]. A new nodal loading model, called the
point as the critical point, or the so-called saddle-node hyper-cone model whose thickness represents the
bifurcation. Static voltage stability analysis can provide uncertainty of future loading, has been proposed for voltage
margin index for power system operators. Stability margin stability assessment of electric power system, and the
is the distance between the current operating point and the formulation of worst cases based on this model, as well as
maximum load point. Stability margin is a useful index related numerical methods, has been described in [7].
because it indicates how close the power system is to Sobierajski used the P – Q curve to estimate the probability
voltage collapse. of the critical voltage violation under the assumption that
active and reactive power at a given load bus are uniformly
Deterministic voltage stability analysis requires specific distributed [8]. Schellenberg proposed a cumulant-based
value for loads. In practice, the loads vary from time to method to solve a maximum loading problem incorporating
time. It is impossible to carry out conventional voltage a constraint on the maximum variance of the loading
stability study for every possible or probable combination of parameter [9]. This method takes advantage of some
bus loads because of the extremely large computational properties regarding saddle-node bifurcations to create a
effort required. Probabilistic methodology may be most linear mapping relationship between random bus loading
suitable to analyse the system uncertainties. Probabilistic variables and all other system variables so that the cumulants
analysis considering the uncertainties in loads, generation of system variables are evaluated and the probabilistic density
or network has been performed in literatures. The method function (PDF) of stability margin is obtained.

530 IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530– 537
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-gtd.2009.0071
www.ietdl.org

This paper combines the probabilistic technique and the to obtain the solution of power flow near the critical point.
Jacobian approach to determine the probabilistic The difficulty is overcome by introducing other parameter
characteristics of stability margin and nodal voltages at the and equation [12, 13]. This makes the power flow Jacobian
maximum load points, and then applies the maximum non-singular at the voltage collapse point.
entropy technique to study the distribution of stability
margin considering the load variations. Entropy, a Although the continuation power flow method can obtain
conception in information theory, is a measure of the critical point and stability margin without the calculation
uncertainty. An entropy application is to deal with problems difficulty near the critical point, this method only obtains an
involving the determination of unknown distributions but approximate solution and is time consuming.
with available information such as expected values or other
statistical functions [10, 11]. Like the cumulant method, 2.2 Non-linear programming method
maximum entropy method is a general method and is not
restricted to a special distribution. With the moments of The voltage collapse point is the maximum load point.
random variable of stability margin obtained in terms of Calculating the maximum load is represented as a static non-
probabilistic Jacobian approach, the distribution can be linear programming problem. The objective function is to find
determined with maximum entropy. the maximum load factor with the constraints of power flow
equations and other inequalities. The problem is presented as
This paper is arranged as follows. In Section 2, three
prevailed methods to obtain static voltage stability margin max l
are first revealed. Probabilistic voltage stability margin and s:t: f (V )  lB  S0 ¼ 0 (2)
its calculation are proposed in Section 3. The method of g  h(V )  l
maximum entropy is introduced in Section 4. The
proposed probabilistic method is investigated on two test The equalities are the ac power flow equations at different load
systems and the results are compared with Monte Carlo in level. The inequality constrains contain any variable limitations
Section 5. Section 6 gives conclusions. as well as reactive power limits of generators. The merit of this
method is that some constrains can be considered easily.

2 Static voltage stability analyses 2.3 Jacobian method at critical point


Static voltage stability analysis still attracts researchers’ attention A salient characteristic of the critical point is that the
for its simple calculation and intuition. This analysis regards Jacobian matrix J of power flow equations is singular; J has
the maximum load point as the critical point. The distance a zero eigenvalue but the corresponding eigenvectors (left
between the current operating point and the maximum load and right) are non-zero.
point is called voltage stability margin. Methods to determine
the maximum load, such as continuation power flow [12, Based on this property, the exact critical point of power
13], non-linear programming method [14, 15] and Jacobian system is directly obtained by solving the extended power
method [16, 17], have been used to obtain the saddle-node flow equations as follows
bifurcation and stability margin. The pros and cons of these
methods are discussed as below. f (V )  lB  S0 ¼ 0 (3a)

J TV w ¼ 0 (3b)
2.1 Continuation power flow
The model is described as follows wT w  1 ¼ 0 (3c)

f (V )  lB  S0 ¼ 0 where w is the left eigenvector of JV with respect to zero


(1) eigenvalue and T stands for transpose. Equation (3a)
l0  l  lcritical
describes the power flow and (3b) and (3c) ensure the JV is
singular at the critical point. This method, also called point
where V is the vector of nodal voltages, B describes load of collapse method, is a particular case of non-linear
increases and S0 is the normal injection power vector programming method.
(including active and reactive power). l is the load factor.
l ¼ 0 corresponds to normal load level and l ¼ lcritical
corresponds to critical load level. 3 Stability margin assessment by
To access the critical point and stability margin, the iterative
probabilistic approach
calculation starts from a known solution, and the predictor and There are always uncertainties in power systems, such as the
corrector technique is used to obtain the subsequent solution at variations of loads and the generators outputs. In this paper,
different load levels. As the Jacobian matrix of power flow loads are regarded as random variables. Based on the Jacobian
equations is singular at the maximum load point, it is difficult approach mentioned in Section 2.3, the nodal voltages, the

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530 – 537 531
doi: 10.1049/iet-gtd.2009.0071 & The Institution of Engineering and Technology 2010
www.ietdl.org

left eigenvectors and hence stability margin are also random The expectation of (5) is expressed as (6)
variables.

For the sake of convenience, the Jacobian approach method F 1 (V 1 V 1 ,..., V 1 V n ,..., V n V 1 ,..., V n V n )
is combined with probabilistic method to determine the þ F 1 (DV1 DV1 ,..., DV1 DVn ,..., DVn DV 1 ,..., DVn DV n )
characteristics of stability margin. If the voltages in (3) are
expressed with rectangular coordinates, every expression is  l B  S 0 ¼ 0 (6a)
the second-order function of nodal voltage and left
eigenvector. Equation (3) can be represented as
F 2 (V 1 w
 1 ,..., V 1 w
 n ,..., V n w
 1 ,..., V n w
 n)
F 1 (V1 V1 , ... , V1 Vn , ... , Vn V1 , ..., Vn Vn )  lB  S0 ¼ 0 (4a)
þ F 2 (DV1 Dw1 ,..., DV1 Dwn ,..., DVn Dw1 ,..., DVn Dwn )
¼0 (6b)
F 2 (V1 w1 , ... , V1 wn , ... , Vn w1 , ... , Vn wn ) ¼ 0 (4b)

F 3 (w  1 ,..., w
 1w  n ) þ F 3 (Dw1 Dw1 ,..., Dwn Dwn )  1
 nw
F 3 (w1 w1 , ..., wn wn )  1 ¼ 0 (4c)
¼0 (6c)

where Vi and wi , for i ¼ 1 to n, respectively, stand for the


elements of voltage vector and left eigenvector. S0 is As DVi DVj ¼ CVi Vj , DVi Dwj ¼ CVi wj , Dwi Dwj ¼ Cwi wj , (6)
composed of initial injections of active and reactive power can also be represented with covariance of random
PGj0 2 PLj0 and QGj0 2 QLj0 for j ¼ 1 to N. In this study, variances of nodal voltages and left eigenvector. With the
PLj0 and QLj0 are active and reactive power of load at bus j covariances, the means of nodal voltages, stability margin
and are random variables. PGj0 and QGj0 are active and and left eigenvector at critical point can be obtained by
reactive power of generator at bus j and are not random solving (6) with Newton – Raphson method.
variables. S0 is random vector. B describes load increase. In
this paper, B is expectation of S0 , that is B is composed Omitting the second-order terms of (5) and (6), the
by injections of active and reactive powers PGj0  P Lj0 and linearised relationships among nodal voltage V, stability
QGj0  QLj0 for j ¼ 1 to N. l stands for stability margin. At margin l, left eigenvector w and the nodal load S0 are
the critical point, the load at bus j can be represented by shown as
PLj ¼ PLj0 þ lP Lj0 and QLj ¼ QLj0 þ lQLj0 .
2 3 2 3
If (4) is expanded at the means of nodal voltages, stability DV DS0
margin, the left eigenvector and nodal injection J 4 Dw 5 ¼ 4 0 5 (7)
Dl 0

F 1 (V 1 V 1 ,..., V 1 V n ,..., V n V 1 ,..., V n V n )


where
þ F 1 (DV1 V 1 ,..., DV1 V n ,..., DVn V 1 ,..., DVn V n )
þ F 1 (V 1 DV1 ,..., V n DV1 ,..., V 1 DVn ,..., V n DVn ) 2  3
@F 1 
þ F 1 (DV1 DV1 ,..., DV1 DVn ,..., DVn DV1 ,..., DVn DVn ) 6 @V   0 B 7
6 V ¼V 7
6 @F  
@F 2  7
 lB  DlB  S0  DS0 ¼ 0 (5a) 6 2 0 7
J ¼6 7
6 @V V ¼V @w w¼w 7
6  7
4 @F 3  5
F 2 (V 1 w
 1 ,..., V 1 w n ,..., V n w
 1 ,..., V n w 0 0
 n) @w w¼w
þ F 2 (DV1 w  1 ,..., DV1 w  n ,..., DVn w  1 ,..., DVn w
 n)
þ F 2 (V 1 Dw1 ,..., V 1 Dwn ,..., V n Dw1 ,..., V n Dwn ) The deviations from the means of random variables nodal
þ F 2 (DV1 Dw1 ,..., DV1 Dwn ,..., DVn Dw1 ,..., DVn Dwn ) voltage, left eigenvector and stability margin is obtained by
rearranging (7) as
¼0 (5b)
2 3 2 3
DV DS0
F 3 (w  1 ,..., w
 1w  n)
 nw 4 Dw 5 ¼ J 1 4 0 5 (8)
þ F 3 (w
 1 Dw1 ,..., w n Dwn ) Dl 0
þ F 3 (Dw1 w 1 ,..., Dwn w n)
þ F 3 (Dw1 Dw1 ,..., Dwn Dwn )  1 ¼ 0 (5c) The covariances of random variables nodal voltage, left

532 IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530– 537
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-gtd.2009.0071
www.ietdl.org

eigenvector and stability margin are obtained as 4 Maximum entropy method


2 3 Entropy is a measure to tackle uncertainty. The entropy of a
C VV C Vw CV l PDF p(x) is defined as [11]
6 7
C ¼ 4 C wV C ww C wl 5 ð
C lV C lw C ll H ¼  p(x) ln p(x) dx (11)
02 3 1
DS0
B6 7   C
¼ J 1 E @4 0 5 DST0 0 0 A(J 1 )T An application of entropy is to determine an unknown
0 distribution based on the principle of maximum entropy,
2 3 provided that expected values and its other statistical
C S0 S0 0 0 functions are given. The solution by maximum entropy, in
6 7
¼ J 1 4 0 0 0 5(J 1 )T (9) its general form [18], is
0 0 0 ð
max H ¼  p(x) ln p(x) dx
T ð
The nth moment for [ DV Dw Dl] is computed
from (8). As a result, the nth moment for s:t: E{wn (x)} ¼ wn (x)p(x) dx ¼ mn , n ¼ 0, . . . , N
[V w l]T can be obtained using following equation
(12)
X
n1
r where m0 ¼ 1,w0(x) ¼ 1 and wn(x),n ¼ 1, . . . , N are N
E(X n ) ¼ (X )n þ (DX )n þ Cnr X (DX )nr (10)
r¼1
known functions, and mn , n ¼ 1, . . . , N are N given
expectation data. The classical solution of this problem is
given by
The probabilistic characteristics of critical points can be
described by probabilistic characteristics of stability margins " #
X
N
and nodal voltages. The procedure of probabilistic critical p(x) ¼ exp  ln wn (x) (13)
points calculation is as follows. n¼0

Step 1: Calculate the probabilistic power flow at a load level, The N þ 1 Lagrange parameters l ¼ [l0 , . . . , lN ] are
and obtain the initial voltages and covariances of voltages CV. obtained by solving the following N þ 1 non-linear equations
ð " #
Step 2: Calculate the eigenvalues and eigenvectors of Jacobian X
N

matrix of power flow equation. The left eigenvector wn (x)exp  ln wn (x) dx ¼ mn , n ¼ 0, .. ., N (14)
n¼0
corresponding to minimum real eigenvalue is the initial left
eigenvector.
For illustration, the maximum entropy is applied to a simple
exponential distribution with a PDF of
Step 3: Compute the mismatches in (6); if the mismatches
meet the error requirement, go to step 6, otherwise continue.
f (x) ¼ lelx (15)
Step 4: Form the Jacobian matrix J of extended power flow (6)
with respect to means of nodal voltage, stability margin and where x  0, l . 0. If l ¼ 4, 0  x  4.
left eigenvector.
With the PDF, expectations of random variable functions,
Step 5: Calculate the correction of means of nodal voltage, for example the geometrical moments, can be calculated as
listed in Table 1. The objective is to determine the
stability margin and left eigenvector and correct them; go
to step 3. distribution of random variable only with known function
wn(x), n ¼ 0, . . . , 4 and expectations of random variable
functions Efwn(x)g ¼ mn in Table 1. By solving the five
Step 6: Calculate the covariances of nodal voltage, left
eigenvector and stability margin according to (9). Substitute non-linear equations in (14), the five Lagrange parameters
the covariances in (6), and calculate the mismatches. If the l ¼ [l0 , l1 , l2 , l3 , l4] in (13) are obtained. The PDF of
mismatches meet the error requirement, stop, otherwise go
to step 4. Table 1 Functions and their expectation of random variable

n 0 1 2 3 4
Consequently, the means and covariances of both nodal
2 3
voltages and stability margin can be determined from the wn(x) 1 x x x x4
probabilistic computation. The other moments of stability
mn 1 0.25 0.125 0.09374 0.09371
margin can be obtained in terms of (10).

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530 – 537 533
doi: 10.1049/iet-gtd.2009.0071 & The Institution of Engineering and Technology 2010
www.ietdl.org

exponential curve almost coincide, and accurate result can


be obtained when N ¼ 4. This shows that maximum
entropy method can determine any unknown distribution if
enough available information of random variables is known.

In this paper, according to (10), the moments of


stability margin and nodal voltage can be computed.
Therefore wn(x) ¼ x n, and mn for n ¼ 1, . . . , 4 are available
information to determine the distribution stability margin.

Figure 1 PDF of exponential distribution 5 Applications of probabilistic


random variable x is (13) with five known Lagrange approach and maximum entropy
parameters l ¼ [l0 , l1 , l2 , l3 , l4 ]. The proposed method is investigated on two test systems: the 39-
bus system and the 57-bus system. Normal distribution is one of
The PDF obtained by maximum entropy is then compared the most common random distributions. All real and reactive
with known PDF of exponential distribution in Fig. 1. Note loads in this study are treated as Gaussian random variables.
that for this simple case, the derived curve and the actual The loads are independent random variables with means at the

Table 2 Stability margin for the 39-bus system

sL Probabilistic method Monte Carlo Difference (%)


23 23
Mean Variance (10 ) ISM index of 99% Mean Variance (10 ) Mean Variance
(a) 0.025m 1.3513 0.2248 1.3163 1.3510 0.2252 0.0222 0.1776
(b) 0.05m 1.3500 0.8993 1.2801 1.3495 0.9036 0.0371 0.4759
(c) 0.075m 1.3480 2.0238 1.2431 1.3472 2.0428 0.0594 0.9301
(d) 0.1m 1.3450 3.5989 1.2052 1.3436 3.7263 0.1042 3.4189

Figure 2 Probabilistic distributions of stability margin with different sL of the 39-bus system
a sL ¼ 0.025m
b sL ¼ 0.05m
c sL ¼ 0.075m
d sL ¼ 0.1m

534 IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530– 537
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-gtd.2009.0071
www.ietdl.org

Table 3 Stability margin for the 57-bus system

sL Probabilistic method Monte Carlo Difference (%)


23 23
Mean Variance (10 ) ISM index of 99% Mean Variance (10 ) Mean Variance
(a) 0.025m 0.80157 0.3769 0.7872 0.80158 0.3769 0.0012 0.0000
(b) 0.05m 0.80140 1.5074 0.7727 0.80142 1.5074 0.0025 0.0000
(c) 0.075m 0.80112 3.3916 0.7582 0.80116 3.3914 0.0050 0.0059
(d) 0.1m 0.80073 6.0293 0.7435 0.80078 6.0053 0.0061 0.3996

normal bus loading, denoted by m, from the original system. The probabilistic Jacobian method as listed in Table 2. From
variance of each load has been chosen such that 99% confidence Table 2, it is observed that although the mean of stability
in within +10% from the normal loading value [9]. The load margin has insignificant changes, the variance of stability
standard deviation sL ¼ 0.04m has been used for probabilistic margin increases with load standard deviation sL (from
transient stability analysis in [19]. In order to study the effect 0.025m to 0.1m) as expected. Based on the moments, the
of uncertainty of load on stability margin, different standard PDFs obtained by the maximum entropy approach are
deviations of loads from 0.025m to 0.1m are, respectively, plotted in Fig. 2. In Table 2, a stability margin index ISM
analysed. For simplicity, there are no limitations for reactive of, say, 99% confidence is introduced. For example,
power of generation or bus voltage levels considered. All the ISM ¼ 1.3163 in (a) implies that if the system load is
loads are constant power models. The probabilistic stability increased by 1.3163B (B is the load expectation in (4)),
margin and its distribution are determined by applying 99% of scenarios will not experience voltage collapse (i.e.
probabilistic Jacobian method (Section 3) and maximum 99% of area under the PDF in Fig. 2a has a margin greater
entropy (Section 4). Monte Carlo simulations, consisting of than 1.3163) for a system load standard deviation
1000 samples, are used to validate the proposed method. sL ¼ 0.025m. Of course, if the load variance increases, ISM
will be reduced.
5.1 Case study on the 39-bus system [20]
With different variance of loads, the means and variances of To validate the probabilistic point of collapse method, the
stability margins are calculated from (6) and (9) using mean and variance are compared with those obtained by

Figure 3 Probabilistic distributions of stability margin with different sL of the 57-bus system
a sL ¼ 0.025m
b sL ¼ 0.05m
c sL ¼ 0.075m
d sL ¼ 0.1m

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530 – 537 535
doi: 10.1049/iet-gtd.2009.0071 & The Institution of Engineering and Technology 2010
www.ietdl.org

Monte Carlo method based on 1000 deterministic loadflows 8 References


(Section 2.3) as listed in Table 2. Results show that their
differences in the mean computation are very small, and the [1] KUNDUR P.: ‘Power system stability and control’
difference in variance is only noticeable with high load (McGraw-Hill, New York, 1994)
uncertainty.
[2] TAYLOR C.W.: ‘Power system voltage stability’ (McGraw-
To validate the maximum entropy method, the kernel Hill, New York, 1994)
smoothing density estimation is used to obtain the actual
distribution of stability margin (in Fig. 2) from 1000 [3] ZHANG P., LEE S.T.: ‘Probabilistic load flow computation
stability margin samples. From the curve obtained Monte using the method of combined cumulants and Gram –
Carlo, it can be seen that the stability margin is not normal Charlier expansion’, IEEE Trans. Power Syst., 2004, 19, (1),
distribution. The probabilistic distributions of the stability pp. 676– 682
margins obtained by the two approaches are compared with
different sL shown in Fig. 2 and the results show that their [4] WANG K.W., CHUNG C.Y. , TSE C.T., TSANG K.M.: ‘Improved
differences are acceptably small. probabilistic method for power system dynamic stability
studies’, IEE Proc. Gener. Transm. Distrib., 2000, 147, (1),
5.2 Case study on the 57-bus system [21] pp. 37– 43
The 57-bus system includes four generating units and three
synchronous condensers. The means and variances of [5] TSE C.T., WANG K.W., CHUNG C.Y., TSANG K.M.: ‘Robust PSS design
stability margin with different sL are listed in Table 3 and by probabilistic eigenvalue sensitivity analysis’, Electr. Power
the probabilistic distributions are shown in Fig. 3. As Syst. Res., 2001, 59, (1), pp. 47–54
expected, the so-called stability margin index ISM will be
reduced with increase of load standard deviation sL (i.e. [6] HATZIARGYRIOU N.D., KARAKATSANIS T.S.: ‘Probabilistic load
with increase of system uncertainty). The computation flow for assessment of voltage instability’, IEE Proc. Gener.
results are then compared with Monte Carlo simulation Transm. Distrib., 1998, 145, (2), pp. 196– 202
and the difference between two methods are quite small.
[7] KATAOKA Y.: ‘A probabilistic nodal loading model and
worst case solutions for electric power system voltage
6 Conclusion stability assessment’, IEEE Trans. Power Syst., 2003, 18,
Methods of voltage stability margin index calculation had been (4), pp. 1507 – 1514
well established, but these methods are based on deterministic
system load flows. This paper proposes a new probabilistic [8] SOBIERAJSKI M., FULCZYK M.: ‘Voltage stability study by p – q
approach to evaluate this important margin index by taking curve with rectangular probability distribution of bus load’.
into account the uncertainty of system loads. With the Eighth Int. Conf. on Probabilistic Methods Applied to Power
assumption that all loads are independent Gauss distributions, Systems, Iowa State University, Ames, Iowa, September
the prevailed Jacobian method to evaluate stability margin is 2004, pp. 894 – 899
modified such that the moments of stability margin and critical
nodal voltage can be obtained. Maximum entropy technique is [9] SCHELLENBERG A., ROSEHART W., AGUADO J.A.: ‘Cumulant-based
then applied to determine the distribution of the stability stochastic nonlinear programming for variance constrained
margin. As expected, the stability margin will be reduced with voltage stability analysis of power systems’, IEEE Trans.
higher load uncertainty, implying that the stability margins Power Syst., 2006, 21, (2), pp. 579 – 585
obtained based on traditional deterministic loadflows are too
‘optimistic’. The proposed approach is tested on two test [10] PAPOULIS A.: ‘Probability, random variables, and stochastic
systems. To validate the proposed probabilistic algorithm, the processes’ (McGraw-Hill, New York, Boston, 2002, 4th edn.)
distributions of the stability margin are compared with Monte
Carlo technique (by 1000 computation simulations using [11] ZELLNER A., HIGHFIELD R.A.: ‘Calculation of maximum entropy
deterministic method) and the computation results are found distributions and approximation of marginalposterior
to be quite close. As a conclusion, the proposed probabilistic distributions’, J. Econ., 1988, 37, (2), pp. 195–209
approach (based on single computer run) is very effective and
pragmatic to compute the voltage stability margin under [12] AJJARAPU V., CHRISTY C.: ‘The continuation power flow:
system uncertainties. a tool for steady state voltage stability analysis’, IEEE
Trans. Power Syst., 1992, 7, (1), pp. 416 – 423
7 Acknowledgments [13] IBA K., SUZUKI H., EGAWA M., WATANABE T.: ‘Calculation of
This work is supported by Research Grants Council of Hong critical loading condition with nose curve using homotopy
Kong under project PolyU 5212/03E and the Hong Kong continuation method’, IEEE Trans. Power Syst., 1991, 6,
Polytechnic University. (2), pp. 584– 593

536 IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530– 537
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-gtd.2009.0071
www.ietdl.org

[14] IRISARRI G.D. , WANG X., TONG J., MOKHTARI S.: ‘Maximum [18] MOHAMMAD-DJAFARI A.: ‘A matlab program to calculate
loadability of power systems using interior point the maximum entropy distributions’. Maximum entropy
nonlinear optimization method’, IEEE Trans. Power Syst., and Bayesian methods: proc. of the eleventh
1997, 12, (1), pp. 162– 172 international workshop on maximum entropy and
bayesian methods of statistical analysis, Seattle,
[15] PARKER C.J., MORRISON I.F., SUTANTO D.: ‘Application of an Washington, 1991, pp. 221– 233
optimisation method for determining the reactive margin
from voltage collapse in reactive power planning’, IEEE [19] ABORESHAID S. , BILLINTON R., FOTUHI-FIRUZABAD M. :
Trans. Power Syst., 1996, 11, (3), pp. 1473– 1481 ‘Probabilistic transient stability studies using the method
of bisection [power systems]’, IEEE Trans. Power Syst.,
[16] CANIZARES C.A. , ALVARADO F.L.: ‘Point of collapse and 1996, 11, (4), pp. 1990– 1995
continuation methods for large AC/DC systems’, IEEE
Trans. Power Syst., 1993, 8, (1), pp. 1 – 8 [20] PAI M.A.: ‘Energy function analysis for power system
stability’ (Kluwer Academic Publishers, Boston, 1989)
[17] CHIANG H.-D., JEAN-JUMEAU R.: ‘A more efficient formulation for
computation of the maximum loading points in electric power [21] http://www.ee.washington.edu/research/pstca/pf57/
systems’, IEEE Trans. Power Syst., 1995, 10, (2), pp. 635–646 pg_tca57bus.htm, accessed February 2009

IET Gener. Transm. Distrib., 2010, Vol. 4, Iss. 4, pp. 530 – 537 537
doi: 10.1049/iet-gtd.2009.0071 & The Institution of Engineering and Technology 2010