Received December 6, 2019, accepted December 24, 2019, date of publication December 31, 2019, date of current version

January 9, 2020.
Digital Object Identifier 10.1109/ACCESS.2019.2963280

A Probabilistic Assessment Method for Voltage

Stability Considering Large Scale Correlated
Stochastic Variables
JING ZHANG 1 , (Member, IEEE), LUQIN FAN1 , YING ZHANG2 , GANG YAO2 , PEIJIA YU3 ,
GUOJIANG XIONG1 , KE MENG4 , (Member, IEEE), XIANGPING CHEN1 ,
AND ZHAOYANG DONG4 , (Fellow, IEEE)
1 School of Electrical Engineering, Guizhou University, Guiyang 550025, China
2 Guizhou Power Grid Company, Guiyang 550001, China
3 College of Computer Science and Technology, Guizhou University, Guiyang 550025, China
4 School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, NSW 2031, Australia

Corresponding authors: Peijia Yu ([email protected]) and Xiangping Chen ([email protected])

This work was supported in part by the National Natural Science Foundation of China under Grant 51867005 and Grant 51867007, in part
by the Science and Technology Foundation of Guizhou Province under Grant [2016]1036, in part by the Guizhou Province Science and
Technology Innovation Talent Team Project under Grant [2018] 5615, in part by the Guizhou Province Reform Foundation for
Postgraduate Education under Grant [2016]02, in part by the Science and Technology Foundation of Guizhou Power Grid Company under
Grant 066500KK52170037, and in part by the Science and Technology Foundation of Guizhou Province under Grant [2018]5781.

ABSTRACT Voltage stability has always been one of the most important concerns. As the increasing
integration of large-scale renewable energy sources in power systems, the correlation between load demands
and renewable energy systems becomes more and more complex and important for probabilistic voltage
stability. There are two significant issues for probabilistic voltage stability assessment: (i) how to choose
the reasonable power increment direction which determines the reliability of voltage stability assessment
when considering the actual operating characteristics of the power system; and (ii) how to obtain the samples
characterized with the specified distribution and the desired correlation. We propose methodologies to define
the reasonable power increment direction with theoretical proof. Moreover, power method transformation
combined with Latin hypercube sampling and twice-permutation technique is proposed for probabilistic
voltage stability assessment. Case studies with two modified IEEE test systems show that the proposed
method is accurate and efficient.

INDEX TERMS Correlation, Latin hypercube sampling, Monte Carlo simulation, probabilistic voltage
stability, singular value decomposition, stochastic variables, twice-permutation.

I. INTRODUCTION affecting power system planning and operations. The effects

Voltage stability has always been one of the most impor- of these uncertainties need to be thoroughly explored in order
tant concerns in power system planning and operations [1]. to fully ensure system security. Therefore, the probabilistic
Over the past decades, deterministic voltage stability was voltage stability analysis was studied and the results pro-
studied in many previous literatures [2]–[4]. Load margin vided rich information to discover more useful insights than
is an efficient voltage stability index which is convenient the deterministic analysis did [9], [10]. For the probabilistic
to be understood and easy to be used for operators [5], [6]. voltage stability assessment, the covariance matrix between
For the deterministic voltage stability evaluation, the smallest stochastic variables can be used to measure the most proba-
load margin associated with the closest bifurcation point ble bifurcation point with an iterative method [11]. Authors
is always applied [7], [8]. However with the integrating of in [12] evaluated voltage stability with global sensitivity
large-scale renewable energy sources in power systems and analysis, in which the load increment direction is assumed
load variations, there are more and more stochastic factors as normal distribution. Authors in [13] calculated the volt-
age collapse point by increasing only one single bus load
The associate editor coordinating the review of this manuscript and and remaining the other constant. Authors in [14] adopted
approving it for publication was Arash Asrari . nonconforming load model as the load direction. The load

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 5407
 J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

model is assumed as normal distribution to estimate voltage the line and bus VSIs [19]. The max load margin is always
stability [15]. Authors in [16] estimated voltage stability with adopted to evaluate power system voltage stability as it is an
a direction by predicting load-path. efficient index, which is convenient to be understood and easy
However, two issues are still not solved well based on to be used for operators [20]. Optimization based methods are
the previous literature. Firstly, the correlation between the popular to calculate the max load margin [21]. In this paper,
stochastic variables are ignored, which are more and more the model based on optimal power flow is adopted to calculate
complex and important for the probabilistic voltage sta- the max load margin. The optimization model is as follows,
bility analysis. Secondly, the probabilistic voltage stability which is a nonlinear optimal problem and the interior point
is studied without fully explaining the actual power incre- method is used for solving it.
ment direction. The stochastic variables and the correlation
between them are not carefully considered when choosing max λ (1.1)
X
the power increment direction. Under such circumstances, subject to : Vi Vj (Gij cos θij
the load margin assessment may not truly reflect the actual + Bij sin θij ) + PLi − PGi − PRi = 0 (1.2)
stochastic characteristics of the system and therefore cannot X
be effectively used to evaluate probabilistic voltage stability. Vi Vj (Gij sin θij − Bij cos θij )
In order to deal with these two problems, based on a reason- + QLi − QGi − QRi = 0 (1.3)
able power increment direction for generators and loads con- PGi = P0Gi + λPdGi (1.4)
sidering the stochastic variables and the correlation between
PLi = P0Li + λPdLi (1.5)
them, we propose the power method combined with Latin
hypercube sampling and twice-permutation technique (PLT) QLi = Q0Li + λQdLi (1.6)
in this paper. The power increment direction is proven in PGi,min ≤ PGi ≤ PGi,max (1.7)
theory that the actual stochastic characteristics of the power QGi,min ≤ QGi ≤ QGi,max (1.8)
system are maintained. The PLT realizes the nonlinear cor-
Vi,min ≤ Vi ≤ Vi,max (1.9)
relation matrix transformation between the real stochastic
variables and the standard normal variables, and obtains the Pij = Vi Vj (Gij cos θij + Bij sin θij ) − tij Gij Vi2
samples of large scale input stochastic variables with desired (1.10)
correlation coefficients. 1
Qij = Vi Vj (Gij sin θij −Bij cos θij )+(tij − )Bij Vi2
The rest of this paper is organized as follows. Models of 2
typical stochastic sources and assessment of load margin are (1.11)
introduced in Section II. The method used to generate the P2ij + Q2ij ≤ Sij,max
2
(1.12)
samples of the stochastic variables with desired correlation
is proposed in Section III. The performances of the proposed In the optimization model above, the objective is to
method are studied with a modified IEEE 14-bus system maximize the load margin λ while satisfying power
and a modified IEEE 118-bus system in Section IV. Finally, flow constraints (1.2)-(1.3) and other operating con-
Section V concludes the paper. straints (1.4)-(1.12). PGi and QGi are the active power and
reactive power of traditional generator at bus i respectively.
PLi and QLi are the active power and reactive power of the
II. PROBABILISTIC VOLTAGE STABILITY MODEL
load at bus i respectively. PRi and QRi are the active power
A. STOCHASTIC SOURCES
and reactive power of the renewable energy source at bus
There are more and more stochastic sources in power systems
i respectively. P0Gi , P0Li and Q0Li are the base power of the
than before. Modeling of the stochastic characteristic and the
traditional generator and load at bus i respectively. Vi is the
correlation between the sources is essential for probabilistic
voltage of bus i. θij is the phase angle difference between bus
voltage stability assessment. Generally, the wind generation,
i and bus j, Gij and Bij are the conductance and susceptance
the solar generation and the load are described as PQ buses
of the line connecting the bus i and bus j respectively. Pij and
and the power factor is fixed. The active power of the wind
Qij are the active and reactive power flow from bus i to bus
generation, PW , is decided by the wind speed modeled as
j. tij is the transformation ratio of line from bus i to bus j.
the Weibull distribution, the active power of the photovoltaic
In this optimal model, the renewable energy sources keep
generation, PPV , is typically modeled as Beta distribution,
without power increment due to their intermittent outputs.
and the active power of the load, PL , is commonly modeled
While the powers of traditional generators and loads increase
as normal distribution [17], [18]. The correlation matrix, R,
with load margin λ and the power increment direction of PdGi ,
is symmetric.
PdLi , QdLi . As illustrated in Fig. 1, the load margin is sensitive
to the power increment direction. Different power increment
B. VOLTAGE STABILITY MODEL directions of generator and load power lead to different load
The index of load margin is one kind of overall voltage margins.
stability indices (VSIs), and in general the accuracy of the The actual operating characteristics of the power sys-
overall voltage stability indices (VSIs) is better than that of tem need to be considered to determine the reasonable

5408 VOLUME 8, 2020

J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

MCS is always computational expensive, the result obtained

by MCS is often used as a benchmark compared with other
methods. Obtaining accurate samples of stochastic variables
with desired correlation is most important for MCS, which
is focused in this paper. When the improvement of the com-
putational cost of MCS is needed, there are methods to be
selected [24].
With the rapidly increasing on the number of stochas-
tic variables in the power system, the correlation between
the stochastic variables cannot be ignored for probabilis-
FIGURE 1. Illustration of voltage stability load margin for different tic analysis. Cholesky decomposition is a feasible tech-
direction of generator and load power. nique and was widely used to deal with the correlation
between variables in condition of positive definitive corre-
power increment direction used. The correlation between the lation matrix [16], [17] [24]–[26]. But the correlation matrix
stochastic variables reflects the operating characteristics of is not always positive definitive in general especially when
the power system. In order to maintain the correlation of the the number of the correlated variables is large [27], which
input stochastic variables, the direction of base generator and makes the method infeasible. In order to deal with this prob-
load power is adopted. lem, the polynomial transformation was used [28]. However,
Theorem 1: the correlation coefficient of the stochastic the polynomial expansion terms are difficult to be solved
loads maintains invariable when loads increase with the direc- especially for large scale variables. An optimal permuta-
tion of base load power. tion technique based on Evolutionary algorithm is proposed
Proof: the correlation coefficient between any load i and in [27]. However, the evolutionary algorithm is time con-
load j with the increment direction of base load power is sumed. We proposed a twice-permutation technique based on
singular value decomposition (SVD) in [29]. This method is
rij ((1 + λ)PLi , (1 + λ)PLj )
valid and efficient even for non-positive definitive correlation
Cov((1 + λ)PLi , (1 + λ)PLj ) matrix.
= √
D((1 + λ)PLi ) D((1 + λ)PLj )
p
In this paper, the power method combined with Latin
(1 + λ)2 Cov(PLi , PLj ) hypercube sampling and twice-permutation technique is pro-
= √ posed for probabilistic voltage stability assessment. In this
(1 + λ)2 D(PLi ) D(PLj )
p
proposed method, power method based transformation real-
Cov(PLi , PLj )
= √ p = rij (PLi , PLj ) izes the nonlinear correlation matrix transformation between
D(PLi ) D(PLj ) the real stochastic variables and the standard normal vari-
Theorem 2: the correlation coefficient between the stochas- ables. Sampling efficiency is improved by Latin hypercube
tic loads and the renewable energy sources maintains invari- sampling (LHS). The desired correlation coefficients are
able when loads increase with the direction of base load maintained by the twice-permutation technique with high
power. efficiency and reliability.
Proof: in this situation, only the load power increases with
the direction of base load power. The correlation coefficient A. POWER METHOD BASED TRANSFORMATION
between any load i and renewable energy source j is Let Xi be the stochastic variable of any distribution, it can
be expressed as standard form below by transforming it’s
rij ((1 + λ)PLi , PRj ) expected value to zero.
Cov((1 + λ)PLi , PRj ) (1 + λ)Cov(PLi , PRj )
= √ = √ Xi − E(Xi )
D((1 + λ)PLi ) D(PRj ) (1 + λ) D(PLi ) D(PRj )
p p
Yi = (2)
Var(Xi )
Cov(PLi , PRj )
= √ p = rij (PLi , PRj ) where E(·) is the expected value function and Var(·) is the
D(PLi ) D(PRj )
variance value function. Yi is the standard form of Xi , which
has expected value of zero and variance value of one.
III. MATH GENERATION CORRELATED STOCHASTIC The standard stochastic variable can be approximated with
SAMPLES FOR PROBABILISTIC VOLTAGE STABILITY polynomials of mth-order as below [30].
ASSESSMENT
m
Power system probabilistic analysis is a powerful tool to dis- X
Yi = cir Zir (3)
cover the power system stochastic characteristics [22], [23].
r=0
Monte Carlo simulation (MCS) is a traditional and reli-
able mathematical technique for probabilistic analysis and where Zi is a standard normal distribution. It is the third-order
MCS with simple random sampling (SRS) has been used polynomial transformation when m = 3 [31], and is fifth-
in power system probabilistic analysis [17], [24]. Though order polynomial transformation when m = 5 [32]. cir is the

VOLUME 8, 2020 5409

 J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

polynomial coefficient obtained by solving equations with m where P, E and R∗ satisfy SVD equation (8). Q, D and RS
associated moments of Yi as shown below. satisfy SVD equation (9). RS is the correlation matrix of
Z Z Xm samples S.
E[Yim ] = Zim Yi dZi = Zim cir Zir dZi (4) 1
 1 0
r=0 R∗ = PEP0 = PE 2 PE 2 (8)
0
where E[Yim ] is the
 
mth-order moment of the stochastic vari- 1
RS = QDQ0 = QD 2 QD 2
1
(9)
able Yi , which can be calculated when the marginal distri-
bution or the samples of the stochastic variable are given. Permutating each row of S according to matrix S∗ , a sample
In this paper m is set as 5 and the simplifying equations are matrix Z with the same rank correlation matrix to that of
referenced in [32]. matrix S∗ is generated [34]. Then the correlation matrix of
When samples of the correlated non-normal stochastic Z is close to R∗ .
variable Xi with correlation matrix R are desired, the interme- Finally the stochastic samples X with approximate desired
diate correlation matrix R∗ of the standard normal variables correlation matrix R is calculated by using (2) and (3) with
Zi with the form in (5) must be calculated first. The equations substituting Z. Each column in the sample matrix X forms
about each nondiagonal correlation coefficient rij in R and its one set of samples to be used as inputs to the deterministic
corresponding intermediate correlation coefficient rij∗ in R∗ max load margin (1).
are built in [32].
∗ ∗
 
1 r12 · · · r1n C. PROBABILISTIC VOLTAGE STABILITY ASSESSMENT
 r∗ ∗  METHOD
 21 1 · · · r2n
R∗ =  . . . (5)

 .. .. .
··· . 
Once the stochastic samples with specified distribution and

∗ ∗ desired correlation are obtained, the probabilistic voltage sta-
rn1 rn2 ··· 1
bility assessment can be carried out by repeatedly determin-
B. SAMPLING AND PERMUTATION istic calculating. Then the probability characteristics of max
load margin can be obtained by statistics. The flowchart of the
In order to improve the probabilistic voltage stability assess-
probabilistic voltage stability assessment method is shown
ment efficiency, the samples of the n stochastic variables with
in Fig. 2. The corresponding calculating steps are below.
a size of k shown in (6) can be obtained by LHS method.
  Step 1: Input cumulative distribution function of each
s11 s12 · · · s1k stochastic variable and the desired correlation matrix R.
 s21 s22 · · · s2k  Step 2: Generate n×k sample matrix S by LHS.
S= . .. ..  (6)
 
 .. Step 3: Calculate S∗ by using (7).
. ··· . 
Step 4: Obtain Z by permutating S according to S∗ .
sn1 sn2 · · · snk
Step 5: Calculate samples X by using (2) and (3).
The quality and accuracy of the probability analysis solu- Step 6: Solve deterministic max load margin (1) repeatedly
tions can be affected by the correlation coefficients between with all stochastic samples X.
the stochastic variables. Therefore, the samples for proba- Step 7: Calculate the probability characteristics of max
bilistic voltage stability assessment must have the desired load margin results by statistics.
correlation coefficients. Cholesky decomposition based per- The following mean square error index of the correlation
mutation method is always adopted due to its less calculating matrix is used to measure the statistics accuracy of the sam-
cost. However, there must be two assumptions to be satis- ples.
fied. The first assumption is that samples S is independent, n
and the second assumption is that the correlation matrix of 1 X
ρcorr = 2 (rij − rijx )2 (10)
stochastic variables is positive definite [33]. In fact, the inde- n
i,j=1
pendence assumption of initial samples S generated by LHS
is always violated [26], and the desired correlation matrix where rij is the ith row and the jth column element of the
is not positive definite sometimes especially when there desired correlation matrix, rijx is the ith row and the jth column
are large scale stochastic variables in power systems [27]. element of the correlation matrix of the samples, and n is the
Therefore, the twice-permutation technique in [29] is used in total number of the stochastic variables.
this paper. The twice-permutation technique can decorrelate The relative errors of means and standard deviations are
initial samples generated by LHS and handle with the desired adopted in (11) and (12). They are used to measure the
correlation matrix, which even is non-positive definite. statistics accuracy of the max load margin.
The twice-permutation technique is described in (7).
µa − µs
εµ = × 100% (11)
 ∗
S1  1    0 µa
S∗ =  ...  = PE 2 Ŝ = PE 2
1 −1
 1
QD 2 S (7) σa − σs
 
∗ εσ = × 100% (12)
Sn σa
5410 VOLUME 8, 2020
J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

TABLE 1. Parameters of wind farms and photovoltaic plants.

TABLE 2. Details of load parameters.

where m is the total number of tests. µρcorr , µεµ and µεσ are
the means of the average errors of m tests, respectively. σρcorr ,
σεµ and σεσ are the standard deviations of average errors of
m tests, respectively.
FIGURE 2. Flowchart of the probabilistic voltage stability assessment
method.
IV. SIMULATION ANALYSIS
where the µa and σa are the accurate mean and standard In order to investigate the performance of the proposed prob-
deviation of the max load margin respectively, which are abilistic voltage stability assessment method, studies on two
obtained from MCS with SRS of a large enough sample size. modified IEEE test systems are explored. Matlab is used and
µs and σs are the mean and standard deviation of the max load the codes are performed with 1.60GHz CPU.
margin obtained by the method shown in Fig. 2.
Due to the random sampling process, error indices of multi- A. THE MODIFIED IEEE 14-BUS SYSTEM
ple tests are adopted to evaluate the steadiness of the proposed The standard IEEE 14-bus system is show in [35]. For the
method as below. modified test system, two wind farms are connected to bus
m 4 and bus 5 and two photovoltaic plants are connected to bus
1X
µρcorr = ρcorr (i) (13) 9 and bus 10 respectively. The power factor of each wind farm
m
i=1 is set as 0.95 and that of each photovoltaic plant is set as 1.0.
m
1 X The thermal limit of lines is set to be 100 MVA. Lower and
µεµ = εµ (i) (14)
m upper bounds of bus voltage are set as 0.8 p.u. and 1.2 p.u.,
i=1
m respectively.
1 X
The detail parameters of the wind farms and the photo-
µεσ = εσ (i) (15)
m voltaic plants are given in Table 1.
i=1
v The Details of the load parameters are shown in Table 2 The
u m
u1 X reactive power of load is determined by the fixed power
σρcorr =t (ρcorr (i) − µρcorr )2 (16)
m factor.
i=1
v There are 15 stochastic variables in total in this case. And
u m the correlation matrix of the stochastic variables is as the same
u1 X
σεµ = t (εµ (i) − µεµ )2 (17) in [29] and the corresponding intermediate correlation matrix
m
i=1 with power method transformation is non-positive definite.
v
u
u1 X m The stochastic variables on the buses connected directly are
σεσ =t (εσ (i) − µεσ )2 (18) assumed dependent. The correlation coefficients between dif-
m ferent types of variables are all considered.
i=1

VOLUME 8, 2020 5411

 J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

FIGURE 4. Error indices of 100 tests for max load margin (µεµ and µεσ ).

B. THE MODIFIED IEEE 118-BUS SYSTEM

The standard IEEE 118-bus system is introduced in [35].
This test system is divided into four areas covering
buses 1–31, 32–58, 59–92 and 94–118, respectively. For the
modified test system, four wind farms and four photovoltaic
plants are connected in every area. The wind farms are all
with the same capacity of 40 MW and the photovoltaic plants
are all with the same capacity of 20 MW. The wind farms are
connected to buses 1, 4, 6, 8, 32, 34, 36, 40, 62, 70, 72, 73, 99,
104, 105, 107. The photovoltaic plants are connected to buses
15, 18, 19, 24, 42, 43, 55, 56, 74, 76, 77, 85, 110, 112, 113,
116. The other parameters of wind farms and photovoltaic
plants are the same as those of the 14-bus study case above.
The mean of load active power is set as the deterministic load
active power of the standard test system and the standard
variation is set as 5%. As in the 14-bus study case above,
the load reactive power is determined by fixed power factor.
There are 131 stochastic variables in total in this case. The
FIGURE 3. Error indices of 100 tests for the correlation matrix.
correlation matrix of the stochastic variables is the same as
The probabilistic results of 10 000 times of MCS with SRS that in [29] and the corresponding intermediate correlation
are assumed to be accurate and used to calculate the error matrix with power method transformation is non-positive def-
of solutions obtained by PLT, power method transformation inite. The correlation coefficients between different variables
combined with SRS and twice-permutation (PST), and Nataf are all considered.
transformation. Different sample sizes are studied. The error indices of PLT and PST with sample size
The 100 tests mean square error index of the correlation 1500 are given in Table 3. The same conclusions as those
matrix is shown in Fig. 3. The results without considering in the modified IEEE 14-bus system can be drawn from the
correlation matrix RS of samples S are also shown in Fig. 3. simulation results.
The error indices of 100 tests for different types of output
stochastic variables are given in Fig. 4 and Fig. 5. Moreover, C. RESULTS DISCUSSION
the error indices with the sample size of 1500 are given It can be seen from Fig. 3 that twice-permutation technique
in Table 3. is valid even when the correlation matrix is non-positive
With sample size of 1500, the probability density and definite. PLT has smaller errors compared with PST and
cumulative distribution of load margin are shown in Fig. 6. Nataf transformation. More error produces when the stochas-
The results of PLT is closer to those of MCS. tic samples obtained by LHS is assumed independent. The

5412 VOLUME 8, 2020

J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

TABLE 3. Error indices comparisons (k = 1500).

PST and Nataf. The same conclusions can also be drawn from
the error indices with sample size 1500 given in Table 3.
The simulation results show that power method trans-
formation combined with Latin hypercube sampling and
twice-permutation technique can address the correlations
FIGURE 5. Error indices of 100 tests for max load margin (σεµ and σεσ ).
between the input stochastic variables and provide more
accurate results stably, even when the correlation matrix is
not positive definite. In addition, the consuming time of the
twice-permutation for 14-bus system and 118-bus system
are 0.035s and 0.167s respectively, and consuming time of
the once-permutation for 14-bus system and 118-bus system
are 0.031s and 0.161s respectively. The twice-permutation
technique is efficient.

V. CONCLUSION
In this paper, a reasonable power increment direction of
generator and load power is proposed for probabilistic volt-
age stability assessment. With the proposed direction the
actual operating characteristics of the power system are main-
tained from the viewpoint of the correlation between stochas-
tic variables. In order to obtain the samples characterized
with the specified distribution and the desired correlation
between large scale stochastic variables for probabilistic volt-
age stability assessment, a method based on power method
transformation combined with Latin hypercube sampling and
twice-permutation is proposed. The power method transfor-
mation realizes the nonlinear correlation matrix transforma-
tion between the real stochastic variables and the standard
FIGURE 6. Cumulative distribution and probability density of max load
margins. normal variables. The twice-permutation technique decorre-
lates initial samples and handles with the desired correlation
matrix, which even is non-positive definite. The method is
small error of the correlation matrix shows that the adopted tested with modified IEEE 14-bus and modified IEEE 118-
power increment direction reflects the correlation between bus systems. The simulation results show that the proposed
the stochastic variables and shows the actual operating char- method is valid and feasible.
acteristics of the power system.
In Fig. 4, µεµ and µεσ of the max load margin with PLT REFERENCES
are all smallest, which shows that the results with PLT is [1] G. Andersson, P. Donalek, R. Farmer, N. Hatziargyriou, I. Kamwa,
more accurate than those with PST and Nataf when the same P. Kundur, N. Martins, J. Paserba, P. Pourbeik, J. Sanchez-Gasca,
sample size is adopted. In Fig. 5, σεµ and σεσ of the max R. Schulz, A. Stankovic, C. Taylor, and V. Vittal, ‘‘Causes of the 2003
major grid blackouts in north america and europe, and recommended
load margin with PLT are also all smallest, which shows means to improve system dynamic performance,’’ IEEE Trans. Power
that the results with PLT is more stable than those with Syst., vol. 20, no. 4, pp. 1922–1928, Nov. 2005.

VOLUME 8, 2020 5413

 J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

[2] Z. Liu, W. Yao, and J. Wen, ‘‘Enhancement of power system stability using [24] Y. Chen, J. Wen, and S. Cheng, ‘‘Probabilistic load flow method based on
a novel power system stabilizer with large critical gain,’’ Energies, vol. 10, nataf transformation and latin hypercube sampling,’’ IEEE Trans. Sustain.
no. 4, pp. 449–462, Apr. 2017. Energy, vol. 4, no. 2, pp. 294–301, Apr. 2013.
[3] J. Zhang, J. Y. Wen, S. J. Cheng, and J. Ma, ‘‘A novel SVC alloca- [25] H. Chen, ‘‘Initialization for NORTA: Generation of random vectors with
tion method for power system voltage stability enhancement by normal specified marginals and correlations,’’ INFORMS J. Comput., vol. 13,
forms of diffeomorphism,’’ IEEE Trans. Power Syst., vol. 22, no. 4, no. 4, pp. 312–331, Nov. 2001.
pp. 1819–1825, Nov. 2007. [26] H. Yu, C. Chung, K. Wong, H. Lee, and J. Zhang, ‘‘Probabilistic load flow
[4] P. Kundur, ‘‘Definition and classification of power system stability evaluation with hybrid latin hypercube sampling and Cholesky decompo-
IEEE/CIGRE joint task force on stability terms and definitions,’’ IEEE sition,’’ IEEE Trans. Power Syst., vol. 24, no. 2, pp. 661–667, May 2009.
Trans. Power Syst., vol. 19, no. 3, pp. 1387–1401, May 2004. [27] X. Xu and Z. Yan, ‘‘Probabilistic load flow evaluation considering corre-
[5] S. Greene, I. Dobson, and F. Alvarado, ‘‘Contingency ranking for voltage lated input random variables,’’ Int. Trans. Elect. Energ. Syst., vol. 26, no. 3,
collapse via sensitivities from a single nose curve,’’ IEEE Trans. Power pp. 555–572, Mar. 2016.
Syst., vol. 14, no. 1, pp. 232–240, Feb. 1999. [28] H. Sheng and X. Wang, ‘‘Applying polynomial chaos expansion to assess
[6] S. Greene, I. Dobson, and F. Alvarado, ‘‘Sensitivity of the loading margin probabilistic available delivery capability for distribution networks with
to voltage collapse with respect to arbitrary parameters,’’ IEEE Trans. renewables,’’ IEEE Trans. Power Syst., vol. 33, no. 6, pp. 6726–6735,
Power Syst., vol. 12, no. 1, pp. 262–272, Feb. 1997. Nov. 2018.
[7] Y. V. Makarov, J. Ma, and Z. Y. Dong, ‘‘Non–iterative method to deter- [29] J. Zhang, G. Xiong, K. Meng, P. Yu, G. Yao, and Z. Dong, ‘‘An improved
mine static stability boundaries,’’ in Proc. IEEE Lausanne Power Tech., probabilistic load flow simulation method considering correlated stochas-
Jul. 2007, pp. 349–354. tic variables,’’ Int. J. Elect. Power Energy Syst., vol. 111, pp. 260–268,
[8] I. Dobson and L. Lu, ‘‘New methods for computing a closest saddle node Oct. 2019.
bifurcation and worst case load power margin for voltage collapse,’’ IEEE [30] T. C. Headrick and R. K. Kowalchuk, ‘‘The power method transformation:
Trans. Power Syst., vol. 8, no. 3, pp. 905–913, Aug. 1993. Its probability density function, distribution function, and its further use
for fitting data,’’ J. Stat. Comput. Simul., vol. 77, no. 3, pp. 229–249,
[9] S. Q. Bu, W. Du, and H. F. Wang, ‘‘Probabilistic analysis of small–
Mar. 2007.
signal rotor angle/voltage stability of large–scale AC/DC power systems as
[31] A. I. Fleishman, ‘‘A method for simulating nonnormal distributions,’’
affected by grid–connected offshore wind generation,’’ IEEE Trans. Power
Psychometrika, vol. 43, no. 4, pp. 521–532, Dec. 1978.
Syst., vol. 28, no. 4, pp. 3712–3719, Nov. 2013.
[32] T. C. Headrick, ‘‘Fast fifth-order polynomial transforms for generating uni-
[10] Y. Kataoka, ‘‘A probabilistic nodal loading model and worst case solutions
variate and multivariate nonnormal distributions,’’ Comput. Statist. Data
for electric power system voltage stability assessment,’’ IEEE Trans. Power
Anal., vol. 40, no. 4, pp. 685–711, Oct. 2002.
Syst., vol. 18, no. 4, pp. 1507–1514, Nov. 2003.
[33] Z. Q. Xie, T. Y. Ji, M. S. Li, and Q. H. Wu, ‘‘Quasi–Monte Carlo
[11] M. Mittelstaedt, S. Bauer, and A. Schnettler, ‘‘An iterative method for
based probabilistic optimal power flow considering the correlation of wind
determining the most probable bifurcation in large scale power systems,’’
speeds using copula function,’’ IEEE Trans. Power Syst., vol. 33, no. 2,
IEEE Trans. Power Syst., vol. 32, no. 4, pp. 2966–2973, Jul. 2017.
pp. 2239–2247, Mar. 2018.
[12] X. Xu, Z. Yan, M. Shahidehpour, H. Wang, and S. Chen, ‘‘Power system [34] M. Stein, ‘‘Large sample properties of simulations using latin hypercube
voltage stability evaluation considering renewable energy with correlated sampling,’’ Technometrics, vol. 29, no. 2, pp. 143–151, May 1987.
variabilities,’’ IEEE Trans. Power Syst., vol. 33, no. 3, pp. 3236–3245, [35] Power Systems Test Case Archive. Accessed: 2018. [Online]. Available:
May 2018. http://www.ee.washington.edu/research/pstca
[13] R. Preece, N. C. Woolley, and J. V. Milanovic, ‘‘The probabilistic collo-
cation method for power–system damping and voltage collapse studies in
the presence of uncertainties,’’ IEEE Trans. Power Syst., vol. 28, no. 3,
pp. 2253–2262, Aug. 2013. JING ZHANG received the Ph.D. degree from the
[14] E. Haesen, C. Bastiaensen, J. Driesen, and R. Belmans, ‘‘A probabilistic Huazhong University of Science and Technology,
formulation of load margins in power systems with stochastic generation,’’ Wuhan, Hubei, China, in 2008. He is currently a
IEEE Trans. Power Syst., vol. 24, no. 2, pp. 951–958, May 2009. Professor with the School of Electrical Engineer-
[15] B. Tamimi and S. Vaez-Zadeh, ‘‘An optimal pricing scheme in electricity ing, Guizhou University.
markets considering voltage security cost,’’ IEEE Trans. Power Syst.,
vol. 23, no. 2, pp. 451–459, May 2008.
[16] M. Perninge, V. Knazkins, M. Amelin, and L. Soder, ‘‘Risk estimation of
critical time to voltage instability induced by saddle–node bifurcation,’’
IEEE Trans. Power Syst., vol. 25, no. 3, pp. 1600–1610, Aug. 2010.
[17] B. R. Prusty and D. Jena, ‘‘A critical review on probabilistic load flow stud-
ies in uncertainty constrained power systems with photovoltaic generation
and a new approach,’’ Renew. Sustain. Energy Rev., vol. 69, pp. 1286–1302, LUQIN FAN is currently with the School of Elec-
Mar. 2017. trical Engineering, Guizhou University.
[18] X. Wang, T. Wang, H.-D. Chiang, J. Wang, and H. Liu, ‘‘A framework for
dynamic stability analysis of power systems with volatile wind power,’’
IEEE J. Emerg. Sel. Topics Circuits Syst., vol. 7, no. 3, pp. 422–431,
Sep. 2017.
[19] J. Modarresi, E. Gholipour, and A. Khodabakhshian, ‘‘A comprehensive
review of the voltage stability indices,’’ Renew. Sustain. Energy Rev.,
vol. 63, pp. 1–12, Sep. 2016.
[20] A. Sode-Yome, N. Mithulananthan, and K. Lee, ‘‘A maximum loading
margin method for static voltage stability in power systems,’’ IEEE Trans.
Power Syst., vol. 21, no. 2, pp. 799–808, May 2006.
[21] O. Alizadeh Mousavi and R. Cherkaoui, ‘‘Maximum voltage stability mar- YING ZHANG received the Ph.D. degree from
gin problem with complementarity constraints for multi–area power sys- Guizhou University, in 2012. She is currently with
tems,’’ IEEE Trans. Power Syst., vol. 29, no. 6, pp. 2993–3002, Nov. 2014. Guizhou Power Grid Company.
[22] Q. Xiao and S. Zhou, ‘‘Probabilistic power flow computation using quadra-
ture rules based on discrete Fourier transformation matrix,’’ Int. J. Elect.
Power Energy Syst., vol. 104, pp. 472–480, Jan. 2019.
[23] X. Lin, Y. Jiang, S. Peng, H. Chen, J. Tang, and W. Li, ‘‘An efficient
Nataf transformation based probabilistic power flow for high-dimensional
correlated uncertainty sources in operation,’’ Int. J. Elect. Power Energy
Syst., vol. 116, Mar. 2020, Art. no. 105543.

5414 VOLUME 8, 2020

J. Zhang et al.: Probabilistic Assessment Method for Voltage Stability Considering Large Scale Correlated Stochastic Variables

GANG YAO received the M.D. degree from KE MENG (Member, IEEE) received the Ph.D.
Guizhou University, Guiyang, Guizhou, China, degree from the University of Queensland,
in 2011. He is currently with Guizhou Power Grid Brisbane, QLD, Australia, in 2009. He is currently
Company. with the School of Electrical Engineering and
Telecommunications, University of New South
Wales.

PEIJIA YU received the M.D. degree from the XIANGPING CHEN received the Ph.D. degree in
Dalian Jiaotong University, Dalian, Liaoning, electronic and electrical engineering from the Uni-
China, in 2011. She is currently with the School versity of Newcastle upon Tyne, U.K., in 2013. She
of Computer Science and Technology, Guizhou is currently an Associate Professor with the School
University. of Electrical Engineering, Guizhou University.

GUOJIANG XIONG received the Ph.D. degree ZHAOYANG DONG (Fellow, IEEE) received the
from the Huazhong University of Science and Ph.D. degree in electrical engineering from The
Technology, Wuhan, Hubei, China, in 2015. He is University of Sydney, Sydney, NSW, Australia,
currently with the School of Electrical Engineer- in 1999. He is currently a Professor with the
ing, Guizhou University. School of Electrical Engineering and Telecommu-
nications, University of New South Wales.

VOLUME 8, 2020 5415