Electrical Power and Energy Systems 82 (2016) 392–399

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Statistical method for on-line voltage collapse proximity estimation

Fredy A. Sanz a,⇑, Juan M. Ramirez b, Johnny Posada c
a
Institución Universitaria Salazar y Herrera, Carrera 70 No. 52-49, Medellín, Colombia
b
CINVESTAV del IPN, Avenida del Bosque 1145, Zapopan, Mexico
c
Universidad Autónoma de Occidente, Calle 25 No. 115-85, Cali, Colombia

a r t i c l e i n f o a b s t r a c t

Article history: This paper is aimed to propose a reliable method for estimating the voltage collapse proximity through a
Received 1 September 2015 model obtained using statistical techniques. In the model building process a database is required,
Received in revised form 17 March 2016 therefore the Voltage Collapse Proximity Index (VCPI) is used to obtain previous readings for different
Accepted 20 March 2016
contingencies and loading conditions. This analytical proposal could be combined with existing
equipment in the power system control centers for future on-line applications. The proposed method
is applied to the IEEE 14-bus test system and the 190-bus Mexican equivalent. Results indicate that
Keywords:
the proposed strategy is a reliable choice.
Statistic
Voltage stability
Ó 2016 Elsevier Ltd. All rights reserved.
Voltage collapse
Estimation

Introduction provide information to power system control center, which fosters

take decisions preventing voltage collapse.
A power system may become vulnerable for several reasons: Although several methods conducting to voltage stability anal-
natural calamities, component failures, protection and control ysis; accurate voltage collapse predictions and rapid voltage stabil-
failures, information and communication failures, instability due ity analysis providing reduction in computation time, which are
to disturbances, human errors, inadequate security assessment needed to successfully prevent voltage collapses and blackouts.
procedures, sabotage, and missing or uncertain information in Researchers have presented some alternatives for on-line pre-
decision making [1]. Causes of instability that are internal to the dicting or analyzing voltage collapse. In [4] a forecasting-aided
civil infrastructure may be reduced by decreasing the probability state estimator is employed to detect the proximity of a voltage
and severity of occurrence through the improved engineering of collapse problem, taking into account system loads trends is pro-
related systems. On the other hand, causes of instability that are posed. The exact voltage collapse point is determined through
external to the infrastructure (e.g., different contingencies) may the use of an extrapolation technique. Other approach is presented
be reduced by decreasing the severity of occurrence by construct- by [5], the method requires the voltage phasor and angle informa-
ing defender restoration systems [2]. tion at a bus and the network admittance matrix to evaluate the
Power system stability involve the system changes in normal system voltage stability at a bus in the form of Voltage Collapse
operation and changes in the working point caused by distur- Prediction Index (VCPI). This is an impractical situation in a real
bances. Although, small signal security problems studied alter- power system because network configuration may change due to
ations of the power system in the presence of small changes in contingencies that modify the admittance matrix, requiring tech-
the electrical variables, for these reason power grid security niques to know the current state of the power grid.
include major changes commonly called contingencies [3]. In voltage security assessment an Artificial Intelligence-based
Considering that is important the system monitoring to know techniques are applied in [6,7], where Neural Networks and
when it is close to an unsafe zone, in this work is proposed a Genetic Algorithms are used. Starting from the state estimation, a
method to estimate the proximity to voltage collapse. Specifically, given set of mathematical indices is computed to represent a snap-
considering the most important smart grids characteristics, like shot of the current electric system operating point. In [8] a method
for use on-line voltage collapse predications, theoretically proven
for finding voltage collapse point precisely and determining volt-
age stability. Other on-line application is presented in [9], applica-
⇑ Corresponding author.
ble for large scale power systems, the contingencies are defined
E-mail addresses: [email protected] (F.A. Sanz), [email protected]
(J.M. Ramirez), [email protected] (J. Posada).
previously and collapse prediction is possible if the model knows

http://dx.doi.org/10.1016/j.ijepes.2016.03.035
0142-0615/Ó 2016 Elsevier Ltd. All rights reserved.
 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399 393

the contingency, situation that is not always possible especially (i) It requires long computing time and (ii) require load increment
when simultaneous contingencies occur or in a complex power information (because it is no possible real-time load estimation)
systems. [17–19].
Several algorithms have been developed that make use of Researchers have presented some alternatives for on-line
synchrophasor technology for online voltage stability monitoring predicting voltage collapse, in [20] the method requires the voltage
[10–13], however, these methods assume that know the current phasor, and angle information, and the admittance matrix to
network configuration, a situation which in most cases is not evaluate the system voltage stability in the form of VCPI. This is
possible due to different faults that may occur in power grid. an impractical situation in a real power system because network
Considering the above arguments, this paper proposes a consis- configuration may change due to contingencies that modify the
tent method to voltage collapse proximity estimation. The paper is admittance matrix, requiring techniques to know the current state
presented in the following order. Section ‘‘Problem description” of the power grid. Other on-line application is presented in [9,21],
justifies the proposed method; Section ‘‘Voltage stability model applicable for large scale power systems, the contingencies are
and statistical overview” describes theoretic considerations, and defined previously and collapse prediction is possible if the model
Section ‘‘Proposal” presents our contribution; Section ‘‘Results of knows the contingency, situation that is not always possible
190-bus Mexican equivalent power grid” shows the results of the especially when simultaneous contingencies occur in a complex
method applied to 190-bus Mexican power grid equivalent; finally, power systems.
conclusions are presented. Currently, it is relevant propose novel applications to power
system monitoring, ensuring a safe operation, providing tools to
prevent voltage collapse. In this way researchers have proposed
Problem description different techniques that evaluate security conditions and so to
identify corrective actions to maintain proper system operation.
Naturally, voltage stability is a dynamic process, although Specifically statistical methods present a promising alternative in
steady-state assessment the use of steady-state is accepted gener- which VCPI changes could be predicted with measurable on-line
ally. Considering voltage stability theory, the use of static and variables [22].
dynamic methods must provide similar results, provided that the
analysis is not made during a transitional phenomenon. Voltage
Voltage stability model and statistical overview
collapse may happen on power systems which are heavily loaded,
faulted and/or have reactive power shortages. In fact, voltage
The conventional power flow model is generally expressed by
collapse often affects the entire power system, although it usually
equations as [22,23]:
arises in one particular area. Maintaining a system voltage profile
within an acceptable range in power system operations, may dx
¼ f ðx; y; kÞ ð1Þ
improve system security and reliability, and prevents system dt
collapse [14–16]. Operation beyond acceptable range limits leads 0 ¼ gðx; y:kÞ
to voltage instability and ultimately to voltage collapse. Relevance
of voltage collapse indices lies in the fact that they are useful tools where x is the vector of state variables; y is a vector, and k is a factor
for voltage collapse proximity estimation. But, it is necessary to use that change gradually, when power system is moving from an equi-
under on-line context. That means, know in real time all parame- librium point to another until reaching the collapse point are used
ters required to quick index calculation. equations,
Considering that voltage stability in a power system may be
affected by different reasons, highlighting load changes and contin-
PD;i ¼ PD0;i ð1 þ kP;i kÞ ð2Þ
gencies are predominant factors, the voltage stability index repre- Q D;i ¼ Q D0;i ð1 þ kQ ;i kÞ ¼ PD0;i tanðui Þð1 þ kQ ;i kÞ
sents a computational advantage respect to Continuation Power
Flow (CPF). Now to compare voltage stability behavior using CPF where PD,i and QD,i represent the active and reactive power demand
and the Voltage Collapse Prediction Index (VCPI), is presented the at i-th bus, respectively, kP,i and, kQ,i are a constant that specify the
14-Bus IEEE test system, see results in Fig. 1. The CPF is not used rate of change in power generation when k (load parameter) is
under real-time applications due to the following disadvantages: varied; followed to calculation in power demand, as [17,22]:

PG;i ¼ PG0;i ð1 þ kG;i kÞ ð3Þ

where PG0,i is the i-th initial active power generated; kG,i is the con-
stant specifying the rate of change in generation when k is varied.
Commonly, in steady-state analysis are used the following
methods: modal analysis, direct method, optimization method,
and continuation method [13].

Voltage stability indices

The security analysis in power grids usually divide voltage

instabilities in two classes: the first one is due to electrical
variables changes (e.g. power and voltage variations), and the other
one is due to configuration changes (e.g. generators and transform-
ers outages). The Voltage Collapse Proximity Index (VCPI) calculate
voltage stability condition in each bus [20,24,25]. The VCPI index
used the conventional power flow definition, and it calculated
the injected power in each bus k. The calculation of this index
requires the voltage phasor information of buses as well as the
Fig. 1. Comparison between the Global VCPI prediction and CPF method. bus admittance matrix. The VCPI for the k-th bus is defined as,
394 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399

 PN 
 0 
Step 1: variable selection criteria
 m¼1;m–k V m 
VCPIi ¼ 1   ð4Þ
 Vk 
An automatic searching procedure that sequentially develops
Y km
V 0m ¼ PN Vm ð5Þ the ‘‘best” subset of V (bus voltage variables) may be helpful. The
j¼1;j–k Y kj forward stepwise regression procedure is probably the most
widely used. It was developed to economize computational efforts,
where Vk is the voltage phasor at bus k, Vm is the voltage phasor at
as compared with other possible regression procedures. Essen-
bus m, Ykm is the admittance between bus k and bus m, Ykj is the
tially, this searching method develops a sequence regression mod-
admittance between bus k and bus j, k is the studied bus, m is other
els, at each step is added or delete a V variable, which corresponds
bus connected to bus k. The VCPI takes values between [0, 1]. If the
to some bus voltage. The criterion for adding or deleting a V
index is 0, the voltage at bus k is assumed stable and if the index is
variable is stated in terms of error sum squares reduction, coeffi-
1, a voltage collapse occurs. For the operating condition, the Global
cient of partial correlation, t-statistic, or F-statistic [27].
VCPI is defined as the maximum VCPI found midst in all buses at any
Where, the independent variables are the bus voltages, whereas
specific time.
the global VCPI is the prediction variable. By include a new variable
Complete studies have been presented to evaluate the predic-
in model he Root Mean Square Error (RMSE) must be reduced, the
tive ability of VCPI index. Comparison of stability index based on
RMSE is defined by [28]
powers and voltages are presented in [5,20]. Researchers have
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
shown that VCPI is a consistent vulnerability measure, due to it 1 Xn
calculated the precise voltage collapse proximity. RMSE ¼ ðy  y ^ Þ2 ð6Þ
n i¼1

VCPI’s statistical overview where y ^ are the VCPI estimate values and y are the power flow
values, for n samples. The RMSE is calculated considering each
Due to the power system probabilistic behavior, uncertainties Xi inclusion. Thus, the stepwise regression used in variables selec-
associated with load demand, resulting in problems about accurate tion is presented in Fig. 2, lower plot shows RMSE evolution and
load determination. Consequences of random failures of power upper each variable behavior.
system components are generally outside the control of the opera- The lengths of the horizontal lines of each variable indicate the
tor. But, it is not easy to reflect all types of uncertainties associated 95% confidence interval. If p-value is greater than 0.05 is not
with any possible consequences using probabilities. In this way, acceptable because it indicates the level of the expected error.
statistical tools provide arguments to analyze the correlation Notice that variables X1, X2, X3, X6, X7, X8 (bus voltages) are
between variables that are present in the process of determining excluded from the model because observed errors and the
vulnerability of a power system. Specifically in the case of VCPI, p-value are close to 1. The lower plot shows the RMSE for each
depends on the voltage and on the admittance matrix, may be bus voltage included in the model. Thus, for the test system under
displayed if a relationship exist between obtained values through study, voltages at buses 1, 2, 3, 6, 7, and 8 are not transcendental
VCPI on each bus and the given value in the load parameter k. from voltage collapse viewpoint. Each power system will require
Through the analysis and design of experiments theory, if any a similar analysis to verify those buses which are transcendental.
factor has random effects, this is ordinarily conducted with the Thus, for the IEEE 14-buses test system bus voltages V4, V5, V9,
aim to draw inferences, not on the specific factor levels included V10, V11, V12, V13, and V14 are selected as independent variables.
in the experiment, but mainly on the variability of the population
(factor is analog to variables, e.g. load levels, contingencies, etc.). Step 2: factors values criteria
Inferences on data have been gathered from all factor levels e.g.
each load value or load range is considered a level. Inferences on Robustness may be limited by the modeling conditions, being
random factor, considering that each factor may have many possi- able to guarantee a specific immunity only under considered
ble levels, but only a random sample of levels is included in the conditions. However, model validity may be assessed against not
data. In the power system vulnerability context is clear that considered phenomena. Initially, the proposed method is applied
random behavior is present, because parameters like load and con- to simulations data, obtained using power flow computation (or
tingencies have not predictable occurrence. a mixed computation of CPF method with N  1 line contingency),
subjecting the system to different load conditions and single line
Proposal contingencies simultaneously; bus voltages and the VCPI for each
condition are obtained. Therefore, model validity it should provide
Considering the difficulties presented in real applications of good vulnerability estimates for considered cases. If there is a dif-
voltage-collapse-proximity on-line estimation, and considering ferent condition, a good estimate is not guaranteed, hence the
the assumption that data can be measured in this context, one importance of taking different scenarios into account [29]. For
could affirm that it is possible through statistical modeling to the above reason, factors, levels, and ranges are limited by voltage
obtain an estimation of voltage collapse proximity [26]. Classical collapse incidence un data base is built under these considerations.
power system stability techniques and stability indices, presented
satisfactory simulations results, however some proposals assume Step 3: model criteria
that the power grid contingency is known, i.e. the admittance
matrix is known. On the other hand, some heuristic techniques The best option to select the model is based in a specific metric
are used, but predictions lack of reliability indicators. For these about model errors [30,31]. Four different models are considered:
reasons the need to propose methods that do not require further full model (FM), interactions model (IM), pure quadratic model
network information and allow a confidence level of predictions (PM), and linear model (LM), Fig. 3. Residual plots behavior are
becomes important, due to the size of the networks, and unex- helpful to determine whether one model is to be preferred to
pected load changes, and the simultaneously multiple contingen- another using some specific criteria, such as error terms evalua-
cies occurrence. Suggesting a new reliable on-line method, useful tion. The appropriate metric is based on normality of data, in this
with online measurable variables, which has a high degree of case in Fig. 4 a non-normality conduct is presented in each plot.
confidence. Therefore the RMSE is a proposed to evaluate the models, results
 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399 395

Fig. 2. Variables selection: where the main buses are identified.

For each
scenario:

Linear models
= + 1
0 1 + 2 2 +⋯
⋯+ −1 , −1
Bus Voltage
of n bars
Exponential
Multivariable
model = dl
0 exp( 1 )+
(Vulnerability
estimator)

Vulnerabilty Logistic models

indicator
0
= +
1+ 1 exp( 2 )

Fig. 3. Different type of models.

are below: RMSELM = 0.0748; RMSEPM = 0.0658; RMSEIM = 0.0555; Finally, the last metric is compared with the Mean Square Error
RMSEFM = 0.0536. (MSE) inferring about model validation, where MSE ¼ RMSE2 [28],
Thus, evaluating all RMSEs obtained, all values are of the same unlike n is the sample size (n⁄). In the analyzed case n = 985 and
rank and are close to each other. Additionally, considering the linear n⁄ = 197. Now, in the studied system, MSPRLM = 0.0068 and
model has less parameters, for these reasons, the linear model may MSELM = 0.0055, where the validation criteria indicates that
be appropriate for this test system, and it is selected. Therefore, the MSPRLM > MSELM, for the above reason it can be said that the
linear model is evaluated to verify that the estimations are consis- selected model is reliable [19].
tent respect to expectations. Its equation is written as,

VCPI ¼ b0 þ b1 V bus 4i þ b2 V bus 5i þ b3 V bus 9i þ b4 V bus 10i

Step 5: data analysis criteria

þ b5 V bus 11i þ b6 V bus 12i þ b7 V bus 13i þ b8 V bus 14i ð7Þ Statistical method are used to data analysis, if the experiment
where coefficients bk are derived based on the regression procedure, has been designed correctly and performed accordingly, the statis-
and Vbus mi represents the i-th sample of voltage magnitude at bus tical model needs not be oversize. Hypothesis test and confidence
m. interval estimation procedures are very useful in data analysis.
Theory indicates that the F test is a powerful tool to infer on model
Step 4: validation criteria validation. In this paper data analysis is studied according to error
terms behavior (normally and not normally distributed). Therefore,
Referring to the proposed method, the model validation proce- the test of significance regression is evaluated using the ANOVA
dure uses a different dataset. A metric named MSPR (Mean Squared test and contrasted with the non-parametric test (Kruskal–Wallis
Prediction Errors), which is defined as [28], Test) F test, defined as [28]:
Pn 
^i Þ2 H0 : l1 ¼ l2 ¼    ¼ lk ¼ 0
i¼1 ðyi  y
MSPR ¼ ð8Þ H1 : not all li are equal ð9Þ
n
where yi is the computed value for the i-th validation case, y^i is the where l1, l2, . . . , lk are the regression variables’ means. Rejection of
predicted value for the i-th validation case, n⁄ it is the sample size. H0 implies that at least one of the variables X1, X2, . . . , Xk contributes
396 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399

Fig. 4. Residual plots for different models: (a) linear model; (b) pure quadratic model; (c) interaction model; and (d) full quadratic model.

considerably to the model. The test for significance regression Considering the influential case presented in the study case a
determine whether a linear relationship exists between the model depiction and its parameters from regression, (before and
response variable VCPI and a subset of regression variables V1, after robust regression), are presented in (10), where Global
V2, . . . , Vk, Rejection of H0 implies that at least one of the bus voltages VCPImodeli is the model structure, Global VCPIfitted is the obtained
(regressor variables) V1, V2, . . . , Vk contribute significantly to the model after robust regression, and Global VCPIRobust is the robust
model and reject H0. If F0 exceeds Fa, k, nk1 (probability that model.
F-value is less than or equal that confidence interval, where a is
Global VCPImodeli ¼ b0 þ b1 V bus 4i þ b2 V bus 5i þ b3 V bus 9i
de confidence interval, and k is the degree of freedom). Alterna-
tively, the p-value approach may be used in hypothesis testing þ b4 V bus 10i þ b5 V bus 11i þ b6 V bus 12i
and, reject H0 if the p-value for statistical F0 is less than). The test þ b7 V bus þ b8 V bus ð10aÞ
13i 14i
is usually summarized in ANOVA test presented in Table 1, where
a p-value = 6.2108e102 is obtained, indicating that a data set has
Global VCPIfitted ¼ 0:9978 þ 0:5170V bus 4  1:1800V bus 5
a reasonable fit to the normal distribution, and the classical ANOVA
test is more sensitive to differences among groups. þ 0:5273V bus 9 þ 0:2686V bus 10

 0:2898V bus 11  0:2208V bus 12

Step 6: improvement criteria þ 0:2801V bus 13  0:8225V bus 14 ð10bÞ

Model improvement depends of remedial measures, like: errors Global VCPIRobust ¼ 0:9369  0:1040V bus 4  0:3919V bus 5
variance, multicollinearity, and influential cases. Linear model was
þ 0:3968V bus 9  0:0983V bus 10
selected previously. Using the described factors in Section ‘‘Step 2:
factors values criteria” it is necessary identify which of them may  0:0120V bus 11  0:3841V bus 12

render the model improvement. Boxplots in Fig. 5 shows evidence þ 0:3955V bus 13  0:6780V bus 14 ð10cÞ
of influential cases, manifested in the outside data from interquar-
tile ranges, these data may induce errors in the prediction [31,32]. It is noteworthy that VCPImodeli is the chosen linear model structure,
Therefore improved model using a robust regression gives better which includes only the required variables for the prediction as
predictions. indicated in Steps 1–4. The VCPIfitted is the result of Steps 5–6, where
the linear model is statistically validated and analyzed. VCPIRobust is
the outcome of robust regression application, where outliers are
removed allowing more accurate prediction.
Table 1
The full robust predictor takes all bus voltages in step 2 into
Summary of the ANOVA test.
account. Fig. 6 shows a graphic of the multivariable predictor,
Source of Sum of Degrees of Mean F- p-Value which illustrates the bus-14 voltage (continuous line), where 95%
variation squares freedom square value
confidence boundary is used (dashed lines around the predictor).
Bus voltage 11.631 7 1.66163 103.36 3.05152e145 Current value of variable is represented by the vertical dashed line,
Error 126.557 7872 0.01608
obtaining the prediction value for a specific condition (horizontal
Total 138.189 7879
dashed line).
 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399 397

Fig. 5. Boxplot for each relevant bus voltage.

Fig. 6. Bus 14: Estimation of bus voltage, their upper and lower limits, and the VCPI.

In order to verify the reliability of the proposition, in Fig. 1 a com- for each model are presented, it is noteworthy that for the four
parison between the Global VCPI prediction and CPF method is models tested, all metrics values are close to each other, and the
made. It is demonstrated that while voltages are going to collapse, RMSE for the linear model is the highest value, therefore when
the VCPI grows inversely proportional to the voltages, indicating the other three models are considered evaluating the parameters
that system is moving into a vulnerable operating condition. amount, the pure quadratic model has least parameters, thus pure
quadratic model it is faster to provide an estimate. For these
Results of 190-bus Mexican equivalent power grid reasons, the quadratic model may be appropriate for this test
system, and it is selected.
The results presented below are from equivalent of the Mexican Model validation and statistical data analysis were performed
power grid, which consists of 190 buses, 49 generators, and 265 as was proposed in the method, the results were satisfactory,
lines, see Fig. 7. The system is subjected to simple contingencies finally model improvement was developed, in which influential
in each line, in addition to contingencies in each generator. The data were found, and therefore a robust regression was performed.
system is subjected to gradual load changes. Power flows are calcu- Finally, in Table 4, it have been evaluated some states of the grid,
lated in each case; in this process database is built, which is used specifically two separated contingencies (lines and generators
for proposing the prediction model. outages) and other status with the network operating under a
The method process was applied to 190-Bus Mexican equiva- normal load condition. Also VCPI index calculations are presented
lent. In variables selection it was determined that only requires for each considered contingency. Finally errors obtained compar-
knowing the voltages of 47 buses to predict voltage collapse in ing the actual value with value obtained through the model are
the system. In step factors, levels and ranges considerations were presented. The results show good behavior model. Similarly, previ-
the same as in the 14 buses case. ous procedure is made after robust regression procedure, which
Furthermore, to assess the predictive capacity have been evalu- shows that the resulting errors are smaller than in the previous
ated different models: pure quadratic, full quadratic, linear, and case, see Table 5. In both tables are considered the same operating
interaction models, evaluation results are summarized in Table 2, conditions, using in the first case the initial model and in the sec-
and the other hand in Table 3 the quantity of parameters required ond case the robust regression
398 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399

Fig. 7. 190.-Bus Mexican equivalent.

Table 4
Table 2 Evaluation of scenarios for the Mexican equivalent using fitted model.
Statistical measures for different models of Mexican equivalent.
Operation condition VCPI computed VCPI predicted Error (%)
Statistical Linear Pure Interaction Full results value
metric quadratic model quadratic
Normal 0.1323915 0.1314937 0.7
RMSE 0.0336 0.0239 0.0199 0.0186 Line 1 outage 0.2557218 0.2605349 1.8
MSE 0.0011 0.0006 0.0004 0.0004 Generator outage at bus 45 0.4238335 0.4297563 1.4

Table 5
Table 3 Evaluation of scenarios for the Mexican equivalent using robust model.
Parameters for different models of the Mexican equivalent.
Operation condition VCPI computed VCPI predicted Error (%)
Linear Pure Interaction Full results value
quadratic model quadratic
Normal 0.1323915 0.1316983 0.5
Quantity of 47 93 1082 1128 Line 1 outage 0.2557218 0.2585357 1.1
parameters Generator outage at bus 45 0.4238335 0.4279531 1
 F.A. Sanz et al. / Electrical Power and Energy Systems 82 (2016) 392–399 399

Conclusions [11] Biswas SS, Vellaithurai CB, Srivastava AK. Development and real time
implementation of a synchrophasor based fast voltage stability monitoring
algorithm with consideration of load models. In: Industry applications society
VCPI is an index that has been widely documented, often annual meeting, 2013 IEEE. p. 1–9. 6–11 October.
proposed as an alternative for on-line prediction. However, it is [12] Hu Fengkai, Sun Kai, Del Rosso A, Farantatos E, Bhatt N. An adaptive three-bus
power system equivalent for estimating voltage stability margin from
requires knowing the electrical grid configuration, a situation
synchronized phasor measurements. In: PES general meeting | conference &
which is not possible when fault as line outages are presented. exposition, 2014 IEEE. p. 1–5. 27–31 July.
For the above reason data obtained by simulation for a test system [13] Maslennikov S, Litvinov E, Vaiman M, Vaiman M. Implementation of ROSE for
on-line voltage stability analysis at ISO New England. In: PES general meeting |
are subjected to statistical tests. Specifically, VCPI dependence was
conference & exposition, 2014 IEEE. p. 1–5. 27–31 July.
assessed in relation to the different bus voltages and load changes. [14] Varshney S, Srivastava L, Pandit M, Sharma M. Voltage stability based
Evidence was found that vulnerability can be treated as a voltage contingency ranking using distributed computing environment. In: 2013
stability problem. Therefore, the use of VCPI index is proposed to International Conference on Power, Energy and Control (ICPEC). p. 208–12.
[15] S. DasGupta, M. Paramasivam, U. Vaidya, V. Ajjarapu, Real-time monitoring of
determine the voltage collapse proximity. short-term voltage stability using PMU data. IEEE Trans Power Syst. http://
The six criterias that constitute the method are defined to build dx.doi.org/10.1109/TPWRS.2013.2258946.
a model, which is able to predict the voltage collapse proximity [16] Leonardi B, Ajjarapu V. An approach for real time voltage stability margin
control via reactive power reserve sensitivities. IEEE Trans Power Syst
using the bus voltage as predictor variables. At each criteria theo- 2013;28:615–25.
retical tools are defined to be used according to the data nature. [17] Nakawiro W. Voltage stability assessment and control of power systems using
The method was explained using the IEEE 14-bus test system computational intelligence Ph.D. thesis. University of Duisburg-Essen; 2011.
[18] Ben-Kilani K, Elleuch M. Structural analysis of voltage stability in power
and applied to 190-bus Mexican equivalent power grid, obtaining systems integrating wind power. IEEE Trans Power Syst. http://dx.doi.org/10.
a suitable model under simulated conditions, using checking data. 1109/TPWRS.2013.2258043.
Proposed method is based on statistical techniques, which [19] Correa GJ, Yusta JM. Grid vulnerability analysis based on scale-free graphs
versus power flow models. Electric Power Syst Res 2013;101:71–9.
enables: consideration of different phenomena and contingencies
[20] Bulac C, Tristiu I, Mandis A, Toma L. On-line power systems voltage stability
that occur in a large power system; evaluation of different models; monitoring using artificial neural networks. In: 2015 9th international
validation of selected model; the data are analyzed statistically, symposium on Advanced Topics in Electrical Engineering (ATEE). p. 622–5.
7–9 May.
and the resulting model may be improved. Therefore, the method
[21] Duraipandy P, Devaraj D. On-line voltage stability assessment using least
allows the construction of reliable models for voltage collapse squares support vector machine with reduced input features. In: 2014
proximity estimation. International Conference on Control, Instrumentation, Communication and
Computational Technologies (ICCICCT). p. 1070–4. 10–11 July.
[22] Power system stability subcommittee special publication (Ed.), Voltage
stability assessment: concepts, practices, and tools. Product number
References SP101PSS, IEEE Power Engenieering Society; 2002. ISBN: 0780378695.
[23] Jesus PDO-D, Alvarez M, Yusta J. Distribution power flow method based on a
real quasi-symmetric matrix. Electric Power Syst Res 2013;95:148–59.
[1] Teng J-H, Chen C-Y, Martinez I, Chen C-F. Power system vulnerability
[24] Kessel P, Glavitsch H. Estimating the voltage stability of a power system. IEEE
assessment considering energy storage systems. In: 2013 IEEE 10th
Power Deliv 1986;1(3):346–54.
international conference on Power Electronics and Drive Systems (PEDS).
[25] Perez Londono S, Rodriguez Garcia L, Lopez YU. Effects of doubly fed wind
p. 903–7.
generators on voltage stability of power systems. In: Proc. Sixth IEEE/PES
[2] Mousavi OA, Bozorg M, Cherkaoui R. Preventive reactive power management
Transmission and Distribution: Latin American Conf. and Exposition (T&D-LA).
for improving voltage stability margin. Electric Power Syst Res 2013;96:36–46.
p. 1–6.
[3] Zio E, Sansavini G. Vulnerability of smart grids with variable generation and
[26] Caro E, Mınguez R, Conejo AJ. Robust WLS estimator using reweighting
consumption: a system of systems perspective. IEEE Trans Syst Man Cybern
techniques for electric energy systems. Electric Power Syst Res
2013;43(3):477–87.
2013;104:9–17.
[4] de Souza A, De Souza JCS, Leite da Silva A. On-line voltage stability monitoring.
[27] Zhou N, Pierre J, Trudnowski D. A stepwise regression method for estimating
IEEE Trans Power Syst 2000;15(4):1300–5.
dominant electromechanical modes. IEEE Trans Power Syst 2012;27
[5] Balamourougan V, Sidhu TS, Sachdev MS. Technique for online prediction of
(2):1051–9.
voltage collapse. IEE Proc Gener Transm Distrib 2004;151(4):453–60.
[28] Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied linear statistical models. 5th
[6] Berizzi A, Bovo C, Cirio D, Delfanti M, Merlo M, Pozzi M. Online fuzzy voltage
ed. McGraw-Hill; 2005.
collapse risk quantification. Electric Power Syst Res 2009;79(5):740–9.
[29] Lahoud N, Faucher J, Malec D, Maussion P. Electrical aging of the insulation of
[7] Sanchez L, Couso I, Viera JC. Online SOC estimation of Li-FePO4 batteries
low-voltage machines: model definition and test with the design of
through a new fuzzy rule-based recursive filter with feedback of the heat flow
experiments. IEEE Trans Industr Electron 2013;60(9):4147–55.
rate. In: Vehicle Power and Propulsion Conference (VPPC), 2014 IEEE. p. 1–6.
[30] Liu L, Jin X, Min G, Xu L. Real-time diagnosis of network anomaly based on
27–30 October.
statistical traffic analysis. In: 2012 IEEE 11th international conference on trust,
[8] Althowibi FA, Mustafa M. On-line voltage collapse indicator for power systems.
security and privacy in computing and communications (TrustCom). p.
In: 2010 IEEE international conference on power and energy (PECon).
264–70.
p. 408–13.
[31] Dufrenois F, Noyer J. Formulating robust linear regression estimation as a one-
[9] Chiang H-D, Li H, Tong J, Causgrove P. On-line voltage stability monitoring of
class LDA criterion: discriminative hat matrix. IEEE Trans Neural Networks
large power systems. In: 2011 IEEE power and energy society general meeting.
Learn Syst 2013;24(2):262–73.
p. 1–6.
[32] Yin S, Wang G. A modified partial robust m-regression to improve prediction
[10] Duong DT, Uhlen K. Online voltage stability monitoring based on PMU
performance for data with outliers. In: 2013 IEEE International Symposium on
measurements and system topology. In: 2013 3rd international conference on
Industrial Electronics (ISIE). p. 1–6.
Electric Power and Energy Conversion Systems (EPECS). p. 1–6. 2–4 October.