See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/230626792

Artificial Neural Network based Approach to Analyze Transient Overvoltages

during Capacitor Banks Switching

Article  in  International Journal of Computer Applications · June 2011

CITATION READS

1 106

3 authors:

Iman Sadeghkhani abbas Ketabi

Islamic Azad University, Najafabad Branch University of Kashan
74 PUBLICATIONS   425 CITATIONS    92 PUBLICATIONS   707 CITATIONS   

SEE PROFILE SEE PROFILE

Rene Feuillet
Grenoble Institute of Technology
94 PUBLICATIONS   1,373 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Fuse saving scheme View project

Optimization of New Geometries of Switched Reluctances Motors Using FEM And Multi Objective Genetic Algorithm View project

All content following this page was uploaded by abbas Ketabi on 10 July 2014.

The user has requested enhancement of the downloaded file.

 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

Artificial Neural Network based Approach to Analyze

Transient Overvoltages during Capacitor Banks Switching
Iman Sadeghkhani1, Abbas Ketabi2, Rene Feuillet3
(1,2)
Department of Electrical Engineering, University of Kashan, Kashan, Iran
(3)
Laboratoire d’Electrotechnique de Grenoble, INPG/ENSIEG, Cedex, France

ABSTRACT paper that can give the maximum switching overvoltage will be
The quality of electric power has been a constant topic of study, helpful to the operator. It can be used as training tool for the
mainly because inherent problems to it can lead to great operators. The proposed ANN is expected to learn many
economic losses, especially in industrial processes. Among the scenarios of operation. To give the maximum peak overvoltages
various factors that affect power quality, those related to in a shortest computational time which is the requirement during
transients originating from capacitor bank (CB) switching in the online operation of power systems.
primary distribution systems must be highlighted. This paper In the proposed ANN we have considered the most important
presents an Artificial Neural Network (ANN)-based approach to aspects, which influence the transient overvoltages such as
estimate the transient overvoltages due to capacitor energization. voltage at capacitor bus before switching, equivalent resistance,
In proposed methodology, Levenberg-Marquardt second order equivalent inductance, equivalent capacitance, line length,
method is used to train the multilayer perceptron. ANN training switching angle, and capacitor capacity. This information will
is based on equivalent parameters of the network. Therefore, help the operator to select the proper condition of capacitor
trained ANN is applicable to every studied system. The switching with transients appearing safe within the limits.
developed ANN is trained with the extensive simulated results, Results of the studies are presented for a partial of 39-bus New
and tested for typical cases. Then the new algorithms are England test system to illustrate the proposed approach.
presented and demonstrated for a partial of 39-bus New England
test system. The simulated results show that the proposed 2. MODELLING ISSUES
technique can estimate the peak values of switching
overvoltages with good accuracy. 2.1 PSB
Simulations presented in this paper are performed using the
Keywords PSB. The simulation tool has been developed using state
Artificial neural networks; capacitor banks switching; switching variable approach and runs in the MATLAB/Simulink
overvoltages. environment. This program has been compared with other
popular simulation packages (EMTP and Pspice) in [4]. The user
1. INTRODUCTION friendly graphical interfaces of PSB enable faster development
Electric power systems have predominantly inductive loads, for power system transient analysis.
so that the systems themselves must supply the reactive power
consumed. The most practical and efficient way for the utility to 2.2 Generator model
supply the reactive power demanded is through the installation In [5] generators have been modeled by generalized Park’s
of capacitor bank (CBs) in the system. The installation of shunt model that both electrical and mechanical part are thoroughly
CB brings benefits concerning the reduction of system charging modeled, but it has been shown that a simple static generator
and electrical losses, system capacity release, and also model containing an ideal voltage source behind the sub-
improvements in the power factor [1, 2]. transient inductance in series with the armature winding
resistance can be as accurate as the Park model. Thus in this
Although various factors influence power quality, the work work, generators are represented by the static generator model.
presented here focuses on transients originating from shunt Phases of voltage sources are determined by the load flow
capacitor bank switching in power systems. results.
The magnitude and shape of the switching overvoltages vary
with the system parameters and network configuration. Even
2.3 Transmission line model
Transmission lines are described by the distributed line
with the same system parameters and network configuration, the
model. This model is accurate enough for frequency dependent
switching overvoltages are highly dependent on the
parameters, because the positive sequence resistance and
characteristics of the circuit breaker operation and the point-on-
inductance are fairly constant up to approximately 1 KHz [6]
wave where the switching operation takes place [3].
which cover the frequency range of harmonic overvoltages
In this paper power system blockset (PSB), a phenomena.
MATLAB/Simulink-based simulation tool [4] is used for
computation of both switching and temporary overvoltages. This 2.4 Load and Shunt Devices Model
paper presents the artificial neural network (ANN) application All of the loads and shunt devices, such as capacitors and
for estimation of overvoltage peaks under switching transients reactors, are modeled as constant impedances.
during capacitor energization. A tool such as proposed in this

5
 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

3. TRANSIENT OVERVOLTAGES 1.8

DURING CAPACITOR ENERGIZATION 1.5
One of the major concerns in power system restoration is the 1.2
0.9
occurrence of overvoltages as a result of switching procedures.

Voltage [p.u.]
0.6
These can be classified as transient overvoltages, sustained 0.3
overvoltages, harmonic resonance overvoltages, and 0
overvoltages resulting from ferro-resonance. Steady-state -0.3
overvoltages occur at the receiving end of lightly loaded -0.6
transmission lines as a consequence of line-charging currents -0.9
(reactive power balance). Excessive sustained overvoltages may -1.2
lead to damage of transformers and other power system -1.5
equipment. Transient overvoltages are a consequence of -1.8
0 0.02 0.04 0.06 0.08 0.1 0.12
switching operations on long transmission lines, or the switching
of capacitive devices, and may result in arrester failures. Ferro- Time [s]
resonance is a nonharmonic resonance characterized by Fig 2: Voltage at bus 2 after switching of capacitor.
overvoltages whose waveforms are highly distorted and can
cause catastrophic equipment damages [7-10].
• Equivalent capacitance of the network
This paper concentrates on the estimation of switching
overvoltages during capacitor energization. The CB switching • Line length
provokes transient overvoltages that theoretically can reach peak
• Closing time of the circuit breaker poles
phase-to-earth values in the order of 2-3 p.u.
• Capacitor bank capacity
The sample system considered for explanation of the
proposed methodology is a 400 kV EHV network shown in Fig.
1. The normal peak value of any phase voltage is 400√2/√3 kV In proposed method, equivalent parameters of the network as
and this value is taken as base for voltage p.u. In the system well as other parameters are used as ANN inputs. Thus, ANN is
studies 400 kV line-to-line base voltage and 100 MVA as a base trained just once for simple system of Fig. 1 and developed
power is considered. Fig. 2 shows the switching transient at bus ANN is applicable to every studied system. For using developed
2 when capacitor is energized. ANN, just studied system must convert to Fig. 1. Section 5 has
more details about proposed method.
In practical system a number of factors affect the
overvoltages factors due to energization or reclosing. In this Source voltage affects the overvoltage strongly. Fig. 3 shows
paper following parameters is considered: the effect of source voltage on overvoltage at different
equivalent resistance. Fig. 4 shows the effect of line length on
• Voltage at capacitor bus before switching
overvoltages at different source voltage. Controlled switching of
• Equivalent resistance of the network high-voltage ac circuit breakers has become a commonly
accepted means of controlling switching transients in power
• Equivalent inductance of the network systems [11]. Fig. 5 shows effect of switching angle on
overvoltages at different equivalent capacitance. Fig. 6 shows
the effect of shunt capacitor capacity on overvoltages at
different equivalent inductance.
As discussed above for an existing system the main factors
which affect the peak values of switching overvoltage are
capacitor bus voltage, equivalent resistance, equivalent
inductance, equivalent capacitance, line length, switching angle,
and capacitor capacity. Here it should be mentioned that a single
parameter often cannot be regarded independently from the
Fig 1: Sample system for capacitor energization study. G: other important influencing factors. The magnitude of the
generator, Reqv: equivalent resistance, Leqv: equivalent overvoltages normally does not depend directly on any single
inductance, and Ceqv: equivalent capacitance. isolated parameter and a variation of one parameter can often
alter the influence of another parameter, in other words there
exists an interaction between the various system and breaker
parameters. This forbids the derivation of precise generalized
rule of simple formulae applicable to all cases [12]. So an ANN
can help to estimate the peak values of switching overvoltages
generated during reactor energization. An ANN is programmed
by presenting it with training set of input/output patterns from
which it then learns the relationship between the inputs and

6
 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

2.2 1.8
Reqv = 0.003 p.u. Ceqv = 1.282 p.u.
2.1 1.7
Reqv = 0.006 p.u. Ceqv = 2.5 p.u.
2
1.6
1.9
1.5
Voltage [p.u.]

Voltage [p.u.]
1.8
1.7 1.4
1.6
1.3
1.5
1.2
1.4
1.3 1.1

1.2 1
0.85 0.9 0.95 1 10 20 30 40 50 60 70 80 90
Source Voltage [p.u.] Switching Angle [deg.]

Fig 3: Overvoltage peak at bus 2 as source voltage while Fig 5: Overvoltage peak at bus 2 as switching angle while
equivalent inductance 0.025 p.u., equivalent capacitor 1.2825 source voltage 0.9 p.u., equivalent resistance 0.003 p.u.,
p.u., line length 200 km, switching angle 20°, and shunt equivalent inductance 0.03 p.u., line length 150 km, and
capacitor capacity 30 MVAR. Reqv is equivalent resistance. capacitor capacity 20 MVAR. Ceqv is equivalent
capacitance.
2.6
S.V. = 0.85 p.u.
2.4 2.4
S.V. = 0.95 p.u.
Leqv = 0.02 p.u.
2.2 2.2 Leqv = 0.035 p.u.

2
Voltage [p.u.]

Voltage [p.u.] 2

1.8
1.8
1.6
1.6
1.4
1.4
1.2
1.2
1
100 150 200 250 300
Line Length [km] 1
10 20 30 40 50
Fig 4: Overvoltage peak at bus 2 as line length while Shunt Capacitor Capacity [MVAR]

equivalent resistance 0.004 p.u., equivalent inductance 0.025 Fig 6: Overvoltage peak at bus 2 as shunt capacitor capacity
p.u., equivalent capacitance 1.2825 p.u., switching angle 20°, while source voltage 0.9 p.u., equivalent resistance 0.003 p.u.,
and capacitor capacity 30 MVAR. S.V. is source voltage. equivalent capacitance 1.8912 p.u., line length 150 km, and
switching angle 30°. Leqv is equivalent inductance.
outputs. In next section an ANN-based approach is described
which can give a acceptable solution of switching transients by composition of the input variables for the proposed neural
the help of which an operator can take a quick decision at the networks has been carefully selected.
time of operation.
Supervised training of ANN is a usual training paradigm for
4.THE ARTIFICIAL NEURAL MLP architecture. Fig. 8 shows the supervised learning of ANN
NETWORK for which input is given to PSB to get the peak values of
The proposal in this work considers the adoption of feed transient overvoltages and the same data is used to train the
forward Multilayer Perceptron (MLP) architecture. A MLP ANN. Error is calculated by the difference of PSB output and
trained with the back-propagation algorithm may be viewed as a ANN output. This error is used to adjust the weight of
practical vehicle for performing a nonlinear input–output connection. Since the switching transient demands a solution
mapping of a general nature [3, 13]. Function approximation by with high precision, the neural network has to be trained
feed forward MLP network is proven to be very efficient, considering a very small stopping criterion. Output values of the
considering various learning strategies like simple back trained neural networks must be capable of computing the
propagation or the robust Levenberg–Marquardt. Its ability to voltages with very good precision. Gradient-based training
perform well is affected by the chosen training data as well as algorithms, like back propagation, are most commonly used for
training scheme. The schematic diagram of the proposed MLP training procedures. They are not efficient due to the fact that
neural networks architecture is shown in Fig. 7. The the gradient vanishes at the solution. Hessian-based algorithms
allow the network to learn more subtle features of a complicated
mapping. The training process converges quickly as the solution

7
 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

Fig 7: Proposed MLP-based ANN architecture.

increased ξ (x) . When a step reduces ξ (x) , µ is divided by β.

Notice that when µ is large the algorithm becomes steepest
descent; while for small µ the algorithm becomes Gauss–
Newton. The LM algorithm is very efficient when training
networks have up to few hundred weights. Although the
computational requirements are much higher for the each
iteration of the LM algorithm, this is more than made up for by
the increased efficiency. This is especially true when high
precision is required.
Fig 8: Supervised learning of ANN.
4.2 Training Artificial Neural Network
All experiments have been repeated for different system
is approached, because the Hessian does not vanish at the parameters. After learning, all parameters of the trained
solution. To benefit from the advantages of Hessian based networks have been frozen and then used in the retrieval mode
training, we focused on the Levenberg–Marquardt (LM) for testing the capabilities of the system on the data not used in
algorithm reported in [14]. learning. The testing data samples have been generated through
the PSB program by placing the parameter values not used in
4.1 Levenberg-Marquardt (LM) Algorithm learning, by applying different parameters. A large number of
Suppose that we have a function ξ (x) which we want to testing data have been used to check the proposed solution in the
minimize with respect to the parameter vector x, where most objective way at practically all possible parameters
variation. Percentage error is calculated as:
N
2
ξ (x) = ∑e
i =1
i (x) (1)
error(%) =
ANN − PSB
× 100 (3)
PSB
where ei (x) is the error for ith input. Then the Marquardt– Neural network is trained with the goal of mean square error
Levenberg modification to the Gauss–Newton method is (MSE) 1e-3. Fig. 9 shows the training of neural network. Results
for a sample test data are presented in Table 1 and Figs. 10-11.
[
∆ x = J T (x)J (x) + µI ]−1
J T ( x )e ( x ) Table 1 contains the some sample result of test data. Values in
column VPSB are the absolute values of peak voltage at bus 2
(2) calculated by PSB program in p.u. where the VANN values are
the values simulated by trained network. Also, Fig.10 shows
where J (x ) is the jacobian matrix. The parameter µ is
overvoltage peak at bus 2 against the line length and Fig.11
multiplied by some factor β whenever a step would result in an shows overvoltage peak at bus 2 against the shunt capacitor

8
 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

capacity.

Table 1. Some sample testing data and output

V [p.u.] Reqv [p.u.] Leqv [p.u.] Ceqv [p.u.] L.L. [km] S.A. [deg.] C [MVAR] VPSB [p.u.] VANN [p.u.] errorV [%]
0.778 0.0055 0.025 1.2825 125 20 30 1.2589 1.2397 1.5261
0.947 0.0065 0.025 1.2825 325 20 30 2.2192 2.2029 0.7326
1.006 0.0035 0.025 1.2825 325 20 30 2.3791 2.3354 1.8367
0.856 0.0045 0.025 1.2825 175 20 30 1.5191 1.5741 3.6205
0.997 0.0065 0.025 1.2825 275 20 30 2.0904 2.0643 1.2495
0.999 0.0035 0.025 1.2825 275 20 30 2.1173 2.1496 1.5236
0.997 0.0055 0.025 1.2825 225 20 30 1.9829 2.0247 2.1058
0.921 0.0035 0.025 1.2825 125 20 30 1.4702 1.4895 1.3146
0.884 0.003 0.0225 0.3694 150 15 40 1.8591 1.8521 0.3763
0.887 0.003 0.0225 2.1956 150 15 5 1.0916 1.0813 0.9475
0.892 0.003 0.0275 1.5869 150 65 5 1.2704 1.3046 2.6918
0.892 0.003 0.0275 1.5869 150 85 35 2.0641 2.0921 1.3582
0.901 0.003 0.0325 2.8044 150 45 15 1.3656 1.3176 3.5174
0.895 0.003 0.0325 0.9781 150 5 25 1.3541 1.3312 1.6892
0.902 0.003 0.0375 1.5869 150 75 5 1.1871 1.1973 0.8557
0.905 0.003 0.0375 2.8044 150 55 40 2.2417 2.2062 1.5834
V = capacitor bus voltage before switching, Reqv = equivalent resistance, Leqv = equivalent inductance, Ceqv = equivalent capacitance, L.L. =
line length, S.A. = switching angle, C = shunt capacitor capacity, and errorV = voltage error.

Performance is 0.000996931, Goal is 0.001 equivalent capacitance 1.1285 p.u., switching angle 20°, and
capacitor capacity 15 MVAR.
Training-Solid line Goal-Dash line

0
10 2.4
PSB
2.2 ANN

2
-2
10
Voltage [p.u.]

1.8

1.6

-4 1.4
10
0 5 10 15 20
23 Epochs 1.2

Fig 9: Squared error against epoch curve. 1

5 15 25 35 45
Shunt Capacitor Capacity [MVAR]
2.8
PSB
Fig 11: Overvoltage peak vs. shunt capacitor capacity at bus
2.6 2 simulated by ANN and PSB while source voltage 0.925
ANN
2.4 p.u., equivalent resistance 0.003 p.u., equivalent inductance
2.2
0.0225 p.u., equivalent capacitance 2.1956 p.u., line length
150 km, and switching angle 45°.
Voltage [p.u.]

1.8
In the next section, the proposed model tested with portion of
1.6 39-bus New England test system. Various cases of capacitor
1.4 energization are taken into account and corresponding peak
1.2
values estimated from trained model.
1
125 175 225 275 325
5. CASE STUDY
Line Length [p.u.] In this section, the proposed algorithm is demonstrated for
two case studies that are a portion of 39-bus New England test
Fig 10: Overvoltage peak vs. line length at bus 2 simulated system, of which its parameters are listed in [15]. The
by ANN and PSB while source voltage 0.9 p.u., equivalent simulations are undertaken on a single phase representation.
resistance 0.0055 p.u., equivalent inductance 0.025 p.u.,

9
 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

In the proposed method, first, studied system must convert to equivalent inductance, and equivalent capacitance are
equivalent circuit of Fig. 1, i.e., values of equivalent resistance,

Table 2. Case 1 some sample testing data and output

V [p.u.] S.A. [deg.] C [MVAR] VPSB [p.u.] VANN [p.u.] errorV [%]
0.768 10 17 1.0841 1.0945 0.9562
0.768 50 17 1.2172 1.2248 0.6249
0.831 50 17 1.2915 1.2581 2.5863
0.831 50 35 1.6316 1.6083 1.4275
0.895 45 15 1.2709 1.2478 1.8184
0.895 90 33 1.8497 1.7651 4.5736
0.937 30 12 1.2802 1.3024 1.7351
0.937 75 24 1.5398 1.5801 2.6159
V = capacitor bus voltage before switching, S.A. = switching angle, C = shunt capacitor capacity, and errorV = voltage error.

and corresponding peak overvoltages are computed from PSB

program and trained ANN. In this case, values of equivalent
resistance, equivalent inductance, and equivalent capacitance are
0.00731 p.u., 0.02513 p.u., and 1.5724 p.u., respectively.
Summery of few result are presented in Table 3. It can be seen
Fig 12: Studied system for case 1. from the results that the ANN is able to learn the pattern and
give results to acceptable accuracy.

calculated. These values are used in trained artificial neural 6. CONCLUSION

network to estimate overvoltages peak. In this paper a ANN approach has been suggested to estimate
the peak overvoltages due to capacitor energization. The
5.1 Case 1 Levenberg–Marquardt second order training method has been
Fig. 12 shows a one-line diagram of a portion of 39-bus New adopted for obtaining small mean square error (MSE) without
England test system. First, equivalent circuit of this system, seen losing generalization capability of ANN. The results from this
behind bus 16, is determined and values of equivalent resistance, scheme are close to results from the conventional method and
equivalent inductance, and equivalent capacitance are helpful in predicting the overvoltage of the other case studies
calculated. In other words, this system is converted to equivalent within the range of training set. The proposed ANN approach is
system of Fig. 1. In this case, equivalent parameters are 0.00385 tested on a partial 39-bus New England test system.
p.u., 0.03129 p.u., and 2.0674 p.u., respectively. For testing
trained ANN, values of voltage at capacitor bus (bus 19) before
switching, switching angle, and capacitor capacity are varied
and in each case, overvoltage peak values are calculated from
trained ANN and system of Fig. 12. Table 2 contains the some
sample result of test data for case 1.

5.2 Case 2
As another example, the system in Fig. 13 is examined. After
converting this system to equivalent circuit of Fig. 1 and
calculating equivalent circuit parameters seen from bus 5, Fig. 13: Studied system for case 2.
various cases of capacitor energization are taken into account

Table 3. Case 2 some sample testing data and output

V [p.u.] S.A. [deg.] C [MVAR] VPSB [p.u.] VANN [p.u.] errorV [%]
0.754 70 20 1.3325 1.3584 1.9426
0.754 70 42 2.0094 1.9796 1.4835
0.822 15 42 1.7158 1.7785 3.6539
0.822 55 33 1.5418 1.5084 2.1657
0.879 90 21 1.5788 1.5921 0.8401
0.879 90 14 1.1526 1.1323 1.7592
0.925 30 12 1.2697 1.3103 3.1964
0.925 60 40 1.9712 1.9391 1.6285
V = capacitor bus voltage before switching, S.A. = switching angle, C = shunt capacitor capacity, and errorV = voltage error.

10
 IJCA Special Issue on “Artificial Intelligence Techniques - Novel Approaches & Practical Applications”
AIT, 2011

7. REFERENCES [8] Ketabi, A., Ranjbar, A.M., and Feuillet, R., “Analysis and
[1] Coury, D.V., Santos, C.J., Oleskovicz, M., and Tavares, Control of Temporary Overvoltages for Automated
M.C., “Transient analysis concerning capacitor bank Restoration Planning,” IEEE Trans. Power Delivery, vol.
switching in a distribution system,” Electric Power System 17, 1121–1127, 2002.
Research, vol. 65, 13–21, 2002. [9] Sadeghkhani, I., “Using Artificial Neural Network for
[2] Hwang, C., and Lou, J.N., “Transient analysis of Estimation of Switching and Resonance Overvoltages
capacitance switching for industrial power system by during Bulk Power System Restoration,” M.Sc.
PSpice,” Electric Power System Research, vol. 45, 28–38, dissertation, Department of Electrical Engineering,
1998. University of Kashan, 2009.
[3] Thukaram, D., Khincha, H.P., Khandelwal, S., “Estimation [10] Sadeghkhani, I., Ketabi, A., and Feuillet, R., “Estimation
of switching transient peak overvoltages during of Temporary Overvoltages during Power System
transmission line energization using artificial neural Restoration using Artificial Neural Network,” in Proc.
network,” Electric Power System Research, vol. 76, 259– IEEE 15th International Conference on Intelligent System
269, 2006. Applications to Power Systems, Curitiba, Brazil, 2009.
[4] Sybille, G., Brunelle, P., Hoang, L., Dessaint, L.A., and Al- [11] Sadeghkhani, I., Ketabi, A., and Feuillet, R., “New
Haddad, K., “Theory and applications of power system Approach to Harmonic Overvoltages Reduction during
blockset, a MATLAB/Simulink-based simulation tool for Transformer Energization via Controlled Switching,” in
power systems,” in Proc. IEEE Power Eng. Soc. Winter Proc. IEEE 15th International Conference on Intelligent
Meeting, 774–779, 2000. System Applications to Power Systems, Curitiba, Brazil,
2009.
[5] Duro, M.M., “Damping Modelling in Transformer
Energization Studies for System Restoration: Some [12] Cigre Working Group, “Switching overvoltages in EHV
Standard Models Compared to Field Measurements,” in and UHV systems with special reference to closing and
Proc. IEEE Bucharest Power Tech Conference, Bucharest, reclosing transmission lines,” Electra, no. 30, 70–122,
Romania, 2009. 1973.
[6] Boliaris, P.G., Prousalidis, J.M., Hatziargyriou, N.D., and [13] Hagan, M.T., Menhaj, M.B., “Training feedforward
Papadias, B.C., “Simulation of long transmission lines networks with the Marquardt algorithm,” IEEE Trans.
energization for black start studies,” in Proc. 7th Neural Network, vol. 5, 989-993, 1994.
Mediterranean Electrotechn. Conf., 1093–1096, 1994. [14] Haykin, S., Neural Network: A Comprehensive
[7] Adibi, M.M., Alexander, R.W., and Avramovic, B., Foundation, 2nd ed., Prentice Hall, 1998.
“Overvoltage control during restoration,” IEEE Trans. [15] Wunderlich, S., Adibi, M.M., Fischl, R., and Nwankpa,
Power Syst., vol. 7, 1464–1470, 1992. C.O.D., “An approach to standing phase angle reduction,”
IEEE Trans. Power Syst., vol. 9, 470–478, 1994.

11

View publication stats