1

Novel Algorithm for Online Voltage Stability

Assessment Based on Feed Forward Neural Network
S. Kamalasadan, Member IEEE, A. K. Srivastava, Member, IEEE, D. Thukaram, Senior Member, IEEE
AbstractThis paper presents an Artificial Feed Forward
Neural Network (FFNN) approach for the assessment of power
system voltage stability. A novel approach based on the inputoutput relation between real and reactive power, as well as voltage
vectors for generators and load buses is used to train the Neural
Net (NN). The input properties of the feed forward network are
generated from offline training data with various simulated
loading conditions using a conventional voltage stability algorithm
based on the L-index. The neural network is trained for the Lindex output as the target vector for each of the system loads. Two
separate trained NN, corresponding to normal loading and
contingency, are investigated on the 367 node practical power
system network. The performance of the trained Artificial Neural
Network (ANN) is also investigated on the system under various
voltage stability assessment conditions.
As compared to the computationally intensive benchmark
conventional software, near accurate results in the value of Lindex and thus the voltage profile were obtained. Proposed
algorithm is fast, robust and accurate and can be used online for
predicting the L-indices of all the power system buses The
proposed ANN approach is also shown to be effective and
computationally feasible in voltage stability assessment as well as
potential enhancements within an overall energy management
system in order to determining local and global stability indices.
Index Terms Artificial Neural Network, Voltage Stability,
Quasi Newton and Levenberg Marquardt Algorithms, Back
Propagation method.

I. INTRODUCTION

OLTAGE stability analysis is concerned with the ability

of assessing the power system to maintain acceptable
voltages at all system buses under normal conditions and after
being subjected to disturbances. Several works have been
conducted previously for the prediction of voltage stability and
proximity to collapse conditions based on conventional
techniques, using sensitivity indices and based on continuation
methods [1]-[10]. Continuation method based voltage stability
analysis techniques are fairly accurate but hampered by the
fact of taking longer computational time being based on
repetitive power flow. For online voltage stability, index suits
well, if it is accurate indicator of voltage stability margin and
can be calculated fast. Some of the works focus on static
voltage stability analysis using such instability measuring
Sukumar Kamalasadan is with the Division of Engineering and Computer
Technology, University of West Florida, FL 32514 USA (e-mail:
skamalasadan@uwf.edu ).
A K Srivastava is with the Department of Electrical and Computer
Engineering, Mississippi State University, Box 9571, Mississippi State, MS
39762, USA (e-mail: srivastava@ece.msstate.edu ).
D. Thukaram is with the Department of Electrical Engineering, Indian Institute
of Science, Bangalore 560012, India (email: dtram@ee.iisc.ernet.in )

1-4244-0493-2/06/$20.00 2006 IEEE.

indicators as that of the popular L-index [6], [7]. This index

gives sufficiently accurate as well as practical means of the
assessment, and can express the stability analysis in simple and
operator friendly way.
Application of Artificial Neural Network (ANN) to the
above-mentioned problem has attained increasing importance
mainly due to the efficiency of present day computers.
Moreover real-time use of conventional methods in an energy
management center can be difficult due to their significant
large computational times. One of the main features, which can
be attributed to ANN, is its ability to learn nonlinear problem
offline with selective training, which can lead to sufficiently
accurate online response. ANN approach to voltage stability
assessment and improvement has been proposed and various
neural network combinations have been used for solving the
problem in [10]-[18]. The ability of ANN to understand and
properly classify such a problem of highly non-linear
relationship has been established in most of them and the
significant consideration is that once trained effectively ANN
can classify new data much faster than it would be possible
with analytical model. Research work presented in [12] has
very large error and while other work either have been tested
on very small system or they are not fast enough for online
application.
This paper presents an online voltage stability assessment
method using the feed forward neural network. In this method
feed forward neural network is trained for the L indices values,
which is a scalar measure of the voltage stability for all the
power system buses during normal and contingent situations.
Main advantage of the proposed method is that the voltage
stability indices for all the buses in the power system can be
calculated using the trained Artificial Neural Network at every
time instant unlike the other techniques. The easiness in
calculating the stability indices using Index L is exploited for
learning the voltage profile of any complex system by ANN.
Thus the stability margin and voltage profile locally for
individual buses as well as the global stability margin and
improvement measures of the power system can be assessed at
the same time with the proposed technique. Another feature of
the proposed method is its ability in developing L indices of all
the system buses during both normal and contingent situations
using the trained ANNs. This aspect has not been considered
as a single problem so far in the earlier research works.
The trained ANN is then tested on the practical 367 bus
system to prove its practical use Using MATLAB neural
network toolbox. The approach was found to be extremely
useful to use as energy management software for online
establishment of voltage stability margins and to find out the

2
associated limits at each bus.
II. THE PROPOSED ANN ARCHITECTURE AND FUNCTIONS
Fig. 1 shows the proposed feed forward neural network. The
architecture consists of an input layer, a hidden layer and an
output layer. Input vector is fed to the input layer for all the
buses of the selected system. The network weights and biases
are adjusted using the dotprod function in the toolbox and
adaptation is done using adaptwb function, which changes
both weights and biases. First to obtain the best combination of
number of neurons, training and transfer function a test system
is developed and the toolbox parameters are applied. The
following Toolbox functions are analyzed i)-Neural network
architecture and types, ii) Training functions, iii) Activation
functions, iv) Learning function, v) Initialization functions, and
vi) Performance functions

the system based on L index [5]. The algorithm developed for

finding the L index is used for producing the input data for
training the neural network. The operational load flow and
optimization of reactive power control variables using the LP
technique are important functionalities in this algorithm.
The mathematical consideration of the voltage stability
algorithm and the description of the operational load flow are
briefly discussed here. Consider a system where n = total
number of the buses, with 1, 2,KK , g generator buses,

g + 1, g + 2,KK , g + s SVC buses, g + s + 1,KK , n the

remaining buses ( r = n g s ) and t = number of OLTC
transformers.
First load flow solution is obtained incorporating the
generator control and load characteristics. Using the load flow
results the L-index is computed as:

Lj = 1

i=g

ji

i =1

Vi
Vj

(1)

where, j = g + 1,KK , n and all the terms within the sigma

on the RHS of equation are complex quantities. The values

F ji are obtained from the Y bus matrix given by:

I G Y GG Y GL V G
(2)
L = LG
LL L
I
Y
Y
V


G
L
G
L
where, I , I , V and V represent currents and voltages
at the generator and load nodes.
Rearranging the above-mentioned equations yields:

V L Z LL
G = GL
I K
where, F

Fig. 1. Proposed Feed Forward Neural Network (FFNN) architecture

Initially from the selected input data set, the minimum and
maximum values are found out. Then the FFNN architecture is
developed defining the number of neurons, training and
transfer functions. Further the input data is normalized and
filtered for redundancy. Then the network training parameters
such as learning ratio (0.08), learning increment (2.00),
number of epochs (400), Parameter Goal (1e-5), Minimum and
Maximum Gradient (1e-10 and 1e-10 respectively) have been
set based on the extensive analysis done. After training the
network for the set of samples, network performance is
evaluated with a new set of non-simulated data and the output
with the L-index values obtained from the conventional
algorithm is compared. The trained network nomenclature is
assigned and further a new set of NN is trained for
contingencies.
III. THE VOLTAGE STABILITY ASSESSMENT ALGORITHM
This paper uses the enhancement of the voltage stability of

LG

F LG

Y GG

IL
G
V

(3)

= Y LL Y LG are the required values.

The L-indices for a given load condition is computed for all

load buses, and thus the complex equation for the L index for

j th node can be written as:

Lj = 1

i=g

F
i =1

ji

Vi
ji + i j
Vj

(4)

For the stability, the index L j , must not be violated for any
of the nodes ' j ' . Hence the global indicator L describing the
stability of the complete subsystem is given by L = maximum
of L j for all j (load buses). Thus the indicator ' L ' is the
quantitative measure of the estimation of the distance of the
actual state of the system to the stability limit. The maximum
of the L-index gives the proximity of the system to voltage
collapse. An L-index away from 1 and close to 0 gives the
improved stability margin. The stability margin can also be
obtained as the distance of L from a unit value i.e. (1-L). Thus
the L-index gives a scalar number to each load bus and
performs consistent results. The stability margin and the
weakest bus can also be identified after obtaining the indices of

3
all the buses. Further the reactive power support measures and
the improved power flow can be performed in the system.
The algorithm developed is based on the operational load
flow, which will overcome the deficiencies of the conventional
load flow model. The model developed is suitable for
incorporating such various control strategies as Generator
prime mover response, Automatic Generation Control (AGC),
Generator tripping, Load shedding, and Load characteristics.
IV. THE ANN ALGORITHM AND FUNCTIONS
The proposed network architecture is a three layer feed
forward structure including input, output and hidden layer
using a back propagation algorithm. Following algorithmic
steps describes in detail the approach used for investigating the
different parameters and functions in the MATLAB toolbox.
Step 1: A conventional voltage stability algorithm is run with
the test system for simulated loading conditions. Using this
first the base case and the maximum loading conditions of the
test system are determined using the conventional software [5].
Then the load conditions are varied from base case till full load
and training samples are generated.
Step 2: Create a database for the input vector in the
following

form

PgT QgT VgT Pl T QlT Vl T

where,

Pg , Qg , Pl and Ql are the real and reactive power in

generator as well as load buses respectively and Vg and Vl
are bus voltage at generator and load buses. Further, create
target vector in the form of L-indices for the corresponding
input vectors.
Step 3: Find the minimum and maximum values of the input
vector, remove redundancies and normalize to suit to train the
selected feed forward neural network.
Step 4: Select the set of training parameters such as number
of epochs, learning increment and rate, performance goal with
Mean Squared Error (MSE) and minimum and maximum
gradient.
Step 5: Train the network based on a set of transfer
functions and number of neurons. The number of neurons in
each layer is varied initially and optimum combination is found
out depending on the training period and performance error.
Step 6: Find the most suitable combination of the activation
function. Behavioral accuracy depends on the uniformity in
values of L-indices at all the buses. It can happen that the
network gives output, which is accurate for some buses but
may be unacceptable on some others.
Step 7: Change the training function keeping same transfer
functions and optimum number of neurons in each layer.
Step 8: Find the most suitable network based on the
simplicity least possible Mean Square Error and computational
speed. Further use various test functions to confirm the
effectiveness of the proposed neural network. At this state the
functions and all the parameters are finalized for a particular
combination.

V. PERFORMANCE COMPARISON OF THE TRAINING FUNCTIONS

Keeping the suitable activation functions, each of the
training function is analyzed for two different conditions.
TABLE 1
PERFORMANCE COMPARISON FOR VARIOUS TRAINING FUNCTIONS
Type of
Description of
Training Perf.
CPU
Network
Training Pattern
Function Goal
Time
(MSE)
(sec)
10
Epochs
CONDITION 1
Traincgb 0.0068
5
(12,10,1)
Tansig,Tansig
Trainoss 0.0077
4
FeedForward
Purelin
Trainbfg 0.0065
8
Traincgf 0.0110
5
Traincgp 0.0072
6
Trainscg 0.0169
4
Trainrp 0.0145
3
Traingdx 0.0732
3
CONDITION 2
(24,20,1)Tansig,Tansig,Purelin Traincgb 0.0018
9
(32,30,1)Tansig,logsig,Purelin
Trainoss 0.0059
15
Feedforward
(18,18,1)Tansig,Tansig,Purelin Trainbfg 0.0046
8
(24,20,1)Tansig,logsig,Purelin
Traincgf 0.0037
10
(24,20,1)Tansig,Tansig,Purelin Traincgp 0.0038
12
(24,20,1)Tansig,Tansig,Purelin Trainscg 0.0032
9
(24,20,1)Tansig,Tansig,Purelin
Trainrp 0.0075
4
(30,25,1)Tansig,Tansig,Purelin Traingdx 0.035
4
(30,25,1)Tansig,Tansig,Purelin
Trainlm 0.0002
4
Legend :- (x,x,x) Tansig,Tansig,Purelin --> (Nh,Nh,No) Transfer Function

Initially the training was conducted for same network with

constant number of neurons in all the layers to identify the
performance of each function and also to calculate the CPU
time. It was observed that the computational time for
TRAINGDX and TRAINRP functions are same (3 seconds)
but the Mean Square Error was 0.0732 and 0.01454
respectively. Also function TRAINCGB found to be the
optimal as the error function is 0.00679 and the computation
time is 5 seconds for ten epochs.
The NN neurons are then varied in order to improve the
network error function performance. It can be seen that
TRAINLM function found to be the optimal as though the
number of neurons are more than TRAINCGB it gave better
performance goal of 0.000227 with a computational time of 4
seconds for ten epochs. Table I compiles the performance of
each of these training functions for same number of neurons
(cond.1) and best suitable neurons in each layer (cond.2). It
was also found that Quasi Newton Algorithm based on
Levenberg Marquardt (LM) training algorithm performs the
best considering the criterias as Mean Square Error for the Lindex at each bus, training time and overall accuracy.
Moreover, sigmoid transfer functions (tansig) for the input
and hidden layer and the linear transfer function (purelin) for
the output layer was the most suitable.
VI. ALGORITHM ASSESSMENT
Figure 2 gives the functional flowchart of the overall voltage
stability monitoring and assessment tool.

G1
119

115
T12

123
T13

54

52

55

1
T1

T11

111

G10

G2

G3

G4

50
2

T10

57

3
T2

139
T151

140

T4

T3
143

141

258

136

G6

G9

56
T14

11
T9

129

G5
7

T6

58

155

G7

G8

154

T5
149

157

147

T17
9
T7

10
T8

59

62

145

T16
61

63

267
Fig. 2. Functional Flowchart of the Voltage Stability Monitoring and Assessment
Tool

For the study systems the least square minimization VAR

optimization program is performed and the L indices are found
for each bus without changing the controllers (i.e. without
improving the voltage levels). The output obtained from the
VAR optimization program for different loading conditions are
taken as the data base for training the neural network. The
injecting real and the reactive power for both the generator and
the load buses along with the voltage for the load and the
generator buses respectively are considered as the input vector
for each bus. Output vector is the L-index for all the buses.
The network is trained for different samples for all the buses
as a cell array at a time for obtaining the target L-indices fed as
the output for training. Further a separate similar neural net is
used to train the contingencies of the power system network.
An error function is defined and the training is conducted to
obtain satisfactory results. Once tested satisfactorily, the
trained network is used to find the L-indices online and thus
the stability margin, weakest bus and proximity of voltage
collapse.

92

65

66

266
153
265

T18
166

Fig. 3. One Line Diagram of a 367 node practical system

VII. SIMULATION RESULTS
A practical power system network is used to test the ability
of the proposed neural network architecture. This system
consists of 367 nodes with the total generation and load
schedule as shown in the table 2. The system loads are
distributed into all the areas in a southern Indian state. The
total generation is divided into seven zones. The voltage
profile of the proposed system is as in figure 4. For training
purpose the original system is represented in the equivalent
form with a 45 bus system. Figure 3 shows the one line
diagram of the reduced system. Table 3 shows the equivalent
generation and load schedule.

5
TABLE 2
TOTAL GENERATION AND LOAD SCHEDULE FOR 367 NODE PRACTICAL SYSTEM

Description

MW

MVAR

1712.76
1621.06
--

646.31& -198.61
970.77
435

1712.23
1621.06
91.28
--

613.29
970.77
113.25
470.85

Per-Unit Bus Voltage

Sr.No

A. Scheduled Generation and Load

1
Generation Scheduled
2
Total Load
3
Total SVC
B. System Grid Total
1
Total Generation
2
Total Load
3
Total Power Loss
4
Shunt Capacitance

Voltage profile
(Equivalent Practical System)

MW

0.95
0.85

609.00&189.50
584.76
255
398.24
584.76
68.53
255.00

10

15

20

25

30

35

40

45

Fig. 5. Voltage profile of the reduced 367 node practical system

MVAR

B. Scheduled Generation and Load

1
Generation Scheduled
1594.99
2
Total Load
1568.48
3
Total SVC
-B. System Grid Total
1
Total Generation
1604.95
2
Total Load
1568.48
3
Total Power Loss
36.49
4
Shunt Capacitance
--

Bus Numbers

L-Indices Values
(Equivalent Practical System)
0.4

L-Index for each bus

Description

1.05

TABLE 3
TOTAL GENERATION AND LOAD SCHEDULE FOR REDUCED PRACTICAL
SYSTEM

Sr.No

1.15

0.3
0.2
0.1
0
0

10

15

20

25

30

35

40

45

Bus Numbers
V o lta g e P r o file fo r 3 6 7 N o d e p r a c tic a l S y s te m

Fig. 6. L-Index profile of the reduced practical system

1 .1 5

L-Index values
(Lightly Loaded Conditions)

1 .0 5

0 .9 5

0 .9

358

337

316

295

274

253

232

211

190

169

148

127

85

106

64

43

22

0 .8 5

B u s N um bers

Fig. 4. Voltage profile of a 367 node practical system

Case 1:
The developed equivalent 45-bus system from 367-bus
system represents ten generators, eighteen transformers and
forty-four transmission lines other than the loads. The loads
are combined into thirteen equivalent forms.
The details of the system grid total and the individual bus Lindex values used for training and testing are shown in the
table 3 and figures 6 respectively. The voltage profile shows a
variation in the voltage per-unit voltage magnitude which is as
high as 1.116 at bus number 28 and 0.864 at bus number 18.
Similarly the L-index values show the same effect as these
scalar values moves away from zero in the event of voltage
deviation. Thus these values are still very good indicators of
the bus voltage profile.

L-Index Values

Voltage in p.u

1 .1

0.6
0.5
0.4
0.3
0.2
0.1
0
-0.1

10

15

20

25

30

35

40

45

Bus Numbers

Fig. 7. L-Index profile of the reduced practical system for a lightly loaded
condition

The load is varied for a light loaded condition and the input
vector for training the L-index is obtained for ten percent to
hundred and forty five percent of the base loading. Once the
proposed ANN architecture is well trained, (confirming it with
the post regression analysis) the intelligent stability indicator is
compared with the L-index value from the conventional
software. The results obtained for the non-simulated conditions
are as in the figure 7 and 8. Please note that the system loading
has been performed in a uniform manner getting the data set
for each five percent increase on all the bus. Then the load
changes are done in a random manner at each buses and the
data set for training has been extracted which has been referred
as abnormal way of loading. The L-index value and the error
for the load variation for an abnormal case is as in figure 9 and
10. Figures 8 and 10 shows the trained ANN structure is
capable of generating appropriate L index values during the

6
normal loading conditions as well as abnormal loading within
the maximum error of 0.05. This once again shows that once
trained effectively the proposed ANN structure can be used in
the practical system online irrespective of the system size.

very high. Around bus number 158 the voltages are low again.
Thus these are the areas, which are violating the voltage limits.
For the ANN training it is observed that for the reduced system
the L-index value is learned considerably accurate with a single
ANN.

Absolute Error in L-Index values

(Lightly Loading Conditions)

Voltage profile
(Equivalent Practical System with
contingencies)

0.05
0
-0.05
0

10

15

20

25

30

35

40

45

Per-Unit Bus Voltage

Absolute Error

0.1
1.15
1.05
0.95

-0.1

0.85
0

Bus Numbers

10

15

20

25

30

35

40

45

Bus Numbers

Fig. 8. Absolute Error L-Index profile of the reduced practical system for a
lightly loaded condition (Trained ANN and Conventional software)

Fig. 11. L-Index profile of reduced system for a contingency condition

L-Index values
(Abnormal Conditions)

L-Index values
(Contingency Conditions)
0.2

L-Index Values

L-Index Values

0.25
0.2
0.15
0.1
0.05
0

0.15
0.1
0.05
0
-0.05

-0.05

10

15

20

25

30

35

40

10

45

15

20

25

30

35

40

45

Bus Numbers

Bus Numbers

Fig. 12. L-Index profile of reduced system for a contingency condition

Fig. 9. L-Index profile of the reduced practical system for an abnormal load
condition

Absolute Error in L-Index values

(Contingency Conditions)
0.04

Absolute Error

Absolute Error

Absolute Error in L-Index values

(Abnormal Conditions)
0.03
0.02
0.01
0
-0.01
-0.02
-0.03
-0.04

0.02
0
-0.02
-0.04
-0.06 0

10

15

20

25

30

35

40

45

-0.08
Bus Numbers

10

15

20

25

30

35

40

45

Bus Numbers

Fig. 10. Absolute Error L-Index profile of the reduced practical system for an
abnormal loading condition (Trained ANN and Con. software)

Case 2:
In this case line 17-20 (equivalent to 59-63 in figure 3) is
deleted and simulated as line outage. Initially this line was
carrying 222.57 MW and 77 MVAR under base loading
condition. This was supplying to the load of 270 MW in bus
number 40 (157). The voltage profile is as in figure 11. The
voltage profile shows that the voltages at bus numbers from 58
to 66 is below one per unit and especially on bus number 61,
63 and 66 the voltages are very low. Like wise the voltage
magnitude from the bus numbers 115 to 136 the voltages are

Fig. 13. Absolute Error L-Index profile of the reduced practical system for a
contingency condition (Trained ANN and Conventional software)

This confirms that the ANN can be used for voltage stability
assessment quite efficiently when training is done properly.
When the load is light the trained value found to be having
more error. For moderate loading and for heavily loaded
condition the absolute error is less. The comparison of the Lindex values for a non-simulated condition as in figures 12 and
13 shows the effectiveness of the proposed ANN structure. As
the trained ANN does not need any further computation, the
proposed method is suitable especially on an energy
management center to perform online analysis. Further the
computation time for the offline training was around 120
seconds.

7
VIII. CONCLUSION
An artificial neural network technique for on line assessment
of power system voltage stability using a developed training
algorithm for all system buses has been presented with detail
steps involved with MATLAB neural network toolbox. Unlike
other reported techniques, the main advantage of the proposed
method is that the voltage stability indices for all the buses in
the power system can be calculated using the trained artificial
neural network at every monitoring period. The stability
margin and voltage profile for individual buses, the global
stability margin, as well as possible improvement measures of
the power system can be assessed at the same time during both
normal and contingent situations using the trained ANN.
Training and testing results form all cases, including
contingencies on a practical power systems network shows that
the proposed ANN algorithm is capable to learn and perform
as a tool for online voltage stability analysis by measuring the
L-indices for all the vulnerable buses.
REFERENCES
[1]

[2]

[3]

[4]

[5]

[6]
[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

N. Yorino, H. Sasaki, Y. Masuda, et al., An investigation on voltage

stability problems, IEEE Trans. on Power Systems, Vol7, No2, May 1992,
pp. 600-607.
Y. Tamura et.al., Relationship between voltage instability and multiple
load flow solutions in electric power systems IEEE Trans. on Power
Systems, Vol. PAS 102, No.5, May 1993, pp. 1115-1125.
H. Sun, X. Zhou and R. Li, Accuracy analysis of static stability indices
based on power flow model, IEEE/PES Transmission and Distribution
conference and Exhibition: Asia and pacific, 15-18 Aug. 2005, pp. 1-7.
P-A Lof, G. Anderson and D. J. Hill, Voltage Stability indices of the
stressed power system IEEE Trans on Power Systems, Vol. 8, No.1, Feb
1993, pp. 326-335.
Bansilal, D. Thukaram and K. Parthasarathy Optimal reactive power
dispatch algorithm for Voltage stability improvement Electrical Power
and Energy Systems, Vol. 18, No.7, pp. 461- 468, 1996.
D. Thukaram et. al., Voltage stability Improvement: Case Studies of Indian
Power networks Electric Power Systems Research, 44 (1998), pp. 35-44.
T. Esaka, Y. Kataoka, T. Ohtaka, S Iwamoto, Voltage stability preventive
control using a new voltage stability index, International conference on
Power System Technology, POWERCON 2004, Vol. 1, 21-24 Nov., pp.
344-349.
V. Ajjarapu and C. Christy, The continuation power flow: a tool for steady
state voltage stability analysis, IEEE Trans. On Power System, Vol. 7,
Issue 1, Feb. 1992, pp. 416-423.
C.A. Canizares, A. C. Z. DeSouza and V.H. Quintana, Comparison of
performance indices for detection of proximity to voltage collapse, IEEE
Trans on Power system, Vol.11, No.3, August 1996, pp. 1441-1450.
A. C. Z De Souza, C. A. Canizers, V. H. Quintana, New Techniques to
speed up voltage collapse computations using tangent vectors, IEEE Trans.
On power System, Vol. 12, Issue 3, Aug. 1997, pp. 1380-1387.
B. Jeyasurya, Artificial Neural Networks for on-line voltage stability
assessment, IEEE PES Summer Meeting, 2000, Vol. 4, July 2000, pp.
2014-2018
Bansilal, D. Thukaram and K. H. Kashyap, Artificial neural network
application to power system voltage stability improvement, Conference on
convergent technologies for Asia-Pacific region, TENCON 2003, Vol. 1,
15-17 Oct. 2003, pp. 53-57.
M. Pandit, L. Srivastava, J. Sharma Fast voltage contingency selection
using fuzzy parallel Self-Organizing Hierarchical Neural Network, IEEE
Trans. On Power system, vol. 18, issue 2, May 2003, pp. 657-664
M. La Scala, M. Trovato and F. Torelli A neural Network based method
for voltage security monitoring, IEEE Trans on Power systems, Vol. 11,
No.3, August 1996, pp. 1332-1341.
A. A. El-Keib and X. Ma, Application of artificial neural networks in
voltage stability assessment, IEEE Trans on Power Systems, Vol.10, No.4,
November 1995, pp. 1890-1896.

[16] D Chen and R. R. Mohler, Neural-network-based load modeling and its

use in voltage stability analysis, IEEE Trans. On Control System
Technology, Vol. 11, Issue 4, July 2003, pp. 460-470.
[17] V. R. Dinavahi and S. C. Srivastava, ANN based voltage stability margin
prediction, IEEE PES Summer Meeting, 2001, Vol. 2, 15-19 July 2001,
pp. 1275-1280.
[18] S. Charabarti and B. Jeyasurya, On-Line voltage stability monitoring using
artificial neural network, Large engineering system conference on power
engineering, 2004, LESCOPE 2004, 28-30 July 2004, pp. 71-75.
S. Kamalasadan (SM01-M04) received his B Tech. degree in Electrical and
Electronics in 1991 from the University of Calicut, India, M. Eng in Electrical
Power Systems Management in 1999, from the Asian Institute of Technology,
Bangkok Thailand and his PhD in Electrical Engineering from the University of
Toledo, Ohio in 2004. He is currently serving as an Assistant Professor at the
University of West Florida. He has held industrial positions for six years and has
teaching and research experience for four years. He is the recipient of
Outstanding Teaching Award from the Graduate Student Association University
of Toledo, 2001- 2002 and from the College of Engineering, 2002-2003. He has
also received Summer research award during 2005 from the University of West
Florida. His research interests include Intelligent and Autonomous Control,
Power Systems dynamics, Stability and Control, AI applications in control of
Dynamic Systems and Real-time Embedded Systems Applications. He is a
technical reviewer of IEEE Transactions of Power Systems, IEEE Transactions of
Fuzzy Systems, International Journal of System Science and several other
conferences.
A. K. Srivastava (SM00-M05) received his Ph.D. degree from Department of
Electrical and Computer Engineering, Illinois Institute of Technology, Chicago,
USA in 2005, B. Tech. from Harcourt Butler Technological Institute, Kanpur,
India in 1997 and M. Tech. from Institute of Technology, Varanasi, India in 1999.
He is working as Assistant research professor at Mississippi State University
since September, 2005. He worked as research assistant and teaching assistant at
Illinois Institute of Technology, Chicago, USA and as Senior Research Associate
at Electrical Engineering Department at the Indian Institute of Technology,
Kanpur, India as well as Research Fellow at Asian Institute of Technology,
Bangkok, Thailand. His research interest includes power system security, voltage
stability analysis, power system modeling and artificial intelligent application in
power system. Dr. Srivastava is member of IEEE, Power Engineering Society,
Sigma Xi and Eta Kappa Nu. He is recipient of several awards and serves as
reviewer for IEEE transaction on power system, international journals and
conferences.
D. Thukaram (SM90) received the B.E. degree in Electrical Engineering from
Osmania University, Hyderabad in 1974, M.Tech degree in Integrated Power
Systems from Nagpur University in 1976 and Ph.D. degree from Indian Institute
of Science, Bangalore in 1986. Since 1976 he has been with Indian Institute of
Science as a research fellow and faculty in various positions and currently he is
Professor. His research interests include computer aided power system Analysis,
reactive power optimization, voltage stability, distribution automation and AI
applications in power systems.