Design of FNN AVR for Enhancement of Power System

Stability Using Matlab/Simulink

ASLAM PERVEZ MEMON*, MOHAMMAD ASLAM UQAILI **, AND ZUBAIR AHMED MEMON ***
RECIEVE ON 27.02.2012 ACCEPTED ON 21.06.2012

ABSTRACT
A simple technique of excitation voltage control with NNAVR (Neural Network Automatic Voltage
Regulator) is proposed in this paper. Popular type of ANN (Artificial Neural Networks) known as RBF
(Radial Basis Function) architectures with OLS (Orthogonal Least Square) algorithm is suggested to
design AVR in order to prove its applicability and suitability. This proposed technique is implemented
considering as SMIB (Single Machine Connected to Infinite Bus) system with linearized model of
synchronous machine and its excitation system using Matlab/Simulink. The simulation results of RBF
AVR, when compared with conventional AVR controllers show better performance, improve the transient
and small signal stability of the system and above all its response is more suitable in case of load
changing conditions.
Key Words:

Excitation System, Automatic Voltage Regulator, Synchronous Machine, Stability,

Radial Basis Function, Matlab/Simulink.

1.

INTRODUCTION
t is pertinent to maintain the voltage and frequency
of the system within permissible boundaries in order
to keep the reliability of the system.

Excitation control system is defined as the field current to

maintain excitation of synchronous machine and include
exciter and AVR. When the changes in rotor angle () or
frequency (f) occur, real power (P) is affected which is
controlled by turbine-governor system. Voltage magnitude
disturbs reactive power (Q) which is controlled by exciterAVR system. Therefore LFC (Load Frequency Control)
system controls active power and reactive power is
controlled by AVR and both (LFC & AVR) are installed
separately for every generator.
*
**
**

The better performance of AVR can produce quick

response, simple maintenance and high current facilities
to the system thus the Q and generated voltage can be
controlled efficiently. Excitation system contributes to the
effective voltage control and enhances the system
stability. Its quick response can improve transient and
small signal stability of the system. [1-4,7].
Conventional controllers are mostly fixed gain controllers,
they are settled at particular timings with particular load
conditions, they become failure when load conditions
change from light to heavy loading.
Secondly due to fixed gain parameters, linear modeling
techniques are used. Most popularly PID (Proportional

Assistant Professor, Department of Electrical Engineering, Quaid-e-Awam University of Engineering, Science & Technology,
Nawabshah.
Meritorious Professor, Department of Electical Engineering, Mehran University of Engineering & Technology, Jamshoro.
Associate Professor, Department of Electrical Engineering, Mehran University of Engineering & Technology, Jamshoro.
Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
529

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink

Integral and Derivative) controllers implemented which

are linear and time invariant. They need re-tuned (which
very difficult process) when time varying conditions
occurs. Most importantly, if the PID controller is unable to
deal with the complex process, no matter how you tune it,
the system will not work.
Power system is non linear, complex and subjected to
different kind of disturbances. [6-10].
In order to have high gains, fast acting, automatic settlings,
adaptive to load changing, tackling the nonlinearity and
complexity, the implementation of neural network based
ARVR is proposed in order to show the suitability and
enhancement in the stability of the system by controlling
the terminal voltage with an efficient manner.

2.

ARTIFICIAL NEURAL NETWORKS

ANNs and its applications in power system problems is

not a new topic of research, because it has been suggested
in many research areas with fast growing interest.
Literatures indicate that ANN is swiftly drawing the
attentions and recognition amongst the power system
researchers. They are enormously useful in the area of
electrical engineering within few years [11-12].
They have the capability of modeling the complex
relationships which makes NN better to traditional
controllers. Traditional controllers always require a
comprehensive knowledge and skills about mathematical
model of the controlled system. But neural networks
controllers do not require such knowledge and skills and
they can handle such complex systems very simply and
efficiently. Training process teach them to map inputoutput relationships of the system.
ANNs are universal function approximators, having the
capability of approximating any continuous nonlinear
functions to arbitrary accuracy [13]. They have also
robustness, parallel architecture and fault tolerant
capability [14], furthermore ANNs have great flexibility

because they have been built on the actual mathematical

formulations consisting a great versatility and logical
mathematical techniques [15-16]. ANN architectures have
been divided into feedback NNs, cellular NNs &
feedforward NNs the most popular and known as multilayer
NNs and further sub divided into two categories as:
multilayer perceptron NNs and radial basis function NNs
[17-18].
The RBF network possesses a universal approximation
and best approximation properties which have no
botheration of selecting required number of hidden layer
neurons. The learning properties of radial basis networks
are quicker than that of multilayer perceptron networks;
however, RBF usually take more number of hidden layer
neurons as compared to MLP [19-21].

3.

PROPOSED SCHEME
METHODOLOGY

AND

ANNs are competent of learning from off-line simulation

data and can then be trained to reproduce the behaviour
of the system under various loading conditions.
This work studies the off-line simulations using a
conventional controller like PID to control the system for
normal and heavy loading conditions to obtain the input/
output data of the synchronous machine. This data will
be utilized for training FFNN (RBF).

3.1

Model For FFNN Applications

The proposed model for the application is considered as

SMIB as shown in Fig. 1. The simulation model will be
developed from the block diagram (Fig. 2) of governor
(LFC) and AVR excitation system of synchronous
generator with their linearized equations and transfer
functions [3-5].
The state space equations for the complete simulation
linearized model of synchronous generator with LFC
(governor) and AVR excitation controller system (with
proper assumptions [3-4] and Appendix-B) are given below:

Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
530

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink

And from definition of [PM Anderson].

The linearized equations for the synchronous machine

are given by [3].

& =

E q =

K 3E FD
1 + K 3d0s

K 3K 4

The complete state-space description of the system is

given by:

1 + K 3d0s

Te = K1 + K 2 E q

1
K
3 d0
K2
E&q j
& 0
& K 6 K R
V& =
V&1 R
&3 0
VR

E&FD
0

Vt = K 5 + K 6 Eq

And

1
K

E q 4 + 1 E
E&q =
K
FD
3 d0
d0 d0
From the torque equation we have:

K K
D
T
& = m 1 2 Eq

j j j

K4

d0
K1

0
1

J
1
0

0
K 5K R

R
0

KA

KF

F E

KA

A
1
E

0
0
E Tm
q

0

j
0
0
V1 + 0
K F (SE + K E ) V3
0

VR K A

F E
E VREF
FD A

0
0

(SE + K E )

d0

Here the system is described with excitation system. The

state variables are:

t
x = E q

V1 V3

VR

E FD

The driving functions are VREF and Tm assuming that is

zero. All the parameters of the system are described in per
unit system and their numerical values are mentioned in
[3-5,22-23].

FIG. 1. SMIB SYSTEM

Excitation
System

Con. Field

Automatic Voltage
Regulator

Voltage
Sector

Steam
G

Tarbine
Pv

pa, Qc

pic

Valve Control
Machines
Pi

Load Frequency
Control

Frequency
Sector

FIG. 2. BLOCK DIAGRAM WHICH IS USED FOR LINEAR MODEL WITH GOVERNOR AND AVR SYSTEM
Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
531

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink

3.2

Methodology of RBF Network

structures are generated. In first hidden layer 19 neurons

with radial basis transfer function as activation function

The potentiality and applicability of radial basis networks

are chosen. One neuron with linear transfer function

for this application is being investigated, because they

having 1.7 spread constant and 0.000001 error goal are

have distinctive properties of simple network structure,

trained for this research work.

efficient learning way, and best approximation, which make

radial basis networks more powerful tool than the other

4.

SIMULATIONS RESULTS

types of neural networks.

For the simulation results, MATLAB 7.13, Simulink Version
The radial basis networks consist of three utterly different

7.8 and Neural Network Toolbox 7.0.2 (R2011b) have been

layers. The input layer or first layer consists of a number

utilized.

of units fastened to the input vector. The units constituted

by hidden layer or second layer have an overall response

4.1 Terminal Voltage response

function, mostly a Gaussian function. The function of each

Terminal voltage (Vt) and frequency/speed deviation ()

responses are shown in Figs. 3-6 respectively.

class is computed by third layer.

Diversity of different algorithms has been proposed in the
latest literature for choosing the proper radial basis network
centres. We prefer the universal approximator OLS
algorithm for this research work [20].
OLS learning algorithm generates radial basis network,
which have a hidden layer, smaller than that of radial basis
network with arbitrarily chosen centres.
Radial basis networks are used to fairly accurate function.
They include neurons to the second or hidden layer
awaiting it meets the precise MSEGF (Mean Square Error
Goal [20-21].

Time settling for terminal voltage and frequency deviations

are fixed at 0.4 and 20 seconds till the responses passes
their oscillations and become stable. Simulations results
indicate the responses of RBF AVR and PID LFC controllers
in order to see the enhancement in stability of power
system.
Fig. 3(a-b) shows the Vt response (a) without controller
and (b) with conventional PID AVR controller. Fig. 3 (c-d)
shows Vt responses (c) with RBF AVR (d) with RBF AVR
and PID LFC controllers. Fig. 3(d), clearly indicates that
there is no change in Vt response having PID LFC
controller. Because in Fig. 3(c), RBF AVR is showing and
excellent improvement which is exactly same in Fig 3(d). It

For this application two-layer network has been created.

means there is no impact of LFC controller on AVR

The input or first layer takes radial basis transfer function

excitation system of synchronous machine. These all

neurons which calculates its weighted inputs and its net

input with net product. The second layer takes linear
transfer function neurons and calculates its weighted
inputs with dot product and its net inputs with net sum.
First and second both layers have biases.
At first no neuron is available in radial basis transfer
function and FF network architectures with two-layer

responses have been combined in Fig. 3(e) and Fig. 4 in

large.

4.2

Frequency/Speed Deviation Responses

Frequency deviation responses without LFC controller

(a), with PID AVR (b), PID-LFC and RBF-AVR (c) and RBFLFC and RBF-AVR (d) are shown in Fig. 5(a-d) and
combined responses are illustrated in Fig. 5(e) respectively.

Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
532

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink

The in large view of Fig. 5(e) is shown in Fig. 6. These all

responses show good improvement in stability but again
prove that there is no coupling impact between AVR and
LFC controllers. Hence both RBF AVR and RBF LFC
controllers have their excellent own responses separately
which is verified from their results because transient, small
signal and dynamic stability improves efficiently.

The suggested method of designing RBF AVR is easy to

implement with the help Matlab/simulink software. In order
to validate the simulated results, the propose AVR is
compared with conventional AVR.
These results show that the RBF AVR controller has
promising satisfactory generalization applicability and
suitability as well as accuracy.

using Matlab/Simulink and neural network toolbox has

RBF AVR is more suitable where various types of

disturbances at different times occur as in case of power
system. It also ensures superior responses at different
load conditions. Due to fast acting and settling, proposed
AVR is more suitable for transient and small signal stability
of the system.

been successfully demonstrated. The proposed technique

ACKNOWLEDGEMENTS

5.

CONCLUSIONS

RBF AVR controller with OLS algorithm, in SMIB with

linearized model of SM and excitation system as model

controls terminal voltage and reactive power thereby

improving transient stability of the system and eliminates
the drawbacks of conventional AVR control system.

FIG. 3(a-e). TERMINAL VOLTAGES RESPONSES

Authors acknowledge with thanks the higher authorities

and Department of Electrical Engineering, Quaid-e-Awam
University of Engineering, Science & Technology,
Nawabshah, Sindh, Pakistan ,and Mehran University of
Engineering & Technology, Jamshoro, Sindh, Pakistan,
for continuous encouragement and time to time facilities
when required during this research work, which is an
attempt for Ph.D work.

FIG. 4. COMBINED ALL TERMINAL VOLTAGES (IN LAGER)

Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
533

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink
0.05
0
-0.05

0.05

6
8
10 12 14 16
(A) WITHOUT CONTROLLERS

18

20

Frequency/Speed Deviation

-0.05

8
10 12 14
(B) WITH PID-AVR

16

18

20

6
8
10 12 14
(C) PID-LFC & RBF-AVR

16

18

20

6
8
10 12 14
(D) RBF-LFC & RBF-AVR

16

18

20

0.05
0

-0.05
0.05
0

-0.05
0.05
0
-0.05

8
10 12 14 16 18 20
Time (Seconds)
(E) COMBINED RESPONSES WITH CONTROLLERS

FIG. 5(a-e). FREQUENCY/SPEED RESPONSES

FIG. 6. COMBINED ALL FREQUENCY RESPONSES (IN

LARGE)

APPENDIX-A
List of Notations
A,B,C
System polynomials
D
Damping coefficient
EB or V Infinite bus bar voltage
Ed
Direct axis voltage
Efd or EFD Field voltage
E q
Quadrature axis voltage
Et or Vt Generator terminal voltage
H
Inertia constants
IF
Field current
Constants of the linearized model
K1-K6
KA
Regulator amplifier gain
Damping factor
KD
KE
Exciter gain
Stabilizing transformer gain
KF
Generator gain
KG
Kg
Governor gain
KP, K1, KD Proportional, integral and derivative (PID) controller
gains
KR
Sensor gain
KT
Turbine gain
Le or Xe Transmission line inductance or reactance
P or P e Generator real or electrical power
Pg
Governor power
PL
Load real power
Pm
Mechanical power input
Q
Reference real power
QL
Generator reactive power
rF
Load reactive power
QL
Field resistance
R
Speed regulation of governor
Re
Transmission line resistance
(s)
Shows Laplace transformation
Te
Electrical torque
Tm
Mechanical torque
Vd,Vq
Direct and quadrature axis voltages
Ve
Error voltage
vF
Field voltage
VF
Exciter voltage
VL
Load voltage
VR
Regulator amplifier voltage
Vrefs
Reference voltage
VS
Sensor voltage
Xd,Xq
Direct and quadrature axis reactances
xd,zq
Direct and quadrature transient axis reactances

Generator rotor angle

Regulator amplifier time constant

E
Exciter time constant
F
Stabilizing transformer time constant
G
Generator time constant
g
Governor time constant
R
Sensor time constant
T
Turbine time constant

Angular speed
R
Rated angular speed
ref
Reference speed

Shows change
d
Direct axis flux linkages
q
Quadrature axis flux linkages [2-4]

Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
534

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink
APPENDIX-B

REFERENCES

The numerical values of various components as well as other

constants required developing simulation model.

[1]

Schleif, F.R., Hunkins, H.D., Martin, G.E., and Hattan,

E.E., "Excitation Control to Improve Power Line
Stability", IEEE Transactions on Power Systems,

For turbine and governing systems

Volume 87, pp. 1426-1434, 1968.

Kg=1.0, g=0.2, KT=1.0, T0.5m R=0.05

[2]

deMello, F.P., and Concordia, C., "Concepts of

Synchronous Machine Stability as Affected by Excitation

SMIB (Synchronous Generator and Constants of Linear Model)

Control," IEEE Transactions on Power Systems,

D=0.1, H=5, K 1=1.5, K T=0.2, K 4=1.4, K 5=-0.1, K 6=0.5, K G=0.8,
G=1.5
Values for excitation model
KE=1.,0, E=0.4, KA=1.10, A=0.1, KR=1.0, R=0.05
K1 is the change in electrical torque for a small change in rotor
angle at constant d axis flux linkage; i.e.the synchronizing torque
coefficient

K1 =

Te

[3]

Anderson, P.M., and Fouad, A.A., "Power System Control

and Stability", Iowa State University Press, Iowa, USA,
1977.

[4]

Kundur, P., Power System Stability and Control, 4th

Edition, McGraw-Hill Inc., 1994

Eq

[5]

Saddat, H., Power System Analysis, 1st Edition,

McGraw-Hill Inc., 1999.

K2 is the change in electrical torque for small change in the d axis

flux linkage at constant rotor angle

K2 =

Volume 88, pp. 316-329, 1969.

[6]

Hsu, Y.Y.S., Shyue, W., and Su, C.C., "Low Frequency

Oscillations in Longitudinal Power Systems, Experience

= 0
Eq

Te

with Dynamic Stability of Taiwan Power System", IEEE/

PES, Winter Meeting, New York, 1986.

d0 is the direct axis open circuit time constant of the machine.

[7]

Moussa, H.A.M., and Yu, Y.N., "Optimal Power System

Stabilization Through Excitation and/or Governor

K3 is an impedance factor and

Control", IEEE Transactions on Power Systems,

( )]

Volume 91, pp. 1166-1174, 1972.

K3 final value of unit step VF response = lim E t = 0

t
K4 is the demagnetizing effect of a change in the rotor angle (at
steady state)
K4 =

1
K3

( )]

Vt
Eq

IEEE Transactions PAS, Volume 102, pp. 1738-1746,

()

1983.
[9]

Yu, Y.N., and Siggers, C., "Stabilization and Optimal

Control Signals for a Power System", IEEE TP-531
Summer Power Meeting and EHV Conference, July, 1970.

= Eq0

[10]

= 0
Eq

ANSI/IEEE Std 122 (1985), IEEE Recommended

Practice for Functional and Performance Characteristics
of Control Systems for Steam Turbine-Generator Units,

K6 is the change in terminal voltage Vt for a small change in the d

axis flux linkages at constant rotor angle, or [2-4]

K6 =

Chan, W.C., and Hsu, Y.Y., "An Optimal Variable Structure

Power System Stabilizer for Power System Stabilization",

lim E t v F = 0, = u t
t

K5 is the change in the terminal voltage Vt for a small change in

rotor angle at constant d axis flux linkage, or

K5 =

[8]

The Institute of Electrical and Electronics Engineers,

USA, 1985.

Vt

[11]

Neibur, D., "Artificial Neural Networks for Power

Systems", Report by TF 38.06.06, Electra No. 159,
April, 1995.

Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
535

Design of FNN AVR for Enhancement of Power System Stability Using Matlab/Simulink
[12]

Howard, S., "Neural Networks in Electrical Engineering",

[19]

Memon, A.P., "Artificial Neural Network Applications

in Electrical Alternator Excitation Systems", M.Phil,
Thesis, Mehran University of Engineering &
Technology, Jamshoro, Sindh, Pakistan, 2002.

[20]

Powell, M.J.D., "Radial Basis Functions for Multivariable

Interpolation", A Review, Proceedings IMA Conference
on Algorithms for the Approximation of Functions and
Data, RMCS, pp. 143-167, Shrivenham, UK, 1985.

[21]

Chen, S., Cowan, C.F.N., and Grant, P.M., "Orthogonal

Least Squares Algorithm for Radial Basis Function
Networks", IEEE Transactions on Neural Networks,
Volume 2, No. 2, pp. 302-309, 1991.

[22]

Gurrala, G., and Sen, I., "A Modified Heffron-Phillip's

Model for the Design of Power System Stabilizers",
Power System Technology and IEEE Power India
Conference, 2008.

[23]

Prathap, D., and Krishna, H., et al., "Design of Power

System Stabilizer To Improve Small Signal Stability by
Using Modified Heffron-Phillip's Model", International
Journal of Engineering Science, & Technology,
Volume 3, No. 6, pp. 4888-4896, June, 2011.

[24]

MATLAB Neural Network Toolbox Version 7.0.2

(R2011b), User's Guide, The Mathworks Inc., 2011
http://www.mathworks.com, (Visited on December, 2011.

Proceedings of the ASEE New England Section, Annual

Conference 2006.
[13]

[14]

[15]

Strefezza, M., and Dote, Y., "Radial Basis Neural Network

Adaptive Controller for Servomotor", IEEE,
Transactions on Power Systems, Volume 2,
pp. 413-417, 1993.
Warwick, K., Irwin, G.W., and Hunt, K.J., "Neural
Networks for Control and Systems", 1st Edition, Peter
Peregrinus Ltd., UK, February, 1992.
Suykens, J.A.K., "Artificial Neural Networks for
Modelling and Control of Non-Linear Systems", 1st
Edition, Kluwer Academic Publishers, 1996.

[16]

Fausett, L., "Fundamentals of Neural Networks,

Architectures, Algorithms and Applications", 2nd
Edition, Prentice Hall, 1994.

[17]

Tanaka, K., "An Approach to Stability Criteria of Neural

Network Control Systems", IEEE Transactions on Neural
Networks, Volume 7, No. 3, pp. 629-642, May, 1996.

[18]

Unar, M.A., and Murry, D.J., "Automatic Steering of

Ships Using Neural Networks", International Jouranl
Adaptive Control Signal Process, Volume 13,
pp. 203-218, 1999.

Mehran University Research Journal of Engineering & Technology, Volume 31, No. 3, July, 2012 [ISSN 0254-7821]
536