0% found this document useful (0 votes)
97 views6 pages

Project 2 PDF

This paper presents the development and an application Artificial Neural Network (ANN) based method for predicting of voltage stability index in power system network. The main feature of this paper is to introduce the approach in developing Neural network by using programming and tts accuracy. The advantages of the developed Neural network are its fast convergence of error, fast training and good estimation for testing and small MSE error (0,0029p.u)

Uploaded by

SiddharthDsouza
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
97 views6 pages

Project 2 PDF

This paper presents the development and an application Artificial Neural Network (ANN) based method for predicting of voltage stability index in power system network. The main feature of this paper is to introduce the approach in developing Neural network by using programming and tts accuracy. The advantages of the developed Neural network are its fast convergence of error, fast training and good estimation for testing and small MSE error (0,0029p.u)

Uploaded by

SiddharthDsouza
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

National Power and Energy Conference (PECon) 2003 Proceedings, B angi , Malaysia

37

Development of Artificial Neural Network for


Voltage ,Stability Monitoring
S, Sahari,

A. F, Abidin, and T K, Abdul Rahman

Abstract-- This paper presents the development and an


application Artificial Neural Network (ANN) based method for
predicting of voltage stability index

in power system network.

The main feature of this paper is to introduce the approach in


developing

Neural

Network

backpropagation

by

using

programming and tts accuracy In predicting voltage stability

index in the power system network. The training developed

programming langua ge, The advantages of the developed

neural network are its fast convergence of error, fast training


and good estimation for testing and small MSE error

(0,0029p.u). In ANN application, selection of input variable


is an important aspect, heuristic knowledge is usually
required and no systematic method is available (6).

Pd and

In this study, input variable are determined in a


systematic way, In which, the selection input variable are

individual load bus as output information.

L information covering stability at each


The generalization
capability of the developed neu ral networks under large

This study used the voltage stability index L developed by

number of random operation of loading conditions has been

T.K Abdul Rahman etel [5) in order to measure the stability


of the system, The train ing data was obtained by running the

neural network was accomplished by using Real Power


Reactive Power

Qd

at each load bus as input information,and

voltage stability index

tested.

Fast

prediction

performance,

accurate

evaluation

and

good

for voltage stability index have been obtained.

Results of tests conducted on

IEEE 6-

bus

test system are

presented and discussed.

Keywords--voltage

to the stability conditions in load buses,

load flow program at no-load and on-load condition. The


bus

voltages for this condition will

be substituted in

formulation of the voltage stability index, L [5]. 109 set

patterns of data have been successfully generated with

stability, voltage

stability

Network.

i ndex, Neural

numerous loading conditions. 60 patterns set of data were

randomly selected for the training and 49 patterns for the

testing process.

1.

INTRODUCTION

II.

Voltage stability is the ability of a power system to

maintain steadily acceptable bus voltage at each node under

normal operating conditions, after load increase, following

system increase configuration changes or when the system is


subjected to disturbanc e , The progressive and

being

uncontrollable drop in voltage results eventually in a wide

spread voltage collapse. The phenomenon voltage instability


is attributed to the power system operation at its maximum
transmissible power limit, shortage of reac tive power
resources and inadequacy of reactive power compensation
tools. The main factors contributing to the voltage collapse
are the generators reactive power l imit, voltage control
limits, load characteristics, reactive compensation device

characteristics and its actions. Evaluation teclmiques based

on load flow full or reduced Jacobian analysis such as,


singular

based on its e ffe c t

value

decomposition,

.Eigenvalue

calculations,

sensitivity factor, and modal analysis are time consuming


for a large power system

[1 l-[4].

In power system analysis, voltage stability problems are


not new to the electric utility industry and commercial area.

Since voltage instable has been responsible for several


network collapses, many techniques have been

major

d eve loped to identify critical power system and lines,


Voltage stability index [5) is a new method to determine the

static voltage stability of load buses in a power system for a

certain operating condition and hence identify load buses


that close to voltage collapse,
The present work purposes

neural network based

method for voltage stability monitoring in a power system


network, As an attractive method for voltage stability

prediction the neural network was developed in B orland C

S. Sahri, A. E Abidin and T. K. Abdul Rahman are with FaGulty of


Electrical Engineering. MARA University of Technology
0-7803-8208-0/03/$17.00 2003 lEEK

VOLTAGE STABILITY INDEX

A voltage stability index developed by T.K Abdul


Rahman [5) was utilized in the voltage stability analysis to
indicate stability index at each load buses system. This

formula index was derived from voltage equations used in

the Improved Distflow technique. The voltage stability

index is terms as symbol L and given by Equations I;

Stability illdex, L=

(I)

Where L is the stability index, Vo is the no load voltage and


VL is the load voltage, In order to maintain a stability of
voltage condition in the system network, the value of L at
any load bus must be kept less than 1,0, If the value of L

evaluated at load bus approaches 1,0 the volta ge at the


referred load bus going to become instable condit ion and if

L is equal to 1,0 the voltage is referred as collapse


conditions. This index will be used in the voltage stability
analysis caused by heavy loading conditions, It will be
assigned as the targ eted output of the ANN.
IIL THE PROPOSED ANN BASED APPROACH
A multilayer feedfoward Artificial Neural Network with
error backpropagation learning was used. The developed

network consists of three basic elements,


I. Neural Network architecture
2, Suitable backpropagation learning algorithm,

3. Method of training and testing,

38
A.

satisfactory for estimation voltage stability index. A bipolar

Neural Network Architecture


The topology of a developed network consists of input

two hidden layers, and one output layer. The


architecture of developed network can be seen as Fig. 1.
lay er,

sigmoid activation function was utilized.

r n this pnper, voltage stability index is utilized in the six

b uses IEEE test system. Four neurons are positioned in the


output layer to implement prediction task.

B.

:.,)-,
"

"
<!o

ANN Algorithm
In this paper, the development ANN machine is based

on mathematical algorithm of back propagation neural

network. Deve lopment of backpropagatioll ANN is consists

of three levels of process

Fig. I. Developed ANN archileclure

a.

forward

b.

back propagate of error

c.

w ei ght and bias up.da te

The flow chart in Fig. 4 shows the process involved in

Appropr iate selection of input variable is the key to the

development backpropagatiOIl Artificial Neural Network.

succes s of ANN appl ication. Usually heuristic knowledge is


required in choosing input variable. Many probabilities can

be related to the Voltage stability problem in a power


system, but this paper focusi ng on the pro blem caused by

the heavy loading of reactive power by the consumer sink.


Based power flow analysis; loading conditions are utilized
to find corresponding stability at each load bus by solving
load flow equation and stability index formulation.
To. declare input variable and targeted output in C

l an gu age

is an important part to ensure that the ANN

mach ine is capable to identify the position of load buses and

stability index L properly. In this paper input variable and

targeted output are arranged in matrix forms, which defined

by using for loop array technique , and can be visualized in

Fig. 2 and Fig. 3. Symbol m is stand for number patte rn .

ANN input =
m

Fig. 2. Input arrangement in

matrix form III x

ANN target =

Fig.4. Flow charI 10 build ANN backpropagation machine

m.

Fig. 3. Targeted output arrangement in

matrix form III x

C.

Training and Testing


The training process was performed in order to train the

Computational power of the ANN is a r esult of addition


of hidden layer. But there are no general guidelines to

detennine number of hidden layers and neurons . Although

many applications have proved that ANNs with one single


layer is' sufficient to solve any function approximation

problem, some problem may be easier to solve using an

ANN with two hidden layers. Our studies demonstrated that

two hidden layers with 40 and 32 ne urons each are quite

de velop ed network with a set of input and targeted output.

As being mentioned in introduction section , 60 tra ining


patterns were randomly selected from the voltage stability

results. The network was trained until it achieved a very

small MSE typically 0.0029 p.u.. The network , which is


fully trained with lowest error, is capable to be used in the
testing process. If the network does not converge in the
training phase, it implies that the network is not fully trained
and it is not capab le to perform the desired task [7].

39
The

generalization capabilities

of proposed Neural

Network have been investigated by using 49 testing patterns


(different from the training pattern). The reliable network
should be able to produce a very good output results. Beside
the overall MSE error, the performance of developed neural
network can also be justified by calculating the absolute
error for every testing patter and by using line regression as
shown in Figures 7, 9, 11 and 13. To further identify the
accuracy of the developed network, the developed network
must be indicating correlation coefficient, R closest to unity
(1.0000). Unity value of R indicates zero absolute and RMS
error.
IV. RESULTS AND DISCUSSION

TAI3LE I

NEURAL NETWORK CONfiGURATION AND PROPERTIES


No. of neu rons
configuration
Type 01' activation
each neuron

(40,32,4)
S igmoid activation

funclion

Leamin.& rate

function

0.3

Momentum ralC

0.19

No.

8823 iteration

of EOPoch

MSE Error

0.0029

Training pallem

60

Testing pattem

49

The test system presented in this study is the IEEE six-bus


system, for which the single line diagram is depicted in Fig.

5.

The

regression

analysis

and

prediction

output

of

developed Neural Network covering stability index at the


whole load bus in an IEEE 6 bus test system is shown in
Figure 7-14. Figures 7, 9,

11

and 13 shows the closeness of

the actual output produced by the network with respect to


targeted output. The results from the testing network are
shown in Table 2.

In Table 2, the results showsa good

agreement between output and targeted output which were


obtained in predicting voltage stability index during nonna!
operating condition.

TABLE 2
PREDICTION VOLTAGE STAI3lLITYINDEX DURING NORMAL
OPERATING CONDITIONS
Input

Bu.

Fig. 5. IEEE 6bus lest system

no

The IEEE 6 bus system is composed of a slack bus

(I)

and one voltage controlled bus (2) and four load buses (3, 4,
5 and 6). As a result, the developed Neural Network has 8

PL

QL

Outllllt

T...gct

0.1485

0.1482

38.0000

5.0000

10.0000

2.0000

0.0480

5
6

20.0000
30.0000

5.0000

0.1484
0.1196

3.0000

0.0477
0.1485
0.1193

inputs representing real and reactive power of each load


buses in the system and four outputs corresponding to the
value of L-index of each load bus.
The

L-index corresponding

to

the slack. bus

and

voltage-controlled bus are not consi.dered in the input and


output list since they are always zeros as long as the bus
voltages remain controlled.

The

configuration

of

developed

neural

network

consisted of two hidden layers with 40 neurons in hidden


layer 1,32 neurons in hidden layer 2 and 4 neurons in output
layer with binary sigmoid activation function was utilized at

all neuron. The selection of this configuration is based of its


training speed and its convergence accuracy. Figure 6 shows
the performance of the neural network in minimizing the
MSE error during training process.

Fig.

01

0.'
00
Oi

7.

Prediction L index on load bus 3 generated from neuron

I.

. ....... 11 ......

Tftt"oetO.148Z

0.1117
O.ll1S
0.2944
0.3911

0.1571
0. 1 4133
0.1798
O.!5H

BUS 110

Ou't-P""t.
SO.
O.IH
0, ZCO
0.2970
O.39U;
O. I5-SIS
o.l.as
o . eo-l
O.Z 5J{i
0.'

f).1I5C

0.0027

0.1'116

O.OOOS
-0.00090.000:':
0.0007
O.Oo;lOS

0.11<1':;

0.6721

0.671'1

0.0007

0."/1"/0

0.710S

O. Z061

0.12:59
O. ('r".-.:']

0.1191

O. HOI

O.QOlO

0.50'11

f1,nnnt

0.5041
0.4471

0,.)12.')

O. lOBO;

O.OOJ

0.0001
0.0041

O.Z9;::

0.0007

O.OOLO
0.0003
0.000::

0.317)

(t. 1 1)5
O. 1"Z
n.I)':'i'

0.0[100;
O.OOll
0.00::1

0.20': 1

0.00':6

0.01';1

0.00)9

n. }1.)':1

O,riU30

0 .0007

0.3;:-:: :::

o .(l.'"

J.b!:.erroL'

O. C!'li

Q.;"; (7t,)

Q. J215-

O.OOSO

0.0'101I) 11'1 (11'

Q.O)5t

O. ()'162

0.0'100

0.0002

0.H'17

,(1.0006

0.7S91

O.7&:!O

0.0026

n. f!I('1

1).('1(;

(L,(}Ot]S

OS058

0.50--18

U.OOIO

0.I'IZ8

0,1)66

0.(106?

O. S772

0.6799
0."/593
0."';4
0.8208
o.e ....o
0.861'1

0. 7517

0.ou03
IJ .O(J()
O.QO'l()
0.'1087
<l.OQI!'
0. 0006

O. Den

0.0019

0.67'150.-6"1")9
0.77L2'
o.e"o
o. B651

0.1<"3
0.092.
0.222S

0.'87'
O 61S
O.Ol
0.227'1

0.)10$8

0.0010

a.QO'.)?
0.00L8
0.0000
0.0053
0.003'
O.OQI'l'5

0:

L'17'1
0, 19n
0.2008
(),H1(l
0 . .2 61"
O.Z:."! I
O .lI'7!;
O .!1(.7

O.H!'I
0.'::(11';
o I 200
n, .":S) l
O.:!S1'
0.::9] 6
o.z,,,
O.i!'le

40

I) .OQi)S

O.:

0. l'/?2

..

0.0010

0.1'1,6
O. 19J 3
0.190'5-

t1, 17-00

.. ..

0.0031
0.0(6)

Q.O30

;: t: :;'"

0.0010

0.1383
0.1107

0,

..

Itlll't

O.QOQ

O.075

1'), o:-.:t J

0, ):171

().()Oll
0.0()5S
(I.oOSS
-o.nOOI)

OutPJt.

0.191$
O. 19:0S.
O,(ln-
.,...,.
(.1,

0.0011

0.0'197
O.93:J
O . ,.,S

O. Z 112
0.e7-11
0.9367

. ...... .,. lIt.IS 110 4

0.0477
0,0";'(5";

0.0006

O.IU

0.71'0

Tnr",.,!t

Ab.r:-t'01
O.OOI
O.OOI

(I.000

0."/181

. r .. .. . .,"

'it ' It

0.0001

0.0315

0.Q)58

0 . 00 1l
0.000-1
0 . 1.1 0-1 ]

0,2291

O.ZZ:l7

o .OO5l

o.ne'll

0,010'-:

0.001.1

0.0976

O.OJ7Z

0.09SI

O.IQ05

0,005'1

D.SPOt

0.S800

0.000"
:O.OOl

0.7221

o.ooo!'

o.ooo:!
O.OO5

o.:!ee

0.2"9.;

O. -)01 ?

0.1000

O.O83

O83-Z9

0.0007

0."'004

0.0957

0.09.0

0 .0003

0.(3))

0.1798

0,1(0)

O.()I)05

0.12'10

0.ll1::

0.7228

0.:'409

0.)9
o . ... ., ;:
0.l4)
o. . . u
o.l7_e
0.24119

0.545'8

0.S!05

0.5'100
0 . 1 '17 )
0.3'179
0.)10'18
0.17S1

0.1'150
0.4194

Fig. 8.

0.1752
0.4191'5

0 . 000-6
0.0001

<l.ooo"

e.OO1S
(l. 0003
o.oooa
-o.OOOj

, OOI.

0.0785

0.0010

0.20;'1'1

0.266e

0.0009

O. lfH
a .046!'

0.1436
O.Olll

o. 19"'

0.191$7

0.0/33

0.0690

o.ooe
0, olile

O.ono

0.0053
0,0007
0.00'12

0. 10 1 9

0.000\5

0.0971

0.097)

0.0002

O. Ii 52

O.IHe

0.0003

0.1271

0.0000

0409?oJ

Q. 1 t.

bus 3

0 ,I)OJ"I-

O.ons

0.0701

0.0009
o.oooe

Resull ofprediclion on

O.ISO)

O. 14i5

Fig.IO. Result of prediction OIl

bus 4

.
.
. .5._--
1 .-____._--r-R

o. f r I. i. j. f +/:$
..

08

(1,2

.'

Fig. 9.

Accuracy prediction L index on load bus 4

generated from neuron

11.

.
""'''Lt .. .. ..... ,! ... .! ..... ..'... ' "
,

.! .....

02

0,9

'CIO

Fig.

!.

0,1

Prediction

0.3

0.4

Il5

0.6

1),1'

o.I!I.

L - index on load bus 5 generated from neuron J

.............

o: ..

TarQct

em; 110 5

(Ut PUt

.. ....... .........

0.14'101
o. ?Z

O. H&S
O.:!?Ol
O.)SJ
Q.35M
0.516)
O.fSH
0.0170
O.IS3;'

O.S1f.f

0.0006
O.OOOJ

(1.0001

O.441
o.He)
0.107]

0.15)0

0.000.

0.0940

o . elJ
o.o,:;:

(I.OOS)

O.)I
O.t"16-4
0.5"0

O.(>(10

0.2976
0.33:'10
o. :;593
Q.';976

0.0001
O.ooos

O. 180

0.0006

0.0106

0.$650

O.I.HOO

0.29$3

0.16';-1"

O. I"'''

0.0005

1J.1S1?

D.197

0 . 210

0.0(1)

I). 859;;;

0.8::8
O. )o;::.
O.170'?

O.6!
o.s:-z .
o.:)ole
O. I"O

O."6&

0,,';03

0.5750

0.57(13

O. '}'10

0.8;:01
0, '17907
O. 71J':!

0."7:8:J
O. "1117

0.017.'1

0.9765

0.99.0
O.ali{.

o. es.

0.0108
0.0000

(I.8SH
O.7L-e

0.0010

(1. 718'3
0.-1999

0 " 9 1
O. ]160
0.171)
0.30360.5601
0.8,)28

0.0005

0. 00 1

O.lt9::
O.lo,,;

0.00)6

0.6::92

0.6126
D.1917
0.153:6o. en I

0.813"
0.19:2:

0.e4S
0.1S6
O. 5"J
a.33le

O.lJoS
o. 'P.
O.33Z0
Q.16St9
O.l221
0.9854
0.0391
0.)::9,

Fig. 12.

Result

0_:::")8S

1).tJOr),";

O.OOH
0.0009

0.,361

0.2)04
O. i SO.
0.09-12
O.BS';

0."I5l0
0.O40

0.0001

0,0012
0.0092
0.0000

0.(1)-16

'J.Or);;L
(j.IJOJJ

0.1)')8
0.0019
O.4)Oc}?
0.('11)1--:

0.000 ..
l"'t. Q("nli
0.000'1
o .0Ul;

0.0001
1J.0n.""!1
n,01J71

0.0020
r).OO1

O,OO)t.>

0.0011

0.0005

O.OOOti
0.000.

f).OO19
0.0009
0.0007
0.0-02Q
-O.OIJOriO.OOO

O.OOOS

O.OOH

0.1006
0.'07S

O.409S

0.00t7

0.3:991'

0.4001

0.0001

0.181

i).:n70

0.39'59-

Fig.

of prediction on bus 5

0.000.

0.0012
O.OOQS
0 .0008
0.0(0)

O. '"9 1
0.10:8
o .:!OO:;8

o. 9) 6
0.10]9
0.2061

0.0003

0.0')02
0.(11)0"

0.10 ..

0.1976

0.0001

0.'000<1

0.3"154

0.228i

0.0001

0 . 170 !
O. )41
0.9762O.O:!94
0.3291

(t.tJOO ..

O,OOO?
O.ooos

O.OOOZ

0.1 SJ'

Q.QO .. '::
(t.OO'31

0.2 SOli
o. ::Si3-!.
O. ::J98

O.2Sei
0.2.389
0.14S7

0.0004

0.5809

o .'J002

0.1<);::9
0.D9-11
O. ;:?t
Q.)3)'?
o. 5'89
O.':::?7i

O. )-;'131
O.OOl

0.2S12

0.001. 1

0.H91

O.3:l3.
O ... "':'. 1

O,3'J5
0.0906-

0.0006

0.1166

1').1)0'::7
a .JOO!)2

Q.)J 11

0.) 172
O. "']9

0.0010

O.ll';')l

O.-1H'?

1).000.,
0.01)053

O.O":t
O. q t0
0).3-t0;;1
o. 19)

0.28(>5
O.3Z2
O. )SlL
0,309.

0.0001

fJ.3':S

o. :;89-1
O. J'1JZ
0.:: ss
o 19' 1':"
0. >10,;

0.2072

G.OOIS

0.9!HJ
D.gep
t),91340

0.9B91

41
to ,

.'j).!".-el;":r;o["

o. J 15;

0.;:; ';;'-1

O.OOl-'>

0.9763:

"'

0.0021

O.19l}

o. z S6
0.1926

II! :4 II'

0.117.2

0.1'191

0.J4;

..

t)'UtHlt

O.o{o'3-11

0.1519
0.1138
O.19Z
0.2906

0.0006
0.0007
0.0009
0.001'

0.11210

1).9"3,?

o. '18-9"

0.0003
0.0003
O.O{'lO7

. ..

O.lSt;S
O. '1':51:

f). "SS]

0.0037

03 rro

O.156i
O. J a 1(1

O.lSSS
0.)548
o. )7e

0.0010
0.0026

0.3S9"

O,):?O,
O."1e:

Tar..,et
O. 11)

Ab.:!!l.error

0.0011

b.3'\9-5

0.1676

:Io:t t11- .,

14. Result of prediction all bus

V.

CONCLUSION

The development ANN for voltage stability monitoring


was

presented.

The

two

layers

backpropagation

neural

network using C programming language has been employed


for voltage stability monitoring. From the lesting process, it

DlS .

shows that the proposed method is capable in predicting

voltage stability index for the load bus in power system


network. The proposed method was tested us ing the IEEE 6-

0.1

bus test system and indicated a good agreement between

[I.Hi

targeted output and ANN output.

0.1
0.05
DDS

Fig.

I J.

Prediction

[11

0,15

0.2 O,;;S 0.3


TlrgllljllfOU

L - index on

load

. -- I ::I: Lm.
0.35

0.04

0....5

0.6

bus 6 generated from neuron 4

V!. REFERENCES
[I]

P. A.

Lof,

et aI., "Voltage Stability Indices

Systems", IEEE Irans.

For

On Power System, Vol. 8. no_

Stressed Power
I, Feb 1993, pp.

326-335.
[21

[3]

[4]

[5]

B. Gao, ct aI., "Voltage Stabili ly Evalualion Using Modal Analysis",


IEEE trans. On Power System. Vol. 7, nO..4, Nov 1992, PI'. 15291542.
N. Flatabo,

et

aI.,

"Voltage

Stability

Condition in a Powe
IEEE trans. on

TransmissiQn Calculated by Sensitivity Method",

Power Systems, Vol.5, no. 4,Nov. 1990, pp. 12861293.


M. M. Begovic, A. O. Phadke, "Control of Voltage Stability using
sensitivity Analysis", IEEE trans. On Power Systems, Vol. 7, no. I,
Feb 1992, pp. 1 14123.
T. K Abdul Rahman and G. B Jasmon, "A New Technique for
Voltage

Stability Analysis

In

Power System and

Improved

42
For Distribution Network", Intrcll1ation.1
Conference on EMPD, Vol.2, pg 7 I 4-719, Nov. I 995.
A.A. ElKeib, X.Ma, "Application of Artificial Neurat Networks in
Voltage stability Assessment" IEEE tr3ns. on power system. 1<)<)5.
l. Musirin, T. K. Abdul Rahman, "Security Assesmcilmt Using
Combinatorial Fonn Neural Network Topology (CrNNT)", 3"
lASTED International Conference on !'ower System 2-5 Sep' 20U3,
Marbella Spain.
Loadnow

[6)
[7)

Algorithm

You might also like