Distance Transmission Lines Protection Based on Recurrent Neural Network

glibdoadingAI and Robotics

Oct 20, 2013 (3 years and 7 months ago)

106 views



International Conference on Science, Technology and Innovation for Sustainable Well
-
Being
(STISWB), 23
-
24 July 2009, Mahasarakham University, Thailand

1

Distance Transmission Lines Protection
Based on Recurrent Neural Network


Saichoomdee
, S
.

,

Oonsivilai
,

A.

,

Marungsri, B.

,

Kulworawanichpong, T.


and

Pao
-
La
-
Or, P
.


Alternative and Sustainable Energy Research Unit, Power and Control Research Group,

Scho
ol of Electrical Engineering, Institute of Engineering, Suranaree University of Technology,

111 University Street, Muang District, Nakhon Ratchasima 30000, Thailand



E
-
mail:
saichoomdee@gmail.com,ana
n
t@sut.ac.th
,

bmshvee@sut.ac.th

,

thanatch@sut.ac.th
,
padej@sut.ac.th
,



Abstract



The
adaptation

of protective distance relay to determine the presence and location of faults in a
transmission based o
n the measure

of fixed settings such as line impedance has been achieved by the
application

of several different techniques. However
, for the modern power systems a

fast, accurate
and robust technique for real
-
time purposes is required. In this paper, the application o
f recurrent
neural network in transmission line protection is demonstrated. The method
applies the power system
using

voltage and current signals to learn the hidden relationship existing in the input patterns. It is
observed that the proposed technique is

able to ident
ify the precise fault direction
more rapidly.
System simulations studied show that the proposed approach is able to detect the direction of
a fault
on a transmission line swift
ly and correctly,
th
erefore

suitable for the real
-
time purposes.




Keywords:

recurrent neural network; distance protection; power system; relaying.

1.
Introduction

Transmission

line

system
is

regarded
with
great
importance in power system
.

Fault
s

that

occur

frequently with
tran
s
mi
ss
i
o
n

lines

system
, sho
uld affect elect
ricity user
s
.

F
aults,
aforementioned may be

cause
d by neither a
single person, animal or natural occurrences
.

Thus
to prevent

and
decrease

damage that

would

happen
,

must

syst
ematically protect

the
tran
s
mi
ss
i
o
n

line system
.

T
ransmiss
i
on line
system

using di
stance relay

is very popular
.

I
n
this current, using an intelligent system

solves

a problem i
n the remedy of
power system
widely.

F
or example
,

dynamic load modeling
[
8
]
,

short term load forecasting
[
9
]
,

stabili
ty

in

power system
[10
]
for
tran
smiss
ion lines

protection

have
been
using neur
al network
,
could

test by

distance

relay

[5
-
6] neural

network,

electric base
will n
ot be used in
calculation
but,

the path calculated

will be used
which

is
obtained by
the format of learning or
the ability
to memorize

of neu
ral network
;

accompanied with flexibility in itself

makes
neural network very interesting
.

W
e can use
neural network in learning and memorizing the
format of fault

and the format of condition
changing

of power

system
.

Although this
might force relay
,
it

pr
otect
s

tran
smission
lines, with increasing precision

(
zone 1)

and
can be applied accompanied with

dis
tance
prevention normally which, hypothetically
wi
ll
help testify the protection of transmiss
ion lines

become much more accurate
.

This paper proposed

dista
nce transmission
lines protection based on recurrent neural
network.


2.

Neural Network

T
he sys
tem assembles a mode format from
the human brain
.

T
he
duty and the work of
neural,

be
could
built

large
-
sized
and could
teach the
system for the lead gone are us
able,
especial
ly
,
the principle works of neural, will
find the

weight
value in work system of neural,

the
n

the
c
omparison output beg for neural, that
get with target value that fix
.

I
f
the value
output is not

equal to target

value,

t
he system
of neural wil
l find the value of the weight until
it reaches

the value of output
, t
he new
sub
stitute
value
is equal to target value
.



International Conference on Science, Technology and Innovation for Sustainable Well
-
Being
(STISWB), 23
-
24 July 2009, Mahasarakham University, Thailand

2

For the

neural

to be built give with the
capability to

learn
.

Mu
st have input and output
data to use in

comparison
. I
nspires the
use
ord
ered
pair
s in traini
ng network

[11
-
12
]
.


2.1
Recurrent Neural Network

Recurrent

network

is
the
connections

interval neural in feedback
,

recurrent neural
network structure is shown

in

Fig 1.



Fig
.
1
:

Recurrent
n
eural
n
etwork

s
t
ructure



Fig.1

Show
s

recurrent neural network
structure
;

Input patterns

have

P
1
,P
2
,…,P
R
,
a
1
(k)

to
be output of


hidden layer 1 and
to
be
input
of hidden layer 2

,
a
2
(k)

to
be input of
hidden layer 2

,
a
3
(k)

to
be
final output.

Have
ƒ
1
, ƒ
2

and
ƒ
3

are transfer

function,
a
1
(k),
a
2
(k)

and
a
3
(k)

can get from the algebraic equation
as follow
ing could
:







(1)







(2
)









(3
)



Whereas
:





;

weights

value
connections
between





input
layer

with

hidden layer

1



;

weights

value
connections between





hid
den layer

1 with hidden layer 2



;

weights

value
connections between




hidden layer 1 with

output layer




;

bias

value

in hidden layer

1




;

bias

value

in hidden layer

2





;

bias

value

in
output layer


3


Training neural network by
gradient descent
algorit
hm
with

tan sigmoid

transfer
function

u
sing n
eural
n
etwork
t
oolbox
of matlab

program

[3]









(4)


Where
as
:





n
;

summation output




b

;

bias adjust.


3
.

Application Recurrent

Neural




Network for

Transmission Lines




Protection

Fig.
2
show
s

recurrent neural network

for

t
ransmission

l
i
nes
p
rotection structure.
Fig.
3

the

recurrent
neural

network
for classifying

the
input patterns into expected categories.

There
are three input signals required at the input
layer
in recurrent
neural

network
:
V
,

I, and X
.
V is voltage, I is current, and
X

is

apparent
impedance
,

the measurement

of the faulted
transmission line
.

The output consists of
recurrent
neural

network which has a
continuous
-
value ou
tput in the region

[0
, 1
].Output

1 indicate
s

tripping,

0
indicates
non
-
tripping
.





Fig.
2
:

Recurrent

neural network

for


transmission lines

protection structure




Fig.
3
:

Input
patterns
classify

of recurrent
neural network



International Conference on Science, Technology and Innovation for Sustainable Well
-
Being
(STISWB), 23
-
24 July 2009, Mahasarakham University, Thailand

3

4.

Simulation

and Result



Training patterns

and
test pat
terns

g
o
t

from
fault
simulation

on
transmission

line
of
power


system


study


using



MATLAB

and

SIMULINK
.

Fig.
4

depicts

the
115

kV,
5
0 Hz
si
mulated system one
-
line diagram
. The other

related parameters of the simula
ted system are
show
n

in

Table 1
.




Fig.

4
:

One
-
line diagram of simulation system


Table
1
:

The parameters of the simulation


t
ransmission

system


Bus 1

:

Voltage 1
15 kV , 50 Hz


Equivalent source impedance


Z
1

= 0.00499 +j0.03384
p.u. /km


Z
0
=

0.00425+j0.0369 p.u. /km

Length of transmission line
:


㄰1


km

䱩湥⁣潮ot慮t
:

Z
1

= 0.085811+j0.36204 p.u./km

Z
0
= 0.25485+j1.4223 p.u./km

Bus 2

:


Load=

㌮㌫3j㈮㈠2VA


T桥h t敳t will 扥bi渠 wit栠
f慵at
潣o畲r敮e攠
獩m畬慴i潮

慴 t桥h 摩獴慮a攠

〬㈰ⰴ〬㘰ⰸ0
ⰸ㐬㠸ⰹ㈬㤶,
慮a

㄰〠
%

of the
total line length
.

Every the distance fault
occurrence
has

fault
resistance


0.000001,5,10
,

15
, 20, 25,30,35,40

and 50 ohms
respectively.

A
lready lead the data has that

go to test with
the recurrent neural network.

Zone 1
pro
tection
is

80

% of the total line length.

Fig. 4 shows
the phase A current waveform,

and the phase
A voltage waveform for the single phase to
ground fault,

the fault resistance is 10

oh
ms
,


at
times 20 ms
-
60

ms
.

Some part calculation
s

of
pattern

data
tests for the recurrent neural
network in case of fault resistance

5 oh
ms are
show
n

in Table
2
.




Fig

.
5
:

Voltage and current
waveform

a
-
g

fault


Table

2
:

T
he
pattern

data
tests
in

case
of fault




resistance

10

ohm


Distance

Magnitude

A
ngle

(%)

VA

IA

VA

IA

0

114820.0

18635.00

-
0.400

-
8.612

20

114780.0

13840.00

-
0.302

-
19.739

40

114810.0

10910.00

-
0.251

-
22.768

60

114840.0

9108.50

-
0.224

-
22.813

80

114860.0

7996.70

-
0.208

-
22.112

84

114860.0

7844.70

-
0.206

-
22.055

88

114870.0

7718
.50

-
0.204

-
22.064

92

114870.0

7621.60

-
0.202

-
22.149

96

114870.0

7559.00

-
0.201

-
22.314

100

114870.0

7537.70

-
0.200

-
22.544














International Conference on Science, Technology and Innovation for Sustainable Well
-
Being
(STISWB), 23
-
24 July 2009, Mahasarakham University, Thailand

4

Table

3
:

Accuracy
transmission

line protection


Distance

(%)

Accuracy

(%)

Error


(%)

0

100

0

20

100

0

40

100

0

60

100

0

80

10
0

0

84

90

1
0

88

100

0

92

100

0

96

100

0

100

100

0


5.
Conclusion

This paper analyze
s recurrent neural
network intelligent computational techniques
application associated with protective distance
relay for transmission line protection.

B
e the
one
method

transmission line
prevention
,

t
he
back manages to apply
recurrent neural
network

in transmission

line

prevention
. From
testing it is correct in

90
transmission line
percentage preventions
,

from the data tests 100

all the data set
tests
.

T
hus
,

recurrent

neural
network

with

the ability to relay one type
in
transmission

line

prevention
.

Therefore

recurrent neural network

c
ould

be used as an
effective tool for real
-
time digital relaying
purposes.

That

might allow

distance relay

work
more accur
acy

and
precision
.


6.
References

[1]


Anderson

P.M.
, Power
System P
rotection,
McGraw
-
Hill, 1999.

[2]

Warwic
k
,K.,

Ekwue
,A.
and

Aggarwal
,R.
,

Artificial
Intelligence T
echniques in
P
ower System, London, Institution of
Electrical Engineers, 1997.

[3]

Howard B. Demuth, Ma
rk Beale, Neural
Network Toolbox for

Use

with
MATLAB
, 1998
.

[4]

Martin T. Hagan, Howard B. Demuth
Mark Beale, Neural Network

Design,
Oklahoma State University, 1996
.

[5]

Qi W
.
,

Swift, G.W
.
,

McLaren, P.G.,
Castro
,

A.V.
, An

A
rtificial
N
eural
Network A
pplication to
D
istance

P
rotection
,

Intelligent

System Application
to

Power

Syst
ems International
Conference, p
p
.

226
-
230, 1996.

[6]

Coury,

D. V. and

Jorge,

D. C.,

Artificial
N
eural
N
etwork
A
pproach to

D
istance
P
rotection o f
Transmission L
ines
,

IEEE
Transactions on

Power
Del
ivery, pp

102
-
108, 1998
.

[7]

Wu
, L.C.,
Liu
,

C.W. and
Chen

C.H.
,
Modeling
and

T
esting of a
D
igital

D
ista
nce
R
elay U
sing
MATLAB
/SIMULINK
,

IEEE
Transactions on
Power Delivery,

pp.253

259,

2005

[8]

Oonsivilai, A. an
d El
-
Hawary, M.E
,

Power
S
ystem
D
ynamic
L
oad

M
odeling
U
sing
A
daptive
-
N
etwo
rk
-
B
ased
F
uzzy
I
nference
S
ystem
,

In Proceedings of the
IEEE Canadian Conference on Electrical
and Computer Eng
ineering
,
pp
.

1217
-

1222.
1999.

[9]

Oonsivilai, A. and El
-
Hawary
,
M.
E.
,

Wavelet
N
eural
N
etwork
B
ased
S
hort
T
erm
L
oad
Fo
recasting of E
lectr
ic Power
System Commercial L
oad
,
In Proceedings
of the IEEE Canadian Conference on
Electrical and Computer Eng
ineering,

pp.

1223
-

1228,
1999.

[10]

Oonsivilai, A. and El
-
Hawary,
M.E.
,
A

Self
-
O
rganizin
g Fuzzy Power System
S
tabilizer
. In Pro
ceedings of the IEEE
Canadian Conference on Electrical and
Computer En
gineering,

p
p
.

197
-

200,

1999.

[11]

Oonsivilai, A
.
,

Boonwuitiwiwat, R
.,

Kulworawanichpong, T.

and

Pao
-
La
-
Or, P.
,

Artificial
Neural

Network

A
pproach to
E
lectric
F
ield
A
ppr
oximation around

O
ve
rhead
P
ower
T
ransmission
L
ines
,

EuroPes

2007.
Spain.
2007

[12]

Oonsivilai,

R.,

and

Oonsivilai,

A.,
Probabilistic
Neural

N
etwork
C
lassification for Model
β

Glucan
Suspensions.

Proceeding of the 7
th

WSEAS Int
ernational

Conference

on
Simulation, Modeling and Optimi
zation,
pp. 59
-
164, 2007.