TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 11, Novembe
r
2014, pp. 76
0
3
~ 761
2
DOI: 10.115
9
1
/telkomni
ka.
v
12i11.64
96
7603
Re
cei
v
ed
Jul
y
16, 201
4; Revi
sed Septe
m
ber
18, 201
4; Acce
pted
Octob
e
r 1, 20
14
Identification and Classification of Power System
Faults using Ratio Analysis of Principal Component
Distances
Alok Mukh
er
jee, Palash Kundu, Arab
inda Das
*
Rese
arch F
e
ll
o
w
,
De
partme
n
t of Electrical En
gin
eeri
ng, Jad
a
vpur U
n
ivers
i
ty,
188, Ra
ja S. C. Mullick Ro
ad,
Kolkata 7
00 0
3
2
, India
*Corres
pon
di
n
g
author, e-ma
i
l
: adas_
e
e
_
ju
@
y
ah
oo.com
A
b
st
r
a
ct
Power system reliability o
per
ation
has
been one of the
m
o
st
vit
a
l topics under res
e
arch. The
pow
er system
netw
o
rk, mostl
y
the long
tran
smiss
i
on l
i
nes
is often subj
ec
t
ed to different
types of faults
lea
d
in
g to
mal
oper
ation
of
p
o
w
e
r flow
. The
id
ea
of
a
re
lia
ble
prot
ection
system is
to
most
accur
a
tely
an
d
efficiently
id
ent
ifying th
e fa
ult, classifyi
ng
an
d the
loc
a
ti
ng
of fault. T
h
is
p
aper r
epr
esent
s the a
p
p
licati
o
n of
dyna
mic phas
o
r
s in the form of Princip
a
l Co
mp
on
ent A
naly
s
is (PCA) to identify fault in a
three phas
e o
n
e
end fed
150 k
m
lo
ng ra
dia
l
p
o
w
e
r system trans
missi
on l
i
n
e
. In the propo
sed w
o
rk, (1/4) cycle pre-fau
l
t an
d
(1/2) cycle
po
st fault lin
e vo
ltages
hav
e b
een
extrac
ted from
El
ectro
m
agn
etic
Transi
ent Progr
a
m
mi
n
g
(EMT
P) simula
tion. T
h
e
pro
p
o
s
ed
alg
o
rith
m
i
s
traine
d
usin
g
only
on
e set
of receiv
in
g e
nd
data c
a
rryin
g o
u
t
fault o
n
ly
at th
e
mid
poi
nt of t
he
l
i
n
e
to
gen
erate fa
ult si
g
natures
usi
ng
PCA. T
he e
i
g
e
n
vectors
and
the
score matrix th
us obtai
ne
d co
rrespo
ndi
ng to
the three
p
has
es usin
g the a
bove a
nalys
is have b
e
e
n
utili
z
e
d
to construct th
e co
mp
one
nt d
i
stances, w
h
ich
have
be
en
a
n
a
ly
z
e
d
usi
ng r
a
tio an
alysis to
extract the si
mi
la
r
features of any
particul
a
r fault
indivi
du
ally.
Ke
y
w
ords
: power system
faults, PCA,
EMTP, MATLAB, ra
tio analysis
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Fault identification and cl
a
ssifi
cation of
faul
ts is on
e of the most vital appro
a
c
he
s o
f
power
syste
m
stability an
d quality ma
n
ageme
n
t th
e
s
e days.
Lon
g
tran
smissio
n
lines alon
g
with
the large
sca
l
e po
we
r
systems a
r
e
the
most
sp
atial
l
y extended t
e
ch
nical sy
st
ems
and
fairly
often a
r
e
mo
st vulnerable
to min
o
r a
s
well a
s
seve
re
faults
sin
c
e
th
ey are m
o
stly
expo
sed
to t
h
e
different atm
o
sp
heri
c
h
a
zard
s [1]. The
objective
of power syste
m
fault
analysis i
s
to provide
enou
gh fault i
n
formatio
n to
the dete
c
tion
mechani
sm t
o
und
erstan
d
the re
ason
s t
hat ca
used th
e
system to de
viate from normal op
eration and to
re
store the n
o
rmal operatio
n as qui
ckly
as
possibl
e. Circuit bre
a
kers and other p
r
otective
rela
ying mech
ani
sm
s are to b
e
given pro
p
e
r
sign
als to a
c
tivate the tri
pping
ci
rcuit at the
corre
c
t in
stants.
Hen
c
e,
prom
pt and
a
c
curate
detectio
n
of
fault along
with preci
s
e
fault
distan
ce me
asure
m
ent have b
een p
r
a
c
tice
d by
sci
entist
s
in o
r
de
r to en
sure system
as
well a
s
prote
c
tive equip
m
ent safety an
d in pre
s
e
n
t era it
has b
e
come
one of the mo
st promi
s
in
g chall
enge
s of
the powe
r
sy
stem stu
d
y [2- 5].
Powe
r syste
m
fault analysis al
gorith
m
shoul
d be d
e
sig
ned in
such a
way so that it
sho
u
ld be hi
ghly efficient
and accu
rat
e
as well
as it should ha
ve general a
pplicability and it
sho
u
ld b
e
sui
t
able for re
al t
i
me u
s
ag
e. Rese
ar
che
r
s h
a
ve develo
p
e
d
so ma
ny m
a
thematical a
nd
comp
utationa
l tools fo
r th
e
dete
c
ti
on,
classificatio
n
a
nd lo
cali
zatio
n
of ele
c
tri
c
al
po
wer sy
ste
m
faults in conj
unction with relaying and protective
devices in order t
o
re
store
system stability at
the earlie
st [2]. Artificial Neu
r
al Network
(ANN) al
ong with
Neu
r
o Fu
zzy System and Wa
velet
based fa
ult p
a
ttern
re
cog
n
i
tion techniq
u
e
s
have
bee
n exten
s
ively used
the
s
e
days [3]. But
the
ANN impl
em
entation requ
ires
a larg
e
numbe
r of tr
aining d
a
ta a
s
well
as trai
ning cy
cle
s
, thus
requi
rin
g
hea
vier amount
of data and comp
utati
ona
l burde
n [6, 7]. Other techniqu
es in
clu
de
wavelet transform with other me
thods such
as P
r
obabilisti
c N
eural Net
w
ork
(P
NN), adaptive
resonance theory, adapt
ive neural fuzzy inference system,
and
support vector machi
n
es [8]-
[10]. Fuzzy lo
gic h
a
s al
so
been
co
mbin
ed with
di
screte Fou
r
ie
r transfo
rm, ad
a
p
tive re
son
a
n
c
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 76
03 – 761
2
7604
theory, pri
n
ci
ples
of estim
a
tion an
d ind
epen
dent co
mpone
nt
ana
lysis
to enha
nce pe
rform
a
nce
[11].
Princi
pal Co
mpone
nt Analysis (P
CA) a usef
ul st
atistical te
ch
nique that h
a
s foun
d
appli
c
ation in
variou
s field
s
of engi
nee
ring study
[12
,
13]. It is, n
o
w-a-days, q
u
ite a com
m
on
method u
s
in
g
multivariate
statistical pro
c
ed
ure
co
ntro
l techniq
ue a
nd pres
ently use
d
extensiv
ely
in powe
r
system operatio
n, analysis a
nd cont
rol. Dynami
c
pha
sor a
nalysi
s
like PCA offe
r
several a
d
va
ntage
s ove
r
conve
n
tional
method
s in
several
are
a
s of fault anal
ysis li
ke fa
st
er
nume
r
ical si
mulation
s, a
s
they prod
uce
slo
w
e
r
variation eve
n
on very f
a
st chan
ge
s o
f
instanta
neo
u
s
inp
u
ts allo
wing fo
r larger
step
sizes in
nume
r
ical expe
rim
ents a
nd fa
ster
simulatio
n
s [
1
]. PCA transform
s a set
of multivariate data to a lower dim
e
n
s
ion o
r
thog
o
nal
space,
retaini
ng the most
variability of
the ori
g
i
nal
i
n
formation, thus pr
oviding with the most
signifi
cant di
rection
of vari
ati
on of the
given multidi
r
ection
al data
set hi
ghlight
ing b
r
oadly t
he
simila
rities an
d differences
betwe
en th
e
variou
s ty
pe
s of faults.
Be
cau
s
e
of the
simplification
of
the analysi
s
and re
du
ced
and orth
ogo
n
a
l dimen
s
ion
a
lity of the d
a
ta obtaine
d, PCA has be
en
use
d
exten
s
i
v
ely in po
we
r sy
stem
an
alysis,
espe
cially in fault
detectio
n
, cl
assificatio
n
a
nd
distan
ce pred
iction whe
r
e multiple
dim
e
nsio
nal
d
a
ta are obtain
ed rega
rdi
ng
volt
age,
current and
to so
me
extent, power an
d
freq
uen
cy of
the th
ree pha
se or
si
ngle
p
has
e
system.
The
dime
nsi
o
n
of the fault d
a
ta dou
ble
s
whe
n
the te
st is pe
rform
e
d on d
oubl
e
circuit line
s
.
In this rega
rd
,
Comp
one
nt analysi
s
h
e
lp
s quite
accu
rately to red
u
c
e d
a
ta dim
e
nsio
n d
r
asti
cally and e
n
a
b
les
easi
e
r, fa
ster
and to
a g
r
ea
t extent, fairly accu
ra
te
co
mputation [2],
[13-1
7
]. Thu
s
PCA
ha
s b
een
cho
s
e
n
as th
e prima
r
y co
mputation too
l
in the propo
sed
work.
In the pro
p
o
s
ed work, a si
mple techniq
ue for
fault id
entification
a
nd cl
assification in a
radial
sy
stem
ha
s be
en d
e
velope
d he
re. This PCA
ba
sed fa
ult
detectio
n
sch
e
me h
a
s be
en
adopte
d
h
e
re
ba
sed
o
n
th
e patte
rn i
ndi
ce
s a
n
d
fault
sign
ature
s
ge
nerate
d
from
pha
se volta
g
e
s
only. The p
r
o
posed
work i
n
vestigate
s
t
he ap
plicat
io
n of dynami
c
pha
sors to b
a
l
anced a
s
wel
l
as
unbal
an
ced
poly-ph
ase
p
o
we
r system
s.
EMTP-AT
P
simul
a
tion
software h
a
s
b
een
u
s
ed
to
develop a
nd
simulate th
e long tra
n
smi
s
sion lin
e mo
del. 15 blo
c
ks of 10 km each ha
s be
en
interconn
ecte
d in a
radi
a
l
netwo
rk in
a 40
0 kV
-150
km lo
n
g
thre
e ph
a
s
e
singl
e ci
rcuit
transmissio
n l
i
ne. The
simu
lation
wa
s foll
owe
d
by
anal
ysis
of the th
ree p
h
a
s
e volt
age
wavefo
rm
for cl
assificati
on of fault types u
s
in
g PCA
algor
ith
m
in t
he MATLAB
environ
ment.
Different type
s
of faults alon
g with the he
althy conditio
n
of
the network u
nde
r varying fault locations have b
een
tested fo
r
bet
ter robu
stne
ss of
the
prop
ose
d
p
o
wer system
prote
c
tion algo
rith
m
which sho
w
s
very much
ap
pre
c
iabl
e pe
rforma
nce in p
r
ompt det
e
c
ti
on within h
a
lf cycle after th
e occu
rre
nce of
the fault. In
the prop
osed work, EMTP si
mulati
on ha
s been restri
cte
d
to unconta
m
inated he
althy
sign
al o
n
ly, i.e. only h
ealth
y voltage
wav
e
form i.e.
pu
re si
nu
soid
al
signal
of 50
Hz free f
r
om
a
n
y
harm
oni
cs h
a
s
bee
n co
nsi
dere
d
he
re.
In the first p
hase of thi
s
pap
er
ori
e
n
t
ation, simul
a
tion p
r
o
c
e
s
s in
detail
h
a
s
bee
n
discu
s
sed he
re along
with the
meth
od
s of
data
co
ll
ection. In the
fol
l
owin
g
stage
s, the p
r
op
ose
d
fault detecti
on alg
o
rithm
has bee
n
describe
d
elabo
rately
with ne
ce
ssary figures
and
comp
utation
s
of the results obtaine
d there fr
om hav
e been carrie
d out. Finally, the utility a
nd
useful
ne
ss
of the co
mpo
n
ent analy
s
is i
n
po
wer sy
st
em fault dia
g
nosi
s
in
rel
a
tion to the
re
sults
obtaine
d ha
s been di
scu
s
sed.
2. Sy
stem Design
In the pro
p
o
s
ed
wo
rk, a
400
kV 150
km long th
ree
pha
se tra
n
smissi
on lin
e
model i
s
experim
ented
to obtain
th
e ne
ce
ssary
voltage a
n
d
cu
rrent
waveform
of the
different fa
ult
con
d
ition
of
a radial
po
wer
system
n
e
t
work. In
th
e
prop
osed
alg
o
rithm, o
n
ly t
he
re
ceiving
end
three p
h
a
s
e
voltage wave
form ha
s be
e
n
taken
as th
e only experi
m
ental data.
The si
ngle lin
e
schemati
c
di
agra
m
of the
EMTP simul
a
ted sin
g
le
source, on
e e
nd fed ra
dial
powe
r
sy
stem
netwo
rk is
sh
own i
n
Fig
u
re
1. The volta
g
e
wavefo
rm
s
are m
onito
re
d from th
e re
ceiving
end
o
n
ly
for different types of faults
occurrin
g at
different locati
ons of the tra
n
smi
ssi
on lin
e.
As sh
own in
the Figu
re 1,
the singl
e si
de fed 40
0kV
50Hz po
we
r transmissio
n
system
con
s
i
s
ts of a
n
AC voltage
sou
r
ce as t
he
only po
wer source. Fi
fteen three p
hase Line
Cable
Con
s
tant
s (L
CC) blo
c
ks of
10 km ea
ch
are
con
n
e
c
te
d in casca
de
for structu
r
in
g the si
mulati
on
model of a
1
50km l
ong
overhe
ad tran
smissi
on lin
e [18, 19]. The
freque
ncy d
e
pend
ent ‘JM
a
rti’
model ha
s be
en ado
pted h
e
re a
s
the ba
sic L
C
C bu
ildi
ng blo
ck of th
e power sy
stem netwo
rk [18-
21]. The line
resi
stan
ce i
s
taken to
be 2
0
Ohm/km; the fault re
si
stance is ta
ken
to be 10O
h
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Identification and
Cl
assification
of Powe
r System
Fau
l
ts usin
g Rati
o… (Alo
k Mu
khe
r
jee
)
7605
For 1
0
different types of
faults o
c
curred at di
ffere
nt location
s,
the faulty voltage si
gnal
s
are
colle
cted an
d analyze
d
in ATP analyzer. The sampling fre
q
uen
cy is ke
pt fixed at
2000
sampl
e
s/
cy
cl
e,
i.
e.
f
o
r t
he 50Hz
sinu
so
idal si
gn
al, each
cycle of 20ms i
s
sam
p
led into 200
0
discrete valu
es, yielding a
sampli
ng fre
q
uen
cy of 100
kHz.
Figure 1. Sch
e
matic Di
ag
ram of the Simulated Singl
e Fed Lon
g T
r
an
smi
ssi
on
Line
2.1. Building the Sy
stem
and Clas
sifi
er Algorithm
The p
r
op
ose
d
po
wer sy
stem long
tran
smissio
n
line
fault detecti
on an
d cl
assification
algorith
m
is
d
e
sig
ned
ba
se
d on the
Prin
cipal
Com
p
o
nent di
stan
ce
s of ea
ch
of the three p
h
a
s
e
s
obtaine
d on pro
c
e
ssi
ng th
e different fault volt
age waveform
s co
rrespon
ding t
o
different fault
types. After the prote
c
tion
algorithm is design
ed, it is trained u
s
ing trai
ning
data set. In the
prop
osed work, only one set of receivin
g end thr
ee p
hase voltage
data co
rrespo
nding to he
althy
con
d
ition and
ten different types of fault (e.g
. single-line-to
-groun
d fault for phase A, phase
B
and p
h
a
s
e
C, doubl
e line f
ault for lin
e AB, line BC an
d line
CA, do
uble-li
ne-to
-g
roun
d fault fo
r
line AB, line BC and line
CA and finall
y
three pha
se
symmetri
c
a
l
fault) cond
u
c
ted at 70
km
from
sen
d
ing
end,
i.e. almost th
e midpoi
nt of the tran
smi
s
sion li
ne is t
a
ken
as th
e o
n
ly training
d
a
ta
.
After training
the PCA ba
sed protectio
n
algorith
m
intende
d for fau
l
t detection
a
nd cl
assification,
it is tested
extensively u
s
ing in
dep
en
dent data
se
ts con
s
i
s
ting
of different fault types a
n
d
con
d
ition
s
ca
rrie
d
out at
different dist
ances
from t
he sen
d
ing
end. The fault location
s are
cha
nge
d for
ten differe
nt
types of faul
ts to
inve
stig
ate the effe
cts of the
s
e f
a
ctors
on th
e
perfo
rman
ce
of the propo
sed
prote
c
tio
n
method.
T
he EMTP si
mulation m
o
del of the
ra
dial
system u
s
ed i
n
the prop
ose
d
work is
sho
w
n in Figu
re
2.
Figure 2. EMTP Simulation Model of the Radi
al, Single End Fed,
Long T
r
an
smi
ssi
on Lin
e
for a
Single Line to
Groun
d Fault
in Phase A
2.2. Training of Classi
fier
Algorithm
The co
mpon
ent analysi
s
techni
que ha
s been a
dop
ted here to identify and cla
ssif
y
different type
s of fa
ults. T
he receiving
end th
ree
ph
ase
sample
d
voltage d
a
ta
corre
s
p
ondin
g
to
one healthy condition
an
d ten
different
t
y
pes of
f
aults ca
rri
ed
out a
l
most
at
the midpoint of
t
he
150
km l
ong
transmissio
n
line h
a
s bee
n
take
n a
s
th
e
only
set of t
r
aining
data.
For
ea
ch
of the
waveforms, q
uarte
r cy
cle p
r
e fault a
nd h
a
lf cycl
e
po
st
fault voltage
data ha
s b
e
e
n
take
n for
ea
ch
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 76
03 – 761
2
7606
waveform. Since
sam
p
ling
freque
ncy h
a
s b
een ta
ke
n as
200
0 sa
mples/
cycle,
the total sam
p
le
data for
ea
ch
set of fault
a
nd ea
ch
pha
se co
me
s out t
o
be
1500
sa
mples. T
h
u
s
t
he trai
ning
da
ta
is a set of 11
× 3 vectors (10 ty
pes of faults an
d 1 h
ealthy con
d
ition of the thre
e pha
se volta
g
e
waveform) e
a
ch
con
s
i
s
tin
g
of 1500 d
a
ta. The dat
a thus taken
has be
en u
s
ed fo
r traini
ng
purp
o
se
only. Since
Prin
cipal
Comp
on
ent Analy
s
is ha
s
bee
n a
dopted
he
re
to analy
z
e
the
voltage d
a
ta,
the p
r
inci
pal
comp
one
nts i
.
e. the mo
st
useful
directi
ons of va
riati
on of th
e
sca
rre
d
data a
nd th
e
score
value
s
are
obtain
e
d
as th
e
def
inin
g pa
ram
e
ters. In the
propo
sed
work, o
n
ly
the two m
a
jo
r pri
n
ci
pal
co
mpone
nts
ha
ve been
take
n to buil
d
the
pro
p
o
s
ed
al
gorithm i.e.
o
n
ly
the two m
o
st
prima
r
y correlated di
re
cti
ons
and
the
two corre
s
po
nding
dire
ctio
nal sco
r
e
dat
a.
Thus the
cal
c
ulatio
n has
been rest
rict
ed to two dimensi
onal a
n
a
lysis only which ha
s yiel
ded
satisfa
c
to
ry result
s. Using
the two dire
ctional
score
values, the
distan
ce ve
ct
ors
have b
e
e
n
cal
c
ulate
d
.
This
distan
ce vecto
r
is
actually a
m
eas
ure
of th
e differen
c
e
or di
stan
ce
of the
waveforms of
each type of fault from th
e healthy sig
nal. This dist
ance com
put
ation has b
e
e
n
carrie
d out fo
r ea
ch of the three
pha
se
s
and a tota
l of
3 sets
of 11 d
a
ta i.e. 33 trai
ning data
have
been
comput
ed, ea
ch
correspon
ding
to differe
nt
type of fa
ults
of the th
ree
pha
se
s. Anot
her
importa
nt poi
nt to be noted
here in this
context
is that all the data h
a
ve been
sca
l
ed with re
sp
ec
t
to the pea
k
value of the
voltage sig
n
a
l
at heal
thy condition
so t
hat even if the voltage le
vel
cha
nge
s o
r
i
n
any othe
r
situation cau
s
i
ng a
cha
nge
in the voltage
or
curre
n
t le
vel, a scaled
or
prop
ortio
nal
value of the corre
s
po
ndi
ng fault voltage is al
way
s
used, thu
s
gene
ralizi
n
g
the
algorith
m
to suit any of voltage or
cu
rre
n
t
level.
2.3. Testing
of Clas
sifier
Algorithm
Next, the test data have
been ta
ken
as t
he th
ree
phase volta
ge wavefo
rm
of the
unkno
wn faul
t or the healt
h
y conditio
n
. Thus the te
st data is a
3×1
500 mat
r
i
x
. This unkn
o
wn
data is comp
ared
with th
e
same
he
althy wavefo
rm u
s
ed for t
r
aini
n
g
and
the
sco
r
e mat
r
ix is th
us
obtaine
d for t
he u
n
kno
w
n t
y
pe. The
sco
r
e value
s
are
obtaine
d fro
m
co
mpon
ent a
nalysi
s
a
nd th
e
corre
s
p
ondin
g
di
stan
ces
from the
he
althy pha
se
s are
plotted
for the
three p
hases.
The
distan
ce
s of each of the phases
a
r
e co
mpared with
the ten different training f
ault data and
the
nearest di
sta
n
ce ma
ppin
g
is obtained f
o
r ea
ch. Thu
s
proximity analysi
s
has b
een appli
ed
here
to com
pare th
e expe
riment
al data
with t
he fault
si
gna
tures to find o
u
t the ne
are
s
t
distan
ce
or t
he
maximum si
milarity with a
n
y of the ten
differ
ent types of fault and the pure
(he
a
l
thy) signal.
3. Simulation Resul
t
s
After simulati
ng the po
we
r system tran
smis
sio
n
line
model an
d a
c
qui
ring the
requi
site
data from EM
TP-ATP
simu
lation
softwa
r
e follo
wed
by
ATP d
r
a
w
d
a
ta p
r
o
c
e
ssin
g
, the
pro
p
o
s
ed
prote
c
tion alg
o
rithm is impl
emented in M
A
TLAB
environment. The
fault data thus obtaine
d ha
ve
been
processed
by the
prop
osed
al
gorithm
an
d
detaile
d
a
n
alysis
ha
s b
een ca
rri
ed out
depe
nding o
n
which the te
st data have been catego
ri
zed
into different types of faults or no fa
ult
type.
Figure 3 sho
w
s the sco
r
e
plot of PCA
whi
c
h
have b
een used a
s
the fault sign
ature
s
in
the p
r
op
osed
wo
rk. Only
the two
prima
r
y directio
ns
or ve
ctors i.e
.
the two m
o
st vital p
r
in
ci
pal
comp
one
nts
of the score matrix are ta
ken h
e
re
to e
x
ecute 2
D
co
mputation of
the three
pha
se
system. T
h
e
s
e figu
res
sho
w
the
cl
uste
r
of differe
nt
training
fault d
a
ta at diffe
re
nt co
ordinate
s
in
the p
r
inci
pal
co
mpon
ent
axes pla
n
e
.
Thu
s
com
b
ining
all th
e three
pha
se P
C
scores
corre
s
p
ondin
g
to
ea
ch typ
e
, a
3D plot
have b
een
o
b
tained
an
d
shown in
Fig
u
re 4.
Using
th
ese
th
r
e
e d
i
ffe
r
ent p
h
a
s
e
sc
or
e va
lu
es
an
d ta
k
i
ng
th
e h
ealth
y p
h
a
s
e
a
s
th
e
r
e
fer
e
nce
,
PC
d
i
s
t
an
ce
of each type
of faults h
a
ve bee
n
comp
uted fo
r th
e t
h
ree
ph
ases
and h
a
ve b
e
en tabul
ated
in
Table 1 an
d g
r
aphi
cally rep
r
esented in Fi
gure 5.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Identification and
Cl
assification
of Powe
r System
Fau
l
ts usin
g Rati
o… (Alo
k Mu
khe
r
jee
)
7607
Figure 3. Prin
cipal
Comp
on
ent Analysis
Score
Data of Three
Phase Recei
v
ing End Voltage
for 10 Type
s of Faults and
Healthy Cond
ition
Figure 4. 3D
Plot of Score
Data Obtai
n
e
d
after
Carrying PCA
for 10 Types
of Faults, He
althy
Con
d
ition an
d Test Data for the Th
ree
Phase
s
Table 1. Di
stance Matrix F
o
rme
d
by Analysis
of the
PCA Scores:
Dista
n
ce Matrix of the
Thre
e Phase Voltage Data
Processe
d by PC
A Based
Propo
s
e
d
Algorithm of He
a
l
thy
Con
d
ition an
d Different Ty
pes of Fa
ults
and Te
st Faul
t Data
Fault T
y
pe
Phase A
Phase B
Phase C
Health
y 0
0
0
A-G 18.5972940
8
3.89798694
6
4.07974065
9
B-G 3.94297645
7
15.5982920
1
4.13244083
7
C-G 4.30624724
9
4.30678011
16.3221794
6
AB 16.1635287
7
16.1619939
1.24E-14
BC 2.09E-15
11.6013671
5
12.2705606
2
CA 16.2872345
8
5.58E-15
17.2484232
1
AB-G 18.1715204
17.7926351
5
3.74053233
2
BC-G 3.13405966
5
13.0712815
1
16.9743982
1
CA-G 20.9390186
3
3.22787233
3
15.6427256
8
ABC 20.2100686
6
15.2959240
5
16.3725085
4
Test Data
3.01746091
9
3.02520429
12.1077589
7
Figure 5. PCA Distan
ce Pl
ot for the 11 Traini
ng Data
Set and the Test Data
Thus,
ob
se
rvation of
Figu
re 4
sho
w
s th
at the
un
kno
w
n te
st
data
plot is
clo
s
e
s
t to the
C-
G fault than
any othe
r type, i.e.
the vector
dista
n
ce
of the un
kn
own te
st dat
a is le
ast
wit
h
C-
pha
se-to
-
g
r
o
und fault tha
n
com
pared
to any other
type. When
vector an
al
ysis o
r
simil
a
rity
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 76
03 – 761
2
7608
analysi
s
is carri
ed out wit
h
the sco
re
plot, t
he above con
c
lusi
o
n
is rea
c
h
e
d
mathematically.
Whe
n
data
correspon
ding
to different l
o
catio
n
s
but
of the sa
me f
ault we
re ta
ken an
d comp
ared,
the di
stan
ce
comp
uted
wa
s
comi
ng
out
to be
mini
m
u
m
with the
particula
r typ
e
of fa
ult un
der
test, i.e. in
each ca
se
showi
ng the true re
sult, although the
PC dista
n
ce
s co
unted to
be
some
wh
at different for different fault locations.
4. Analy
s
is o
f
the PCA O
u
tpu
t
and Pr
otec
tion Alg
o
rithm De
sign
The topol
ogy
used h
e
re is the ratio a
nalys
i
s
of the obtaine
d d
i
stan
ce
s. On
close
observation
of the PC di
stan
ce m
a
trix of the
traini
ng data
and
the un
kno
w
n
or te
st data,
a
s
given in
Ta
bl
e 1, it
ca
n b
e
ob
se
rved
that for
ea
ch
of
the
th
ree
pha
se
s,
the
ratio of
the
PC
distan
ce
s i.e
.
the diverg
ences
of the volt
age
waveform
s fro
m
the pu
re
sign
al of the
corre
s
p
ondin
g
pha
se fo
r t
he traini
ng d
a
ta to the un
kno
w
n fa
ult a
t
different di
stance
s
vary i
n
a
definite patte
rn for
ea
ch p
a
r
ticula
r type
o
f
fault, e.
g. sa
y, for sin
g
le li
ne to g
r
o
und
fault for p
h
a
s
e
A, the ratio of the three phase distan
ce vector
of the training di
stance data (h
ere bei
ng 70
km
from the
sen
d
ing en
d) to
the three
p
hase di
stan
ce vector fo
r
all the differe
nt distan
ce
s
are
extremely an
alogo
us to e
a
ch
other
a
nd also vary
in a ce
rtain
pattern a
s
the distan
ce
is
increa
sed
or decre
ase
d
away from th
e trainin
g
p
o
i
nt. On ob
se
rvation of Ta
ble 1, it can
be
observed
tha
t
the no
-fault
or
healthy
condition
is ta
ken
a
s
the
referen
c
e
wh
ere
ea
ch P
C
A
distan
ce valu
e is ta
ken a
s
zero or
refe
ren
c
e a
nd
with respec
t to this
reference value, all the
other
dista
n
ce value
s
h
a
ve be
en d
e
riv
ed. Figu
re
4
sho
w
s the
PCA di
stan
ce
plot of the
differen
t
training
data
along
with th
e test d
a
ta, a
ll with
respe
c
t to the he
althy con
d
ition
whi
c
h i
s
comi
ng
out to be
ze
ro fo
r all the
three
pha
se
s. So, in o
r
d
e
r to
esta
blish the a
bove
observatio
n
in
algorith
m
form, three different ratios
has b
een
ta
ken for all the
twelve co
ndi
tions, i.e. ele
v
en
training d
a
ta as well as
the tes
t
data as
follows
:
a)
Ratio 1 is the
ratio of the PC dista
n
ce
of pha
se A to the PC distan
ce of phase B,
b)
Ratio 2 is the
ratio of the PC dista
n
ce
of pha
se B to the PC distan
ce of phase C,
c)
Ratio 3 is the
ratio of the PC dista
n
ce
of pha
se C to th
e PC distan
ce of phase A,
Thus,
when t
he above three ratio
s
are
compute
d
say for this pa
rticula
r
test d
a
ta, the
results a
r
e
co
mputed a
nd t
abulate
d
in T
able 2. Plotting the value
s
i
n
gra
phi
cal fo
rm, Figu
re 6 i
s
obtaine
d.
Table 2. Rati
o Matrix Formed by the PC Di
stan
ce
s of the Three
Phase Volta
g
e
Data Pro
c
e
s
sed
by PCA Base
d Propo
se
d Algorithm of Healthy Con
d
ition and
Different Types of
Faults an
d Te
st
Fault Data
Fault T
y
pe
Ratio 1
Ratio 2
Ratio 2
Health
y 1
1
1
A-G 4.77099957
9
0.95544968
9
0.21937281
B-G 0.25278257
7
3.77459536
1.04805110
6
C-G 0.99987627
4
0.26386060
3
3.79034888
5
AB 1.00009496
8
1.30E+15
7.70E-16
BC 1.81E-16
0.94546349
7
5.86E+15
CA 2.92E+15
3.24E-16
1.05901484
5
AB-G 1.02129449
9
4.75671203
2
0.20584586
5
BC-G 0.23976682
5
0.77005861
1
5.41610563
3
CA-G 6.48694138
9
0.20634973
7
0.74706107
1
ABC 1.32127150
9
0.93424437
8
0.81011642
3
Test Data
0.99744038
1
0.24985666
6
4.01256529
7
On cl
ose ob
servatio
n of
the ratio m
a
trix
thus fo
rm
ed, the key feature
s
which have
exploited in the desi
gn of the prop
osed
algorit
hm ca
n be observe
d. For instan
ce, here sing
le
line to groun
d
fault for phase C (SLG-C) at 30km
dista
n
ce from the
sen
d
ing en
d has be
en taken
as th
e te
st d
a
ta an
d the
corre
s
p
ondin
g
PCA
dista
n
c
e
data
co
m
e
s
out to
be
3.0175:
3.02
52:
12.107
8 as shown in the final ro
w of Table 1.
This
signifie
s
that A and B phase
s
have be
en
disturbed
by the fault to so
me extent, bu
t the majo
r u
ndulatio
n ha
s occurred i
n
the third lin
e, i.e.
in C ph
ase
which ha
s been di
re
ctly faulted
to grou
nd. Thi
s
is
what can be exp
e
c
ted
theoreti
c
ally.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Identification and
Cl
assification
of Powe
r System
Fau
l
ts usin
g Rati
o… (Alo
k Mu
khe
r
jee
)
7609
Figure 6. Rati
o of PCA Dist
ances of the
Thre
e P
hases
for Different
Types
of Faults
us
ed for
Traini
ng, Test
Data and the
Healthy Co
n
d
ition
Again, it can
be ea
sily
see
n
t
hat the
sa
me fault i.e.
SLG-C
when
taken
du
ring t
he trai
ning
purp
o
se at the midpoint of
the transmission lin
e,
the PC distan
ce
values for th
e three volta
g
e
waveforms correspon
ding
to the th
ree
p
hases we
re
4
.
3062: 4.3
068
: 16.322
2 a
s
sho
w
n
in
ro
w
4.
In this
contex
t, it may be noted that
all
the types
of faults a
nd th
e test fault u
nder test a
r
e
all
comp
ared
wit
h
the
sam
e
h
ealthy si
gnal
to find o
u
t PCA distan
ce
s
o
f
each type
with re
spe
c
t to
a
fixed referen
c
e, for ea
ch
of the three pha
se
s. In
th
is parti
c
ula
r
case, for both
the testing data
and the corresp
ondi
ng true trainin
g
d
a
ta (he
r
e SL
G-C), the three ratio
s
are
coming o
u
t to b
e
arou
nd 1:0.2
5
:4 as is o
b
served from th
e 4
th
and 12
th
row of Tabl
e
2. This is also observed t
hat
both these ra
tios rem
a
in al
most simil
a
r
and mo
re
a
c
curately, withi
n
a pred
efine
d
toleran
c
e le
vel
as the fa
ult locatio
n
is va
ried. Fig
u
re
6 is t
he
gra
p
h
ical
rep
r
e
s
e
n
tation of Ta
ble 2. From t
h
is
figure, it is al
so o
b
served t
hat test data
line is
m
a
tchi
ng almo
st ide
n
tically with t
he SLG
-
C
(C-G
in fig. 6) line,
i.e. both the li
nes have alm
o
st su
perim
p
o
se
d. On th
e contrary, the test data line is
deviating far away fro
m
all the othe
r types of fa
u
l
t as evide
n
t from Figu
re
6. On simil
a
rity
analysi
s
usi
n
g the least square metho
d
betwee
n
t
he test ratio data and all the twelve different
types of traini
ng ratio dat
a, the minim
u
m deviation
i
s
fou
nd o
u
t. In this
pa
rticul
ar exa
m
ple, t
he
test data
devi
a
ted le
ast f
r
o
m
SLG
-
C line
and
mu
ch
hi
gher value
of
sq
ua
red
erro
r
wa
s o
b
taine
d
for all other types.
The d
ouble
li
ne faults are
not sh
own in
t
he figure be
cau
s
e it i
s
o
b
s
erve
d fro
m
7
th
, 8
th
and
9
th
row of Ta
ble 1 that for this type of fault, t
he two fault involving phases a
r
e de
viated the most,
i.e. their PC distan
ce
s a
r
e mu
ch
hig
her
whe
r
ea
s the third h
e
a
lthy pha
se
has
almo
st zero
deviation fro
m
healthy co
ndition
sho
w
i
ng that this
p
a
rticul
ar
pha
s
e ha
s be
en l
e
ast di
sturbe
d by
the parti
cula
r type of fault, e.g. for line
-
to-line
fa
ult be
tween li
ne
s A and B, the P
C
di
stan
ce
s for
A and B
ph
a
s
e
s
a
r
e
16.1
6352
877
and
16.16
199
39
respe
c
tively and
1.24E-14
(which i
s
al
most
equal to ze
ro
) for the healthy C phase. This feature
h
a
s bee
n expl
oited to easily
and explicitly
to
find out line to line type of faults
dire
ctly in the prop
osed algo
rithm.
But for the line-line to g
r
ou
nd
faults (LLG
),
the third
ph
a
s
e P
C
A di
sta
n
ce
is
not
ex
actly coming
out to be
zero, but
somet
h
ing
much
hig
her
than zero sh
owin
g the inv
o
lvement of
grou
nd in
the
fault, thus in
dire
ctly affect
ing
the ap
parentl
y
the un
distu
r
bed thi
r
d
pha
se, thu
s
m
a
ki
ng a
differe
ntiation b
e
twe
e
n
the t
w
o typ
e
s
very clea
r. Hence for the
groun
d faults, the
ratio a
nalysi
s
ha
s been a
d
opte
d
. Here lies
the
novelty of the
pro
p
o
s
ed
scheme i
n
a
ppl
ying the
ratio
analy
s
is
usi
n
g the P
C
di
stan
ce
s to g
o
o
d
effect to find
out espe
ciall
y
grou
nd faul
ts and
th
is di
rect dou
ble
li
ne
fault cla
s
sification.
Hen
c
e,
the propo
sed
algo
rithm is
most
satisfa
c
torily su
cc
e
s
sful in dete
r
mi
ning a
n
d
cla
s
sifying the fa
ult.
It can also b
e
obse
r
ved fro
m
the same fi
gure that
diff
erent fault
s
h
a
ve different
three ph
ase
PC
distan
ce
s. Although this ratio is differing from
one
type of
fault to
the other, but, for any
particula
r type, this ratio o
f
the three phase vo
ltage
PCA distan
ces are rem
a
i
n
ing all withi
n
a
certai
n limit on varying the physi
cal di
st
anc
e
of fault from s
e
nding end.
This de
sign
of
the
p
o
wer system prote
c
tion
schem
e
ha
s yielde
d
a very g
ood
result of
100% ove
r
all
accuracy a
s
evident fr
om
Table 3
Thu
s
the majo
r poi
nt of intere
st i
s
that by setting
a prop
erly ju
dged limit of these three p
hase PCA
di
stan
ce ratio
s
for the 10 different types
of
faults a
nd th
e
healthy
sig
n
a
l the
exact
kind of f
ault
ca
n be
ea
sily d
e
fined. T
he
e
n
tire P
C
A b
a
s
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 76
03 – 761
2
7610
algorith
m
for i
dentificatio
n
and
cla
s
sifica
tion of faul
t
h
a
s
bee
n
sho
w
n i
n
Fig
u
re
7 in th
e form
of a
flowchart.
Table 3. Re
sults Showi
ng
the Fault Cla
ssifie
r
Pe
rfo
r
mance with O
n
ly One Set of Training
Dat
a
Fault
T
y
p
e
Pure A
G
BG CG A
B
BC
C
A
A
B
G
BCG
C
A
G
A
B
C
Pure 13
0 0 0
0
0
0
0
0
0
0
AG
0
13
0
0
0
0
0
0 0 0
0
BG
0 0
13
0
0
0
0
0 0 0
0
CG
0 0
0
13
0
0
0
0 0 0
0
AB
0
0 0 0
13
0
0
0 0 0
0
BC
0
0 0 0
0
13
0
0 0 0
0
CA
0
0 0 0
0
0
13
0 0 0
0
AB
G
0
0 0 0
0
0
0
13
0 0
0
BCG
0
0 0 0
0
0
0
0
13
0 0
CA
G
0
0 0 0
0
0
0
0
0
13
0
AB
C
0
0 0 0
0
0
0
0 0 0
13
O
v
erall
A
c
curac
y
: 100
%
Figure 7. Flowchart of the
Propo
se
d PCA Based Prot
ection Algo
rit
h
m
Analysis
of Table 4
sho
w
s that the
prop
osed alg
o
rithm works very well to find out
accurately ea
ch of the fau
l
ts occu
rrin
g
at di
fferent lo
cation
s fro
m
the se
nding
end. The fa
u
l
t
cla
ssif
i
e
r
a
c
c
u
ra
cy
be
com
e
s 1
00%
wit
h
t
he pro
p
o
s
e
d
algo
rithm, i.e. the pro
p
o
s
ed cl
assifier
can
su
ccessfully i
dentify each f
ault and
very su
ccessfully cla
ssify it a
c
cordin
g to the
different traini
ng
categ
o
ry. It is also ob
se
rve
d
that even when the pr
ote
c
tion sch
e
me
is tested with
pure re
ceivin
g
end th
ree
ph
ase
voltage
signal
s varyin
g the
mag
n
itud
e
of voltag
e, still the
cl
a
ssifie
r
i
s
able
to
identify it to
be a he
althy system. Th
us the
sy
ste
m
is well eq
uippe
d with t
he ca
pability
of
identifying fault from no-fa
ult.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Identification and
Cl
assification
of Powe
r System
Fau
l
ts usin
g Rati
o… (Alo
k Mu
khe
r
jee
)
7611
Table 4. Re
sults Showi
ng
the Fault Cla
ssifie
r
Perfo
r
mance with V
a
rying Fa
ult Locatio
ns
Location of the F
ault (km)
Total Data
Taken
Correct R
e
sults
Wrong Results
% Accurac
y
10 13
13
0
100
20 13
13
0
100
30 13
13
0
100
40 13
13
0
100
50 13
13
0
100
60 13
13
0
100
80 13
13
0
100
90 13
13
0
100
100 13
13
0
100
110 13
13
0
100
120 13
13
0
100
130 13
13
0
100
140 13
13
0
100
5.
Conclusio
n
A novel a
n
d
effective p
o
w
er sy
stem
prote
c
tion
scheme fo
r th
e ide
n
tificati
on a
nd
cla
ssifi
cation
of faults has
been p
r
op
osed here
for a singl
e side fe
d 400
kV, 50 Hz, 15
0 km l
ong
radial
tra
n
sm
issi
on li
ne. F
o
r thi
s
pu
rpo
s
e, Pri
n
ci
pal
comp
one
nt A
nalysi
s
(PCA
) tool
ha
s be
en
use
d
to ide
a
lize, de
sig
n
an
d implem
ent
the pro
p
o
s
ed
prote
c
tion al
gorithm. In o
r
der to p
e
rfo
r
m
the experi
m
e
n
t, a power
system tran
sm
issi
on lin
e mo
del in EMTP
simulatio
n
software
ha
s be
en
desi
gne
d an
d processed
the data i
n
ATP dra
w
to obtain the
re
ceiving
e
nd fault voltage
waveforms.
MATLAB env
ironm
ent ha
s been
ado
pte
d
for the
de
si
gn of the
pro
posed al
gorit
hm.
The de
sign i
s
made in su
ch a way so a
s
to identify any kind of fault
In orde
r to desig
n the pro
t
ection sch
e
m
e, PCA sco
res h
a
ve bee
n cal
c
ulated f
o
r ea
ch
pha
se u
s
in
g
whi
c
h, PC
di
stan
ce m
a
trix has be
en co
nstru
c
ted. Th
ese
P
C
di
sta
n
ce
s
h
a
ve
be
en
made to
un
d
e
rgo
a
novel
and
sim
p
le
ratio a
nalysi
s
algo
rithm
so
develop
ed t
o
identify fau
l
ts
dire
ctly. The
re
sults
so
obtaine
d rev
eal that the
cla
ssifie
r
h
a
s su
ccessfu
lly
detected and
identified ea
ch type of fault and justified
the cla
ssifi
cat
i
on with 10
0
%
accu
ra
cy.
Based
on th
e PC dista
n
c
e
s
and th
e
ratio anal
y
s
is of these values, the
prop
osed
scheme i
s
so
robu
stly desi
gned that, on
ly one se
t of training d
a
ta has b
een u
s
e
d
whi
c
h proved
to be
sufficie
n
t to pro
d
u
c
e
100%
co
rre
ct result for te
n differe
nt faults carried
o
u
t over the
whole
150
km len
g
th, whe
r
ea
s
most of the
related
wo
ks
have u
s
ed m
o
re th
an o
n
e
training
data
set,
and some ha
ve used
seve
ral.
Thus, the p
r
opo
se
d alg
o
rithm re
qui
res mu
ch le
sser exe
c
uti
on time an
d lesser
mathemati
c
al
com
p
lexity due to the p
r
o
c
essing
of
only
a sin
g
le d
a
ta
set, thus allo
wing th
e ci
rcu
i
t
brea
ke
r to operate mo
re
quickly whi
c
h
is one of
the most impe
rative part of any protecti
on
scheme. T
h
is faster o
p
e
r
at
ion, lesse
r
m
e
mory
requi
rement an
d fin
a
lly and mo
st
importa
ntly, the
highly accu
ra
te ratio analy
s
is p
r
o
c
e
ss a
r
e the novel
area
s of this
resea
r
ch wo
rk. Hen
c
e, it can
be stated that the fault
classifi
er has all the possibilities of
developing an
accurate fault
cla
ssifi
cation
scheme th
a
t
may aid the develop
me
nt of reliabl
e
transi
ent-ba
s
ed
prote
c
tio
n
scheme
s
. It is inten
ded to
ca
rry out fu
rther inve
st
igat
ions to fu
rthe
r confirm th
e
robu
stne
ss a
n
d
flexibility of
the cla
ssifie
r
p
e
rform
a
n
c
e u
nder di
ffe
rent
varying co
nd
itions. The p
r
opo
sed
sche
me
is al
so
very
simple a
nd
ea
sy to im
plem
ent in
re
al ti
me o
peration
,
thus
having
all
the aspe
cts
of
being impl
em
ented in pract
i
cal po
we
r sy
stem net
work.
Referen
ces
[1]
Aleksa
ndar
M
Stankov
ic, T
i
mur A
y
di
n. An
al
ysis
of As
ymmetrical F
a
ul
ts in Po
w
e
r S
y
stems
Usin
g
D
y
namic P
has
ors
. IEEE Transactions on Power System
s
. 200
0; 15(3).
[2]
Qais H Alsafasfeh, Ikhl
as
Ab
d
e
l-Qad
e
r, Ahm
ad M
Har
b
. F
a
ult Cl
assific
a
tio
n
a
nd
Loc
aliz
ation
in
Po
w
e
r
S
y
stems Us
ing
F
ault Sign
atur
es and Pr
inci
p
a
l Com
pon
ents
Anal
ysis.
Sci
e
ntific Rese
arch
, Energy an
d
Po
we
r En
g
i
ne
eri
n
g
. 2012; 4: 506-522 http://dx
.
do
i.org/10.4236/ep
e.2012.46064.
[3]
MA Beg, MK
Khedk
ar, SR
Paraskar, GM
Dhol
e. C
l
a
ssifi
cation
of fa
ult
origi
nate
d
tra
n
sients
in
hig
h
voltag
e n
e
t
w
o
r
k usin
g DW
T
–
PCA a
ppr
oa
ch.
Internati
o
n
a
l Jo
urna
l of
Engi
neer
in
g, Scienc
e a
n
d
T
e
chno
logy
. 2
011; 3(1
1
): 1-1
4
.
[4]
M Kezunovic,
et al. Automate
d F
ault Anal
ysi
s
Usi
ng Ne
ural
Net
w
o
r
k. 9th Annu
al
Co
nfer
ence
for
F
a
ult
and D
i
sturba
nc
e Anal
ys
is, T
e
xas A&M Univer
sit
y
, Col
l
e
ge Station, T
e
xas. 1
994.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 76
03 – 761
2
7612
[5]
M Kezunov
ic, I Rikal
o. Detect and Cl
assif
y
F
aults Usi
ng Ne
ural N
e
ts.
IEEE Computer A
pplic
atio
ns
i
n
Power
. 1996; 9
(
4): 42-47. do
i:10.11
09/6
7
.53
984
6.
[6]
A Jain, A T
hoke, R Patel. Faul
t Cl
assific
a
tion
of Dou
b
le
Circuit T
r
ansmission
Li
ne Us
ing Artific
i
a
l
Neur
al Net
w
o
r
k.
Internation
a
l
Journa
l of Ele
c
trical an
d Elec
tronics Eng
i
ne
erin
g
. 200
8; 1(4): 230-2
35.
[7]
M Sana
ye-P
as
and, H Khor
as
had
i-Z
ade
h.
T
r
ans
missi
on L
i
n
e
F
ault Detecti
on & Phase Se
lectio
n Usin
g
ANN.
Internati
ona
l Conf
erenc
e on Po
w
e
r S
y
stems
T
r
ansie
nts IPST
, Ne
w
Orleans. 20
03:
1-6.
[8]
KS S
w
aru
p
, N
Kamara
j, R Ra
jes
w
ar
i. F
ault
Di-ag
nos
is of
Parall
el T
r
ans
mission
Li
nes
Using
W
a
vel
e
t
Based ANFIS
. Internatio
na
l Journ
a
l of Electr
ical a
nd Pow
e
r
Engin
eeri
n
g
. 2
007; 1(4): 4
10-
415.
[9]
V Malathi, N
Marimuthu.
Multi-Cl
ass Sup
port Vector Machi
ne
Ap
proa
ch for F
ault Classificati
on
i
n
Pow
e
r T
r
ans
miss
ion
Li
ne.
IEEE Interna
t
iona
l C
onfere
n
ce o
n
S
u
stai
nab
le E
nerg
y
T
e
chnolo
g
ie
s
ICSET
, Singap
ore. 200
8; 67-7
1
.
[10]
J Upe
n
d
a
r, C
P
Gupta, GK Sing
h.
Discret
e W
a
vel
e
t T
r
a
n
sform an
d G
enetic A
l
g
o
rith
m B
a
sed
F
ault
Classific
a
tion
of Transmission System
s.
N
a
tion
al Po
w
e
r
S
y
stems C
onf
erenc
e (NPSC
-200
8). India
n
Institute of
T
e
chno
log
y
, Bom
b
a
y
. 2
008: 2
23-
228.
[11]
S Vaslilc a
nd
M kezunov
ic. Fuzz
y
ART
Neural N
e
t-
w
o
rk Algorit
hm for Classif
y
in
g the
Po
w
e
r S
y
ste
m
Faults.
IEEE Tr
ansactions on Power Deliv
ery
. 2005; 20(
2): 1306-
131
4.
[12]
I Jollife. Princi
p
a
l Com
pon
ent
Anal
ys
is. Verla
g
, Spring
er. 19
86.
[13]
Jesús Mi
na,
Cristina
Ver
d
e
.
F
ault D
e
tecti
on for
Lar
ge
Scale
S
y
stem
s Usin
g D
y
n
a
m
ic Princ
i
p
a
l
Comp
one
nts Anal
ys
is
w
i
th
Adaptati
on.
I
n
ternati
o
n
a
l J
ourn
a
l of C
o
mp
uters, Co
mmu
n
ic
ations &
Contro
l
. 200
7; II(2): 185-194
[14] B
Rav
i
kumar, D
T
hukaram, HP
Kh
inch
a. A
pplic
atio
n
of su
pport v
e
ctor
m
a
chi
nes f
o
r fa
u
l
t di
agn
osis
i
n
po
w
e
r transmis
s
ion s
y
stem.
IET
Gener. T
r
ansm. Distrib.
20
0
8
; 2(1): 119
–13
0.
[15]
Ma
yon
g
-gu
a
n
g
, Z
hang
ji. F
ault Di
ag
nosis
base
d
o
n
PC
A and
D-S Evi
denc
e T
heor
y
.
978-
1-42
44
-
2487-0/09 2009 IEEE
[16]
Cha
o
ji
e Z
h
a
n
g
,
Guo He, Sh
u
hai
Lia
ng. PCA
-
base
d
An
alo
g
F
ault Detecti
o
n
b
y
Com
b
in
in
g
F
eatures o
f
T
i
me Domain and Spectrum. 9
78-1-
4244-3894-5/09 2009 IEEE
[17]
Leo H
Chi
a
n
g
, Evan L R
u
ssel
l
,
Ric
hard
D Braatz. F
ault di
a
gnos
is in ch
em
ical pr
ocess
e
s usin
g F
i
she
r
discrimi
n
a
n
t a
nal
ysis,
discr
i
m
ina
n
t parti
al
l
east sq
uar
es,
and
pri
n
ci
pal
c
o
mpo
nent
an
al
ysis.
E
l
sev
i
er
Scienc
e, Che
m
ometrics an
d In
telli
ge
nt Lab
o
r
atory Systems
. 2000; 50: 24
3
–25
2
[18]
Stanisl
av Mis
a
k. Mathematic
al m
ode
l
of el
ectric arc
resp
ecting
Ma
yr t
h
eor
y
in EMT
P
-AT
P
.
Acta
Electrotech
n
ic
a et Informatic
a
. 2008; 8(
3): 66–6
9.
[19]
Lászl
ó Prikl
e
r, Hans Kristi
an
Hø
idal
en. AT
PDRAW
,
versio
n 5.6 Users'
M
anu
al. Prel
imin
ar
y
Rel
eas
e.
200
9; 1.0.
[20]
Y Mendez, H
e
rnan
dez, G Drobn
jak, A Cla
udi, M
Kizilc
a
y
. An Eng
i
n
e
e
r
ing Ap
proac
h
in Mode
lin
g
Lig
h
tnin
g Effects on Mega
w
a
tt-class Onshore
W
i
nd T
u
rbin
es Usin
g EMT
P
and Models. PIER
S
Procee
din
g
s, Marrakes
h
, MOROCCO. 201
1; 20(23).
[21]
Watson N, Ar
rilla
ga J. P
o
w
e
r s
y
st
ems el
ectrom
ag
netic
transie
nts si
mulati
on. T
he Institution
of
Electrical E
ngi
neers, Lo
nd
on.
2003.
Evaluation Warning : The document was created with Spire.PDF for Python.