TELKOM
NIKA
, Vol.14, No
.2, June 20
16
, pp. 741~7
4
7
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i1.2750
741
Re
cei
v
ed
Jan
uary 2, 2016;
Re
vised Ma
rch 18, 2016; A
c
cepted Ap
ril 2, 2016
Fault Diagnosis of Power Network Based on GIS
Platform and Bayesian Networks
Yunfang Xie
*
, Yuhong Zhou, Weina
Liu
Coll
eg
e of Mechan
ical a
nd El
ectrical En
gin
e
e
rin
g
, Agricultu
r
al Univ
ersit
y
o
f
Hebei,
Baod
ing
071
00
1, Hebe
i, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: xy
f
2
0
01_
20
0
1
xd
w
@
16
3.co
m
A
b
st
r
a
ct
In order to det
ermine th
e loc
a
tion of the fau
l
t
comp
on
ents of the pow
er n
e
tw
ork quickly
and g
i
ve
troubl
esho
otin
g so
lutio
n
s, thi
s
pa
per
obta
i
n
s
a s
i
mpl
i
fy
structure
of rel
a
y
protec
ti
on
an
d
circuit-br
eak
er
as
key eq
uip
m
ent
by an
aly
z
i
n
g t
he p
o
w
e
r netw
o
rk topo
logy
of
GIS platform
and
uses th
e B
a
yesi
an n
e
tw
orks
fault dia
g
n
o
sis
algor
ith
m
an
d
finally d
e
si
gn
s the
pow
er n
e
tw
ork fault di
agn
osis
mo
dul
e base
d
on G
I
S
platfor
m
. F
a
u
l
t dia
gnos
is a
l
g
o
r
i
thm bas
ed
on
Bayesi
an
netw
o
rks is
a n
e
w
meth
od
for p
o
w
er netw
o
rk fa
ult
dia
gnos
is w
h
ic
h dea
ls w
i
th the pow
er netw
o
rk fault dia
g
n
o
s
is w
i
th inco
mplete a
l
ar
m si
g
nals ca
use
d
by
the
protectio
n
dev
ice
’
s a
nd the
circuit break
er
’
s
malfu
n
cti
on or refus
a
l
to move, d
e
vice fai
l
ure
of
communic
a
tio
n
and oth
e
r rea
s
ons in the
us
e of Bayesi
an
netw
o
rks meth
od. T
h
is meth
od estab
lis
hes
the
transmissio
n
li
ne fa
ult d
i
a
gno
sis
mod
e
l
by
u
s
ing
No
isy-Or, Noisy-An
d
nod
e
mo
del
an
d s
i
mi
lar BP
n
eura
l
netw
o
rk back prop
agati
on a
l
gorith
m
, an
d o
b
tains the
fau
l
t trust degree o
f
each compo
n
ent by usin
g the
formu
l
a, a
nd fi
nally
d
e
termin
e
s the fa
ult a
ccordi
ng to
th
e fault trust
d
egre
e
. T
he
practical
en
gin
e
e
rin
g
app
licati
on s
h
o
w
s that the se
a
r
ch sp
eed
a
nd
accuracy
of
fau
l
t dia
g
n
o
sis
are
i
m
prov
ed
by
a
pplyi
ng
the
faul
t
dia
gnos
is mod
u
le b
a
se
d on GIS pl
atform a
n
d
Bayesia
n
netw
o
rk.
Ke
y
w
ords
: Bayesian Networks, Faul
t Diagn
osis, Pow
e
r Ne
tw
ork
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
At present, the mai
n
po
wer n
e
two
r
k fa
ult locatio
n
i
s
based
on th
e sche
dulin
g
model.
This
mod
e
l la
ck
s t
h
e n
e
ce
ss
ary
simul
a
t
i
on a
nd d
o
e
s
not provide t
he line
loa
d
t
r
an
sfer sch
e
me,
so the
co
ntro
l of the line is bas
ed o
n
th
e natural on/
off powe
r
. Th
e circuit
-
brea
ker
co
ntrol
st
ate
and li
ne
relati
onship i
s
n
o
t
quantified,
an
d the
pro
b
le
m solvin
g i
s
l
a
ck of
releva
nt mathem
atical
model, so tha
t
the efficien
cy of fault diag
nosi
s
al
gorith
m
is n
o
t high,
and the
r
e i
s
no informatio
n
about the structure of the
overhe
ad line
,
and the stru
cture of the
p
o
we
r network is lack of visua
l
image [1]. Base
d on the
GIS platform
whi
c
h can p
r
ovide the to
pology of the
actual lin
e, th
e
power
net
work
di
agn
osi
s
system solve
s
the abov
e proble
m
s. The
GIS platform expresse
s the
logical co
nne
ction of the
tower, the
ci
rcuit
-
brea
ke
r and the ove
r
head lin
e, so it provide
s
a
simulatio
n
en
vironme
n
t. On this
ba
sis
a
nd u
s
ing th
e
map info
rmati
on of road
an
d co
nst
r
u
c
tio
n
,
we can ma
ke
the fault point positionin
g
more a
c
curate and intuitive.
On the othe
r hand
,
in o
r
der to imp
r
o
v
e the accuracy and the
rapidity of the fault
diagn
osi
s
, many native and foreig
n schol
ars have
broug
ht forward expe
rt system, artifi
cial
neural n
e
two
r
k, fu
zzy pet
ri net a
nd
gen
etic al
gor
ith
m
and
so o
n
[
2
-4]. Mo
st
of these m
e
tho
d
s
can
gain
a
satisfying resu
lt for the a
ccurate
and
co
mplete
sign
al
s that a
r
e
se
nd to the
con
t
rol
cente
r
[5-6]. Ho
wever, in t
he a
c
tual p
r
o
c
e
ss
of
fault diagn
osi
s
it n
eed
s un
ce
rta
i
nty reasonin
g
becau
se
of the la
rge
am
ount of
un
ce
rtai
n kno
w
le
dge and dat
a
which
a
r
e
ca
used by
the
prote
c
tion or brea
ke
r
m
a
lfu
n
ction,
reje
cti
on, chan
nel transmi
ssion
in
terfere
n
ce e
r
rors,
protectio
n
action time
deviation an
d other fa
ct
ors. In m
a
n
y
unce
r
tain
rea
s
oni
ng m
e
thod
s, Bayesia
n
netwo
rks m
e
thod should
be taken a
s
the first con
s
ideratio
n be
cause of
its strict proba
bility
theory fou
n
d
a
tion. In con
s
ideratio
n of th
e flexib
le
cau
s
ual
re
asonin
g
and
dia
gno
sis re
asonin
g
of
Bayesian
net
works, it can
be used in rese
archin
g di
agno
si
s of the fault of the powe
r
net
work
unde
r the in
complete al
arming si
gnal
mode
whi
c
h
i
s
cau
s
ed
by the protectio
n
device’
s an
d
the
circuit brea
ke
r’s malfu
n
ctio
n or refusal to mo
ve, and
the device f
a
ilure
of com
m
unication.
So,
the Baye
sian
netwo
rks met
hod i
s
used f
o
r fa
ult dia
g
n
o
si
s of
po
we
r network i
n
th
is p
ape
r. By t
h
is
method
whi
c
h u
s
e
s
Noisy-Or,
Noi
s
y-And no
de m
odel a
nd
si
milar BP n
e
u
ral
network ba
ck
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 741 – 74
7
742
prop
agatio
n algorith
m
, the transmi
ssion
line fault di
agno
sis mo
del
is establi
s
he
d. The fault trust
degree
of e
a
ch
compo
n
ent is obtain
ed by u
s
i
n
g
the formula,
and t
he fau
l
t is det
ermi
ned
according to the fault trust
degree.
This pap
er di
scus
that
the power network
e
s
tabli
s
he
s
the sim
p
lified
topology
wh
ose
key
equipm
ent is
the relay p
r
ot
ection
and
circuit-b
r
e
a
ke
r
on the ba
si
s
of the GIS platform from t
h
e
point of vie
w
of
engi
ne
ering
ap
plica
t
ion. The
n
,
usin
g Baye
si
an n
e
two
r
ks fault dia
g
n
o
si
s
algorith
m
, the po
wer
net
work
can
qui
ckly a
nd a
ccurately dete
r
mine the lo
cation of the f
ault
point, and the
operatio
n ticket of line
sce
ne ca
n autom
atically gen
erated.
2. Po
w
e
r
Ne
tw
o
r
k Faul
t Diagnosis Ba
sed on GIS Platform
B
e
cau
s
e
t
h
e
mult
i po
we
r
su
pply
ci
r
c
u
i
t
has t
h
e
ch
ara
c
t
e
ri
st
ic
s
of
com
p
lex
s
t
ruct
u
r
e,
multi loop a
n
d
difficult con
t
rol, so the
re
sea
r
ch
on fa
ult cha
r
a
c
teri
stics of po
we
r network ba
se
d
on g
r
id
topol
ogy st
ru
cture
[7]. As th
e f
ault info
rmati
on
com
e
s fro
m
the
po
sitio
n
of th
e
circuit-
brea
ke
r, the
con
n
e
c
tion
relation
ship
a
nd the
ele
c
tr
i
c
al
qua
ntity, the st
ru
ctural
fault an
alysi
s
i
s
different fro
m
the
po
we
r g
r
id
analy
s
is of ot
h
e
r ele
c
tri
c
p
o
w
er a
pplication
softwa
r
e.
The
cha
r
a
c
teri
stic analysi
s
(th
eoreti
c
al lo
ss calculat
ion, flow calculatio
n etc.) of the
general po
wer
grid fo
cu
s
on
the
con
necti
on b
e
twe
en t
he
con
n
e
c
tions
of pri
m
ary
devices. T
h
e
Powe
r
network
fault diagn
osi
s
is
an a
naly
s
is
of the lo
cal po
wer
net
work,
which
only analy
z
e
s
the
con
n
e
c
tion
betwe
en the
electri
c
al
equ
ipment an
d th
e po
wer
netwo
rk i
n
the faul
t area. At the same tim
e
, we
must e
s
tabli
s
h a
co
nne
ctio
n bet
wee
n
va
riety of secon
dary e
quip
m
e
n
t (in
c
ludi
ng
relay protectio
n
and a
u
tomati
c devi
c
e
)
an
d prim
ary d
e
v
ices and
a
variety of eq
uipment [8].
Therefore, th
e
resea
r
ch met
hod ba
se
d o
n
the GIS pla
tform ca
n be
cho
s
e
n
to re
alize fault a
n
a
lysis
better.
The
netwo
rk m
a
n
ageme
n
t of the power n
e
twork ba
se
d o
n
GIS platform provide
s
the topology of the
actual lin
e. A scene
simul
a
tion environ
ment ba
s
ed
on GIS platform i
s
provided abo
ut logi
ca
l
con
n
e
c
tion b
e
twee
n the tower,
circuit-b
r
eaker, and ov
erhe
ad line.
GIS platform
has
reali
z
e
d
the integ
r
ated
manag
eme
n
t of power
net
work d
a
ta. When th
e
actual
circuit
faults occu
r, the
detectio
n
of the circuit-brea
ker
state is the ba
sis of the fa
ult
diagn
osi
s
. In
this p
ape
r,
the ci
rcuit-b
r
eaker
state
monitori
ng te
rminal
of 4G
com
m
uni
cati
on
module
is introdu
ced, a
nd t
he mo
dule
se
nds a
comm
a
nd to the
mo
nitoring
cente
r
to id
entify the
status of the
curre
n
t line whe
n
the circuit-brea
ker i
s
disconn
ecte
d. The circuit
-
brea
ker
state is
achi
eving sy
nch
r
on
ou
s a
nd real
-time displ
a
y
on the GIS platform. Und
e
r the premi
s
e,
the
topology of
multi po
wer
sup
p
ly ci
rcuit is
st
udie
d
,
and the
si
m
p
lified topol
o
g
y of the
rel
a
y
prote
c
tion
an
d ci
rcuit-b
r
ea
ker a
s
the
ke
y equipm
ent i
s
e
s
tabli
s
he
d
.
Finally, the l
i
ne fault lo
cat
i
on
is d
e
termi
n
e
d
by
usin
g t
he Baye
sian
netwo
rks faul
t diagn
osi
s
a
l
gorithm.
Fig
u
re
1
sh
ows the
spe
c
ific p
r
og
ram flow ch
art
.
Figure 1. Fau
l
t diagno
sis fl
ow chart of th
e power net
work b
a
sed on
GIS
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Fault Diag
no
sis of Po
wer
Network Based
on GIS Platform
and Baye
sia
n
… (Y
unfang Xie
)
743
3. The Ba
y
esian net
w
o
r
ks
fault diagno
sis algorith
m
3.1. Nois
y
-
Or model
Noi
s
y-O
r
nod
e in Bayesia
n
netwo
rks i
s
a gene
rali
zation of logic "or". The Noisy-O
r
model i
s
simil
a
r to logi
c "or", when all th
e premi
s
e
s
of
Nj are fal
s
e,
the events
re
pre
s
ente
d
by Nj
are
also take
n a
s
false. But the differe
nt from lo
gic
"or" is that if
a premise
of
Nj is tru
e
, it d
oes
not mea
n
tha
t
the Nj
valu
e is true.
Ni t
hat is
any p
r
ereq
uisite
of
Nj
can
be
se
en a
s
having
a
prob
ability qij that is associ
ated with it and ha
s a
blocking effe
ct. The value of Nj is true wh
en
Ni
is the only prerequisite, then the pr
obability that Nj i
s
true i
s
1-qij.
Set cij=1-qij is the conditional
prob
abilitie
s from no
de Ni
to node Nj. T
hen the deg
ree of belief whe
n
Nod
e
Nj is true
ca
n be
cal
c
ulate
d
usi
ng the formul
a 1.
i
i
ij
j
True
N
Bel
c
Ture
N
Bel
))
(
1
(
1
)
(
(1)
Among them
, Nj is the jt
h Noi
s
y-O
r
node in th
e
netwo
rk;
Ni i
s
Nj'
s
the it
h dire
ct
pre
r
eq
uisite,
also
kno
w
n a
s
the parent node; Bel
ind
i
cate
s the de
gree of beli
e
f. The con
c
ep
tual
view of the Noisy-O
r
no
de
is sh
own in Figure 2.
3.2. Nois
y
-
And model
Noi
s
y- And
n
ode in
Baye
sian net
wo
rks is a
ge
nerali
z
ation
of logi
c "an
d
". The
Noi
s
y-
And model is
simila
r to logic "and", whe
n
all the
prem
ise
s
of Nj are
true, the events rep
r
e
s
e
n
te
d
by Nj
are
al
so taken
as tru
e
. But the
different
from
lo
gic "a
nd" i
s
th
at if a p
r
emi
s
e of
Nj i
s
fal
s
e, it
doe
s not me
a
n
that the Nj
value is fal
s
e.
Ni that
is a
n
y
prerequi
site
of Nj can be
see
n
a
s
havi
ng
a prob
ability qij that is associate
d
with it and ha
s
a blocking effe
ct. The value of Nj is false wh
en
Ni is the
onl
y prerequi
site, then the
prob
ability that
Nj is fal
s
e is 1
-
qij. Set cij=1-qij i
s
the
con
d
itional
probabilitie
s fro
m
node
Ni to
node
Nj. Th
e
n
the de
gre
e
of belief whe
n
No
de
Nj is t
r
u
e
can b
e
cal
c
ul
ated usin
g the formula 2. The co
ncept
u
a
l view of the Noisy- And n
ode is sho
w
n
in
Figure 3.
i
i
ij
j
True
N
Bel
c
Ture
N
Bel
)))
(
1
(
1
(
)
(
(2)
Figure 2. Con
c
eptu
a
l view
of Noisy-Or n
ode
A
B
X
q
a
q
b
A
B
I
a
P
(
I
a
=T
u
r
e
)
=q
a
X
OR
AN
D
P
(
I
b
=T
u
r
e
)
=q
b
I
b
OR
Figure 3. Con
c
eptu
a
l view
of Noisy-A
nd
node
3.3. Parameter Learning
Algorithm fo
r Fault Mode
l
Referen
c
e t
o
the
stan
dard
ba
ck
prop
agatio
n
algo
rithm
for trai
ning
multilaye
r
feedforwa
rd
neural n
e
two
r
ks a
nd
usi
n
g the
gr
a
d
ie
nt de
scent m
e
thod, the
m
ean
sq
uare
error
betwe
en the actual value
and the cal
c
u
l
ated value of
the target variable i
s
mini
mized, so tha
t
the
para
m
eter
of the Bayesia
n
netwo
rk is modifi
ed. T
he gradie
n
t algorith
m
formula of Bayesia
n
netwo
rks pa
rameter a
d
ju
stment is shown belo
w
[9]:
A
B
X
q
a
q
b
A
B
I
a
AN
D
P
(
I
a
=F
a
l
s
e
)
=q
a
X
OR
AN
D
P
(
I
b
=F
a
l
s
e
)
=q
b
I
b
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 741 – 74
7
744
)
(
)))
(
1
(
1
(
))
(
1
(
)
(
))
(
1
(
)
(
And
Noisy
T
N
Bel
c
T
N
Bel
Or
Noisy
T
N
Bel
c
T
N
Bel
c
m
i
m
mj
i
j
m
i
m
mj
i
j
ij
(3)
Among the
m
, cij i
s
the
co
n
d
itional p
r
o
b
a
b
ility from the
node
Ni
to n
ode
Nj, an
d it
s value
rang
e is [0,1]
;
η
is the lea
r
ning rate;
δ
j
is the e
rro
r of
node
Nj. For
the output no
de,
δ
j
is defin
ed
as form
ula 4:
)
(
)
(
T
N
Bel
T
N
j
j
j
(4)
Among the
m
,
)
(
T
N
j
is the
true
b
e
lief wh
en
Nj
is tru
e
that i
s
the jth target
variable;
)
(
T
N
Bel
j
is the p
r
edi
cti
v
e value of b
e
lief wh
en Nj
is true th
at is the jth targ
et variable. F
o
r
hidde
n laye
r
node
s, the
error from
the
node
Nk to
th
e pa
rent
nod
e Nj
can
be
calcul
ated
by the
formula 5.
j
l
l
lk
jk
k
j
l
l
lk
jk
k
j
And
Noisy
True
N
Bel
c
c
Or
Noisy
True
N
Bel
c
c
)
(
)))
(
1
(
1
(
)
(
))
(
1
(
(5)
Among them,
δ
k
is the erro
r of node Nk.
In addition to
the Noi
s
y-Or
and
Noi
s
y-An
d nod
es
,
the
netwo
rk
can also co
ntain a
logi
cal
"non" nod
e. Logic "no
n
" no
de's d
e
g
r
ee o
f
belief can b
e
cal
c
ulate
d
according to the formul
a 6:
)
(
1
)
(
True
N
Bel
True
N
Bel
i
j
(6)
Among them,
Nj is a "non"
node, Ni i
s
the only pare
n
t node.
3.4. Diagnos
tic Meth
ods
Usi
ng real ti
me inform
ation of ci
rcuit brea
ke
r, the
topology of t
he sy
stem b
e
fore a
n
d
after fault is identified by the method
of real
-time t
i
e line analy
s
is. The
n
find the differe
nce
betwe
en the t
w
o top
o
logie
s
, that is po
we
r su
ppl
y inte
rrupted regio
n
. The fault co
mpone
nts m
u
st
be in
the
o
u
tage
are
a
.
After dete
r
mi
ning th
e o
u
tage
are
a
,
p
r
otection
an
d
ci
rcuit brea
ker
informatio
n of each comp
o
nent is broug
ht into t
he fault diagnosi
s
m
odel whi
c
h i
s
modified by the
para
m
eter le
arnin
g
. And t
he fault trust
deg
ree
of
e
a
ch
co
mpo
n
ent is i
n
ferre
d
by u
s
ing t
he
formula
1 an
d 2. The
com
pone
nt whe
n
its fault tr
ust
degree i
s
ab
ove 0.7 is a
d
e
termini
s
tic f
ault
comp
one
nt. And wh
en it
s fault tru
s
t
deg
ree
bet
wee
n
0.1
~
0.
7, it is suspicio
us
of faulty
comp
one
nts. And whe
n
its fault trust deg
ree
bel
ow 0.1
,
it is a non fault compo
nen
t.
4. Case an
aly
s
is
Takin
g
the
lin
e L2
fault a
s
an exam
ple i
n
Figu
re
4, th
e Bayesi
an
n
e
twork fa
ult d
i
agno
si
s
algorith
m
is
checke
d. L2 li
ne fault diag
nosi
s
mo
del i
s
shown in F
i
gure
5. In Figure
5
,
the firs
t
letter indi
cate
s the type of prote
c
tion, where th
e F
is
out of orde
r, the M re
pre
s
e
n
ts the pri
m
a
r
y
prote
c
tion, a
nd the P rep
r
ese
n
ts the fi
rst ba
ckup p
r
otection, a
n
d
the
S rep
r
e
s
ents the
se
cond
backu
p p
r
ote
c
tion. And
seco
nd lette
r
B expre
s
s
b
u
s, L
expre
s
s line; thi
r
d l
e
tter P express
prote
c
tion, di
gital expre
ss
circuit breake
r
se
rial
num
b
e
r. In the high voltage po
wer g
r
id, in o
r
de
r
to isol
ate the
fault sou
r
ce, both
sid
e
s o
f
the fault li
n
e
mu
st h
a
ve
protective
a
c
tion
and
ci
rcuit
brea
ke
r. So
L
2
no
de i
s
Noi
s
y-And
no
de.
On
one
side
of the fa
ult lin
e, all
kind
s of
protectio
n
a
r
e
likely to b
r
e
a
k the
corre
s
po
ndin
g
ci
rcuit b
r
ea
ke
r, they are
Noisy-O
r
n
ode
. Unde
r n
o
rmal
circum
stan
ce
s, at the
sa
me time, the
dispatch
ing
end sh
ould
receive
the a
c
tion sig
nal o
f
the
prote
c
tion
an
d the
co
rre
sp
ondin
g
ci
rcuit
bre
a
ker,
so t
he p
r
ote
c
tion
and it
s
co
rre
spo
ndin
g
ci
rcuit
brea
ke
r con
s
i
s
ting of Noi
s
y
-
And no
de.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Fault Diag
no
sis of Po
wer
Network Based
on GIS Platform
and Baye
sia
n
… (Y
unfang Xie
)
745
B1
L1
CB1
C
B
2
B2
L2
CB3
CB
4
B3
CB5
C
B
6
B4
B5
CB7
CB
8
L3
L4
Figure 4. Sample of tran
smissi
on line
Figure 5. The
fault diagno
sis model of transmi
ssion li
ne
After rando
m initializati
on of the condi
tion
al probability cij betwe
en no
des, the
para
m
eters
o
f
the line faul
t model a
r
e t
r
aine
d
an
d
studied
by usi
ng the
sam
p
l
e
as shown i
n
Table
1 a
nd t
he g
r
adi
ent al
gorithm
form
ula 3
~
6. Th
e lea
r
nin
g
out
come
s
(condi
tional p
r
ob
abi
lity
cij) h
a
ve bee
n re
spe
c
tivel
y
labeled in
Figure 5. Re
peat the trai
n
i
ng sa
mple f
o
r this
gro
up
until
the de
sired
o
u
tput is re
ached. Fo
r
det
ermini
stic fau
l
t sampl
e
s,
t
he trai
ning
o
u
tput is bet
ween
0.7~0.9
5
. And for no fault sampl
e
s, the
training o
u
tp
u
t
is betwee
n
0.0~0.1. As
shown in Tabl
e 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 741 – 74
7
746
Table 1. Training re
sult
s a
nd sam
p
le
s o
f
fault diagno
sis mo
del of tran
smi
ssi
on li
ne
Sample
1 2 3 4 5 6 7 8 9
10
11
12
13
14
15
16
17
18
19
20
21
SLP1
0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0
CB1
0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0
FBP2
0 0 0 0 0 0 1 0 1 0 0 0 0 1 0 0 0 0 0 1 0
CB2
0 0 0 0 0 0 1 0 1 0 0 0 0 1 0 0 0 0 0 1 0
PLP3
0 0 0 1 0 0 0 0 0 1 1 0 0 1 1 0 1 1 0 1 0
CB3
0 1 1 1 1 0 0 1 0 1 1 0 0 1 1 1 1 1 0 1 1
MLP3
0 1 1 0 1 0 0 1 0 0 0 0 0 1 0 1 0 0 0 1 1
FBP5
0 0 0 0 0 0 0 1 0 1 0 0 1 0 0 0 0 0 0 0 0
MLP5
0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0
CB5
0 0 0 0 0 0 0 1 0 1 0 0 1 0 0 0 0 0 0 0 0
MLP2
0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 1 0 0
PLP4
0 0 1 0 0 0 0 0 1 0 0 1 1 0 1 0 0 0 1 0 0
CB4
0 1 1 1 0 1 1 0 1 0 0 1 1 0 1 0 0 0 1 0 0
MLP4
0 1 0 1 0 1 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0
SLP6
0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
CB6
0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
FBP7
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 0 0
CB7
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 0 0
MLP7
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0
SLP8
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1
CB8
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1
Expected
output
0
0.9
0
.8
0.8
0.7
0.7
0.8
0
.8
0.7
0
.7
0 0 0 0
0.7
0
.8
0.7
0
0 0
0.7
Training
r
e
sults
0.0
09
0.9
31
0.8
26
0.8
26
0.7
03
0.7
03
0.8
03
0.8
03
0.7
13
0.7
13
0.0
32
0.0
32
0.0
94
0.0
94
0.7
34
0.8
03
0.7
13
0.0
32
0.0
94
0.0
94
0.7
03
5. Conclusio
n
In this
pap
er, by analy
z
i
ng the
po
we
r net
wo
rk da
ta of GIS pl
atform, a
si
mplified
stru
cture of
relay protectio
n
an
d
circuit-brea
ke
r
as key equi
pmen
t obtaine
d by
re
co
nstructi
o
n
and
simplify. On the b
a
si
s of the
sim
p
lified stru
ctu
r
e, the lo
cati
on of the fau
l
t compo
nent
i
s
determi
ned b
y
using the
Bayesian
net
works fault
diagn
osi
s
alg
o
rithm. Finall
y
, the system
automatically generates th
e operati
on ticket of line scen
e, whi
c
h
can g
u
ide th
e staff to quickly
eliminate th
e
fault, sh
orte
n the
po
wer
outage
ti
me
and im
prove
the reli
ability
of po
wer sup
p
ly.
Examples
show that the fault di
agnosi
s
model has
the characterist
ics of strong versatility, fast
rea
s
oni
ng, hi
gh lea
r
nin
g
e
fficiency a
nd
high fault tol
e
ran
c
e. An
d
the fault tru
s
t deg
ree
of e
a
ch
comp
one
nt is obtain
ed
by
usin
g the
formula, an
d
th
e fault i
s
d
e
termin
ed
acco
rding
to th
e f
ault
trust de
gre
e
. Practi
cal en
gi
neeri
ng ap
pli
c
ation
sho
w
s
that the devel
oped fault dia
gno
sis m
odel
s
are corre
c
t and efficient.
Ackn
o
w
l
e
dg
ements
This work
wa
s su
ppo
rted by Baod
ing Scien
c
e
and Te
chn
o
logy Re
se
a
r
ch a
n
d
Develo
pment
Project (11Z
G029, 11
ZN0
15, 14ZG
004,
12ZG0
27)
Referen
ces
[1]
Qi-feng L
o
n
g
, Gang C
h
e
n
, Xi
ao-q
un D
i
n
g
.
New
Method
o
f
Pow
e
r Netw
ork T
opol
ogy A
nalysis
Base
d
on Object-Orie
n
ted T
e
chn
o
l
o
gy
. Proceedings of the CSU-EPSA
.
2005; 17(
1): 73-77.
[2]
CIGRE TF. Practical Use of Ex
pert Sy
stem
s
in Pl
ann
in
g a
n
d
Operati
on
of
Po
w
e
r S
y
stem
s.
ELECTRA
.
199
3; (2): 31-6
8
.
[3]
Jong
epi
er
AG.
Ne
ural
Netw
orks Ap
pli
e
d
to Al
ar
m Proc
essin
g
. Proce
edi
ng
of 3r
d
S
y
mp
osi
u
m o
n
Exp
e
rt S
y
stem
s Applic
ation to
Po
w
e
r S
y
stem
. 1991: 61
5-62
1.
[4]
Wen FS, Chang CS.
Pr
obabilistic ap
proac
h f
o
r fault-section estim
a
tion in power
system
s
bas
ed on
a
refine
d ge
netic
algor
ith
m
. IEE Procee
din
g
s -
Generati
on, T
r
ansmi
ssi
on
an
d Distrib
utio
n, 199
7; 14
4(2):
160-
168.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Fault Diag
no
sis of Po
wer
Network Based
on GIS Platform
and Baye
sia
n
… (Y
unfang Xie
)
747
[5]
Hai
y
i
ng D
o
n
g
, Xi
aon
an
Li. F
ault Di
ag
nosis
for
Substatio
n
w
i
t
h
Re
dun
da
nt
Protection Confi
gurati
o
n
Based
on
T
i
me-Sequ
enc
e F
u
z
z
y
P
e
tri-Net.
T
E
LKOMNIKA
T
e
lec
o
mmun
icat
ion Co
mputin
g Electron
ics
and C
ontrol
. 2
013; 11(
2): 231
-240.
[6]
Mithun M. Bha
skar, S
y
d
u
l
u
Mahes
w
a
ra
pu.
A H
y
br
id Ge
n
e
tic Alg
o
rithm
Appro
a
ch for
Optimal Po
w
e
r
Flo
w
.
T
E
LKOM
NIKA T
e
leco
mmu
n
ic
ation C
o
mp
utin
g Electr
onics a
nd C
ont
rol.
201
1; 9(2): 211-
216.
[7]
Shua
ng Z
h
ao,
Jian-
w
e
n
Re
n
.
Desig
n
a
nd r
ealiz
atio
n of a
fault di
ag
nosti
c s
y
stem for
e
l
ectric p
o
w
e
r
sy
s
t
e
m
.
East China Electric P
o
wer
. 2003; 32
(11): 19-2
1
.
[8]
W
e
i Z
hu, Da
q
i
ng M
u
. T
w
o
Methods
of P
o
w
e
r
Net
w
ork
T
opolog
y A
n
al
ysis.
Jo
urna
l
of Cha
ngs
ha
Univers
i
ty of Electric Pow
e
r (Natura
l
Scienc
e).
2001; 1
6
(2)
:
13-25.
[9]
So
w
m
ya R
a
m
a
cha
ndra
n
. T
h
eor
y Refi
nem
e
n
t of Ba
yesia
n
Net
w
orks
w
i
th Hid
de
n Va
riabl
es. T
he
Univers
i
t
y
of
T
e
xas at Austin;
1998.
Evaluation Warning : The document was created with Spire.PDF for Python.