TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.5, May 2014, pp
. 3662 ~ 36
7
0
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i5.4206
3662
Re
cei
v
ed Au
gust 24, 20
13
; Revi
sed
De
cem
ber 1
6
, 2013; Accepte
d
Jan
uary 4, 2014
An Effective Approach of
AC Signal Detection in DC
Power System Based on Wavelet Neural Network
Zhong
y
u
an Zhang
1
, Wei
Zheng
2
, Chao Ma
2
, Yuchen Zhang*
3
, Shuaibing
Li
4
1
Gansu Electri
c
Po
w
e
r Cor
p
o
r
ation, No. 8 of
Be
ibi
ng He R
o
ad, Che
ng-
gua
n District, 7300
30, Lanz
ho
u,
Chin
a, +
86-09
31-4
956
10
6
2
Gansu Electri
c
Po
w
e
r Res
e
a
r
ch Institute, No. 648 of
Xi-ji
n
East Road, Qi-l
i He District, 73
005
0, Lanz
ho
u
,
Chin
a, +
86-09
31-4
956
10
6
3
Gansu Ruite
Electric Po
w
e
r Scienc
e & T
e
chno
log
y
C
o
., Ltd, District of Lanzho
u Hig
h-tec
h
Deve
lopm
ent
,
730
01
0, Lanz
h
ou, Chi
na, +
8
6
-
093
1-49
56
10
6
4
School of Aut
o
matio
n
& Elec
trical Eng
i
ne
eri
ng,
Lanz
ho
u Ji
ao T
ong Univ
e
r
sit
y
, Anni
ng W
e
st Road,7
3
0
0
70
Lanz
ho
u, Chin
a, +
86-093
1-4
9
561
06
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: z
y
c9
818
@16
3
.com
A
b
st
r
a
ct
T
h
is pa
per d
e
v
elo
p
s an
d ex
peri
m
e
n
tal
l
y d
e
monstrates
a
n
AC sig
n
a
l
d
e
tection
metho
d
in D
C
power system
with a c
o
mbination
of
a novel
detection c
i
rcuit and the wa
velet neural network
m
e
thod. A
i
m
s
at deal
ing w
i
th
the travellin
g w
a
ve and fault
signal ca
nn
ot be detecte
d ac
curately w
i
th the unc
ertainty
of
sign
al v
e
loc
i
ty w
hen AC s
i
gn
al i
n
ject
ed. In
w
h
ich, t
he i
n
j
e
cted AC s
i
gn
al
in D
C
p
o
w
e
r
system
is d
e
te
cted
via the
curre
n
t
transformers
and
the v
o
lt
age tr
ansfor
m
ers distri
bute
d
in
differe
nt current l
o
o
p
s.
T
he
acqu
ired s
i
g
n
a
l
is take
n as
origi
n
a
l
fault si
gna
l w
h
ile t
h
e
sub-b
and
en
e
r
gy functio
n
of
w
a
velet pack
e
t
deco
m
positi
o
n
is used as se
cond
ary c
hara
c
teristic and th
e mi
ni
mu
m d
i
s
t
anc
e is use
d
as the criterio
n
in
W
NN meth
od.
Simulati
ons a
r
e carrie
d
out
to de
mons
tr
ate the correctn
e
ss and v
a
li
dit
y
of the prop
o
s
ed
meth
od
an
d w
h
ich
gives
a
go
od co
nsiste
ncy
w
i
th exper
i
m
e
n
tal test. Res
u
l
t
s show
that th
e AC si
gn
als c
a
n
be detect
ed an
d locate
d in D
C
pow
er system accur
a
tely w
i
th the prop
ose
d
met
hod a
p
p
l
i
ed.
Ke
y
w
ords
: AC
signa
l, w
a
vele
t neural n
e
tw
ork (WNN), DC pow
er system, fault detecti
on
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
In powe
r
sy
stem and
sub
s
t
a
tions of la
rg
e ent
erpri
s
e
s
and plant
s, the DC po
we
r system
is e
s
sential t
o
the op
erati
on, monito
rin
g
and
pr
ote
c
t
i
on of sub
s
tations. T
he no
rmal ope
ratio
n
of
electri
c
al
equ
ipment in
su
b
s
tation o
r
p
o
w
er gri
d
woul
d be
affected
once a
seve
re fault o
c
curred
like
A
C
sig
n
a
l
inje
ction,
DC g
r
ou
ndin
g
etc. [1,
2]. Of all these powe
r
failures, the AC signal
mixed faults own the largest
propo
rtion, which re
sult in
the outage of a whol
e plant and
gene
rato
r tri
pping
out [3
-5]. Thu
s
, re
levant te
ch
ni
cal me
asure
s
are devel
o
ped in
ord
e
r to
prevent the
AC sig
nal
s inject into
DC powe
r
sy
ste
m
and its in
sulatio
n
syst
em, is of a
grea
t
signifi
can
c
e f
o
r the safety and sta
b
ility of DC po
wer
system.
Curre
n
tly, the fault detectio
n
in DC p
o
we
r
system m
a
inly focu
se
d
on DC p
o
we
r supply
like uninte
r
ru
ptible power
sup
p
ly syste
m
(UPS)
an
d
high voltage dire
ct cu
rre
nt (HVDC)
syst
em.
To the forme
r
, there are th
ree different p
r
acti
cal impl
e
m
entation
s
for fault detecti
on:
(1) Th
e bala
n
cin
g
brid
ge
method. In
[6], a balanci
ng brid
ge m
odel is sugg
ested to
detect the i
s
o
l
ation fault in DC p
o
wer
system, in
whi
c
h the voltage
and cu
rrent
variou
s in sh
unt
resi
stan
ce
re
sults in
an u
n
balan
ce of th
e brid
ge,
and
the output si
gnal can b
e
d
e
tected. But this
method cann
ot classify
the fault categori
e
s.
(2) The
un
bal
anced
brid
ge
method
is u
s
ed for g
r
ou
nd
ing fault
dete
c
tion i
n
p
o
siti
ve and
negative el
ectrode
s of
DC
power
syste
m
, the faults
can
be
dete
c
ted when
the
voltage d
r
op
s of
isolatio
n resi
stan
ce
s i
s
eq
ual [7], but al
so
ha
s the
same
sho
r
tag
e
with th
e m
e
thod int
r
od
u
c
e
d
above.
(3) AC
signa
l injection is
use
d
for an ungroun
ded
battery syste
m
, and use
s
an AC
sign
al
sou
r
ce
to impo
se
an
AC voltag
e t
o
groun
d o
n
t
he u
ngroun
d
ed b
a
ttery st
ri
ng. The
fa
ct that
an AC cu
rren
t signal is u
s
ed for dete
c
tion purpo
se
s make
s this a
n
attractive o
p
tion for larg
er
battery sy
ste
m
s, i.e. an
AC current
CT i
s
u
s
e
d
for dete
c
tion
and th
e d
e
t
ection level
is
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Appro
a
ch of AC Signal De
tection in DC
Powe
r Syste
m
… (Zhong
yuan Zha
n
g
)
3663
indep
ende
nt
of the DC b
a
ttery power l
e
vel. T
he
su
ccess of
su
ch a dete
c
tion
method
ha
s not
been verifie
d
to date [8, 9].
Other studies are about
HVDC fa
ult
detection with utilizations
of filtering techniques,
wavelet tran
sformation, ne
ural net
wo
rk
based me
tho
d
and the
co
mbination of
the wavelet a
n
d
neural net
wo
rk
name
d
wavelet neu
ral
netwo
rk (WNN) meth
od.
The traveli
n
g wave
and
the
cha
nge
rate
o
f
voltage o
r
current a
r
e u
s
ed in
HVDC
system
mainly con
c
e
n
trated
in
fault
lo
catio
n
[10-12]. Ge
n
e
rally, there i
s
no effe
ctively met
hod for the detectio
n
of mixed AC si
gnal
s in
DC
power sy
ste
m
. The curre
n
t protectio
n
approa
ch is
to avoid laying the AC and
DC ca
ble
s
si
de by
side, battery
cabl
es
wea
r
on the metal sleeve
s
an
d other p
r
ote
c
tive measu
r
e
s
.
This p
ape
r p
r
ese
n
ts the A
C
si
gnal
dete
c
tion meth
od
and a
novel
acq
u
isitio
n circuit in
DC p
o
wer
system, firstly. And then giv
e
s an
onli
ne
fault detectio
n
app
roa
c
h
with WNN
ap
plied
on the
ba
se
s of
a th
eoretic a
nalysi
s
of r
egul
ar i
s
olatio
n d
e
te
ction
metho
d
and
wavele
t
transfo
rmatio
n com
b
ine
s
with re
gula
r
a
nalog
sign
al acq
u
isitio
n m
e
thod. Finally
, simulation
s
and
experim
ents
are u
s
ed to verify the co
rre
c
tne
ss of the
prop
osed met
hod.
2. Implementation of Sign
al Acquisitio
n
There are th
ree patterns o
f
AC signal in
jects
into
DC
power sy
ste
m
, the first way is AC
power
sou
r
ce
injects
between DC p
o
sit
i
ve el
ectrode
and g
r
ou
nd, the se
co
nd
way is AC po
wer
sou
r
ce
inje
ct
s
betwe
en DC negative
el
ectrode and gr
o
und a
nd the
l
a
st way is A
C
power
so
urce
inject
s bet
we
en the p
o
siti
ve bus
and
negative b
u
s.
In this pa
p
e
r, we u
s
e
AC source
with
maximum am
plitude 22
0V (RMS
) a
s
test signal, t
he
sign
al dete
c
tion structu
r
e
can b
e
seen
in
Figure 1.
Figure 1. Flow Dia
g
ra
m of AC Signal Detection in
DC Powe
r Syst
em
Figure 2. Sch
e
matic Di
ag
ram of AC Signal Dete
ction
in Positive an
d Neg
a
tive Electro
d
e
s
The Fig
u
re
2
pre
s
ent
s two
pattern
s for
AC sig
nal a
c
quisitio
n
, the
first pattern u
s
e
s
R1,
C1, R2, IL1, IL2 and VL2 as pa
rt of AC signal
a
c
qui
sition ci
rcuit.
Once an AC sign
al mixed in
either p
o
sitiv
e
or n
egative
electrode
of DC
po
wer
sy
stem, it woul
d make up
a
circle
circuit
with
R1,
C1,
R2,
C2
and
R3. If the o
u
tput of
IL1 e
qual
s to
IL2 a
nd th
e
voltage d
r
op
i
n
R3 i
s
zero,
we
can poi
nt out that the AC signal is mixed
betw
een p
o
sitive electrod
e and gro
und
. In the secon
d
St
a
r
t
D
a
ta
A
c
q
u
i
s
i
ti
o
n
S
i
gn
al
W
a
v
e
A
n
al
y
s
i
s
D
a
ta
I
d
e
n
ti
f
i
c
a
ti
o
n
A
C
S
i
gn
al
D
e
t
e
c
t
ed
?
S
end
A
l
ar
m
M
e
s
s
age
St
o
p
YE
S
NO
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3662 – 36
70
3664
pattern, R4,
C3, R5, C4, R6, IL3, IL4, VL3 and VL4
are used to make u
p
ano
ther ci
rcl
e
circuit.
The lo
cation
of AC si
gnal
mixed ca
n
be de
cid
ed b
e
twee
n ne
ga
tive electro
d
e
and g
r
ou
nd
or
betwe
en p
o
si
tive bus a
nd
negative b
u
s
by makin
g
an
analysi
s
in v
o
ltage d
r
op
s
of R4 a
nd
R5.
Whe
r
e IL* an
d VL* respe
c
tively stand for the cu
rrent sen
s
o
r
and th
e voltage se
n
s
or.
3. WNN Ba
s
e
d Fault Ana
l
y
s
is
3.1. Wav
e
let Trans
f
orm T
h
eor
y
Wavelet fun
c
t
i
on is
co
nstru
c
ted th
roug
h
a se
rie
s
of b
a
si
c tran
sfo
r
mation with
a
mother
wavelet funct
i
on. Let
()
t
be
a squa
re inte
grabl
e functi
on, that is
2
()
(
)
tL
R
.
If its Fourier
trans
form
()
can satisfy the following
com
patibility condition:
2
|(
)
|
R
d
(1)
Then
()
t
is call
e
d
a basi
c
wa
velet or mother wavel
e
t function. We m
a
ke tra
n
sl
atio
n and scale
for wavelet func
tion, th
e tr
an
s
l
a
t
ion
fa
c
t
or
, and the
scale facto
r
(al
s
o kn
own a
s
t
he expa
nsi
on
fac
t
or)
a
, s
o
that we get func
tion:
12
,
()
0
,
a
t
ta
a
a
R
(2)
As the transl
a
tion facto
r
b
and the
scal
e facto
r
a
a
r
e co
ntinuo
us variable
s
, th
eir valu
e
can
be
po
siti
ve or ne
gative; so
,
()
a
is call
ed contin
uou
s wavelet
fu
n
c
tion (al
s
o
ca
lled
the
mother wavel
e
t
function).
Wavelet t
r
an
sform
calcula
t
es the
inn
e
r pro
d
u
c
t bet
wee
n
the
si
g
nal
()
x
t
with mother
wavelet func
tion:
*
1
(,
)
(
)
(
)
x
t
f
ax
t
d
t
a
a
(3)
Equivalent expre
ssi
on in time domai
n is given as:
*
(,
)
(
)
(
)
2
jt
x
a
f
ab
X
a
e
d
(4)
Whe
r
e
0,
a
R
,
()
X
and
()
are the Fou
r
i
e
r tran
sfo
r
m of
()
x
t
and
()
t
, res
p
ec
tively.
3.2. The The
oretical
Anal
y
s
is of WNN
1
x
2
x
n
x
n
h
1
h
2
h
1
y
m
y
Figure 3. Top
o
logy of Wav
e
let Neu
r
al Network
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Appro
a
ch of AC Signal De
tection in DC
Powe
r Syste
m
… (Zhong
yuan Zha
n
g
)
3665
The WNN i
s
a variety of two techniq
ues
an
d inh
e
rits the adv
antage
s of the neu
ral
netwo
rk an
d
wavelet tran
sformation. T
h
e WNN t
opol
ogy is
ba
sed
on BP n
e
twork; th
e tran
sfer
function
of hidden laye
r n
ode
s is the
m
o
ther
wavele
t
function; a
n
d
the network
sign
al is p
r
io
r to
transmissio
n
while
error i
s
backp
rop
a
g
a
tion in t
he t
r
ainin
g
p
r
o
c
e
ss. T
he n
e
twork topolo
g
y is
sho
w
n i
n
Fig
u
re 1. In Fi
gu
re 1,
12
,
,
...,
n
x
xx
is the in
put vecto
r
;
12
,
,
...
,
l
y
yy
is the pre
d
icte
d
output;
ij
w
and
kj
w
are the
weig
hts conn
ecting eve
r
y layer; and
j
h
is
mother
wavelet func
tion [16].
For the in
pu
t signal
seq
uen
ce
12
(,
,
.
.
.
,
)
n
x
xx
x
, the output of the hidd
en la
yer is
cal
c
ulate
d
as:
1
(
)
,
1
,
2
,
...
,
,
n
ij
i
j
i
j
j
wx
b
hj
h
j
m
a
(5)
Whe
r
e
()
hj
is out
put value fo
r t
he n
ode
j
in t
he hid
den
lay
e
r;
j
h
is
the mot
her
wavelet func
tion;
ij
w
is
weight
co
nne
cting the
input layer a
nd hid
den l
a
yer;
j
b
is
the shift fac
t
or, and
j
a
is
the
s
t
ret
c
h fac
t
or for
j
h
.
The output of
the output layer is calcul
ated as:
1
(
)
(
)
,
1
,
2
,
.
..,
,
m
ik
i
yk
w
h
i
k
l
(6)
W
h
er
e
()
hi
is the
output val
u
e
for
nod
e
i
i
n
the hi
dde
n l
a
yer;
ik
w
is
wei
g
ht co
nne
cting
the hid
den
la
yer a
nd
outp
u
t layer;
l
an
d
m
are
the
numbe
r
of n
ode
s fo
r o
u
tp
ut layer an
d
the
hidde
n layer, respe
c
tively.
For WNN,
the
updatin
g wei
ght
algo
rithm is si
milar to B
P
netwo
rk; the gradie
n
t me
thod i
s
use
d
to updat
e mother
wav
e
let function
para
m
eter
s a
nd co
nne
ctio
n weig
hts bet
wee
n
the layers,
makin
g
the p
r
edi
ction outp
u
t closer a
n
d
closer
to the
desired outp
u
t. The weig
hts of WNN
and
the para
m
ete
r
s of wavelet function a
r
e u
pdated a
s
foll
ows.
(1)
Cal
c
ulatin
g the predi
cti
on error of
WNN.
1
()
(
)
,
m
k
ey
n
k
y
k
(7)
Whe
r
e,
()
y
k
is the predi
cted o
u
tput value,
()
y
nk
is the expe
cte
d
output valu
e for the network.
(2)
Upd
a
ting
the weig
hts o
f
WNN a
nd the param
ete
r
s of wavelet functio
n
acco
rding to
the predi
ction
erro
r
e.
(
1
)
(
)
(
1)
(
1
)
(
)
(
1)
(
1
)
(
)
(
1
)
,,,
,,
,
iii
i
i
i
i
i
i
n
k
nk
nk
k
k
k
k
k
k
ww
w
a
a
a
b
b
b
(8)
Whe
r
e
(1
)
,
i
nk
w
,
(1
)
i
k
a
and
(1
)
i
k
b
are cal
c
ul
ate
d
by the network p
r
e
d
ictio
n
erro
r.
3.3. Fault De
tec
t
ion
w
i
th
WNN
3.3.1. Featur
e Selection a
nd Extra
c
tio
n
To the AC si
gnal with a freque
ncy of 50Hz,
once inj
e
cted into DC power sy
stem, the
voltage
an
d current can be
dete
c
ted
a
s
the
AC si
g
nal
co
ntains ha
rmonic with
di
fferent o
r
de
rs.
As is well kn
own, the WNN is of signifi
cant effe
ct in
extraction of
singula
r
co
mpone
nt. In this
pape
r, we use
the
fu
nda
m
ental comp
on
ent
0
i
of the i
n
j
e
cted
AC si
g
nal
cu
rre
nt a
s
the
o
r
igin
a
l
sign
al for faul
t recog
n
ition.
The wavelet coeffici
ents o
f
sub band
s is
obt
ai
ned
b
y
wavelet pa
cket tran
sform, and
use
s
the fun
c
tion of the
wavelet coeff
i
cient
qu
ad
rat
i
c sum a
s
the quad
rati
c chara
c
te
risti
c
E
of
fault pattern reco
gnition.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3662 – 36
70
3666
22
,,
0
()
(
)
sn
jn
ml
E
fd
k
d
i
(9)
Whe
r
e
m
d
is the
averag
e di
stance of ea
ch
fault cla
ss,
l
a
nd
m
stand fo
r the o
u
tput d
i
mensi
on o
f
WN
N,
s
and
j
stand fo
r th
e dimen
s
io
n
of fault eigen
vector, a
nd t
he p
r
opo
rtion
a
l gain
10
10
k
[15].
3.3.2. WNN Architec
ture
Design
T
h
is p
a
p
e
r
us
es
th
e lo
ose
WNN
mo
de
l as
tra
i
n
i
ng Mo
de
l as
is
s
h
ow
n in
F
i
gu
r
e
5
.
In
whi
c
h, ba
ck
prop
agatio
n (BP) netwo
rk i
s
selecte
d
for neural network, the
excit
a
tion functio
n
of
hidde
n layer
use
s
tan
gent
Sigmoid fun
c
t
i
on, the ex
citation fun
c
tion
of output l
a
yer u
s
e
s
a
line
a
r
function
an
d t
he n
e
two
r
k training f
unctio
n
is ada
ptive learni
ng rate algorith
m
.
Th
e
net
work uses
the wavelet energy functi
on
E
of the bottom sub
-
ban
d ene
rg
y as input after a wav
e
let
decompo
sitio
n
of the origin
al fault signal.
Figure 4. Structure of
WNN for Fault Identification
The no
de
s
of input lay
e
r a
r
e e
q
u
a
l to vector dimen
s
ion
s
of wavelet
packet
decompo
sitio
n
su
b-ban
d e
nergy fu
nctio
n
of the o
r
ig
i
nal fault
sign
al, in this
pap
er, the n
u
mb
er o
f
layers
are o
n
l
y
one as u
s
in
g BP while th
e numb
e
r of
hidde
n neu
ro
ns a
r
e 2*3
+
1
=
7 a
s
there a
r
e
three
pattern
s of A
C
sign
al inje
cts into
DC p
o
we
r
sys
tem. In this paper, as
the fault
c
l
as
s
i
f
i
er
has t
h
re
e pa
tterns, n
a
mel
y
3
c
and traini
ng sampl
e
s
,1
,
2
,
3
j
Tj
X
, the output
T
can be
donate
d
as:
12
3
(0
,
0
)
,
(0
,1
)
,
(
1
,1
)
TT
T
TT
T
Whe
r
e
1
T
stan
d
s
for th
e first
way of AC
si
gnal inj
e
ct
s i
n
to DC po
we
r sy
stem, na
mely AC si
gn
al
inject
s bet
we
en DC
po
sitive elect
r
od
e
and g
r
o
und,
2
T
is the
se
co
n
d
way
of AC sign
al inje
ct
s
betwe
en DC
negative ele
c
trode an
d ground a
nd
3
T
is AC sign
al in
jects b
e
twe
e
n
the positiv
e
bus a
nd ne
ga
tive bus.
3.3.3. AC Signal Detectio
n
The detail p
r
oce
dure for
AC sign
al de
tection is: Th
e training
sa
mple
X
is diffieren
ce
betten the
fu
ndame
n
tal
compon
ent
*
0
N
kz
i
of AC
sig
nal i
n
jecte
d
into
DC ci
rcuit a
n
d the
the
fundame
n
tal comp
one
nt
0
N
kz
i
of the p
a
rasiti
c
AC
signal
inh
e
rente
d
in
DC
circuit; the
sub
-
ba
nd
energy fun
c
tion u
s
e
s
th
e q
uadratic ch
aracteri
stic
E
m
entione
d in
section
A. The
traini
ng
sam
p
l
e
X
is deco
m
po
sed into
3
lay
e
rs by u
s
ing
orthog
onal
wavelet functio
n
, that is:
21
3,
3
2
3
0
2(
2
1
)
n
n
ln
kl
Xd
u
t
z
(10)
Whe
r
e
*
00
X
ii
is the
or ori
g
inal
si
gnal,
32
2
n
u
is the
k
grou
p of wa
velet packet
basi
s
in the 3
layer of th
e
space
2
()
VL
R
of
X
, a
nd
3,
n
l
d
is the
wa
velet ba
se
s coefficient
co
rresp
ondi
ng to
the
k
gro
up of
X
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Appro
a
ch of AC Signal De
tection in DC
Powe
r Syste
m
… (Zhong
yuan Zha
n
g
)
3667
Set the sub-band
ene
rgy
function
E
di
scribe
d in
(9
) as th
e inp
u
t of WNN, th
e
i
no
de
output of the WNN can be
written a
s
:
1
m
io
i
j
i
j
j
y
ww
o
(11)
Whe
r
e
j
o
is the output of output layer.
If
Y
is an n dimention
ou
tput
vecto
r
conri
s
po
ding
to the
ori
g
inal
fault
sign
al
X
, the
i
th
state vari
able
of fault categ
o
ry satisfie
s
ij
tT
T
. Whe
r
e
j
T
stan
ds fo
r the
j
th
way of AC
si
gna
l
injectio
n pattern, T is the space of faul
t pattern o
r
AC signal inj
e
cti
on pattern.
4. Simulations and Expe
riments Verification
4.1. Experiments o
f
WNN
Figure 5. Fun
damental
Co
mpone
nt of AC Signal
Figure 6. The
3 Layer Wavelet Dep
o
sitio
n
Re
sults
As is mentio
ned
above, t
he trai
ning
sample
X
i
s
d
i
ffierence b
e
tten the fu
nd
amental
comp
one
nt of
AC si
gnal i
n
j
e
cted i
n
to
DC circuit
a
nd th
e funda
menta
l
com
pon
ent
of the pa
ra
sitic
AC sign
al in
here
n
ted in
DC
circuit. Figure 5
sho
w
s the a
naly
z
ed
sign
al which i
s
origi
n
ally
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
0
0.
1
0.
2
0.
3
0.
4
T
i
m
e
(
s
ec
onds
)
A
m
pl
i
t
ud
e (A
)
0
0.
5
1
1.
5
2
2.
5
x 1
0
4
0
0.
2
0
.
4
0
2
000
4
000
60
00
800
0
100
00
120
00
-5
0
5
x 1
0
-3
D
e
ta
i
l
co
e
f
f
i
ci
e
n
t
cd
1
0
1
000
2
000
30
00
400
0
500
0
60
00
-0
.
5
0
0.
5
D
e
t
a
i
l
c
oef
f
i
c
i
ent
c
d
2
0
50
0
1
000
1
500
2
000
2500
3000
-0.
5
0
0.
5
T
r
a
i
ni
ng
s
a
mp
l
e
X
D
e
t
a
i
l
c
oef
f
i
c
i
ent
c
d
3
50
0
10
00
1
500
2
000
2
500
3000
0.
5
1
A
p
pr
ox
i
m
at
i
on c
o
ef
f
i
c
i
ent
c
a
3
Ti
m
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3662 – 36
70
3668
colle
cted
by usin
g the fau
l
t detection
circuit p
r
o
p
o
s
ed in Fig
u
re
2 with b
e
twe
en DC p
o
siti
ve
electrode
an
d gro
und. Th
e red li
ne sta
nds fo
r the training
sam
p
l
e
X
, the blue
line sh
ows t
he
fundame
n
tal comp
one
nt of the parasiti
c
AC sig
nal inherented in
DC
circuit an
d the gree
n line
stand
s for th
e fundam
ent
al com
pon
ent
of AC sig
nal
injected i
n
to
DC
circuit. In this pa
pe
r, we
use
200
01
sample d
a
ta for the 3 l
a
ye
r wavel
e
t de
positio
n, and
the de
comp
osition
re
sult
s is
s
h
ow
n
in
F
i
gu
r
e
6
.
In pra
c
tise, with different
AC si
gnal i
n
jectio
n patt
e
rn
s, the
re
sistan
ce to g
r
ound i
s
distin
ct, so, the definition of
T
is
:
12
3
00
1
=
1
00
k
1
M
,
=
1
00
k
,
=
1
M
01
1
gg
g
TR
T
R
T
R
(12
)
Whe
r
e
g
R
stan
ds for the
re
sista
n
ce to g
r
oun
d of di
fferent AC
sig
nal inje
ction
pattern
s a
s
is
refered ab
ove.
Table 1. Fault
Covera
ge Reco
gnition Sa
mple with AC Signal Inject
ed
Fault resistance to ground
P
T
Fault resistance to ground
P’
15
F
R
9
0.8577
10
[0,1]
T
20
F
R
9
0.6005
10
50
F
R
9
0.5623
10
[0,1]
T
70
F
R
9
0.3992
10
10
0
F
R
9
0.3846
10
[0,1]
T
200
F
R
9
0.2157
10
500
F
R
10
0.10
25
10
[0,0]
T
600
F
R
9
0.9522
10
10
k
F
R
10
0.28
33
10
[0,0]
T
15
k
F
R
10
0.20
74
1
0
20
k
F
R
10
0.78
12
10
[0,0]
T
30
k
F
R
10
0.52
37
1
0
75k
F
R
11
0.122
5
1
0
[0,0]
T
70
k
F
R
10
0.9041
10
90
k
F
R
12
0.17
27
10
[1,1]
T
10
0
k
F
R
12
0.1041
10
Table 1 give
s the input trai
ning sample
s
P
, the output sample
s
T
and the te
st sampl
e
s
of WNN
P
, where
F
R
is the fa
ult resi
stan
ce
to groun
d a
nd the inp
u
t training
sam
p
le is the
lowe
st
sub
-
b
and
ene
rgy f
unctio
n
value
.
In this
pap
er, we
sel
e
ct
15
,
50
,
100
an
d
500
as
the low fault resi
stan
ce val
ues, an
d
10
k
,
20
k
,
75k
and
90
k
as the hi
gh fault re
sist
ance
values. F
o
r t
he traini
ng
sample
P
, we
selec
t
150
k
,
300
k
,
700
k
and
1M
as the
high
fault resi
stan
ce value
s
, an
d
20
,
70
,
20
0
and
600
as the low fault resi
stan
ce val
ues.
Figure
7
(
a
)
re
pre
s
ent
s
th
e
t
e
st re
sult
of
WNN, whe
r
e
“O”
an
d “
☆
” i
s
the
o
u
tput
of WNN
with trainni
ng
sample
s.
(a) Fault c
l
ass
i
fic
a
tion tes
t
of WNN
(b) Abs
o
lute
error of Fault
c
l
as
s
i
fer tes
t
Figure 7. WNN Te
st Re
sult
s wi
th Different Types
of Fault
1
2
3
4
5
6
7
8
-0
.
2
0
0.
2
0.
4
0.
6
0.
8
1
1.
2
S
a
m
p
l
e
num
ber
F
aul
t
c
l
as
s
i
f
i
c
a
t
i
on
Y1
Y2
1
2
3
4
5
6
7
8
-0
.
0
1
0
0.
0
1
0.
0
2
0.
0
3
0.
0
4
0.
0
5
0.
0
6
0.
0
7
0.
0
8
0.
0
9
S
a
mp
l
e
n
u
mb
e
r
Te
s
t
e
r
r
o
r
error o
f
Y
1
error o
f
Y
2
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Effective Appro
a
ch of AC Signal De
tection in DC
Powe
r Syste
m
… (Zhong
yuan Zha
n
g
)
3669
The sim
u
latio
n
output
Y
1
and
Y
2
of WNN are:
1
2
0.003
2
0
.000
0
0
.000
2
0
.000
5
0
.001
8
0
.0065
0.000
9
1
.0108
1.080
5
1
.00
5
2
0
.9942
0.0010
0.002
2
0
.0
051
0.0014
1
.
0
117
Y
Y
Y
And the simul
a
tion error
E=
T
-
Y
is:
0.00
32
0.00
00
0.00
02
0.00
05
0.00
18
0.00
65
0
.
0
009
0
.
0
108
0.08
05
0.00
52
0.00
38
0
.
0
010
0.00
22
0.005
1
0
.0
014
0.01
17
E
Whi
c
h is
sho
w
n in Figu
re
7(b
)
, it can be
seen that
|
|
0.08
05
a
e
.
4.2. Circuit Simulation an
d Test
T
o
ve
r
i
fy th
e
c
o
rr
ec
tn
ess
of W
N
N
me
th
od
, a
n
AC
s
i
gn
a
l
in
jec
t
io
n te
s
t
is
c
o
nd
ucte
d
,
th
e
results
can
b
e
se
en in Ta
ble 2. Figu
re
8(a
)
is
the t
e
st ci
rcuit si
mulation result con
ducte
d
in
MATLAB/Simulink, whe
r
e the
AC sign
a
l
is
inje
cted
betwe
en p
o
si
tive bus
and
gro
und. Fi
g
u
re
8(b
)
sh
ows t
he tested A
C
signal
wave
whe
n
inje
cte
d
into DC
po
wer
system
b
y
using a
HIOKI
8861
-50
wav
e
memory recorde
r
.
Table 2. Te
st of AC Signal Injecte
d
into Positive Bus to
Groun
d
AC voltage amplitude injected (R
MS) (V)
10.4
100
Voltage bet
w
een
positive electrod
e
and negative el
ectrode DC
(V)
24.5
24.5
Voltage bet
w
een
positive electrod
e AC (V)
10.4
100.2
Voltage bet
w
een
negative electrode AC (V)
10.4
100.2
Voltage on R4 (V
)
3.4
32
Voltage on R5 (V
)
3.4
31.8
Voltage on R6 (V
)
7.0
66.8
(a) Simulation wave with 100V (RMS
) AC
sign
al injecte
d
(b) A
c
qui
red
wave with 1
0
0
V (RMS
) AC signal
injected
Figure 8. Signal Wave
s of a 24V DC Po
we
r Syste
m
with AC Sign
als Inje
cted
From Fig
u
re 8 we can
se
e that once
a
AC sign
al in
jected into
DC po
wer
syst
em, the
develop
ed A
C
sig
nal d
e
te
ction devi
c
e
can d
e
tect th
e fault sign
al
s a
c
curately,
whi
c
h reali
z
e
s
an
acu
r
rate dete
c
tion of AC fa
ult signal
o
n
ce injecte
d
into DC p
o
wer systems.
5. Conclusio
n
This pa
per u
s
e
s
WNN a
s
an effective appro
a
ch to detect AC signal
s in DC powe
r
system, in whi
c
h, the sub
-
ba
nd en
ergy func
tio
n
of wavele
t decomp
o
sit
i
on is used
as
0
10
20
30
40
50
60
70
80
-2
00
-1
50
-1
00
-50
0
50
10
0
15
0
20
0
Ti
m
e
s
(
s
)
A
m
pl
i
t
ude
(
V
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3662 – 36
70
3670
eigenve
c
tor.
The fault sign
al is accu
ratel
y
detec
ted, cl
assified and f
u
lly re
-de
m
on
strated with
the
premi
s
e of no
additional ha
rdware, and also reali
z
ed
synchro
n
ization and a
c
curacy of AC sig
nal
detectio
n
, wh
ich is of gre
a
t
importan
c
e to the
development of fault detection techn
o
logy in DC
power sy
ste
m
.
Referen
ces
[1]
YANG Xi
u, C
H
EN Ch
en. Impact of HVD
C
F
aul
ts on T
o
rsio
nal Oscil
l
a
tion of T
uebi
ne Gen
e
rator’s
Shafting.
Pow
e
r System T
e
ch
nol
ogy
. 20
04; 28(2
2
): 6-10.
[2]
Jae-D
o
Park,
Jared
Can
d
e
l
a
r
ia. F
ault D
e
tection
an
d Iso
l
atio
n in
Lo
w
-
Voltag
e DC-B
u
s Microgr
i
d
S
y
stem.
IEEE Transactions on Power Deliv
ery
. 2013: 28(
2): 779-7
87.
[3]
Sey
e
d Behzad Na
deri, Mehdi Jafari.
Im
pa
ct o
f
Brid
ge
type
Fa
ul
t C
u
rre
n
t L
i
m
i
te
r o
n
Po
we
r Syste
m
Transie
nt Stab
ility
. 7th
Intern
ation
a
l
Co
nfer
ence
on
El
ectrical
an
d El
ect
r
onics
Eng
i
n
e
e
rin
g
, Bursa,
T
u
rkey
. 2
011:
148-
152.
[4]
S Czapp. T
he Impact of DC Earth F
ault C
u
rrent
Sha
pe
on T
r
ipping
of Resid
ual C
u
rr
ent Devic
e
s.
Electron
ics an
d Electrica
l
En
gin
eeri
n
g
. 20
0
8
; 4(84): 1-4.
[5]
M Shid
diq
Yun
u
s, A Abu-S
i
ad
a.
Imp
a
ct of D
C
-Link
F
ault o
n
the Dy
na
mic
Performanc
e o
f
DF
IG
. IEEE
Confer
ence
on
Po
w
e
r an
d En
erg
y
Soci
et
y
G
ener
al Meeti
n
g
,
San Dieg
o
, Cana
da. 20
12: 1
-
6.
[6]
W
u
Z
hengr
on
g. Discussi
ons
on Bal
anc
ed
brid
ge d
e
tectio
n in the s
ubst
a
tion
app
licati
ons.
Electric
Pow
e
r Science
and Eng
i
n
eeri
n
g
. 200
9; 25(1
0
): 56-58.
[7]
Hua
ng
Ha
iho
n
g
, Quan
C
h
e
n
g
. Dev
e
lo
pme
n
t
of di
stri
bute
d
DC grou
ndi
ng
detectin
g
s
y
ste
m
bas
ed
on
differenti
a
l c
u
r
r
ent
de
te
cti
n
g me
th
od
.
Jour
nal
of El
ectro
n
ic Me
asur
e
m
ent a
nd Instru
me
nt
. 20
09;
23(1
1
): 36-4
1
.
[8]
G Buigues, V Valverde.
S
i
gna
l i
n
jecti
o
n
techn
i
qu
es f
o
r fau
l
t loc
a
ti
on
in
distrib
u
tion
netw
o
rks
.
Internatio
na
l C
onfere
n
ce o
n
Ren
e
w
a
ble E
n
ergi
es an
d Po
w
e
r Qual
it
y
,
Sa
ntiag
o
de C
o
m
postel
a
, Spai
n.
201
2: 1-6.
[9]
Xi
ao
jie
Z
h
a
ng,
Baoh
ua
Mao.
W
a
velet
Neur
a
l
N
e
t
w
ork-b
a
se
d Sh
ort-T
e
rm Passen
ger
F
l
o
w
F
o
rec
a
stin
g
on Ur
ba
n Ra
il
T
r
ansit.
T
E
L
K
OMNIKA Indones
ian
Jour
n
a
l of
El
ectrica
l
Engi
ne
erin
g
. 201
3;
11(
12):
737
9-73
85.
[10]
VR Shee
lav
ant
, Vija
ya C. W
a
velet bas
ed F
a
ul
t Detectio
n M
e
thod for U
ngr
oun
de
d Po
w
e
r
S
y
stem
w
i
t
h
Bala
nced a
nd
Unb
a
la
nce
d
lo
ad.
Internatio
n
a
l Jour
nal of E
l
ectr
ical a
nd C
o
mputer En
gin
eeri
n
g
. 201
1;
1(1): 9-16.
[11]
A Lizk
an
o, J
R
u
iz.
Eval
ua
ti
o
n
o
f
A Ne
w Propo
sa
l fo
r Arci
n
g
Fa
u
l
t De
te
cti
on Me
th
od
Ba
sed o
n
Wa
vel
e
t
Packet Analysis
. Po
w
e
r E
ngi
neer
ing
Soci
ety S
u
mmer M
e
eti
ng, V
anco
u
v
e
r, Can
a
d
a
. 2
011;
3: 13
28-
133
3.
[12]
Michal
ik M, R
ebiz
ant. Hi
gh-
i
m
ped
ance
faul
t detectio
n
in
distrib
u
tion
net
w
o
rks
w
i
t
h
us
e of
w
a
vel
e
t-
base
d
al
gorith
m
.
IEEE
Transactions on Power Deliv
ery
. 20
06; 21(4): 1
793
-180
2.
[13]
Ran
d
y
K Yo
un
g. W
a
velet T
heor
y
a
nd
Its App
licatio
ns. Sprin
ger. 199
3: 222-
223.
[14]
Luo
Ha
ng, W
a
ng
Hou
j
u
n
. Re
search
ab
out ti
me-ser
ia
l pred
i
c
tion bas
ed on
rel
a
xin
g
w
a
ve
l
e
t
ne
ural-
net
.
Co
mp
uter Engi
neer
ing a
nd Ap
plicati
ons
. 2
0
0
8
; 44(8): 16-1
9
.
[15]
Mei W
ang, Yuan-B
i
ng H
ou.
Modeli
ng of
cabl
e fault system
. Proce
e
d
i
ngs of 20
04
Internatio
na
l
Confer
ence
on
Mechin
e Le
arnin
g
an
d C
y
ber
netics, Shan
gh
ai, Chi
na. 20
04
; 3283-3
2
8
6
.
[16]
Li
Xin
w
u,
Gua
n
Pe
ngc
hen
g.
A Nov
e
l A
l
g
o
rith
m of
Net
w
ork
T
r
ade Custom
er Cl
assific
a
tio
n
Bas
ed
o
n
F
ourier Bas
i
s F
unctions.
T
E
LKOMNIKA Indon
esia
n Jour
nal
of El
ectric
al Eng
i
n
eeri
n
g
.
2013; 1
1
(11)
:
643
4-64
40.
Evaluation Warning : The document was created with Spire.PDF for Python.