Internati
o
nal
Journal of P
o
wer Elect
roni
cs an
d
Drive
S
y
ste
m
(I
JPE
D
S)
V
o
l.
5, N
o
. 4
,
A
p
r
il
201
5, p
p
.
56
8
~
57
5
I
S
SN
: 208
8-8
6
9
4
5
68
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJPEDS
Power Quality Improvement in
Distribution System using ANN
Based Shunt Active Power Filter
Jar
upul
a S
o
m
l
al, Ve
nu
Gop
a
la
Rao.Man
n
a
m,
Narsimh
a
Rao.Vutlapal
li
Departem
ent
of
Ele
c
tri
cal
and
E
l
ectron
i
cs
Eng
i
ne
ering,
K L
Univ
ers
i
t
y
,
Guntur,
I
NDIA
Article Info
A
B
STRAC
T
Article histo
r
y:
Received J
u
n
4, 2014
Rev
i
sed
No
v
29
, 20
14
Accepted Dec 20, 2014
This paper focuses on an Artificial Ne
ural Network (ANN)
controller based
Shunt Active P
o
wer Filter (S
APF) for m
itigating
the h
a
rm
onics of the
distribution s
y
s
t
em. To increase th
e performance of the convention
a
l
controller and take advantage of
sm
art contro
llers, a feed
fo
rward-ty
pe
(trained by
a back propagation
algorithm) ANN-based t
echnique is
implemented in shunt
activ
e power filters for
pro
ducing th
e
contr
o
lled pu
lses
required for IGBT inverter. Th
e proposed approach mainly
work on the
principle of
cap
acitor en
erg
y
to
main
tain th
e D
C
link vo
ltage
of a shunt
connected f
ilter
and thus r
e
duces the
tr
ansien
t r
e
sponse time w
h
en th
ere
is
abrupt vari
ation
in the load
. The
entir
e power s
y
stem block set model of the
proposed scheme has been d
e
veloped
in MATLA
B environment.
Simulations
are
carr
i
ed ou
t b
y
us
ing
M
A
TLAB, it
is
no
tic
ed t
h
at th
e %
T
HD is
reduc
ed
t
o
2.
27% from 29.71% by
ANN controlled fi
lter.
The
simulate
d experimental
res
u
lts
als
o
s
how that th
e nov
el co
n
t
rol meth
od is not only
eas
y
to be
com
puted and
i
m
p
lem
e
nted,
but
als
o
v
e
r
y
succes
sful in r
e
ducing
harmonics.
Keyword:
Distribution sy
ste
m
Nue
r
al
net
w
or
k
c
ont
rol
l
e
r
Shu
n
t
active po
wer filter
Tot
a
l
ha
rm
oni
c di
st
o
r
t
i
o
n
Copyright ©
201
5 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Jar
upu
la So
m
l
al,
Depa
rtem
ent of Electrical a
n
d
El
ect
ro
ni
cs E
n
gi
nee
r
i
n
g,
K L Un
iv
ersity,
Gree
n Fi
el
ds, Vad
d
es
waram
,
G
unt
ur
Distric
t
, Andhra Pradesh, INDIA.
Em
a
il: j
a
rup
u
l
aso
m
u
@
k
l
un
iversity.in
1.
INTRODUCTION
In
rece
nt
y
ear
s wi
t
h
t
h
e e
x
p
a
nsi
o
n
of
p
o
w
e
r sem
i
cond
uc
t
o
r t
e
c
h
n
o
l
o
gy
, p
o
w
er el
ect
r
oni
cs
base
d
devices s
u
c
h
a
s
adjusta
b
le-s
peed drive
s
, arc
furnac
e, s
w
i
t
c
hed
-
m
ode p
o
w
er s
u
ppl
y, un
in
terrup
tib
le p
o
wer
supply etc are
em
ployed in
vari
ous
fields [1]
,
[5]
-
[
7]
. So
m
e
of
these
c
o
nve
r
ters not only
inc
r
ease reactive
currents
, but a
l
so produce ha
rm
onics
in
th
e so
urce cu
rren
t
.
Du
e to
th
e
harm
o
n
i
cs, th
ere
m
a
n
y
lo
sses
in
th
e
p
o
wer syste
m
.
To
mitig
ate
th
e h
a
rm
o
n
i
cs
, th
ere are d
i
fferen
t
so
lu
tion
s
are
p
r
op
o
s
ed
and
u
s
ed
b
y
research
ers
in
literatu
re such
as lin
e co
nditio
n
e
rs, p
a
ssive filters, ac
tiv
e filter, etc.,
Firstly, co
nv
en
tion
a
l p
a
ssiv
e
filter are
u
s
ed
for eli
m
i
n
atio
n
o
f
th
e
harm
o
n
i
cs; b
u
t
th
ese p
a
ssi
v
e
filter h
a
v
i
ng
some d
i
sad
v
a
n
t
ag
es; su
ch
as larg
e in
size ,fix
ed
h
a
rm
o
n
i
c co
m
p
ensatio
n
,
weigh
t
an
d
reson
a
n
ce
o
ccurren
ce. The ab
ov
e drawback
s of p
a
ssiv
e filter
can
b
e
ov
ercome b
y
th
e concep
t
o
f
acti
v
e
p
o
wer filte
r app
r
o
a
ch
.
Shun
t-typ
e
activ
e
p
o
wer filter
(SAPF) is
use
d
to elim
inate the c
u
rrent
ha
rm
onics. T
h
e SAPF t
o
p
o
l
ogy
i
s
co
n
n
ect
ed i
n
pa
ral
l
e
l
f
o
r
cu
rre
nt
har
m
oni
c
co
m
p
en
satio
n. Th
e shun
t activ
e power filter has th
e cap
ab
ility to
m
a
in
tain
th
e m
a
in
s
cu
rren
t
b
a
lan
c
ed
an
d
si
nus
oi
dal
aft
e
r com
p
ensat
i
o
n rega
r
d
l
e
ss o
f
whet
her t
h
e
l
o
ad i
s
no
n-l
i
near an
d u
n
b
a
l
a
nced o
r
bal
a
nced
.
Recent technol
ogical de
velopments of
switc
hing de
vices a
nd a
v
ailability
of ine
x
pensive
controlling
de
vices,
e.g., DSP-field-programmable-gate-a
rray-ce
n
tere
d system
,
accom
p
lish a
n
active
powe
r line c
o
nditioner, a
n
a
tural o
p
tion
to
co
m
p
en
sate fo
r
h
a
rm
o
n
i
cs. Th
e con
t
ro
ller is th
e h
eart or p
r
im
ary co
mp
on
en
t o
f
th
e
SAPF
sy
st
em
. C
onve
nt
i
onal
P
I
an
d
PID c
o
nt
rol
l
e
rs are
used t
o
ext
r
act
t
h
e f
u
ndam
e
nt
al
com
ponent
o
f
t
h
e l
o
ad
cu
rren
t thu
s
facilitat
i
n
g
red
u
c
tio
n
of h
a
rm
o
n
i
cs an
d
si
m
u
lt
an
eou
s
ly
con
t
ro
llin
g
d
c
-si
d
e
cap
acito
r v
o
ltag
e
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PED
S
I
S
SN
:
208
8-8
6
9
4
Po
wer Qu
a
lity Imp
r
o
vemen
t
i
n
Distrib
u
tion
S
y
stem
u
s
i
n
g ANN Ba
sed
S
hun
t Active… (Jaru
p
u
l
a
S
o
mla
l
)
56
9
the voltage
source
inverter.
Recently
, diffe
r
ent
AI techni
que
s c
ontrollers
are
use
d
for shunt active
powe
r
filters.
Th
e m
a
j
o
r research
work
s are related
with
co
n
t
ro
l ci
rcu
it d
e
sign
.Th
e
targ
et is to
o
b
t
ai
n
reliab
ility
cont
rol
al
g
o
ri
t
h
m
s
of t
h
e ref
e
rence c
u
r
r
e
n
t
and a
qui
c
k
r
e
sp
onse
pr
oce
d
u
r
e t
o
get
t
h
e cont
r
o
l
si
g
n
a
l
and
si
m
u
ltan
e
o
u
s
ly
qu
ick
con
t
ro
llin
g
d
c
-si
d
e cap
acito
r vo
ltag
e
o
f
th
e
vo
ltag
e
sou
r
ce i
n
v
e
rter. Th
e
Artificial
Neu
r
al
N
e
t
w
or
ks (A
N
N
s) ha
ve been
systematically applied to electrical
en
gi
nee
r
i
n
g [
2
-
3
]
.
T
h
i
s
m
e
tho
d
i
s
considere
d
as a new tool to
design SAPF cont
rol ci
rcuit
s
. The ANN
presents two pri
n
cipal cha
r
acteristics
.It
’
s
not
neces
s
a
ry
t
o
est
a
bl
i
s
h speci
fi
c i
n
pu
t
-
o
u
t
p
ut
rel
a
t
i
onshi
ps b
u
t
t
h
e
y
are form
ul
at
ed t
h
r
o
ug
h a l
earni
ng
p
r
o
cess. Moreo
v
e
r, th
e
p
a
rall
el co
m
p
u
tin
g arch
itect
ure in
creases th
e
system
sp
eed
and
reliab
ility [4
].
In th
is
p
a
p
e
r,
a n
e
w SAPF co
n
t
ro
l m
e
th
o
d
b
a
se
d
o
n
ANNs will
b
e
p
r
esen
ted.
Lo
ad vo
ltag
e
s
and
cur
r
ent
s
a
r
e s
e
nse
d
, t
h
e
co
nt
r
o
l
bl
o
c
ks
c
a
l
c
ul
at
es t
h
e
po
we
r ci
rc
ui
t
cont
rol
si
gnal
s
fr
om
t
h
e ref
e
renc
e
com
p
ensation
currents , a
nd t
h
e powe
r circuit in
j
ects th
e co
m
p
en
satio
n
cu
rren
t to
power syste
m
. Th
e article
is p
r
im
ilary fo
cu
sed
on
a syste
m
wh
ich
u
s
es th
e ANN sy
s
t
e
m
and the
re
sults for t
h
e sa
me are discus
s
e
d.
In
th
is p
a
p
e
r, a
shu
n
t
APF
with
a h
y
steresis
b
a
nd
co
n
t
ro
l is
u
tilized
to
co
m
p
ensate th
e
n
o
n
-
li
n
ear lo
ad
s.
2.
CONFIGURATION OF SHUNT ACTIVE POWE
R FILTER(SAPF) AND
ESTIMATION OF
CO
MPEN
SA
TING C
U
R
R
E
NT
Fig
u
re 1
sho
w
s a sh
un
t active p
o
wer
filter,
it co
n
s
is
ts of th
e 3-ph
ase source,
un
iversal
b
r
i
d
g
e
, lo
ad
alo
n
g
with
activ
e filters. A SAPF is to
p
r
odu
ce th
e co
m
p
en
sation
cu
rren
t
.
Th
e no
n-lin
ear lo
ad
is th
e su
m
o
f
source c
u
rrent
and the
harm
onic curre
n
t
.Th
e
obj
ectiv
e is to
g
e
t th
e b
a
lan
c
ed
supp
ly cu
rren
t with
o
u
t
harm
oni
c a
n
d
react
i
v
e c
o
m
pone
nt
s.T
h
e
sui
t
abl
e
cu
rre
nt
i
s
i
n
ject
ed
by
t
h
e S
A
P
F
c
o
r
r
e
s
po
n
d
i
n
g t
o
t
h
e l
o
ad
cur
r
ent
.
T
h
e S
A
PF i
s
desi
gn
ed wi
t
h
A
N
N
cont
rol
l
e
r. T
h
e
pr
op
ose
d
co
n
t
rol
l
e
r, acc
ou
n
t
s for T
HD a
n
d DC
v
o
ltag
e
co
n
t
rol, th
e con
t
ro
ller
h
a
v
e
r
a
p
i
d dyn
amic r
e
sponse in
case
o
f
lo
ad cu
rr
en
t
d
e
v
i
atio
n
.
Th
e
pr
op
er
o
p
e
ration
of t
h
e co
n
t
ro
ller
resu
lts in
th
e
gen
e
ration
o
f
gate sig
n
a
ls
for 3
-
ph
ase inv
e
rter wh
ich
i
n
tu
rn
is
resp
o
n
si
bl
e
fo
r
ge
nerat
i
n
g c
o
m
p
ensat
i
ng c
u
rre
nt
s. T
h
ese
c
o
m
p
ensat
i
n
g
c
u
r
r
ent
s
o
n
i
n
je
ct
i
on t
h
r
o
u
g
h
t
h
e
3-
p
h
a
se in
v
e
rter
resu
lts in
h
a
rmo
n
i
c co
m
p
en
satio
n
of s
ource cu
rren
ts an
d
i
m
p
r
o
v
e
m
e
n
t
o
f
po
wer qu
ality o
n
the
connected power system
[9].
Fig
u
re 1
.
Con
f
i
g
uration
o
f
Shu
n
t
Activ
e Power
Filter
A
gene
ral
f
o
rm
ul
at
i
on
f
o
r t
h
e
l
o
ad c
u
rre
nt
c
o
rres
p
on
di
n
g
t
o
Fi
gu
re
1 i
s
:
)
(
)
(
)
(
)
(
1
1
t
i
t
i
t
i
t
i
h
L
(1
)
݅
ఈଵ
an
d
݅
ఉଵ
are
th
e in
-p
hase
a
n
d
q
u
a
d
rat
u
re
c
o
m
ponents
o
f
the
p
h
ase
cu
rr
e
n
t at
the
f
u
nda
m
e
ntal
fre
que
ncy
res
p
ectively
.
All
othe
r
harm
on
ics are incl
ud
ed in
݅
. Th
e
per
-
p
h
a
se so
ur
ce vo
ltag
e
an
d th
e
cor
r
es
po
n
d
in
g
in-
phase
com
pone
nt
of
the l
o
ad c
u
r
r
ent m
a
y
be c
o
nvey
e
d a
s
:
t
V
t
v
m
s
cos
)
(
(2
)
t
I
t
i
cos
)
(
1
1
(3)
Ass
u
m
i
ng that
harm
onics ca
n
be elim
inated by
the
APF, t
h
e com
p
ensating c
u
rrent
bec
o
m
e
s:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN:
2
088
-86
94
I
J
PEDS
Vo
l. 5
,
No
. 4
,
Ap
r
il 2
015
:
56
8
–
57
5
57
0
t
I
t
i
t
i
t
i
t
i
L
L
L
cos
)
(
)
(
)
(
)
(
1
1
(4)
Whe
r
e
1
i
is the
p
eak m
a
gnitu
de
of
the i
n
-
pha
se
cu
rre
nt that t
h
e m
a
ins sh
o
u
ld
su
pply
a
n
d the
r
ef
ore
nee
d
s t
o
be asse
ssed Once
1
i
assessm
ent is over, the refere
nce current
for the active powe
r
filter m
a
y easily be fi
xed
as pe
r
(4
).
L
i
m
a
y be m
easure
d
u
s
ing
cu
rre
nt se
nso
r
s.
3.
PROP
OSE
D
CO
NTR
O
L
S
T
RATEGIES
3.
1.
Refere
nce Current Calculation:
For refe
rence current
calcula
tion,
insta
n
tan
e
ou
s abc
_
to
_d
q0 tra
n
s
f
o
r
m
a
tion ha
s bee
n
applied
.
Th
e
abc_t
o
_dq0 t
r
ansform
a
tion bl
ock calcu
lates the d-axis,
q-axis, and zero
sequence
quantities in a two-axi
s
rotatin
g refe
re
nce fram
e
for a 3-
Φ
sinus
oi
dal signal. E
q
u
a
tion (5
), (
6
) and (
7
) are use
d
for re
fere
nce
cur
r
ent
calculation,
))
3
2
sin(
)
3
2
sin(
)
sin(
(
3
2
t
I
t
I
t
I
I
C
b
a
d
(5)
))
3
2
cos(
)
3
2
cos(
)
cos(
(
3
2
t
I
t
I
t
I
I
C
b
a
q
(6)
)
(
3
1
c
b
a
o
I
I
I
I
(7)
Whe
r
e
ω
=
rot
a
tion speed (rad/s)
of the rotating
fram
e
3.
2.
Design
o
f
A
N
N
Co
ntr
o
ller
An
Artificial neural network (ANN), is a
m
ode
l (m
athem
a
tical) inspire
d
by
biol
ogical ne
ural
networks. An
ANN consists
of an interlinked collectio
n
of artificial neurons, an
d it
devel
o
p
s
in
f
o
r
m
ation
using a connec
tionist
m
e
thod
to calcula
tion.
It resem
b
les th
e brain in tw
o facets: 1) The
data is accum
u
lated
by
the
netw
or
k
thr
o
u
g
h
the le
arni
ng
p
r
oces
s
and
,
2)
In
te
rn
eur
o
n co
n
n
ection
stre
ngt
hs ar
e em
ploy
ed to
store
the data. T
h
ese
netw
or
ks a
r
e c
a
tego
rized
by
their to
p
o
logy,
the m
a
nner in
which th
ey c
o
m
m
unicate with their
surroundi
ngs, t
h
e m
a
nner in
which th
ey
are guided, and t
h
ei
r capability to
process
inform
ation. ANNs
are
applied to s
o
l
v
e artificial intelligen
ce problem
s
without
necessa
rily cr
eating a
m
odel of a
real
dynam
ic
syste
m
.
The
rapi
d s
p
otting
of t
h
e
dis
t
ur
bance
sig
n
a
l
with
high ac
curacy, fast
processing
of the refe
re
nce
signal,
an
d
hig
h
dy
nam
i
c respo
n
se
o
f
the
c
ont
roller a
r
e t
h
e prim
e pre
r
eq
uisites f
o
r
desi
red
com
p
ensat
i
on i
n
case o
f
A
PF.
The c
o
nve
nt
ional c
ont
roll
er fails to
ac
hieve satis
factorily under
param
e
ter variations
no
nlinea
rity
load
distur
ba
nce,
an
d s
o
fo
rth
For im
proving the perform
a
nce of
the suggested Shunt
Active Po
wer filter, single layer feed
fo
rwa
r
d net
w
o
r
k
(train
ed
by
t
h
e
back
p
r
o
p
a
g
ation alg
o
r
ith
m
)
is seen. Thi
s
netw
o
r
k
co
ns
ists of t
w
o
lay
e
rs a
n
d
their corres
p
onding ne
uron
interconnections
. ‘2’
neurons i
n
input la
yer to receive the i
n
puts.
Hidden
layer
com
p
rises o
f
21
ne
ur
o
n
s t
o
w
h
ich eac
h
o
f
the
p
r
oces
se
d in
p
u
t is fe
d.
The
o
u
tp
ut lay
e
r com
p
rises
of
‘
1
’
neu
r
on
w
h
o
s
e
out
put is t
o
be
calculated as
P
loss
. Activation functions are assigned
for e
ach
of t
h
e layers i
n
or
der to train t
h
em
. Input lay
e
r is given the
Tan-Si
gm
oidal functio
n as activation f
unct
i
on an
d the o
u
t
put
layer is being given t
h
e Pos-Li
near activ
ation function as act
ivation
function.
Figu
re 2 s
h
ow
s the inter
n
al b
l
ocks
of
pr
o
p
o
s
ed ne
u
r
al net
w
o
r
k
[
10]
. T
h
e large data
of
the DC
-lin
k
v
o
ltag
e
fo
r ‘n
’
an
d ‘n
-1’
in
terv
als fro
m
th
e co
nv
en
tio
n
a
l m
e
thod a
r
e
gathe
r
ed and a
r
e st
ored in t
h
e MATLAB
workspace
. This data is used for trai
ning the
ANN. The
data stored in
workspace is being retrieve
d using the
training algorithm
used. The
neurons
in the
input and out
put layers is
alm
o
st a fixed quantity to obta
i
n the
provide
d
input. The
accuracy
of
the
ANN operation is m
o
stly depe
nd
s on
the num
ber of
hidde
n neurons
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PEDS
I
S
SN:
208
8-8
6
9
4
Power Quality Improvement
i
n
Dist
ribution System
usi
n
g
ANN Based
Shunt Active…
(Jarupul
a
Somlal)
57
1
Fig
u
r
e
2
.
In
tern
al b
l
o
c
ks
o
f
pr
opo
sed n
e
ur
al n
e
twor
k
21
1
.
n
n
n
b
i
w
y
(8)
3.
2.
1.
Algorithm
for ANN
Step 1
:
N
o
rm
alize the inputs
and
out
puts
with respe
c
t to
their
m
a
xim
u
m
values
. It is sh
ow
n that the
neural networks work
better if
the inputs and outputs lie be
tween 0-1. T
h
ere are two inputs
give
n by
{P}
2X
2
0
and o
n
e o
u
tput
{
O
}
1X20
in a norm
a
lized form
.
Step 2
:
E
n
ter t
h
e
num
ber
of
inp
u
ts
fo
r a
fe
d
netw
or
k.
Step 3
:
E
n
ter t
h
e
num
ber
of
lay
e
rs.
Step 4
:
Create a
ne
w feed
forward network with
‘tansig
an
d
posli
n’ t
r
ans
f
er f
u
nction
s
.
Step 5
:
T
r
ain t
h
e
netw
or
k
with a lea
r
ni
ng
rat
e
0.
0
2
.
Step 6
:
E
n
ter t
h
e
num
ber
of
e
poc
hs
.
Step 7
:
Enter t
h
e
goal.
Step 8
:
T
r
ain t
h
e
netw
or
k
fo
r
give
n in
p
u
t a
n
d tar
g
eted
o
u
tp
ut.
Step 9
:
Gene
ra
te sim
u
lation o
f
the
gi
ven
net
w
o
r
k
wit
h
a c
o
m
m
a
nd ‘
g
ensi
m
’
The Ne
ural Network is creat
ed w
ith the se
t num
ber of
neurons in the
each layer usi
ng t
h
e above
algorithm
.
At
each traini
ng session, 500 iterations are
done and
6 such a validation checks are taken
out i
n
or
der
to m
i
nim
i
ze the sco
p
e
o
f
er
r
o
r
occ
u
rrence
. T
h
e m
a
in aim
of this
is
to
bring the
perf
orm
a
nce to ze
ro.
The Learni
ng
rate is the m
a
j
o
r consid
eration in the t
r
aini
ng
of the
Arti
ficial Neural Network (c
ha
nge of
interconnection wei
ghts). It
should
no
t be too low that the training ge
ts too
delay
e
d. It s
h
oul
d not
b
e
excessively because
t
h
e oscillations occu
r about t
h
e target
values and the ti
me
needed t
o
converge is too
high
and
the t
r
ainin
g
gets
delay
e
d.
Fo
r t
h
e c
onsi
d
ere
d
c
o
ntro
lle
r,
Ne
ural
Netw
or
k is t
r
aine
d
at a learni
ng
r
a
te o
f
0.
02
. T
h
e c
o
m
p
en
sator
o
u
tp
u
t
depe
n
d
s
on the input an
d its
evol
ution.
Figure
3. Contr
o
l schem
e
f
o
r
A
N
N
contr
o
ller
Figu
re 3 s
h
o
w
s pr
op
ose
d
co
ntr
o
l schem
e
f
o
r A
N
N
, in w
h
ich the loa
d
cur
r
ents
, PC
C
voltage
s an
d
DC bus voltage of shunt active filter
are sensed. The constant DC voltage
is
m
a
intained by the DC voltage
loop. The input of ANN controlle
r is the difference
betwee
n V
DC
a
nd a
ref
e
rence
value
.
T
h
e o
u
tp
ut o
f
A
NN is
responsible for harm
onic
m
i
tigation.
A phase-locked loop
(PLL) sy
nchr
onizes on t
h
e po
sitive-sequence
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-86
94
I
J
PED
S
Vo
l. 5
,
No
. 4
,
Ap
r
il 2
015
:
56
8
–
57
5
57
2
com
pone
nt
of t
h
e
cu
rre
nt
I. The o
u
t
p
ut
o
f
t
h
e
PLL
(a
n
g
l
e
t
) i
s
use
d
t
o
c
o
m
put
e t
h
e
di
r
ect
-axi
s
a
n
d
qua
d
r
at
ure
-
a
x
i
s
com
pone
nt
s
of t
h
e t
h
ree-
p
h
a
se cu
rre
nt
s [
1
1]
. T
h
e
out
put
si
gnal
s
of
A
N
N
c
ont
r
o
l
l
e
r a
n
d
di
rect
axi
s
com
p
o
n
e
n
t
of c
u
rre
nt
f
r
o
m
d-q-
o t
r
a
n
sf
orm
a
t
i
on are c
o
m
p
ared
whi
c
h
pr
od
uces
di
r
ect
axi
s
com
p
o
n
ent
o
f
referen
ce
sign
al. Th
e si
g
n
a
ls
fro
m
d
-
q-o
frame are ag
ai
n
co
nv
erted
to
a-b-c fram
e
are
co
m
p
ared
with
a
filter
current (I
shabc
),
whi
c
h
res
u
l
t
s
i
n
gen
e
rat
i
o
n o
f
refe
re
nce co
m
p
ensat
i
on cu
rre
nt
, w
h
i
c
h i
s
gi
ven as i
n
p
u
t
t
o
t
h
e
hy
st
eresi
s
c
ont
rol
l
e
r.
Fi
gu
re 4 s
h
o
w
s ope
rat
i
n
g pri
n
ci
pl
e o
f
hy
st
eresi
s
ba
nd c
o
nt
r
o
l
l
e
r i
s
t
o
p
r
o
d
u
ce t
r
i
g
geri
ng si
gnal
s
requ
ired
for
switch
i
ng
ON/
OFF
of IGBT’s
of sh
un
t activ
e filter. Th
e
ob
j
ective of t
h
is con
t
ro
ller is t
o
co
n
t
ro
l
th
e co
m
p
en
satio
n
cu
rren
ts b
y
forcing
it to
fo
llo
w t
h
e re
ference ones.
The switching strategies of t
h
e three
-
p
h
a
se in
v
e
rter
will k
eep
th
e cu
rren
ts in
t
o
th
e h
y
steresis
b
a
nd
. Th
e real lo
ad
curren
t
s are
sen
s
ed
and
th
eir n
on
active com
ponents are com
p
ared
w
ith the
refere
nce com
p
ensation curre
nts. The
hysteresis com
p
arator
out
put
s
si
g
n
al
s
are
use
d
t
o
t
u
r
n
on
t
h
e i
nve
rt
er
po
wer
swi
t
c
hes.
.
Fi
gu
re
4.
O
p
er
at
i
ng
pri
n
ci
pl
e
of
hy
st
eresi
s
b
a
nd
co
nt
r
o
l
l
e
r
4.
RESULTS
AN
D
D
I
SC
US
SI
O
N
S
4.
1.
F
o
r
U
n
co
m
p
en
s
a
t
e
d
S
y
st
e
m
THREE
PHA
S
E
S
O
UR
CE
A
B
C
NON L
I
NE
AR
L
O
A
D
Fi
gu
re
5.
U
n
c
o
m
p
ensat
e
d sy
st
em
Fi
gu
re
5 s
h
o
w
s t
h
e si
m
u
l
a
t
i
on ci
rc
ui
t
f
o
r
3
-
pha
se 3
-
wi
re
d
i
st
ri
but
i
o
n sy
st
em
wi
t
h
a 3
-
p
h
ase
vol
t
a
ge
source c
o
nnect
ed to
non linea
r load. Ta
bl
e 1
sho
w
s t
h
e va
ri
ous
pa
ram
e
t
e
rs of t
h
e c
o
nsi
d
e
r
ed sy
st
em
. Figu
re
6
sho
w
s
Wa
ve f
o
rm
s of so
u
r
ce
cur
r
ent
a
nd l
o
ad cu
rre
nt
o
f
unc
om
pensat
e
d
sy
st
em
. It
can be
o
b
ser
v
e
d
f
r
o
m
Fi
gu
re
6
t
h
at
i
n
st
ead
o
f
t
h
e
act
ual
si
nus
oi
dal
wave
fo
rm
, a
hu
ge
di
st
o
r
t
i
on i
n
t
h
e
s
o
urce
cu
rre
nt
c
a
n
be
obs
er
ved
.
A d
e
l
a
y
can be
o
b
ser
v
e
d
i
n
t
h
e
o
u
t
p
ut
wa
ve
fo
rm
, i
t
i
s
caused
due
t
o
a
n
i
n
d
u
ct
o
r
beca
use a
n
in
du
ctor
op
po
ses th
e sud
d
e
n
ch
ang
e
in
th
e
cu
rr
en
t, t
h
ough
th
e supp
ly wav
e
fo
r
m
ch
ang
e
s in
stan
tan
e
o
u
s
ly it
tak
e
s ti
m
e
fo
r th
e in
du
ctor cau
sing
th
e d
e
lay in
th
e wav
e
form
. Fig
u
r
e 7 sh
ows th
e FFT an
alysis o
f
so
urce
current. F
r
om
the FFT a
n
alysis of th
e output wave
form
without filter show
n in Figure
7, the %THD is about
29
.7
1.
Tabl
e 1. Sy
st
em
param
e
t
e
rs
Syste
m
para
m
e
te
rs
Values
Used
Source I
m
pedance
L
=
0.
01e-
3
m
H
Load
R=10
Ω
,
L
=30e-
3
m
H
Active fi
lter
R=0.1
Ω
,L=3
e
-
3
m
H
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PEDS
I
S
SN:
208
8-8
6
9
4
Power Quality Improvement
i
n
Dist
ribution System
usi
n
g
ANN Based
Shunt Active…
(Jarupul
a
Somlal)
57
3
Figu
re
6.
Wa
v
e
f
o
rm
s of l
o
ad
cu
rre
nt an
d so
urce
cu
rre
nt of unc
om
pensate
d
sy
stem
Figure
7. FFT
analysis of s
o
urce c
u
rrent
4.
2.
For
Shun
t
Active Filter with
ANN Con
t
rol
l
er
Figure
8 shows si
m
u
lation circuit of Shunt
Active
Filter.
Figure
9 shows the Sim
u
lati
on
results of
Shunt
Active
Filter with ANN Controller.
From
Figure
9, it can be
observed that
after Shunt Active Filter
with ANN C
ont
roller runs, it reduces the
m
u
ch dela
y
and wa
vef
o
r
m
appears sin
u
soi
d
ally
with fewer
distortions
when com
p
ared t
o
uncom
p
ensat
e
d system
and it also observed
that the ha
rm
onics of the source
current are elim
inated by
injecting the capa
c
itor curre
n
t which ha
pp
ens
because of m
a
in
taining the capacitor
voltage
near t
o
constant. Ca
pacitor
voltage
takes 0.08s
ec
to reach the
steady stat
e. Figure 10 shows FFT
analysis of s
o
urce curre
n
t with ANN c
ont
rol
l
er. From
Fi
gure 10, it can be
seen that
the current total harm
onic
distortio
n re
du
ces
to 2.
2
7
% fr
om
29.
71
%.
Figure
8. Sim
u
lation circuit of
Shunt
hybrid
Active Filter
0
0.
05
0.
1
0.
15
0.
2
0.
25
0.
3
-4
0
-3
0
-2
0
-1
0
0
10
20
30
40
c
u
rre
n
t
(
A
)
s
o
u
r
ce
cu
r
r
e
n
t
0
0.
05
0.
1
0.
1
5
0.
2
0.
2
5
0.
3
-40
-30
-20
-10
0
10
20
30
40
Lo
a
d
c
u
r
r
e
n
t
Ti
m
e
i
n
s
e
c
s
cu
r
r
e
n
t
(
A
)
0
10
0
20
0
30
0
40
0
50
0
60
0
70
0
80
0
900
10
00
0
0.
5
1
1.
5
2
2.
5
F
r
eq
ue
nc
y
(
H
z
)
F
u
nda
m
e
n
t
al (
6
0H
z
)
=
1
6
.
3
2
,
T
H
D
=
29
.
7
1%
M
a
g (
%
of F
u
ndam
ental
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN:
2
088
-86
94
I
J
PEDS
Vo
l. 5
,
No
. 4
,
Ap
r
il 2
015
:
56
8
–
57
5
57
4
Figure
9. Sim
u
lation results
of Sh
unt Active Filter
w
ith ANN
Controller
Figu
re
1
0
.
FFT
analy
s
is o
f
s
o
urce
cu
rre
nt
with A
N
N
c
ont
ro
ller
Table
2. C
o
m
p
ariso
n
of
sim
u
lated res
u
lts
SIMULATED RESULTS
APF
Unco
m
p
ensated S
y
ste
m
ANN
Settling Ti
m
e
(V
DC
)
in Sec.
-
-
0.
08sec
%T
HD 29.
71
2.
27
5.
CO
NCL
USI
O
N
In this paper, a detailed analysis of Shunt
Active Power Filter with ANN controller has been
pr
o
pose
d
t
o
m
i
tigate harm
oni
cs o
f
the
th
ree
pha
se
distrib
u
tion
sy
stem
.
The obtained results
show the si
m
p
lici
t
y an
d the effectivene
ss of the proposed
intelligent cont
roller
under nonlinear load conditio
ns. From
the results, it can be observed th
at t
h
e current tota
l harm
onic distortion
reduces better
with ANN controlled activ
e filter. The si
m
u
lation and experim
e
ntal
results also show that the
new control method is not
only easy to be
calculated and im
ple
m
ente
d, but also
very
effective i
n
re
duci
n
g
harm
onics
ACKNOWLE
DGE
M
ENTS
Following authors are hi
ghl
y supported and encouraged
by th
e followi
ng i
n
stitutio
n by
providi
n
g
sufficient tim
e
and resources.
REFERE
NC
ES
[1]
H Do
ğ
an, R A
kka
ya.
A Simple
Control Scheme
for
Single-Phase Shunt Active
Po
wer F
ilter with
Fuzzy Logic Based
DC Bus Voltag
e
Controller
. P
r
oceed
ings of t
h
e Intern
ation
a
l
Multi Confer
e
n
ce of Eng
i
nee
r
s and Com
puter
Scientists. IMECS. Hong Kong. 2009; II
0
0.
0
5
0.
1
0.
15
0.
2
0.
2
5
0.
3
-4
0
-2
0
0
20
40
CU
RR
E
N
T
(
A
)
SO
URC
E
C
U
RRE
N
T
0
0.
0
5
0.
1
0.
15
0.
2
0.
2
5
0.
3
-4
0
-2
0
0
20
40
LOAD
C
U
R
R
E
N
T
CURREN
T
(
A
)
0
0.
0
5
0.
1
0.
15
0.
2
0.
2
5
0.
3
-4
0
-2
0
0
20
40
F
I
L
T
E
R
CURRE
N
T
CU
R
R
E
N
T
(
A
)
0
0.
0
5
0.
1
0.
15
0.
2
0.
2
5
0.
3
0
20
0
40
0
CA
P
I
CA
T
O
R
V
O
L
T
A
G
E
TIM
E
IN
S
E
C
S
VO
LTA
G
E
(
V)
0
10
0
20
0
30
0
40
0
500
60
0
700
80
0
900
1
000
0
0.
5
1
1.
5
2
2.
5
F
r
equ
enc
y
(
H
z
)
F
u
n
d
a
m
e
nt
al
(
6
0H
z
)
=
3
9
.
9
9
,
T
HD=
2.
27
%
M
ag (
%
of
F
undam
ental
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PEDS
I
S
SN:
208
8-8
6
9
4
Power Quality Improvement
i
n
Dist
ribution System
usi
n
g
ANN Based
Shunt Active…
(Jarupul
a
Somlal)
57
5
[2]
Boren Ren
lin
,
Richard
G Hoft.
P
o
wer el
ect
ron
i
cs inver
t
er with
neural networks
. IEEE Techno
lo
gy update series
.
1997; 12(2).
[3]
YM Chen, RM
O’Connell. Act
i
v
e power line c
onditioner with
a neural network control.
IEEE Tran
s. Ind. Appl.,
1997; 33(4): 113
1–1136.
[4]
J
R
Vázquez, P
a
tri
c
io S
a
lm
ero
n
, F
J
a
vieralc
a
n
tra,
Jain
e Prie
to. A new act
ive power line
conditioner fo
r
com
p
ensation
in
un unbalan
ced
/
d
istorted
el
ec
trical power
s
y
s
t
em
. 14th PSCC, Sev
illa. 2002
.
[5]
N P
echaran
in.
An applica
tion o
f
neural network to
harmonic de
te
ction
in a
c
ti
ve
fi
lter
. Proc. WCC
I- ICNN. 1994;
6:
3756–3756.
[6]
N Pechanr
a
nin,
H Uitsui, M
Sone.
Harmonic d
e
tection
by usin
g neural n
e
twor
k
. Proc.
IEEE I
n
t. Conf
.
Neur
al
Network., 1995,
vol. 2
,
pp
. 923–9
26.
[7]
D Gao,
Xiaoruisun.
A new met
hod to generate current
referen
ce for acti
ve p
o
wer filt
er
. P
r
oc. IEE
E
CIEP
,
Alcapulco, Mexico. 2000
: 99–10
3.
[8]
Avik Bhattach
ar
y
a
, Chandan Ch
akrabort
y
. A Shunt Activ
e Power Filter With Enhanced Perform
ance Using ANN-
Based Predi
c
tiv
e
and Ad
aptiv
e C
ontrolle
rs.
I
EEE
Transactions On Industrial
Electronics.
2011; 58
(2).
[9]
M
r
iul J
h
a,
S
a
t
y
a
P
r
akas
h Dube
y.
Neuro-Fuzzy ba
sed controller
fo
r a Shunt
Active Power
Filter
.
I
EEE
Confer
ence
,
2011.
[10]
M Farahani. Damping of sub s
y
nchronous
oscillations in
power
s
y
stem using st
atic s
y
nch
r
onous
series compensator.
Published in
IET
Gener
a
tion
,
Tra
n
sm
ission & Distribution
.
BIOGRAP
HI
ES OF
AUTH
ORS
Jar
upula Somla
l
,
at pr
es
ent
is
working as
an As
s
o
cia
t
e P
r
ofes
s
o
r in the dep
a
rtm
e
n
t
of EEE
, K L
University
, Gun
t
ur, Andhra Pr
adesh, India. H
e
received B
.
Tech, d
e
gree in
Electrical
and
Electronics Eng
i
neer
ing from J.N.T.University
,
H
y
der
a
bad
,
A.P, India, M.Tech.,(
Electr
i
cal
Power Engineering) from J.
N.
T.
University
,
Hy
der
a
bad
,
A.P, India and currently
working
towards
the Do
ctora
l
degr
ee in
Ele
c
tri
cal
& E
l
ec
tronics
Eng
i
n
eering
at Ach
a
r
y
a N
a
garjun
a
University
, Guntur, Andhra Prad
esh, India. He p
ublished 7 paper
s
in national an
d internation
a
l
journals
and pr
esented
var
i
ous papers
in n
a
tio
nal and
In
ternational c
onferences. His current
research
interes
t
s includ
e activ
e filtering
for
power conditio
n
ing, Fuzzy
Lo
gic and ANN
applications to
p
o
wer quality
.
Ve
nu Gopala Rao M
a
nnam
, F
I
E, M
e
m
b
er I
E
E
E
,
at pr
es
ent
is
P
r
ofes
s
o
r & Head
, dep
a
rtm
e
nt of
Electrical & Electron
i
cs Eng
i
n
eering
,
K L U
n
iversity
, Guntu
r
, Andhra Prad
esh, India. He
rece
ived B.
E. d
e
gree in E
l
ec
tri
cal and E
l
ec
tro
n
ics
Engine
erin
g from
Gulbarga Univers
i
t
y
in
1996, M.E (Electrica
l Power
Eng
i
neer
ing) from
M S Univer
sity
,
Baroda, India in
1999, M.Tech
(Computer Science) f
r
om JNT
University
, Indi
a in
2004
and
Doctoral Degree in
Electr
ical
&
Electronics Engineering
from J.N.
T.University
,
H
y
der
a
bad
,
India in 2009
. He p
ublished more
than 30 p
a
pers
in various Nation
a
l, Intern
ational
Conferenc
e
s
and
J
ournals
. His
r
e
s
earch
inter
e
s
t
s
accum
u
la
te in
t
h
e are
a
of P
o
wer Qualit
y,
Dis
t
ri
bution S
y
s
t
em
, High Voltage E
ngineer
ing and
Ele
c
tri
cal
M
ach
i
n
es
.
Narsimha Rao.Vutlapalli,
a
t
present is pursuing M.Tech
.
in the depar
t
m
e
nt of EEE,
K.L.University
,
Guntur, Andhra Pradesh, India.
He received B
.
Tech
, degr
ee in
Electrical
and
Ele
c
troni
cs Eng
i
neering
from
J.N.T.Un
iv
ersity
, H
y
der
a
bad
,
A
.
P, I
ndia.
Evaluation Warning : The document was created with Spire.PDF for Python.