TELKOM
NIKA
, Vol. 11, No. 2, Februa
ry 2013, pp. 710~716
ISSN: 2302-4
046
710
Re
cei
v
ed Se
ptem
ber 10, 2012; Revi
se
d De
ce
m
ber
27, 2012; Accepted Janu
ary 12, 201
3
A Self-Learning Network Reconfiguration Using Fuzzy
Preferences Multi-Objective A
pproach
Hongbin Su
n*
1
, Chunjun
Zhou
2
1
Schoo
l of Electri
c
al Eng
i
ne
eri
n
g and Inform
ati
on, Jili
n univ
e
r
s
it
y
2
School of Elec
trical Eng
i
ne
eri
ng an
d Informa
tion,
Cha
ngc
hu
n Institute of
T
e
chn
o
lo
g
y
, Ch
angc
hu
n,
130
01
2,Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: hbsun
@mai
l.dhu.e
du.cn
A
b
st
r
a
ct
The pap
er pr
opos
es a self
-learn
i
n
g
evol
ution
a
ry
multi-
age
nt system fo
r distributio
n netw
o
rk
reconfi
gurati
o
n
.
T
he netw
o
rk
reconf
i
gurati
o
n
is mod
e
le
d a
s
a multi-
obj
ec
tive combi
natio
nal o
p
ti
mi
z
a
ti
o
n
.
An auto
n
o
m
ou
s age
nt-entity
cogn
i
z
e
s
the p
h
ysical asp
e
cts
as
op
eratio
n
a
l
states of th
e loc
a
l su
bstation
,
the a
g
e
n
t-entit
ies
establ
ish r
e
lati
onsh
i
p
net
w
o
rk base
d
o
n
the
inter
a
cti
ons to
prov
id
e
service. M
u
lti
p
l
e
obj
ectives ar
e consi
dere
d
for loa
d
ba
lanc
in
g amon
g t
he fee
ders, mini
mu
m deviati
on of th
e nod
es volt
ag
e
,
mi
ni
mi
z
e
the
p
o
w
e
r loss a
nd
branc
h curre
nt
constrai
nt
viol
ation. T
hes
e o
b
jectiv
es are
mo
de
led w
i
th f
u
zz
y
sets to evalu
a
te their i
m
pr
eci
s
e nature
and
one ca
n
prov
id
e the antic
ipat
ed val
ue of e
a
c
h obj
ective. T
h
e
meth
od co
mpl
e
tes the netw
o
rk reconfi
gur
ation b
a
se
d
o
n
the ne
gotiati
on of aut
o
n
o
m
ous ag
ent-e
ntities
.
Simulati
on res
u
lts de
mo
nstra
t
ed that the pro
pose
d
metho
d
is effective in i
m
pr
ovin
g perfo
rma
n
ce
Ke
y
w
ords
: net
w
o
rk reconfigu
r
ation, fu
zz
y
pr
ef
erenc
es, mu
l
t
i-obj
ective, opt
imi
z
a
t
io
n
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The di
stri
bution net
wo
rk reco
nf
iguratio
n problem
is
to find
a radi
al
ope
ratin
g
stru
cture
that minimizes the
syste
m
power l
o
ss an
d
minim
u
m deviation
of the nod
e
s
voltage
wh
ile
satisfying
op
erating
con
s
traints.
In
re
ce
nt years, ma
ny of the
di
st
ributed
ge
ne
rations are
set
up
in the vicinity
of the cu
stom
er, with th
e a
d
v
antage th
at this de
crea
ses transmissi
on lo
sses. T
w
o
obje
c
tives co
nsid
ere
d
a
r
e real
-po
w
e
r
lo
ss
re
du
ct
ion,
maximum n
o
des volta
ge
d
e
viation is ke
p
t
within a
rang
e, and the a
b
s
olute valu
e
of bran
ch
cu
rrents i
s
n
o
t a
llowe
d to exceed thei
r rate
d
cap
a
citie
s
. T
he ra
dial
con
s
traint
and di
screte
natur
e
of the switch
es p
r
event
th
e use of cl
assical
techni
que
s to
solve the re
configuration p
r
oble
m
.
Most
of the
algorith
m
s in
the lite
r
ature
a
r
e
ba
sed
on
heu
risti
c
se
arch
tech
nique
s.
Con
s
id
era
b
le
research
h
a
s be
en
con
ducte
d fo
r l
o
ss minimi
za
tion in the area
of network
reconfigu
r
atio
n of distributi
on
sy
st
em
s.
Dist
ri
but
ion
s
y
s
t
e
m re
c
onf
iguratio
n for loss red
u
ctio
n
wa
s first pro
posed by M
e
rlin
and Ba
ck [1
4
]. The
y
have u
s
e
d
a b
r
an
ch
-and-bou
nd-ty
pe
optimizatio
n tech
niqu
e to determin
e
the minimum lo
ss co
nfiguratio
n. In this method, all network
swit
che
s
a
r
e
first clo
s
e
d
to
form a me
sh
ed net
work. T
he switche
s
a
r
e then
ope
n
ed succe
s
sively
to resto
r
e ra
dial co
nfiguration. The solution pr
oce
dure
start
s
b
y
closin
g all
of the network
swit
che
s
whi
c
h are then op
ened on
e after anothe
r so
as to establi
s
h the optimum flow pattern in
the net
wo
rk [1]. A heu
ri
stic m
e
thod
[2]
have b
een
p
r
opo
sed
to d
e
t
ermine
a
distribution
sy
stem
config
uratio
n whi
c
h would
redu
ce lin
e lo
sses by
u
s
in
g a simplified
formula to calcul
ate the loss
redu
ction a
s
a result of load tran
sfer b
e
tw
ee
n two feede
rs. T
w
o
approximati
on formula
s
for
power flo
w
in
the tran
sfer
of
system lo
a
d
s
were mad
e
to impr
ove the method
o
f
heuri
s
tic [3]
.
Artificial-intelli
gen
ce-ba
s
ed
application
s
in
a minimu
m loss confi
guratio
n hav
e been
prop
ose
d
[4-7]. Da
s [8] has p
r
e
s
ente
d
an algo
rith
m for netwo
rk reco
nfigu
r
ati
on ba
sed o
n
heuri
s
tic
rule
and
fuzzy multi-o
b
jective ap
proach.
Agent-o
riente
d
comp
uting provide
s
a bi
g
po
ss
ibility
of implem
ent
ation fo
r thi
s
probl
em.
The age
nt communitie
s
a
r
e actu
ally group
s of
software co
ope
ratives and communi
cativ
e
bu
t
indep
ende
nt
agent
s, whi
c
h are expe
ct
ed to join
de
cisi
on
s an
d
action
s to a
c
hieve a
com
m
on
goal. The pu
rpo
s
e of the
comm
on go
al
is to provide
a glue to bind individual
s' a
c
tions int
o
a
coh
e
sive
wh
o
l
e. The
pro
p
o
s
ed
archite
c
t
u
re im
pl
em
en
ts simil
a
r
gro
ups of collab
o
rating
software
agent
s.
In o
u
r previo
us work, we
ha
ve
implem
en
ted the
proto
t
ype of a
gen
t-netwo
rk
se
rvice
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
simu
l
sy
st
e
the
d
agen
confi
g
A
ppr
o
impr
o
2.
A
Rec
o
2.1.
T
agen
agen
mobi
l
trans
differ
A
ge
n
serv
i
c
awa
r
e
devi
c
F
2.2
T
with
m
con
s
t
radia
is pr
pre
s
e
K
OM
NIKA
A
S
l
atio
n pl
atfo
r
e
m
s
and
ap
p
d
es
ig
n
of th
e
ts
.
T
h
e
m
a
i
n
g
uratio
n
ba
s
o
ach to
solv
o
vin
g
perfor
m
A
Sel
f
-Lea
o
nfigur
a
t
ion
T
he Agent
-
A
s an
at
o
t
)
in
clu
d
e
s t
h
t itself. Fun
c
l
e ag
en
ts.
mission.
The
A
ge
n
t
ent info
ra
m
n
t-entity. It
i
s
c
e i
s
cre
a
te
d
e
ag
ent
s a
n
c
e a
s
a n
e
tw
o
F
igure 1.
Th
e
T
he model
o
The n
e
t
w
m
inimum lo
s
t
rai
n
ts u
nde
r
l o
perating
s
es
e
n
t
ed in
e
nte
d
as
i
L
i
lo
s
s
r
F
1
S
el
f
-Learni
n
g
r
m [9]. T
he
s
p
li
cat
i
on
s co
n
e
sy
st
em
r
e
n
contri
b
u
ti
o
s
ed o
n
th
e
a
e it. Simula
m
an
ce.
rning
Ev
o
-
Entity
Desi
g
o
mi
c unit of
h
re
e mod
u
le
s
c
tion
is
de
s
i
g
Behavior
c
t
-E
ntity is a
m
tion se
rvice
s
s
a novel c
o
d
out
of the
n
d thei
r envi
o
rk no
de, a
n
e
Structure
o
o
f Recon
f
ig
w
or
k r
e
confi
g
s
s and
mini
m
r
a
ce
rtai
n
l
s
truc
ture of
t
the
literatu
r
i
i
i
i
V
Q
P
r
2
2
2
I
S
g
Net
w
or
k R
e
s
oftware d
e
v
n
forming t
o
e
sulted in
a
g
o
n of thi
s
pa
a
ge
nt-net
wo
r
tion results
d
lutionar
y
M
g
n
Distribution
s
, a
s
sh
own
g
ned to
e
v
al
c
ontain
s
i
n
t
unit
develo
p
s
. Net
w
ork
R
o
mp
uting a
n
interaction
o
ronm
ent [9].
n
d functio
n
a
l
o
f Agent-Enti
t
uration Ne
t
g
u
r
atio
n pro
b
m
um d
e
viati
o
l
oad
patte
rn
t
he system
.
T
r
e in
differ
e
S
SN: 2302-4
0
e
co
nfigurati
o
v
elopme
n
t f
r
FIPA [10]
s
t
g
ent comm
u
pe
r i
s
to p
r
e
r
k.
We pro
p
d
emo
n
s
t
r
a
t
e
M
ulti-
A
ge
n
t
Net
w
or
k R
in Figure 1.
uate the
m
a
t
erface op
e
p
ed b
y
java
R
ec
o
n
f
ig
ur
a
t
d problem
-
s
o
f multiple
a
The ideal
m
merits
refe
r
t
y In
c
l
ud
es
T
Behavior
t
w
ork in Di
s
b
lem
in
a di
s
o
n of the no
. The
ope
r
a
T
he mathe
m
e
nt
way
s
.
I
n
0
46
o
n Usi
n
g Fu
z
r
amework ai
m
t
and
ard
s
for
u
nities, ea
ch
e
sent the
a
u
p
ose a
Fu
zz
y
d that the
p
r
t
Sy
stem
e
c
o
n
figurati
o
Attributes d
e
a
tchin
g
abil
i
t
y
e
ration, inf
o
. Different
A
t
ion is a
c
hi
e
v
s
olving envi
r
a
ware a
gent
s
m
odel woul
d
r
to our previ
o
T
hre
e
Modul
e
s
tribution S
y
s
t
r
ibutio
n
sy
s
de
s voltag
e
a
ting cons
tr
a
m
atical form
u
n
this
pape
r
z
zy Prefe
r
en
c
m
ed at dev
e
intelligent
a
con
s
i
s
ting
u
tonomou
s
s
y
Preferen
c
e
r
opo
sed m
e
t
for Dis
t
ri
o
n, the
A
ge
e
scribe the
c
y
of the me
s
o
rmation i
s
s
A
gent-En
t
ity
v
ed by se
rvi
r
on
me
n
t
wh
e
s
and the
in
d
place the
o
us
work
[1
1
e
s: Attribute,
y
st
em
s
tem is
to fi
n
while sat
i
sf
y
a
in
ts
ar
e
c
u
r
u
lation re
con
f
r
, the
probl
e
c
es
(H
o
n
gb
i
n
e
loping multi
-
a
g
ents. Dev
e
of th
r
e
e
so
f
s
ys
t
e
m
o
f
n
e
e
s Multi-Obj
t
h
od is
effec
t
bution Ne
nt-Entity (
m
c
ha
r
a
c
t
e
r
is
ti
c
s
sage to the
s
u
e
, and
e
may contrib
ce c
o
mpos
i
t
e
re
an appli
teraction be
platform on
1
,12].
Func
tion a
n
n
d a
config
u
y
ing the op
e
r
re
nt ca
pa
ci
t
f
i
g
uratio
n pr
o
e
m form
ulat
n
Su
n
)
711
-
ag
en
t
e
lo
ped
f
tware
e
twork
ecti
ve
t
ive in
tw
ork
m
obil
e
c
of
an
other
e
ne
rgy
ute to
t
ion of
cat
i
on
t
ween
every
n
d
u
ration
e
rating
y and
o
ble
m
ion is
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 2, Februa
ry 2013 : 710 – 716
712
K
k
Q
Q
K
k
Q
Q
K
k
T
T
K
k
T
T
MepF
F
Mep
Subject
F
Mep
F
Mep
Min
j
j
j
k
k
k
k
po
loss
vol
max
max
min
0
)
(
))
(
),
(
(
loss
F
is the m
e
mb
ership fun
c
tio
n
for ac
tive power los
s
es
,
i
r
rep
r
e
s
ent
s the re
si
stan
ce
of
the bran
ch i.
i
P
,
i
Q
repre
s
e
n
t active power
and re
active
pow
er that flowin
g the terminal of the
bran
ch
i.
i
V
re
prese
n
ts the
no
de voltage
of
the termin
al
of bra
n
ch i.
i
L
r
e
p
r
es
en
ts
the n
u
m
be
r
o
f
bran
ch
es. Vol
t
age variation
may be cau
s
ed by
the Dist
ributed G
ene
ration output changi
ng.
N
V
is voltage rat
i
ng;
i
V
is real v
o
ltage of the
system;
vol
F
is the membe
r
shi
p
function fo
r
voltage profil
es.
n
i
N
N
i
vol
V
V
V
F
1
2
2
)
(
(2)
i
P
,
max
,
i
P
represen
t the real
ru
nning
po
we
r and
the
m
a
ximum p
e
rmitted po
we
r of the
trans
former.
i
a
is penalty function p
a
ra
m
e
ter.
po
F
is
the membership func
tion for power.
b
i
i
i
n
i
P
P
K
i
i
i
po
a
P
P
F
1
)
(
max
,
)]
1
(
[
max
,
(3)
Based
o
n
th
e fuzzy
eval
uation fu
ncti
ons, th
e m
u
l
t
io-bje
ctive o
p
timization
model
is
con
s
tru
c
ted t
o
maximize the satisfa
c
tio
n
s of diffe
ren
t
objective
s b
y
adjustin
g
transfo
rme
r
ta
p-
cha
nge
rs
an
d sh
unt capa
citors. The
o
b
jective
s
in
cl
ude voltag
e
profile
s, activ
e
po
wer l
o
sses.
The multio-bj
e
ctive optimization model i
s
rep
r
e
s
e
n
ted
as:
(4)
Whi
c
h T
k
i
s
t
he ratio of tra
n
sformer k;
Qj
is the
cap
a
city of
capa
citors at
nod
e
j,
)
(
F
Mep
is the value to evaluate
F
,
max
k
T
,
max
j
Q
rep
r
e
s
ent
s the threshold of
criteri
on.
3. The Fuzzy
Preference
s Ev
olutionar
y
Algorithm
There are m
any MO solu
tion algorith
m
s allo
wi
ng t
he attainmen
t
of these re
sults, like
SPEA2 [16], PESA-II [17],
NSGA-II [13]. An important
iss
ue in multiple objec
tive
optimiz
ations
is
the ha
ndling
of huma
n
p
r
eferen
ce
s. Fi
nding
all Pa
reto-optim
al
solution
s i
s
n
o
t the final
g
oal.
Suc
h
prefer
enc
e
s
c
a
n us
ually be
r
e
pr
esented
w
i
th
the help of
fuzzy logic
.
B
a
sed on pr
eferenc
e
relation
s [12,
13]
and
ind
u
ce
d o
r
d
e
rs, these li
ngui
stic
cate
go
rie
s
we
re t
r
an
sf
orme
d into
real
weig
hts an
d a weig
hted Pareto do
mina
nce relation
wa
s introd
uced.
In this
pape
r,
the n
o
vel fu
zzy
prefe
r
e
n
ces
ev
olutiona
ry algo
rithm
(FP-EA) i
s
p
r
opo
sed.
Suppo
se tha
t
the size of
evolutionary
populatio
n P is n, and
Pt is t-th generatio
n of the
popul
ation. Qt is a new evolutionary
populatio
n fr
om Pt that is upd
ated b
y
the selecti
on,
cro
s
sove
r an
d mutation
op
erato
r
s, a
nd t
he si
ze
of Q,
is al
so n. L
e
t Rt=Pt
∪
Qt, an
d the si
ze
of
Rt
is 2n.
The
no
n-do
minate
d
set P1 i
s
g
e
n
e
rated
fr
om
Rt, with the
q
u
ick
sort
pro
c
edure. If|P1|>n,
the clu
s
teri
ng
pro
c
ed
ure i
s
used to
red
u
c
e the
size of
P1, and to keep the dive
rsity of P1 at the
same time. T
he
size of P1 will be
n
after the
clusteri
ng process.
is
equally im
portant,
is less
importa
nt,
is much less im
portant,
is not important,
!
is
important.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Self-Learni
ng Net
w
ork Reco
nfiguratio
n
Usi
ng Fu
zzy Prefe
r
en
ce
s (Hong
bin S
un)
713
Definition
1: (Weig
h
ted d
o
m
inan
ce
relat
i
on) F
o
r
a giv
en weight
s–v
e
cto
r
)
....
1
k
w
w
w
summi
ng to
1 and
a real
numbe
r
1
0
, a re
al vecto
r
)
,
)(
....
1
w
x
x
x
k
– dominate
s
a
real
vec
t
or
)
....
1
k
y
y
y
written as
y
x
w
if and only if
)
y
,
(x
I
y
x
i
i
1
k
i
i
w
w
(5)
whe
r
e
y
x
y
x
y
x
I
0
1
)
,
(
The sta
nda
rd definition
of dominan
ce coul
d be
obtained b
y
setting
1
and
k
w
w
n
/
1
...
1
. Note that in the standard
definition of
dom
ina
n
ce it
is req
u
ire
d
that at least
one of
the
i
i
y
x
inequ
alities is stri
ct. Howe
ver thi
s
i
s
n
o
t
a p
r
obl
em
since
the
s
e t
w
o o
r
de
rs
are defin
able
in terms of ea
ch othe
r.
Definition 2: (Weig
h
ted sco
r
e). The n
u
m
ber ni
n
e
is used here for the grad
es of relative
importa
nce b
e
twee
n obje
c
tives becau
se we take
the well-kno
w
n t
e
ch
niqu
e
of analytic hierarchy
pro
c
e
ss
(AH
P
) for refe
ren
c
e. For e
a
c
h
X
x
i
comp
ute wei
ght as no
rmal
ized le
aving score
X
x
j
i
i
j
R
x
SL
R
x
SL
x
w
)
,
(
)
,
(
)
(
(6)
Definition
3: (Fitne
ss ev
aluation
)
. Su
ppo
se the
r
e
are
N i
ndiv
i
dual
s in th
e
cu
rre
nt
popul
ation p
o
p
. The
po
sitive strength
)
(
k
x
S
of each individ
u
a
l
pop
x
k
)
,
2
,
1
(
N
k
is cal
c
ul
ated.
Suppo
se
))
(
(
min
,
,
2
,
1
min
k
N
k
x
S
S
,
)
(
max
,
,
2
,
1
max
k
N
k
d
d
.The fitness of
each i
ndividual
)
,
,
2
,
1
(
N
k
pop
x
k
is cal
c
ulate
d
according to the followi
ng formul
ation:
2
max
min
)
/
(
)
1
)
(
(
)
(
d
d
S
x
S
x
fit
k
k
k
(7)
Algorithm: FP
-EA Algorith
m
Pt ,
t = 0 ;//
Set t = 0. Generate an i
n
itial popul
ation P[t],
for each
X
x
i
comp
ute
weig
ht as no
rmalize
d
leavi
ng score :
X
x
j
i
i
j
R
x
SL
R
x
SL
x
w
)
,
(
)
,
(
)
(
While ( t
≤
T
)
do //T is maximum numb
e
r of generatio
n
s
{
2
max
min
)
/
(
)
1
)
(
(
)
(
d
d
S
x
S
x
fit
k
k
k
//Cal
culate
the fitness
value of e
a
ch individu
al i
n
Pt,
)
,
,
2
,
1
(
N
k
P
x
t
k
Qt = make-n
ew-pop
(Pt ) // Use sele
ction,
cro
s
sov
e
r and m
u
tation to create
a new
popul
ation Qt
Rt = Pt
∪
Qt // Combin
e pa
rent and children pop
ulation
I
f
(|
P
t
+ 1 |
<
= N
)
Th
en {
P
t
+ 1 = P
t
+
1
∪
sel
e
ct
- b
y
- ran
dom
(
Rt - Pt + 1,
N
- | Pt +
1 | ) } // randomly selecte
d
N - | Pt + 1 |
element
s and
joined into Pt + 1
Els
e
if (| Pt +
1 | >
N)
Then {
c
rowdi
ng - dista
n
ce - assign
ment
(Pt + 1)
// Calculate
crowdi
ng di
stan
ce.
Sort (Pt +
1 ,
≥
n) // Sort in desce
nding
o
r
de
r usi
ng
≥
n
Pt + 1 = Pt + 1 [1: N]} // Choose the first
N eleme
n
ts
t = t + 1}
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 2, Februa
ry 2013 : 710 – 716
714
It can b
e
pro
v
ed that the ti
me complexit
y
of
Algorith
m
(FP-EA
) is
less than
O (nlogn
). It
is better than
O (n2
)
in the NSGA.
4. Simulatio
n
and Discu
ssion
In our p
r
evio
us work, we
have imple
m
ented
the prototype
of A
gent-n
etwo
rk service
simulatio
n
pl
atform [11,1
2
], including
software, general obje
c
t
s
and sim
u
l
a
tors in java
. It
sup
port
s
plu
ggabl
e funct
i
ons
and p
r
ovides a
ge
neri
c
ea
sy-t
o-u
s
e p
r
og
ramming API
. It
contri
bute
s
to implement o
u
r app
ro
ach in real de
ploy
ment with mi
nimal modifi
cations. The S
e
lf-
Learning Evolutiona
ry Multi-Agent
System ha
s be
en desi
gne
d
as
a simpl
e
prototype. The
simulatio
n
experim
ent is constructe
d o
n
Wind
ow
s 2
000 op
eratio
n system wit
h
Intel Pentium 4
pro
c
e
s
sor (2.4 GHz) an
d 1G RAM. Th
e prop
os
ed method is te
sted in a 69-bus di
strib
u
tion
system [2] (Fi
gure 1
)
. The
para
m
eters o
f
on-load ta
p
-
cha
nge
r, the test sy
stem i
s
a hypotheti
c
al
12.66
kV syst
em, 69 bu
sses, 5 loopi
ng
bran
che
s
(t
i
e
lines). System data
is gi
ven togethe
r with
the voltage p
r
ofile of the ba
se
config
urati
on.
Re
al po
wer lo
ss re
du
ction is
10.49%
and mini
mu
m
voltage of th
e
syste
m
h
a
s improve
d
. In f
a
ct, the vo
lta
ge p
r
ofile
of the b
a
se
syst
em
configu
r
at
ion
is lo
wer th
an
the usual lo
wer. It is
assumed that
every bran
ch in
the syste
m
i
s
availa
ble f
o
r
bran
ch
-ex
c
ha
nge.
Figure 2. A 69-no
de Di
stri
bution Sy
ste
m
Before Re
configuration.
These graph
s sh
ow very
clear
sep
a
ration of Pareto fronts ob
tained u
s
ing
different
prefe
r
en
ce
s, perfo
rms
well
on the conve
r
gen
ce a
nd the diversity. The traditio
n
a
l method
s solve
the multi-obj
ective a pro
b
lem is to transl
a
te
the vector of obj
ectives into
one obje
c
tiv
e
by
averagi
ng th
e obje
c
tives
with a wei
g
h
t
vector
. The
most profou
nd drawback of traditiona
l
algorith
m
s
i
s
their sen
s
itivity
toward
s weights or
de
mand
l
e
vels. This discu
s
si
on
sugg
est
s
t
hat
the cla
s
sical
method
s t
o
the p
r
o
b
lems
of
net
work
re
confi
guratio
n a
r
e
inade
quate
and
inco
nvenient to
use.
In valley load
con
d
ition, re
verse
po
we
r flow rai
s
e
s
th
e voltage an
d po
wer l
o
sses. After
optimizatio
n, the over voltage is allevia
t
ed,
and the voltage profil
es are impro
v
ed. The po
wer
losse
s
com
p
arison sho
w
s that, the optimal control
sch
eme de
cre
a
se the p
o
we
r losse
s
b
y
redu
cin
g
re
a
c
tive power t
r
an
sferred. A
l
though
the
maximum vol
t
age variatio
n increa
se
s
very
slightly, the integral satisfa
c
tion imp
r
ove
d
evidently.
Table 1. Opti
mal Sets of The Re
co
nfigu
r
ation Result
Power loss
Deviation of the
nodes voltage
1 90.67
0.0215
2 91.69
0.0221
3 95.4
0
.0223
4 96.7
0
.0228
Re
spo
n
se time in
self-le
a
rnin
g evoluti
onary
m
u
lti-A
gent sy
stem
rep
r
e
s
ent
s th
e efficien
cy
ofnegotiatio
n
, In all
10
0 mi
nutes (Fi
g
u
r
e
3),
re
spo
n
se
time d
e
crea
se
slo
w
ly. At the
begin
n
in
g,
respon
se
time re
ache
s 700m
s.
When the
re
q
uest i
s
cha
nged, the
a
gent-e
ntities start
evolutiona
ry l
earni
ng, Ea
ch ag
ent-e
ntity sho
u
ld
neg
otiate with
ot
hers
ba
sed
o
n
the
ca
pabili
ties
that
ca
n be e
x
ecuted. Wit
h
the
time pa
ssi
ng by,
the
high
effe
ctive a
r
e
prese
n
t
ed . Fi
nally, t
h
e
respon
se tim
e
de
cre
a
ses
dram
atically,
until it
rea
c
he
s the mini
mu
m value to b
e
abo
ut 150
ms.
0
1
2
3
4
5
6
7
8
9
10
1
1
1
2
1
3
1
4
15
16
1
7
1
8
19
20
21
22
2
3
2
4
25
26
27
59 60 61
62 63
64 65
66 67
68 69
40 41
57 58
55
56
42
43
44
4
5
4
6
4
7
48
49
50
51
52
5
3
5
4
28
29
30
3
1
3
2
33
34
35
36
37
38
3
9
0
1
2
3
4
5
6
7
8
9
10
1
1
1
2
1
3
1
4
15
16
1
7
1
8
19
20
21
22
2
3
2
4
25
26
27
59 60 61
62 63
64 65
66 67
68 69
40 41
57 58
55
56
42
43
44
4
5
4
6
4
7
48
49
50
51
52
5
3
5
4
28
29
30
3
1
3
2
33
34
35
36
37
38
3
9
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Self-Learni
ng Net
w
ork Reco
nfiguratio
n
Usi
ng Fu
zzy Prefe
r
en
ce
s (Hong
bin S
un)
715
we
ca
n
see
that the ex
cessive mig
r
at
ion b
ehavio
rs d
e
crea
se
for the
ag
ent
-entities, it
al
so
exhibits the le
ss
co
st.
Figure 3. The
Pareto Opti
mal Front in
Re
config
urati
o
n
Figure 4. Re
spon
se Time a
nd Migration
Freq
uen
cy
5. Conclusio
n
The
n
e
two
r
k reconfigu
r
atio
n
is mod
e
led as a
multi-obj
ective combi
national opti
m
ization,
a self-lea
rni
n
g evolutio
nary multi-ag
ent
syste
m
ba
se
d
on
Fu
zz
y Pr
e
f
er
e
n
c
e
s mu
lti-
ob
jec
t
ive
approa
ch h
a
s
be
en p
r
op
ose
d
to sol
v
e the netwo
rk re
co
nfig
uration probl
em
in
a ra
dial
dis
t
ribution
sys
tem. The method
c
o
mpletes
the
network
re
config
uratio
n
ba
sed
on
the
negotiatio
n
of autonomou
s agent
-entiti
e
s. The obje
c
tives consi
d
ered attempt
to maximize the
fuzzy
satisfa
c
tion of the
load b
a
lan
c
ing amo
ng
t
he feed
ers, minimization
of power l
o
ss,
deviation of node
s voltag
e and b
r
an
ch cu
rre
nt co
nstr
ai
nt violation su
bje
c
t to radial net
work
stru
cture. Simulation resu
lts demo
n
stra
ted that
the prop
osed me
thod is effect
ive in improvi
n
g
perfo
rman
ce.
Referen
ces
[1]
D Shirm
oham
madi, HW
H
o
n
g
, Reco
nfig
ura
t
ion of
el
ectric
distrib
u
tion
net
w
o
rks for res
i
s
t
ive li
ne
los
s
reducti
on.
IEEE Trans. Power Del
. 198
9; 4(2): 1492
–1
498.
[2]
S Civa
n
lar, JJ
Graing
er, H Yi
n, SSH L
ee. Di
st
ributio
n fee
d
e
r reco
nfig
urati
on for l
o
ss re
d
u
ction.
IE
EE
Trans. Pow
e
r Del.
19
88; 3( 3
)
: 1217–
12
23.
[3]
ME Baran, F
F
W
u
. Net
w
ork
reconfig
urati
o
n in
distri
buti
o
n s
y
stems for
loss reducti
o
n
and l
o
a
d
bal
anci
ng.
IEEE Trans. Power Del
. 198
9; 4(3): 1401
–1
407.
[4]
DJ Shi
n
, JO Kim, T
K
Ki, JB Cho
o
, C Si
ngh.
Optima
l
service r
e
stora
t
ion a
nd r
e
co
nfigur
ation
o
f
net
w
o
rks us
ing
genetic a
nd ta
bu searc
h
al
go
rithm.
Int. J
.
El
ect. Power Sys
t. Res.
2010; 7
1
:145
–1
52.
[5]
YT
Hsiao. Mul
t
iobj
ective
evo
l
utio
n pro
g
ram
m
ing m
e
tho
d
for feed
er rec
onfig
uratio
n.
IEEE Trans.
Power Syst
. 2004; 19(1): 5
94–
599.
[6]
YY Hon
g
, SY
Ho. Determ
ina
t
ion of n
e
t
w
or
k config
urati
o
n
consi
deri
ng
multio
bjectiv
e
i
n
distri
butio
n
s
y
stems usi
ng
gen
etic al
gorith
m
.
IEEE
Trans. Power Syst
. 2
005; 20(
2):10
6
2–1
06
9.
M
ep
F
vo
l
)
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
102
))
(
(
))
(
((
vo
los
F
Mep
f
F
Mep
f
M
e
p
F
lo
s
)
M
ep
F
vo
l
)
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
102
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
102
98
100
102
))
(
(
)
(
(
vo
los
F
Mep
f
F
Mep
f
M
e
p
F
lo
s
)
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
M
e
p
F
lo
s
)
M
ep
F
vo
l
)
102
)
(
(
)
(
(
vol
los
s
F
Mep
f
F
Mep
f
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
M
e
p
F
lo
s
M
ep
F
vo
l
)
102
)
(
(
)
(
(
vol
los
s
F
Mep
f
F
Mep
f
M
ep
F
vo
l
)
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
102
)
(
(
)
(
(
vo
l
los
F
Mep
f
F
Mep
f
M
e
p
F
lo
s
M
ep
F
vo
l
)
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
0
.02
0
0.0
2
1
0.0
2
2
0.0
2
3
0.0
2
4
0.0
2
5
90
92
94
96
98
100
102
)
(
(
)
(
(
vo
l
los
F
Mep
f
F
Mep
f
M
e
p
F
lo
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 2, Februa
ry 2013 : 710 – 716
716
[7]
KY Huan
g, HC
Chin. Distri
but
ion fee
der e
ner
g
y
c
onserv
a
tio
n
b
y
usi
ng h
e
u
r
istics fuzz
y
a
p
p
roac
h.
Int.
J. Elect. Pow
e
r Energy Syst
. 2002; 24:
439-
44
5.
[8]
D Das. A f
u
zz
y m
u
ltio
bjectiv
e
ap
pro
a
ch fo
r
net
w
o
rk
reco
nfigur
ation
of
distrib
u
tion
s
y
s
t
ems.
IEEE
T
r
ans. Pow
e
r Del
. 20
06; 21(
1): 202–
20
9.
[9]
Gao, YS Di
ng,
LH R
en. A
no
vel ec
olo
g
ic
al
net
w
o
rk-
base
d
computati
on
p
l
atform as
grid
midd
le
w
a
r
e
sy
s
t
e
m
.
Int. J. I
n
telligent System
s
. 20
04; 19(
1
0
): 859-8
84.
[10]
YS Ding, HB Sun, KR Hao.
A bio-ins
p
ire
d
e
m
erge
nt s
y
ste
m
fo
r intellig
en
t Web service compos
itio
n
and ma
na
gem
ent.
Know
led
g
e
-Base
d
Systems.
20
07; 20(
5
)
: 457-46
5.
[11]
K Deb, A Prat
ap, S Agra
w
a
l.
A fast and
e
litist multi - o
b
j
e
ctive g
e
n
e
tic
alg
o
rithm: NS
GA-II.
IEEE
T
r
ansactio
n
s o
n
Evoluti
o
n
a
ry Co
mp
utation.
2
002; 6(2): 1
82-
197.
[12]
D Cvetkovic,
IC Parmee.
Preference
s
and the
i
r Appl
icatio
n
in
Evolutio
nar
y Multi-ob
jecti
v
e
Optimisation.
IEEE Transacti
ons on Evo
l
uti
onary C
o
mput
ation.
20
01; 6(
1): 42-57.
[13]
XN
Sh
en, Y
Guo, QW
Ch
e
n
. A mu
lti-o
b
j
e
ctiv
e optimiz
ation
ev
oluti
o
n
a
r
y
a
l
gor
ithm
i
n
corp
oratin
g
prefere
n
ce info
rmation b
a
sed
on fuzz
y
l
o
g
i
c.
Co
mp
utation
a
l
Optimi
z
a
t
i
o
n
a
nd App
licati
ons
. 2008; 40:
55-6
4
.
[14]
A Merlin, H Back.
Search
for a
min
i
mal loss oper
ating
spa
n
n
in
g tree c
onfi
gurati
on
in a
n
urba
n p
o
w
e
r
distrib
u
tion sys
tem.
Proc. 5thPo
w
er Sy
st. Compt. C
onf., Cambridge, UK. 1975:1–18.
[15]
YC
Hua
ng.
En
hanc
ed g
eneti
c
algorit
hm
ba
sed fu
zz
y
mu
lti
-
obj
ective ap
pr
oach to distri
b
u
tion n
e
tw
ork
reconfi
gurati
o
n
. Proc. Inst. Elect. Eng., Gen.,
T
r
ansm. Distrib., 2002; 615
–6
20.
[16]
E Z
i
tzler, M L
a
u
mmans,
L T
h
iele.
SPEA
2: Impr
ovin
g th
e s
t
rength
paret
o
evol
ution
a
ry a
l
gorith
m
, TIK
Rep
o
rt No. 10
3.
S
w
iss F
e
d
e
r
a
l Institute of
T
e
chn
o
lo
g
y
(ET
H
), Computer
Engi
neer
in
g an
d Net
w
orks
Lab
orator
y (T
IK). 2001.
[17]
DW
Corne, N
R
Jerram,
JD
Kno
w
les, MJ
Oates.
PESA-II: Region bas
ed select
ion in evolutionary
mu
ltio
bjectiv
e
opti
m
i
z
at
ion.
P
r
ocee
din
g
s of the Genetic an
d Ev
oluti
onar
y Computati
on
Confer
ence
,
GECCO-200
1, Los Altos, CA: Mo
rga
n
Kaufm
ann. 20
01; 28
3
-
290.
Evaluation Warning : The document was created with Spire.PDF for Python.