Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
1
3
,
No.
3
,
Ma
rch
201
9
, p
p.
9
74
~
981
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
3
.i
3
.pp
9
74
-
9
81
974
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
S
ca
l
ar
i
zin
g f
un
ctions in
so
lving mu
lti
-
object
ive pr
oblem
-
an
evolutio
nary app
roac
h
D.
V
as
um
athi
,
S.
Th
ang
av
el
u
Depa
rtment
o
f
C
om
pute
r
scie
n
ce
and
Eng
ine
e
ring
,
Am
rit
a
school
of
engi
n
ee
r
ing,
Am
rit
a
Vishw
a Vid
y
ap
eetha
m
,
I
ndia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
6
,
2018
Re
vised Dec
5
,
2018
Accepte
d
Dec
2
2
, 201
8
Scal
ar
iz
ing
func
ti
ons
had
long
bee
n
observ
ed
for
opti
m
izati
on
of
m
ult
i
-
obje
c
ti
ve
prob
lem
s.
Scalari
z
ing
func
ti
ons
on
m
ult
i
-
obj
ective
pro
ble
m
a
long
with
Diffe
r
ent
i
al
Evol
ut
ion
(DE)
al
gorit
hm
var
ia
n
ts
had
be
en
used
to
an
aly
z
e
the
ef
fect
of
sc
al
ar
iz
ing
func
t
io
ns.
The
m
ai
n
p
urpose
is
to
fin
d
the
b
et
t
er
sca
la
r
iz
ing
func
ti
on
which
ca
n
be
applied
for
opti
m
iz
ation.
T
he
eff
ec
t
ive
soluti
on
of
th
e
m
ult
i
-
obj
ective
pr
oble
m
depe
nds
o
n
the
var
ious
factors
li
ke
th
e
DE
al
gor
it
hm
an
d
the
sca
la
r
iz
ing
func
ti
ons
used
.
Multi
ob
je
c
ti
ve
e
volu
ti
on
a
r
y
al
gorit
hm
(MO
EA)
fra
m
ework
in
j
ava
had
b
e
en
used
for
per
form
ing
the
ana
l
y
sis.
The
Obtai
n
ed
resul
ts
show
ed
that
T
che
b
y
sheff
s
ca
l
ariza
t
i
on
func
ti
o
n
per
form
s
bet
ter
tha
n
the
oth
er
sca
l
ari
z
ing
fun
ct
ions
on
var
io
us
indi
c
at
or
func
ti
ons used
.
Ke
yw
or
d
s
:
Diff
e
re
ntial
evoluti
on
Indicat
or fu
nctions
MOEA
fram
e
work
Mult
i
-
obj
ect
iv
e pro
blem
s
Scal
arizi
ng fu
nc
ti
on
s
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
S.
T
ha
ng
a
velu
,
Dep
a
rtm
ent o
f C
om
pu
te
r
sci
e
nce a
nd E
ng
i
ne
erin
g
,
Am
rita
V
ishw
a
V
idya
peetham
, In
dia
.
Em
a
il
:
s_
than
ga
vel@c
b.
am
rit
a.ed
u
1.
INTROD
U
CTION
Re
centl
y,
Dif
f
eren
ti
al
e
vo
l
ution
has
at
trac
te
d
c
on
si
der
a
bl
e
researc
h
in
te
rest
in
m
ulti
-
obj
ect
ive
op
ti
m
iz
ation
i
n
the
dom
ai
n
of
evo
l
ution
a
ry
al
gorithm
.
O
n
c
om
par
ison
with
tradit
ion
al
al
go
rithm
s,
DE
is
f
ound
to
be
pe
rfo
rm
i
ng
bette
r
on
di
ff
e
ren
t
pro
blem
s.
The
m
a
in
researc
h
iss
ue
fo
c
us
es
on
perform
ance
eval
uation,
fitness
assig
nm
ent
an
d
div
e
r
sit
y
m
ai
ntenance
w
he
re
our
m
ai
n
goal
is
pe
rfor
m
ance
e
va
luati
on
.
Re
ce
ntly
so
m
e
works
on an
al
y
zi
ng
pe
rfor
m
ance ar
e als
o don
e [
1].
A
M
ulti
-
ob
j
ec
ti
ve
prob
le
m
(
MOP)
is
de
fin
ed
as
a
n
op
ti
m
iz
at
ion
pro
bl
e
m
with
se
veral
ob
j
ect
ive
functi
ons
with
decisi
on
var
ia
bl
es in th
e
f
easi
ble r
e
gion
Ω
. M
at
hem
a
ti
call
y i
t i
s g
ive
n
in
equ
at
io
n (
1)
min
(
)
=
(
1
(
)
,
2
(
)
…
.
(
)
)
(1)
wh
e
re
∈
Ω
A
pro
blem
wh
i
ch
has
o
ne
o
r
m
or
e
obj
ect
ive
is
cal
le
d
m
ulti
-
ob
j
ect
ive prob
lem
.
Scal
arizi
ng fun
ct
io
ns,
in
gen
e
ral,
a
re
us
e
d
f
or
dec
om
po
sin
g
a
m
ulti
-
obj
ect
ive
pro
blem
into
sever
al
sin
gle
ob
j
ect
ive
pro
blem
s.
Objecti
ve
f
unc
ti
on
s
are
denot
ed
by
(
)
.
Eac
h
m
ulti
-
obj
ect
ive
pro
blem
will
ha
ve
seve
ral
obj
e
ct
ive
f
unct
ions
and se
ver
al
de
ci
sion
var
ia
ble
s.
Diff
e
re
ntial
Evo
l
ution
(
DE)
is
an
ev
olu
ti
onary
al
gorithm
for
opti
m
iz
ing
a
pro
blem
iterati
vely
for
i
m
pr
ovin
g
t
he
qu
al
it
y
of
the
cand
i
date
s
ol
ution
.
Dif
fer
e
ntial
ev
olu
t
io
n
ha
s
been
fou
nd
t
o
pe
rfor
m
bett
er
on
sta
bili
ty
[2
]
th
us
, d
if
fer
e
ntial
evo
l
ution
is us
ed
f
or p
er
form
i
ng
ex
pe
rim
ent
s
f
or
so
l
ving MOP
us
in
g
sca
la
rizi
ng
functi
ons.
A
num
ber
of
scal
a
rizi
ng
functi
on
s
e
xist
in
li
te
ra
ture
w
hich
a
re
us
e
d
for
pe
rfo
r
m
ing
e
xperim
e
n
ts
on
diff
e
re
nt
m
ulti
-
obj
ect
ive
opti
m
iz
at
ion
al
gori
thm
s.
Penalty
boun
d
i
ntersecti
on
(
PBI
),
Tch
e
bysh
e
ff
a
nd
we
igh
te
d
su
m
are
com
m
on
ly
us
ed
scal
arizi
ng
f
un
ct
io
ns
for
a
ny
ex
pe
rim
ents.
Hen
c
e,
these
a
re
us
e
d
on
t
he
e
xp
e
ri
m
ents
perform
ed
in t
his a
naly
sis.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Sca
l
ar
izi
ng fu
nc
ti
on
s i
n
s
olvin
g mu
lt
i
-
ob
je
ct
iv
e p
r
oble
m
-
an
evoluti
onary
appro
ac
h
(
D.Va
su
m
ath
i
)
975
DE
s
te
ps
inclu
des
i
niti
al
iz
at
i
on,
m
utati
on
,
reco
m
bin
at
ion
an
d
sel
ect
ion
.
I
niti
al
iz
at
ion
par
t
a
ssig
ns
lowe
r
an
d
uppe
r
bo
und
to
e
ach
of
the
ta
r
get
va
riable
in
can
dida
te
s
of
the
popula
ti
on
a
nd
gen
e
ra
te
the
cand
i
dates
wit
hin
that
bounda
ry.
Muta
ti
on
gen
e
rates
do
nor
vecto
r
based
on
weig
hted
dif
fe
ren
ce
of
t
w
o
diff
e
re
nt
ca
nd
i
dates
a
nd
a
dding
with
a
nothe
r
ca
ndidate
,
in
wh
ic
h
t
he
m
et
ho
d
of
sel
ect
ion
of
ca
nd
i
date
de
ci
des
the
ty
pe
of
DE
var
ia
nt
as
DE
/rand
or
DE/
be
st
.
A
T
rial
ve
ct
or
is
pro
du
ce
d,
w
hich
is
de
no
te
d
by
pr
oduce
d
from
do
no
r
ve
ct
or
a
nd
ta
r
get
vecto
r.
Sele
ct
ion
m
echan
is
m
sel
ec
ts
bette
r
s
olu
ti
on
ei
th
er
f
r
om
tria
l
vector
or
ta
rg
et
vecto
r.
Af
te
r
DE
is
a
pp
li
ed
an
d
fin
al
popula
ti
on
i
s
fou
nd,
they
are
ap
plied
wi
th
in
dicat
or
functi
ons
f
or
analy
zi
ng
th
e
perform
anc
e
of
scal
arizi
ng
f
unct
ions.
Z
DT1
ben
c
hm
ark
pro
blem
has
been
us
e
d
in
this
a
na
ly
sis
with
four
di
fferent
DE
al
gorithm
ic
var
ia
nts
s
uch
as
DE/ran
d/1
/bi
n,
DE/r
and/2
/
bin
,
DE/
best/1/b
in
a
nd
DE/best/2/
bin
with
t
he
scal
a
rizi
ng
f
un
ct
i
on
s
li
ke
PBI
,
Weig
hted
su
m
,
Tc
he
b
yshe
ff,
M
od
if
ie
d
Tc
he
bysh
e
f
f
[3
]
.
The
in
dicat
or
f
un
ct
io
ns
s
uch
as
hyper
volu
m
e
an
d
Inve
rted
ge
ner
at
io
nal
distance
(
IGD)
are
us
e
d
for
an
al
yz
ing
the
pe
rfo
rm
ance
of
scal
arizi
ng
functi
ons
on
s
olv
in
g
M
OP
s
a
nd
Tch
ebys
heff
scal
arizat
io
n
functi
on
is
f
ou
nd
to
perfo
rm
b
et
te
r t
han
oth
e
r
scal
arizi
ng fun
ct
i
ons.
The
div
isi
on
of
the
rem
ai
nin
g
pap
e
r
is
as
fol
lows
sect
io
n
2
inclu
des
li
te
ra
ture
s
urvey
w
hi
ch
incl
ud
e
s
basic
c
oncepts
of
m
ulti
-
ob
j
e
ct
ive
pr
ob
le
m
,
scal
arizi
ng
f
unct
ions
s
uc
h
a
s
Tc
heb
ys
heff
(TS),
W
ei
gh
te
d
Su
m
(
WS)
,
Pe
nalty
Bo
und
In
te
r
s
ect
ion
(P
B
I),
Mod
ifie
d
Tch
ebyshe
ff
(MT
S)
,
var
ia
nts
of
D
E
al
gorith
m
s
an
d
ind
ic
at
or
f
or
perform
ance
e
valuati
on.
Sec
ti
on
3
incl
udes
e
xp
e
rim
ent
al
proce
dure
,
Sect
io
n
4
in
cl
ud
es
exp
e
rim
ental
r
esults an
d disc
us
sio
n
a
nd sect
ion
5
c
onta
in t
he
c
o
ncl
us
io
n.
2.
LIT
ERATUR
E SU
RV
E
Y
2.1.
DE vari
an
t
s
a)
DE/r
and/
p
It
is
a
gen
e
ral
s
chem
e
wh
e
re
s
olu
ti
ons
are
pi
cked
rand
om
l
y.
F
or
eac
h
so
l
ut
ion
(
)
,
w
he
re
var
ie
s
from
1
to N
ve
ct
or
y is
obta
in
ed by eq
uatio
n (3)
=
(
1
)
+
(
∑
(
(
2
,
)
−
(
3
,
)
)
=
1
)
(3)
In
t
he
a
bove
e
qu
at
io
n,
1
,
2
,
,
3
,
li
es
in
1
to
N
an
d
th
ey
are
uniq
ue
and
m
utu
al
ly
e
xclusi
ve.
F
is
a
const
ant
facto
r
within
[
0,
2].
Nu
m
ber
of
w
ei
gh
te
d
dif
fe
ren
c
e
is
giv
en
by
.
Wh
e
n
is
1,
it
i
s
DE/ra
nd
/
1
w
here
1
is ra
ndom
ly
c
ho
s
en
.
is t
he
d
onor
vector o
bt
ai
ned
i
n
m
utatio
n.
b)
DE/best/p
This
DE
sc
hem
e
w
orks
sim
il
ar
ly
to
rand/p
sc
hem
e
excep
t
th
at
val
ue
of
1
is
th
e
m
ini
m
u
m
value
of
t
he
var
ia
bles
in
s
ol
ution
for
m
ini
m
iz
at
ion
prob
l
e
m
s.
If
val
ue
of
is
cha
ng
e
d
t
o
2
t
hen
it
is
DE/best/2
.
F
or
each
so
luti
on
(
)
, whe
re
k varies
fro
m
1
to
N vect
or y i
s
ob
ta
ine
d by the e
quat
io
n (4)
=
(
)
+
(
∑
(
(
2
,
)
−
(
3
,
)
)
=
1
)
(4)
In
t
he
ab
ove e
qu
at
io
n,
1
,
2
,
,
3
,
li
es in 1
to N a
nd
t
he
y are uniq
ue
a
nd
m
utu
al
ly
exc
lusive.
C
onsta
nt
factor F,
is
within t
he ran
ge o
f 0 to
2.
is n
um
ber
o
f wei
ght
ed diffe
re
nce.
2.2.
Scala
ri
z
ing
functio
n
Scal
arizi
ng
f
un
ct
ion
s
pe
rfor
m
op
e
rati
ons
on
i
nd
i
vidual
obj
e
ct
ives
with
eac
h
ca
ndidate
of
a
po
pu
la
ti
on
and
pro
duces
a
sing
le
fitness
va
lue.
Fit
ne
ss
is
com
par
ed
am
on
g
the
par
e
nt
a
nd
child
re
n
a
nd
bette
r
s
olu
ti
on
go
es
to the ne
xt
generati
on and
ref
i
ning
happe
ns
f
or each
ge
nerat
ion
pro
duci
ng
a b
et
te
r Paret
o fron
t.
a)
Wei
ghted s
um
This
is
the
ba
sic
scal
arizi
ng
f
un
ct
io
n
use
d
on
al
m
os
t
ev
ery
scal
arizi
ng
ex
per
im
ents
on
diff
e
ren
t
pap
e
rs
[1
]
,
[
7]
and
[
8].
W
ei
gh
te
d
su
m
m
ulti
plies
weigh
t
al
ong
with
obj
e
ct
ive.
Eac
h
weig
ht
sho
uld
be
c
ho
s
en
in
a
way
that
s
um
of
wei
gh
ts
sh
oul
d
be
eq
ua
l
to
1
an
d
a
ny
weig
ht
sho
uld
li
e
between
0
a
nd
1.
It
is
ge
ne
rall
y
represe
nted by
the equat
io
n (5)
(
)
=
∑
(
)
=
1
(5)
Wh
e
re
w
is
we
igh
t a
nd f(x) is
obj
ect
iv
e f
unc
ti
on
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
9
74
–
9
81
976
b)
Tc
heby
s
hef
f
It
is
sli
ghtl
y
di
ff
e
ren
t
f
ro
m
w
ei
gh
te
d
s
um
m
et
hod
in
w
hich
fitne
ss
is
cal
c
ulate
d
with
ea
ch
ob
j
ect
ive
getti
ng
s
ub
t
rac
te
d
from
the
m
ini
m
u
m
value
of
the
obj
ect
iv
e
in
the
w
ho
le
popula
ti
on
wh
i
ch
will
be
m
ult
ipli
ed
al
ong
with
wei
gh
t
w
her
e
any
ra
ndom
weig
ht
wil
l
be
with
in
ra
nge
0
to
1[3].
Tc
he
bysh
ef
f
deco
m
po
s
it
ion
is
m
at
he
m
at
ic
a
lly give
n by the
equ
at
io
n (
6)
ma
x
=
1
,
…
[
(
(
)
−
∗
)
]
(6)
wh
e
re
represe
nt
w
ei
ght,
(
)
is o
bject
ive
functi
on a
nd
∗
is m
inim
u
m
v
al
ue
of the
obj
ect
iv
e
fu
nction.
c)
P
enalt
y
B
oun
d Intersec
ti
on (PBI
)
This
m
et
ho
d
c
al
culat
es
fitnes
s
base
d
on
s
um
of
d1
a
nd
d2
wh
ic
h
c
orres
ponds
t
o
pro
j
e
ct
ion
vecto
r
le
ng
th
(
(
)
−
∗
)
on
weigh
t
vecto
r
w
an
d pe
r
pendicula
r dist
anc
e
f
ro
m
(
)
to
w.
is
pen
al
ty
factor
w
hich
i
s
m
ul
t
ipli
ed
with
2
an
d
a
dded
alo
ng w
it
h d
1. E
quat
ion (
7) d
e
no
te
s the m
easurem
ent o
f
f
it
nes
s u
si
ng PBI
.
min
(
|
,
∗
)
=
1
+
2
s
.
t
∈
Ω
1
=
|
|
(
(
)
−
∗
)
|
|
|
|
|
|
2
=
|
|
(
)
−
(
∗
+
1
|
|
|
|
)
|
|
(7)
∗
represe
nts
m
i
nim
u
m
value
of
the
obj
e
ct
ive
functi
on.
1
m
ea
su
res
co
nver
ge
nce
a
nd
2
rep
res
ents
di
versi
ty
.
Fr
om
[7
]
we
c
om
e
to
kn
ow
that
PBI
is
able
to
co
nv
e
rg
e
c
onve
x
Paret
o
f
ront
la
rg
e
div
e
rsity
cor
res
ponds
t
o
la
rg
e
values
a
nd less
value
corres
ponds t
o
c
onve
r
gen
c
e.
d)
Modifie
d Tch
ebys
he
ff:
Accor
ding
to
[
3]
m
od
ifie
d
T
cheb
ys
heff
is
sam
e
as
weight
ed
Tche
bys
he
ff
e
xcep
t
t
hat
weig
hts
are
div
ide
d
i
ns
te
ad
of
bein
g
m
ultip
li
ed.
It is
giv
e
n
m
at
he
m
at
ic
a
l
ly
in
eq
uatio
n
(
8)
min
∈
Ω
ℎ
(
(
)
|
,
∗
)
=
ma
x
1
≤
≤
{
(
)
−
∗
}
8
)
it
is easy
to
ha
nd
le
nonlin
ear
relat
ion
s
hip
s
by
m
aking
t
his
m
od
ific
at
ion
to
conv
e
ntio
nal
Tche
bysh
e
ff
[
3]
.
3.
RESEA
R
CH
METHO
D
In
this
pa
per,
f
our
scal
a
rizi
ng
f
unct
ions
nam
ely
PBI,
Tche
bysh
e
ff
,
Mod
ifie
d
Tc
he
bysh
e
ff
an
d
Weig
hted
su
m
are
us
e
d
f
or
e
xp
e
rim
ents
with
dif
fer
e
nt
DE
va
riants.
All
t
he
e
xperim
ents
are
pe
rfo
rm
e
d
on
ZDT
1,
a
co
nv
ex
Pa
reto
f
ron
t
pro
blem
with
unif
or
m
weig
ht
0.5
a
nd
0.5
.
Cr
os
s
ov
e
r
rate,
CR
an
d
s
cal
ing
factor,
F
of
the
cro
ss
over
a
nd
m
uta
ti
on
oper
a
ti
on
s
of
D
E
al
gorithm
hav
e
be
en
set
with
0.1
and
0.5
res
pect
ively
.
Popu
la
ti
on size
is set as 100
a
nd
eac
h
ca
ndid
at
e is def
ine
d wit
h
30
decisi
on v
a
riables.
Th
e stoppin
g
crit
eria of
the
al
gorithm
has bee
n set
as
10000
functi
on
evaluati
ons. T
he t
otal
num
ber
of
r
uns
us
e
d i
s
5 f
or
al
l
va
ria
nts
of
diff
e
re
nt scala
r
iz
ing
functi
ons
, A
s
s
how
n
in
Figure
1
.
Figure
1
.
Proce
dure
for per
f
orm
ing
the e
xper
i
m
ent
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Sca
l
ar
izi
ng fu
nc
ti
on
s i
n
s
olvin
g mu
lt
i
-
ob
je
ct
iv
e p
r
oble
m
-
an
evoluti
onary
appro
ac
h
(
D.Va
su
m
ath
i
)
977
The
f
ollo
wing
F
igure
1
repre
sents
the
pr
oce
dure
use
d
for
perform
ing
the
analy
sis
.
To
c
om
par
e
the
perform
ance
in
ver
te
d
generati
on
al
distan
ce
a
nd
hyper
vo
l
um
e
has
bee
n
use
d
i
n
order
to
m
easur
e
t
he
qu
al
it
y
of
non
-
dom
inate
d
so
l
ution o
btained
in dif
fer
e
nt
v
aria
nts b
e
ca
us
e
of d
i
ff
e
ren
t
scala
rizat
ion f
un
ct
io
n.
4.
RESU
LT
S
A
ND
DI
SCUS
S
ION
The
pe
rfor
m
an
ce
res
ults
of
t
he
va
rio
us
D
E
a
lgorit
hm
s
on
t
he
m
et
rics
I
G
D
a
nd
H
V
a
re
gro
up
e
d
f
or
each scala
rizi
ng
functi
ons.
4.1.
Analysis w
ith
va
ri
at
i
on
of in
dicat
or func
ti
on
s
a)
IGD
f
or
w
ei
ghted
sum
F
igure
2
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on I
GD m
et
rics u
sing t
he sc
al
arizi
ng fu
nction we
igh
te
d s
um
.
Figure
2
.
Com
par
is
on of
DE varia
nts w
it
h I
GD m
et
rics and
weig
hted su
m
scala
riza
t
ion f
un
ct
io
n
DE/best/1
sta
r
ts
pe
rfor
m
ing
bette
r
on
init
ia
l
par
t
of
fun
ct
ion
e
valuati
on
but
D
E/best
/2/bin
sta
rt
s
perform
ing
bet
te
r
f
ro
m
functi
on
eval
uation
500.
DE/
rand/
2
a
nd
DE/
rand
/1
init
ia
ll
y
do
e
s
not
pe
rfo
rm
well
but
at
the
en
d
of
m
id
of
78
00
e
valuat
io
ns
DE/
rand/1
an
d
rand/2
sta
rts
perf
or
m
ing
well
and
at
the
en
d
of
10
000
evaluati
ons
DE
/ran
d/1
c
onve
r
ging to
wa
rd
s
the s
olu
ti
on t
ha
n othe
r DE
vari
ants.
b)
IGD
M
od
ifie
d
Tcheb
ys
he
ff
F
igure
3
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/
ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on I
GD m
et
rics u
sing t
he sc
al
arizi
ng fu
nction M
od
i
fied
Tc
he
by
sh
ef
f
Figure
3
.
Com
par
is
on of
DE varia
nts w
it
h I
GD m
et
rics and
m
od
ifie
d Tc
heb
ys
heff scal
arizat
ion
f
un
ct
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
9
74
–
9
81
978
Applyi
ng
I
GD
on
m
od
ifie
d
tc
heb
ys
heff
DE/
best/2
sta
rts
t
o
perf
or
m
bette
r
than
al
l
oth
e
r
var
ia
nts.
I
n
500
f
un
ct
i
on
e
valuati
on
DE/r
and
/
2
sta
rts
to
perform
bette
r
than
best/
1
a
nd
ra
nd
/
1.
At
40
00
e
valuati
ons
be
st/
1
and ra
nd
/
2
cl
as
hes
a
nd b
est
/
1 st
arts p
e
rfor
m
i
ng b
et
te
r
t
her
e
after. DE/ra
nd/1
perform
s b
et
te
r
ti
ll
1
000 f
un
ct
ion
evaluati
on
an
d
after
t
hat
DE/
best/1
an
d
ra
nd/2
sta
rts
pe
rfor
m
ing
bette
r.
At
the
e
nd
ov
erall
be
st/
2
pe
rfor
m
s
bette
r.
c)
IGD
Pen
alt
y B
ou
nd
In
terse
ction
F
igure
4
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on I
GD m
et
rics u
sing t
he sc
al
arizi
ng fu
nction M
od
i
fied
Tc
he
by
sh
ef
f
.
Figure
4
.
Com
par
is
on of
DE varia
nts w
it
h I
GD m
et
rics and
PB
I
Accor
din
g
to
PBI,
DE/ra
nd
/
2
sta
rts
perfor
m
ing
bette
r
i
ni
ti
al
l
y
bu
t
DE/
best/2
ov
e
rtak
es
it
at
15
00
functi
on
e
valu
at
ion
. D
E/ra
nd
/1
trie
s
to
pe
rfor
m
bette
r
tha
n
best/2
at
ar
ound
en
d
of 7
00 ev
al
uatio
ns
.
D
E/best/
2
sta
rts
perf
or
m
i
ng
bette
r
tha
n
best/1
at
7800
evaluati
on
s
a
nd
at
the
en
d
D
E/best/
2
pe
rfo
r
m
s
bette
r
accor
ding
to
IGD
on p
e
nalt
y fact
or of t
heta 1
.
0.
d)
IGD
Tc
heb
ysheff
F
i
gure
5
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on I
GD m
et
rics u
sing t
he sc
al
arizi
n
g f
unct
ion Tc
heb
ys
heff
.
Figure
5
.
Com
par
is
on of
DE varia
nts w
it
h I
GD m
et
rics and
Tc
he
bysh
e
ff
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Sca
l
ar
izi
ng fu
nc
ti
on
s i
n
s
olvin
g mu
lt
i
-
ob
je
ct
iv
e p
r
oble
m
-
an
evoluti
onary
appro
ac
h
(
D.Va
su
m
ath
i
)
979
Applyi
ng
I
GD
m
et
rics
on
Tc
he
bysh
e
ff
scal
a
r
iz
at
ion
f
un
ct
io
n,
DE/best/1
sta
rts
to
pe
rfo
rm
bette
r
f
ro
m
the
first
an
d
at
the
m
id
of
2300
functi
on
ev
al
uation
DE/be
st/
1
and
DE/
be
st/
2
ov
e
rlap
s
and
i
n
the
m
idd
le
of
3500
f
un
ct
io
n
evaluati
on
D
E/best/
2
sta
rts
to
perform
bette
r
and
at
the
e
nd
DE/best/1
perform
s
bett
er
than
al
l
oth
e
r variants
.
4.2.
Analysis w
ith
Va
ri
at
i
on
in
s
calariz
ing fun
ctions
4.2.1
Wei
ghted s
um
F
igure
6
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on
HV an
d IGD
m
et
rics u
sin
g
t
he
scal
arizi
ng fun
ct
io
n
weig
ht
ed
s
um
Figure
6
.
Com
par
is
on of
DE varia
nts w
it
h
weig
hted sum
Accor
ding
to
hy
per
vo
l
um
e
th
e
higher
one
pe
rfor
m
s
bette
r a
nd
acc
ordi
ng
t
o I
GD
lo
wer
one
pe
rfo
rm
s
bette
r.
Th
us
,
ba
sed
on
the
res
ult
obta
ine
d
a
bove
D
E/ran
d/2
pe
rfor
m
s
bette
r
in
wei
gh
te
d
s
um
and
acc
ord
ing
t
o
IGD
rand/2
p
e
r
form
s b
et
te
r.
4.2.2
Tc
heby
s
hef
f
s
calariz
at
io
n
f
unct
i
on
c
ompari
son
F
igure
7
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on
HV an
d IGD
m
et
rics u
sin
g
t
he
scal
arizi
ng
f
un
ct
io
n Tche
by
sh
ef
f.
Figure
7
.
Com
par
is
on of
DE varia
nts w
it
h T
cheb
ys
heff
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
3
,
Ma
rc
h
201
9
:
9
74
–
9
81
980
Accor
ding
to
I
GD
a
nd
hype
r
vo
l
um
e,
DE/be
st/
1
gets
great
er
value
on
hy
pe
r
volum
e
and
low
e
r
valu
e
on IGD
th
us
producin
g bett
er
resu
lt
s.
4.2.3
Modifie
d Tch
ebys
he
ff
F
igure
8
s
ho
ws
the
res
ults
of
di
ff
e
ren
t
DE
var
ia
nts
vi
z.
DE/ra
nd/1,
DE/ra
nd/2,
D
E/best/
1
a
nd
DE/best/2
, on
HV an
d IGD
m
et
rics u
sin
g
t
he
scal
arizi
ng fun
ct
io
n
M
odifie
d
Tc
he
bysh
e
f
f.
Figure
8
.
Com
par
is
on of
DE varia
nts w
it
h
Mod
ifie
d
Tc
he
bysh
e
ff
Fr
om
above
r
esult
it
is
ob
s
erv
e
d
that
on
hype
r
volum
e
DE/best/1
pe
rfor
m
s
bette
r
and
on
I
GD
DE/best/2
pe
rfor
m
s
bette
r
thus
on
e
sin
gle
D
E
var
ia
nt
can
not
be
sai
d
a
s
be
tt
er
thu
s
m
od
ifi
ed
Tc
heb
ys
heff
cannot
be
sai
d
to
p
e
rfor
m
b
et
te
r
on
a
sin
gle D
E
v
a
ri
ant.
4.2.4
PB
I
Fo
ll
owin
g
F
i
gure
9
s
hows
th
e
res
ults
of
dif
fer
e
nt
DE
var
i
ants
viz.
DE/ra
nd
/
1,
DE/
rand/
2,
DE/
best/1
and DE
/best/
2, on
HV an
d IGD m
et
rics u
sin
g
the
scala
rizi
ng
functi
on PB
I.
Accor
ding
to
the
ob
ta
i
ned
re
su
lt
on
PBI,
by
us
in
g
IGD
best/2
pe
rfor
m
s
well
a
nd
hy
per
volum
e
pro
du
ces
value
in
ne
gative
th
us
a
nythi
ng
ca
nnot
be
pre
dicte
d.
T
hu
s
c
orre
sp
on
ding
t
o
t
he
f
ound
e
xperi
m
ental
resu
lt
s,
Tc
he
by
sh
ef
f
pro
duce
s
a
bette
r
res
ul
t
as
bo
th
hype
r
volum
e
and
IGD
sho
ws
s
a
m
e
DE
var
ia
nt
i.e.
DE/best/1 a
nd
plo
t o
n
f
unct
io
n
eval
uation f
or Tche
byshe
f
f
pro
du
ces
bette
r
r
esults o
n DE/best/
1.
We con
cl
ud
e
that Tche
bys
he
ff
s
cal
arizat
ion f
un
ct
io
n per
f
orm
s b
et
te
r
on
MOP.
Figure
9
.
Com
par
is
on of
DE varia
nts w
it
h P
BI
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Sca
l
ar
izi
ng fu
nc
ti
on
s i
n
s
olvin
g mu
lt
i
-
ob
je
ct
iv
e p
r
oble
m
-
an
evoluti
onary
appro
ac
h
(
D.Va
su
m
ath
i
)
981
5.
CONCL
US
I
O
N
The
way
t
he
te
chnolo
gy
has
s
how
n
gro
wth
in
t
he
past
deca
de
is
trem
end
ous
a
nd
unbelie
vab
le
[12].
The
gro
wth
in
Com
pu
ti
ng
te
chnolo
gies,
I
oT
,
A
rtific
ia
l
I
ntell
igence,
AR/VR
et
c.
is
am
azi
ng
[
13,
14
]
.
I
f
it
is
po
s
sible
to
m
a
p
the
giv
e
n
pr
oble
m
as
op
tim
i
zat
ion
ki
nd
pro
blem
,
then
it
is
ver
y
well
that
diff
e
re
ntial
evol
utio
n
al
gorithm
can
be
us
e
d
to
s
olve
that.
S
o
al
on
gs
ide
thes
e,
Ev
olu
ti
onary
al
go
rithm
s
are
al
s
o
gr
ow
i
ng
in
pa
rall
el
to s
olv
e m
ulti
-
disci
plinary
pr
ob
le
m
s.
This
paper
e
xam
ines
the
pe
rfor
m
ance
of
four
di
ff
e
ren
t
scal
arizi
ng
f
unct
ion
s
by
us
in
g
four
D
E
al
gorithm
var
ia
nts
su
c
h
as
D
E/ran
d/1
,
DE/
r
and
/
2,
DE/best
/1
an
d
D
E/best
/2
f
or
s
olv
i
ng
MOP.
E
xperi
m
ental
resu
lt
s
sho
ws
t
he
bette
r
pe
rfo
rm
ance
of
Tc
he
bysh
e
ff
scal
ar
iz
ing
f
un
ct
io
n
as
it
produces
consi
ste
nt
resul
t
on
diff
e
re
nt
in
dica
tors
s
uch
as
I
G
D
a
nd
H
V
f
or
t
he
DE
al
go
rith
m
,
DE/best/1.
More
ov
e
r,
oth
e
r
scal
a
rizi
ng
functi
ons
do
e
s
not
produ
ce
co
ns
ist
ent
r
esults
on
diff
e
r
ent
in
dicat
ors.
In
-
de
pth
a
naly
sis
can
be
pe
rfor
m
ed
by
incl
ud
i
ng
m
or
e MOPs,
m
or
e
in
dicat
or fun
ct
io
ns an
d
m
or
e
evoluti
ona
r
y al
gorithm
s in
f
utu
re
wo
rk.
REFERE
NCE
S
[1]
W
.
Huang
H.
L
i.
“
On
the
diff
ere
n
t
ia
l
evol
u
ti
on
sch
emes
in
MO
EA/
D
,”
Proc.
6th
Int
.
Conf
.
Na
t.
Comput
(
ICNC)
2010
;
6
,
2010
.
[2]
David
hadk
a.
“
MO
EA
Fram
ew
ork:
A
Fre
e
and
Open
Sourc
e
J
ava
Fr
amework
for
Multi
ob
jective
Optimiz
ation
”
.
;
2011
-
2016.
ht
tp://m
oea
fra
m
ewor
k.
org/downloa
ds
.
html
.
[3]
Ma
E
t
Al.
“
On
T
che
b
y
ch
eff
Dec
o
m
positi
on
Appro
ac
hes
for
MO
EA
,
”
IE
EE
transact
ions
on
e
vol
ut
ion
ary
com
puta
ti
on
2018;
22(2):
[4]
S.
Tha
ng
ave
lu
&
C.
Shunm
uga
Vel
a
y
utha
m
.
“
An
inve
stiga
t
i
on
on
m
ixi
ng
h
et
ero
g
ene
ous
di
ffe
ren
t
ia
l
evo
lution
var
ia
n
ts i
n
a
d
ist
ribut
ed
fra
m
ewo
rk
Inte
rn
at
ion
al
”
.
Journal
of
Bi
o
-
Inspired
Comput
ati
on.
2015;
7
.
[5]
As
wani
V,
Prav
ee
n
V,
Th
anga
v
e
lu
S
.
“
Perform
an
ce
an
aly
sis
of
va
ria
nts
of
Diffe
r
e
nti
al
Evol
ut
ion
o
n
Mult
i
-
objecti
v
e
opti
m
iz
ation
pro
ble
m
s
”
.
Ind
ian
J
ournal
of
S
ci
en
c
e
and
Te
chnol
og
y
2015;
8(17):
[6]
David
h
ad
ka.
B
egi
n
ner's
Guide
to
th
e
MOEA
Framewo
rk.
.
2016.
htt
ps://
books.go
o
gle
.
co.i
n/books?
i
d=okUfjwE
A
CAA
J.
[7]
S.
Ji
ang,
S.
Ya
ng,
Y.
W
ang,
X.
Li
u
.
“
Scal
ar
i
zi
ng
func
ti
ons
i
n
dec
om
positi
on
-
base
d
m
ult
iobjective
evol
ut
ion
ar
y
al
gorit
hm
s
”.
I
EEE
transact
ions o
n
ev
o
lut
ionary
c
omputati
on
.
201
7.
[8]
Lot
fi
,
S. & Ka
ri
m
i,
F.
“
A h
y
br
id
MO
EA/DTS
for
Solving
Mu
lti
-
Objec
t
ive
Probl
e
m
s
”
.
Journal
of
AI
&
Data
Mini
ng.
Proc.
6th
Int
.
Co
nf
.
Nat
.
Com
put
(ICNC) 2017;
5(
2).
[9]
David
hadka.
M
OEA
Framework:
A
Fr
ee
and
O
pen
Source
Jav
a
Framework
for
Mult
iobjec
ti
v
e
O
pti
mization
.
201
1
-
2016
[10]
K.
V.
Pri
ce.
An
in
troduc
t
ion
to
diffe
ren
ti
a
l
evolution.
In
D.W.
Corne,
M.
Dori
go,
F
.
Glov
er
(e
ds),
“
New
Id
ea
s
i
n
Optimiza
ti
o
n
”.
Mc
Gr
aw
Hill
,
1
999.
[11]
(2018)
Test
fun
ct
ions
for
op
ti
m
izati
on,
Av
ai
l
able
at:
ht
tps://
en
.
wik
ipe
d
ia.
org/w
i
ki
/Test_f
un
ct
ion
s_for_optim
izati
on
(Acc
essed:
19
A
pril
2018)
[12]
Vasudeva
n,
S.K.
,
Vi
v
ek,
C
.
and
Srivat
hsan,
S.
“
An
intelligent
b
oxing
appl
i
catio
n
through
augmente
d
r
ea
l
ity
for
two
users
–
hum
an
computer
intera
ct
i
on
attempt
”.
Indi
an
Journal
o
f
S
c
ie
nc
e
and
Techn
ology
.
,
8(34)
,
20
15
[13]
Abhishek,
S.N.
,
Vasudeva
n,
S.
K.,
Nair
,
P.R
.
,
Tha
n
gav
el
u,
S.
and
Sundara
m
,
R.
M.D.
“
A
Propos
al
for
Mit
igating
Fis
her
m
en
Killin
g
In
India
n
Se
a
Border
s
through
Te
chno
log
y
–
Ma
rit
ime
Boundar
y
Ide
nti
f
ic
a
ti
on
De
vic
e
”.
Indone
sia
n
Journal
of
Elec
t
rical
Engi
ne
erin
g
and
Compute
r
Sci
en
ce
6(3), pp.704
-
710,
2017
.
[14]
Datt
a
,
B.
S.
,
Gan
apa
th
y
,
R.
,
Vasudeva
n,
S.K
.
and
Abhishek,
“
An
I
nvent
iv
e
and
Inn
ovat
iv
e
Al
te
rn
ate
for
Le
g
acy
Ch
ai
n
Pulli
ng
S
y
st
em
t
hrough
Inte
rn
et
of
Thi
ngs
”
.
S.
N
.
Indone
sian
Jour
nal
of
El
e
ct
rica
l
Engi
ne
ering
and
Computer
Scien
ce
,
6(3),
pp
.
688
-
694
,
2017
.
Evaluation Warning : The document was created with Spire.PDF for Python.