Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
3
,
June
2020
,
pp. 3
261
~
3274
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
3
.
pp3261
-
32
74
3261
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Populati
on based
optimiz
atio
n
alg
or
ithm
s
i
mp
ro
ve
ment usi
ng
the pred
ictive p
articl
es
M.
M. H.
El
ro
by
1
, S.
F.
Mek
ha
mer
2
,
H. E
. A. T
alaat
3
, an
d
M
. A.
M
ou
s
tafa. H
as
s
an
4
1
El
e
ct
ri
ca
l
Eng
in
ee
ring
Dep
ar
tment,
Fa
cul
t
y
of En
gine
er
ing,
Ain
S
hams
Univer
sit
y,
Eg
y
p
t
2,3
Ele
ct
ri
ca
l
Eng
i
nee
ring
Depa
r
tment,
Future
Univ
ersity
,
Eg
y
pt
4
El
e
ct
ri
ca
l
Eng
in
ee
rin
g
Depa
r
tment,
C
ai
ro
Univer
sit
y
,
Eg
y
pt
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
12
, 2
019
Re
vised
Dec
2
,
2019
Accepte
d
Dec
11
, 201
9
A
new
eff
ic
ie
nt
i
m
prove
m
ent
,
ca
l
le
d
Predictive
Pa
rti
cle
Modification
(PPM),
is
proposed
in
thi
s
pape
r
.
Thi
s
m
odifi
cation
m
ake
s
the
p
art
i
cl
e
look
t
o
the
ne
ar
ar
ea
bef
ore
m
oving
towar
d
the
b
est
soluti
on
of
the
group
.
Thi
s
m
odifi
ca
t
i
on
ca
n
be
appli
ed
to
an
y
popu
la
ti
on
al
gori
thm.
The
basi
c
phil
osoph
y
of
PP
M
is
expl
ai
n
ed
in
de
ta
i
l.
To
e
val
ua
te
th
e
p
erf
orm
anc
e
of
PP
M,
it
is
appl
ie
d
to
Parti
c
le
Sw
arm
Optimiza
ti
on
(PS
O)
al
gorit
hm
and
Te
a
chi
ng
Le
a
rni
ng
Based
Optim
iz
a
ti
on
(
TL
BO)
al
gorit
hm
th
en
t
este
d
using
23
standa
rd
ben
chmark
func
t
ion
s.
The
eff
ec
t
iveness
of
the
se
m
odifi
c
at
ions
are
compare
d
w
it
h
th
e
oth
er
un
m
odifi
ed
popul
a
ti
on
opt
imiza
t
io
n
al
gori
thms
base
d
on
the be
s
t
soluti
on
,
ave
r
a
ge
solut
ion, a
nd
conve
rge
n
ce ra
t
e
.
Ke
yw
or
d
s
:
Op
ti
m
iz
ation
Partic
le
Sw
a
rm
Optim
iz
at
i
on
Popu
la
ti
on
op
ti
m
iz
at
ion
Pr
e
dicti
ve
pa
rtic
le
Teachin
g Lea
r
ning Base
d
Copyright
©
202
0
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
:
M. M.
H.
El
roby,
Ele
ct
rical
En
gi
neer
i
ng D
e
par
t
m
ent,
Faculty
of E
ngineerin
g,
Ain Sham
s Un
iversity
, E
gypt.
Em
a
il
:
m
ou
sael
roby@ya
hoo.
com
1.
INTROD
U
CTION
Re
centl
y,
m
any
Me
ta
heu
risti
c
op
ti
m
iz
ation
al
gorithm
s
have
bee
n
de
velo
pe
d.
T
hese
i
nclu
de
Pa
rtic
le
Sw
arm
Op
ti
m
iz
at
ion
(
PSO)
[1
-
5
]
,
Gen
et
ic
Algorithm
(GA)
[
6
-
9
]
,
Def
e
ren
ti
al
E
vo
l
ution
(
DE)
[
10
]
,
An
t
Col
on
y
(
AC)
[
11
]
,
Gr
a
vitat
ion
al
Sear
ch
al
gorithm
(
GSA)
[
12
]
,
Sine
Cosine
Algorithm
(S
CA)
[
13
-
15
]
,
Hybr
i
d
PS
O
G
SA
Al
gorithm
[
16
]
,
A
dap
ti
ve
SCA
integ
rate
d
with
par
ti
cl
e
swar
m
[
17
]
,
a
nd
Teachi
ng
Le
arn
i
ng
Ba
sed
O
ptim
izati
on
(T
LBO
)
[
18
-
20
]
.
The
s
a
m
e
go
al
f
or
t
hem
is
to
fin
d
the
global
opti
m
u
m
.
In
order
to
do
this,
a
heurist
ic
al
go
rithm
s
hould
be
eq
uip
pe
d
with
tw
o
m
a
in
char
a
ct
erist
ic
s
to
e
ns
ure
fi
nd
i
ng
globa
l
op
ti
m
u
m
.
These
two
m
ajo
r
c
har
act
e
risti
cs
are
exp
l
or
at
i
on
a
nd
e
xp
l
oitat
ion
.
E
xp
l
or
a
ti
on
is
the
abili
ty
to
search
w
ho
le
pa
rts
of
th
e
sp
ac
e
wh
e
reas
ex
pl
oitat
ion
is
the
conve
rg
e
nce
a
bili
ty
to
the
best
so
luti
on.
T
he
go
al
of
al
l
Me
ta
he
ur
ist
ic
opti
m
izati
on
al
gorith
m
s
is
to
balanc
e
the
ab
il
it
y
of
exp
l
oitat
ion
a
nd
e
xplo
rati
on
in
ord
e
r
to
fin
d
global
op
ti
m
u
m
.
Acco
r
ding
to
[
21
]
,
exp
loit
at
io
n
and
e
xp
l
or
at
io
n
in
evo
l
ution
a
r
y
co
m
pu
ti
ng
are
not
cl
ear
due
to
la
ke
of
a
ge
ner
al
ly
acce
pted
pe
rcep
ti
on.
In
ot
her
hand,
with
stren
gth
e
ning
on
e
a
bili
ty
,
th
e
oth
er
will
weak
en
a
nd
vice
versa.
Be
cause
of
the
above
-
m
entioned
points,
the
e
xisti
ng
Me
ta
he
ur
ist
ic
op
ti
m
i
zat
ion
al
gorithm
s
are
capab
le
of
so
l
vi
ng
finite
set
of
pro
blem
s.
It
has
bee
n
pro
ve
d
that
there
is
no
al
gorit
hm
,
wh
ic
h
can
pe
rfor
m
gen
eral
e
nough
to
so
lve
al
l
opt
i
m
iz
ation
pro
bl
e
m
s
[
22
]
.
Ma
ny
hydri
de
op
t
i
m
iz
ation
al
gorithm
s
are to
b
al
a
nce t
he ov
e
rall
expl
or
at
io
n
a
nd e
xploit
at
ion abil
it
y.
In
this
stu
dy,
the
pro
po
se
d
m
od
ific
at
ion
increase
s
the
exp
l
or
at
io
n
an
d
m
ake
the
part
ic
le
loo
k
to
the
s
urrou
nd
i
ng
s
pace
be
fore
aff
e
ct
ed
by
th
e
best
so
l
ution.
The
pr
opos
e
d
m
od
ific
at
ion
c
an
be
a
ppli
ed
t
o
a
ny
popula
ti
on
opt
i
m
iz
ation
al
gorithm
s.
The
PSO
is
on
e
of
the
widely
us
e
d
popula
ti
on
a
lgorit
hm
s
du
e
to
its
si
m
plici
t
y,
converge
nce
sp
ee
d,
an
d
abili
ty
of
sea
rch
i
ng
glo
bal
opti
m
u
m
.
Re
centl
y
TLB
O
is
a
new
ef
f
ic
ie
nt
op
ti
m
iz
ation
m
et
ho
d
c
om
bi
ne
bet
ween
te
achin
g
an
d
le
arn
i
ng
phases.
Fo
r
t
he
reas
on
s
li
ste
d
a
bo
ve
this
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
32
6
1
-
32
7
4
3262
m
od
ific
at
ion
ha
s
bee
n
ap
plied
to
PSO
a
nd
TLBO.
T
he
organ
iz
at
io
n
of
t
his
pa
per
is
as
fo
ll
ows:
Sect
i
o
n
2
descr
i
bes
t
he
sta
nd
a
r
d
P
S
O
a
nd
it
s
ex
plorat
ion
pro
bl
e
m
.
Sect
ion
3
descr
i
bes
t
he
sta
nd
a
rd
TLBO.
The
pro
po
se
d
m
od
ific
at
ion
is
pr
e
sente
d
in
Sect
io
n
4
.
Sect
ion
5
des
cribes
t
he
re
s
ults
of
the
pr
opos
e
d
m
od
ific
at
ion
. Sec
ti
on
6
c
onc
lud
es
this
resea
rch.
2.
T
HE STA
N
D
ARD
P
A
RTI
CLE SW
A
R
M
O
PTIMIZ
ATIO
N
2
.1
.
P
art
ic
le
Swarm
Opt
im
iz
at
ion
Algori
th
m
PSO
is
a
popula
ti
on
com
puta
ti
on
al
gorith
m
,
wh
ic
h
is
pro
posed
by
K
enn
e
dy
a
nd
E
berhart
[
1]
.
The
PS
O
was
insp
i
red
from
s
ocial
beh
a
vior
of
bir
d
floc
king.
It
us
es
a
nu
m
ber
of
pa
rtic
le
s,
w
hich
fly
,
arou
nd
the
searc
h
s
pa
ce.
All
pa
rtic
le
s
try
to
fin
d
be
st
so
luti
on.
Me
anwhil
e,
they
al
l
loo
k
at
the
best
pa
rtic
le
in
their
paths
.
I
n
oth
er
wor
ds
,
par
ti
cl
es
co
ns
ide
r
th
ei
r
own
best
s
olu
ti
ons
a
nd
t
he
be
st
so
l
utio
n
ha
s
f
ound
s
o
fa
r
.
Each
par
ti
cl
e
in
PS
O
s
hould
con
si
der
t
he
current
posit
ion
,
the
distance
to
p
best,
the
current
velocit
y,
an
d
the d
ist
a
nce to
global be
st (
gbest
)
to m
od
ify
i
ts p
os
it
ion
.
PSO
was
m
od
el
e
d
as
foll
ow
[
1]
:
+
1
=
+
1
×
×
(
−
)
+
2
×
×
(
−
)
(1)
+
1
=
+
+
1
(
2
)
w
he
re
v
i
t
+
1
is t
he v
el
ocity
o
f pa
rtic
le
i
at
it
erati
on t,
w
is
a we
ig
hting f
unct
ion,
c
j
is a
weig
htin
g fact
or,
ra
nd is a
rand
om
n
um
ber
b
et
ween 0
and
1,
x
i
t
is t
he
c
urre
nt
po
sit
io
n of pa
rt
ic
le
i
at
it
erati
on
t,
pbe
s
t
i
is t
he
pbe
st
of
a
ge
nt
i
at
it
erati
on
t,
gbest
is the
b
e
st solutio
n so
fa
r.
The
fi
rst
par
t
of
(1)
,
pr
ovi
des
ex
plorat
io
n
abili
ty
fo
r
PSO.
The
s
ec
ond
a
nd
t
hird
par
ts
,
1
×
×
(
−
)
and,
1
×
×
(
−
)
rep
r
ese
nt
pri
vate
thi
nk
i
ng
a
nd
c
ol
la
bo
rati
on
of
par
ti
cl
es
res
pe
ct
ively
[
23
,
24
]
.
The
PS
O
is
i
niti
al
iz
ed
with
rand
om
l
y
placin
g
t
he
par
ti
cl
es
in
a
pr
ob
le
m
sp
ace
.
In
each
it
erati
on,
the
pa
rtic
le
s
velocit
ie
s
are
cal
culat
ed
us
in
g
(1).
A
fter
ve
locit
ie
s
cal
cula
ti
ng
,
the
posit
ion
of
par
ti
cl
e can
b
e
cal
culat
ed
as
(
2). T
his
proces
s w
il
l co
ntin
ue un
ti
l m
eet
ing
an
e
nd crite
ri
on.
2
.
1.
1
.
PSO
Ex
ploratio
n Pr
obl
em
The
fi
rst
pa
rt
of
(
1)
,
pro
vid
es
P
SO
e
xplorati
on
a
bili
ty.
Wh
e
n
the
a
lgorit
hm
is
st
arted
,
the
velocit
y
is
init
ia
li
zed
with
ze
ro
value.
Th
us
f
rom
Equ
at
ion
1,
the
Gl
ob
al
Be
st
Partic
le
(G
BP
)
(
i.e.
P
1
i
n
Fi
gure
1
(
a)
)
rem
ai
ns
in
it
s
place
unti
l
the
be
st
global
so
l
ut
ion
i
s
c
ha
ng
e
d
by
a
new
pa
r
ti
cl
e.
This
m
eans
th
e
global
best
par
ti
cl
e
can
no
t
exp
l
or
e
nea
r
area
beca
us
e
it
is
not
exit
ed
by
any
pa
rtic
le
.
In
a
dd
it
io
n,
pa
rtic
le
s
that
arr
ive
from
ano
th
er
places
(P2
-
P5
)
to
th
e
place
o
f
the
gl
obal
best
so
luti
on
wit
h
a certai
n veloci
ty
after
a
nu
m
ber
of ite
rati
on
m
ay
b
e d
am
ped
befor
e
r
eac
hin
g t
he o
pti
m
al
so
luti
on as
sho
wn i
n
Figure
1 (
b
)
.
T
his
ph
e
nom
eno
n wil
l be treate
d usin
g PPM i
n
Sect
io
n
2.
(a)
(b)
Figure
1
.
Parti
cles
at in
itial
and
fin
al iteratio
n
, (
a)
in
itia
l
iteratio
n,
(b
)
fin
al iteratio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Po
pu
l
ation b
ase
d op
ti
miz
atio
n alg
or
it
hms i
mp
r
ove
me
nt
…
(
M. M.
H.
Elro
by
)
3263
3.
THE
STA
NDARD
TE
ACH
ING
LE
ARNI
NG
B
AS
E
D
O
PTIMIZ
ATION
The
T
LBO
m
et
hod
is
base
d
on
the
ef
fect
of
the
te
ache
r
on
the
le
ar
ner
s
.
T
he
te
ache
r
is
c
on
si
der
e
d
a
s
a
global
be
st
le
arn
e
d
per
s
on
(
)
who
s
ha
res
hi
s
knowle
dge
with
the
le
ar
ne
rs.
The
process
of
TLB
O
i
s
div
ide
d
t
o
tw
o
phase
.
T
he
first
phase
c
onsis
ts
of
the
‘
Teacher
P
hase
’
a
nd
t
he
sec
ond
phase
c
ons
ist
s
of
the
‘Lea
r
ner
P
hase’.
T
he
‘Te
acher
Phase
’
m
eans
le
arn
i
ng
f
r
om
the
te
acher
an
d
the
‘
Learn
e
r
Ph
ase
’
m
eans
le
arn
in
g
t
hro
ugh
the
interact
io
n betwee
n
le
ar
ner
s
. TLB
O w
as m
od
el
ed
as
fo
ll
ows
[
18
]
:
3.1.
Te
ac
her
Pha
se
A
le
ar
ner
le
a
r
ns
f
r
om
te
acher
by
m
ov
in
g
it
s
m
ean
to
te
acher
value
.
Lear
ner
m
odific
at
ion
is
expresse
d
as:
3.2.
Le
ar
ner
Pha
se
A
le
ar
ner
le
a
r
ns
new
so
m
eth
in
g
if
t
he
ot
her
le
a
rn
e
r
ha
s
bette
r
know
le
dg
e
t
han
hi
m
.
Learn
er
m
od
ific
at
ion
is
expres
sed
as:
4.
PREDI
CTI
V
E PA
RTICLE
The
m
ai
n
idea
of
the
PPM
ba
sed
on
that
each
it
erati
on
the
par
ti
cl
e
sh
oul
d
look
at
it
s
near
area
a
nd
see
if
it
hav
e
a
value
be
st
than
the
GBP
or
no
t.
If
it
hav
e
val
ue
bette
r
tha
n
GBP,
it
will
be
the
GBP.
T
he
PPM
can
rem
edy
non
-
e
xiti
ng
GB
P
(P1
in
Fig
ure
1
(
a
)
)
a
nd
no
t
wait
unti
l
excit
at
ion
f
r
om
ano
the
r
pa
r
ti
cl
e.
In
a
dd
it
io
n,
it
can
im
pr
ov
e
th
e
vision
of
t
he
par
ti
cl
e
before
m
ov
em
ent
toward
GBP
a
nd
ov
e
rco
m
e
the
j
um
p
ov
e
r nar
r
ow ar
ea le
avin
g g
oloa
bal s
olu
ti
on.
Con
si
der
the
i
niti
al
values
of
the
pa
rtic
le
s
P1
to
P
5,
whic
h
are
sho
wn
in
Figure
2
.
In
the
ne
xt
it
erati
on
,
t
hese
par
ti
cl
es
will
m
ov
e
toward
P
1
(as
it
is
the
GBP
at
this
m
om
ent)
and
ta
ke
posit
ion
s
P1,
P2
to
P5
.
In
ad
diti
on,
th
e
P
3
m
ay
j
um
p
to
P
3
without
co
nver
ge
to
gbest
es
pec
ia
ll
y
wh
en
the
fitness
f
unct
i
on
ha
ve
narrow
a
rea
wi
th
hi
gh
dee
p
va
lue.
I
n
a
ddit
ion
,
t
he
P
1
sti
ll
in
it
s
posit
ion
a
s
it
is
GBP.
T
he
se
phen
om
ena
can
be
treat
ed
i
f
th
e
par
ti
cl
e
try
to
fi
nd
a
best
s
olu
ti
on
(ta
rg
et
)
from
near
are
a
befor
e
m
ov
e
to
GBP
as
show
n
i
n
Figure
3
.
T
his
can
be
done
usi
ng
the
num
erical
gr
a
dient
w
it
h
a
def
init
e
t
arg
et
.
As
su
m
e
the
fitne
ss
fun
ct
ion
(
F
)
is a
li
near
functi
on
near
t
he parti
cl
e posi
ti
on
in
m
at
rix
f
or
m
:
Figure
2
.
Par
ti
cl
es m
ov
e
m
ent
=
1
∶
,
ℎ
≠
(
)
<
(
)
+
1
=
+
(
−
)
+
1
=
+
(
−
)
+
1
.
=
[
1
+
(
0
,
1
)
]
=
(
−
)
W
her
e
is
the
m
ea
n
of
th
e
l
ea
rne
r
and
‘
’
i
s
the
glo
bal
best
(the
teac
h
er)
at
an
y
it
er
at
ion
.
=
1
∶
+
1
=
+
+
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
32
6
1
-
32
7
4
3264
Figure
3
.
Init
ia
l and ta
r
get of t
he parti
cl
e
F
=
AX
+
b
(3)
Using
nu
m
erical
g
ra
dient m
eth
od:
X
new
=
X
old
−
R
∗
dF
dX
(4)
w
he
re
=
[
∆
/
∆
1
∆
/
∆
]
=
′
X
new
is t
he
ne
w po
s
ti
on
of the
pa
rtic
le
in co
l
um
n
form
X
old
is t
he
curre
nt
po
sit
io
n of t
he parti
cl
e
R
is
the
ste
p
siz
e
∆
/
∆
is
cal
culat
ed
num
erical
l
y
near
by c
hange
only
Fr
om
(3
)
∶
F
=
AX
+
b
(5
)
F
=
AX
+
b
(6
)
Fr
om
(5)
a
nd (6)
by s
ubstract
ion
:
X
ne
w
=
−
F
old
−
F
new
A
+
X
old
(7)
w
he
re
F
old
is t
he
curre
nt
fiti
nen
ss
v
al
ue
F
n
e
w
is t
he ne
w
fiti
nenss
value
Fr
om
(4)
a
nd
(
7).
R
=
F
old
−
F
ne
w
A
∗
dF
dX
=
F
old
−
F
ne
w
(
dF
dX
)
′
∗
dF
dX
(8)
X
ne
w
=
X
old
−
F
old
−
F
ne
w
(
dF
dX
)
′
∗
dF
dX
∗
dF
/
dX
(9)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Po
pu
l
ation b
ase
d op
ti
miz
atio
n alg
or
it
hms i
mp
r
ove
me
nt
…
(
M. M.
H.
Elro
by
)
3265
If
F
i
is
the
c
urr
ent
fitne
ss
value
of
the
pa
rtic
le
an
d
F
t
is
the
ta
rg
et
fitness
of
the
pa
rtic
le
(less
t
han
gbe
st
value
)
. It
is
nice to
div
e
d
sea
r
ch
ste
p
s
to
N st
eps
as
foll
ows:
Assume
dist
=
F
i
−
F
t
(10)
f
or eac
h
ste
p
∆
X
=
dis
t
/
N
(
dF
dX
)
′
∗
dF
dX
∗
dF
dX
(11)
X
new
=
X
old
−
∆
X
(12)
The
c
o
m
plete
PPM
al
gorit
hm
bef
or
e
m
ov
ing
to
GBP
is
sh
ow
n
i
n
Ta
bl
e
1
.
I
n
ad
diti
on,
t
he
M
od
i
fied
P
SO
(MPS
O)
a
nd
Mod
ifie
d
TLB
O
(MT
LBO
)
a
re show
n
in
Ta
ble 2
an
d
T
abl
e
3
res
pecti
vel
y.
Table
1
. Gra
di
ent alg
or
it
hm
Set
pa
rti
cle gra
di
ent
pa
ra
m
et
er:
<
gb
est
=
current
po
sition
of
partic
le
=
−
Vtemp
=
0
Ex
ecute
g
ra
dia
nt
a
lg
o
rith
m
:
For
N step
=
−
∆
accord
in
g
to (12
)
=
m
ax
(
,
x
m
in
);
=
m
in
(
,
x
m
ax
);
If
F(
)
<
F(
)
=
Vte
m
p
=
∆
Els
e
=
−
2
∗
Vtemp
=
m
ax
(
,
x
m
in
);
=
m
in
(
,
x
m
ax
);
Vte
m
p
=
Vtemp
End
End
Upda
te
pa
r
ticle
p
o
sitio
n :
If
(
)
<
+
1
=
+
1
=
Vtemp
End
Table
2.
M
od
i
f
ie
d
PS
O
For
ea
ch
pa
rticle
in
itialize pa
rti
cle
End
Ch
o
o
se th
e particle
with
the b
est f
itn
ess
valu
e
o
f
all
th
e particles
as th
e gb
est
Do
For
ea
ch
pa
rticle
Up
d
ate particle
p
o
sitio
n
acc
o
rdin
g
to
+
1
=
+
2
×
×
(
−
)
+
1
=
+
+
1
g
radien
t algo
rith
m
as sh
o
wn
in
Table
1
End
For
each p
arti
cle
Calcu
late f
itn
ess
valu
e
If
the f
itn
ess
valu
e is better than
the b
est
f
itn
ess
valu
e (
p
b
es
t)
in
his
to
ry set cur
rent
v
alu
e as the new p
b
est
End
Ch
o
o
se th
e particle
with
the b
est f
itn
ess
valu
e
o
f
all
th
e particles
as th
e gb
est
While
m
ax
i
m
u
m
it
eration
s o
r
m
in
i
m
u
m
e
rr
o
r
criter
ia
is no
t attai
n
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
32
6
1
-
32
7
4
3266
Table
3
. M
od
i
f
ie
d
TLB
O
For
each p
arti
cle
in
itialize pa
rti
cle
End
Ch
o
o
se th
e particle
with
the b
est f
itn
ess
v
alu
e of
all
th
e particles as
the g
b
est
Do
1
)
Teac
h
e
r
p
h
ase
=
[
1
+
(
0
,
1
)
]
=
(
−
)
=
1
∶
+
1
=
+
g
radien
t algo
rith
m
as sh
o
wn
in Table.
1 f
o
r
i
+
1
.
2
)
learner
ph
ase
=
1
∶
,
ℎ
≠
g
radien
t algo
rith
m
as sh
o
wn
in T
ab
le
1 f
o
r
i and
j
(
)
<
(
)
+
1
=
+
(
−
)
+
1
=
+
(
−
)
g
radien
t algo
rith
m
as sh
o
wn
in T
ab
le
1 f
o
r
i
+
1
.
Ch
o
o
se th
e particle
with
the b
est f
itn
ess
valu
e of all
th
e parti
cles as
the g
b
est
While
m
ax
i
m
u
m
i
teration
s o
r
m
in
i
m
u
m
er
ror c
riter
i
a is
no
t attained
5.
E
X
PERI
MEN
TAL RES
UL
TS A
ND DIS
CUSSIO
N
The
sta
nd
a
rd
PSO,
P
SOS
GSA,
SC
A,
TLBO,
M
PS
O,
a
nd
MTL
BO
with
the
par
am
et
er
in
Table
4
[
25
-
28
]
hav
e
e
xecu
t
ed
30
inde
pe
ndent
r
uns
over
each
be
nch
m
ark
f
unct
io
n
f
or
sta
ti
sti
cal
a
naly
sis.
As
s
how
n
in
T
able
5
,
M
PSO
and
MTL
B
O
outpe
rfor
m
ed
al
l
of
the
ot
her
a
lgorit
hm
s
with
re
gard
t
o
t
he
qual
ity
of
the
s
olu
ti
on
s
for
al
l
fu
ncti
on
s
.
I
n
co
ntras
t,
the
oth
e
r
al
gorithm
s
pr
od
uc
ed
poor
res
ults
on
certai
n
fun
ct
ion
s
and
acc
ur
at
e
r
esults
on
ot
hers.
T
his
fi
nd
i
ng
ref
le
ct
s
t
he
e
f
fici
ent
pe
rfo
rm
ance
of
the
M
PSO
an
d
MT
L
BO
in
com
par
ison
wi
th
the
oth
e
r
unm
od
ifei
ed
al
gorithm
s.
In
a
dd
it
io
n
,
Fig
ur
e
4
t
o
Fig
ure
1
1
s
how
a
com
par
is
on
betwee
n
MPS
O
an
d
MTL
B
O
an
d
al
l
the
oth
e
r
al
gorith
m
s
fo
r
the
c
onve
r
gen
ce
rate
fo
r
t
he
fitnes
s
ver
s
us
the
it
erati
on
s.
These
fi
gures
s
how
that
MPS
O
an
d
MTL
BO
outpe
rfor
m
s
al
l
the
oth
er
unm
od
ifei
ed
al
gorithm
s
in term
s o
f
t
he c
onve
rg
e
nce
s
peed with
an a
ccur
at
e s
olu
ti
o
Table
4.
Algori
thm
s p
aram
et
e
r
Alg
o
r
it
h
m
P
a
ram
e
ter
PSO
C
1
=
C
2
=
2
w
d
a
m
p
=
0
.9
P
S
OG
S
A
G0=1
,
C
1
=
0
.5
,
C
2
=
1
.5
S
C
A
a
=
2,
r2
=
(2
*
p
i)*r
a
n
d
,
r3
=
2
*
ran
d
,
r4
=
r
a
n
d
TLBO
T
F
=
ran
d
i([1
2]
)
MP
S
OA
C
1
=
C
2
=
2
,
wda
m
p
=
0
.9
,
N=
5
MT
L
B
O
G0=1
,
C
1
=
0
.5
,
C
2
=
1
.5
,
N=
5
Ma
x
Ve
lo
c
it
y
=
0
.2*
(
Va
rMax
-
Va
rMi
n
)
,
Min
Ve
lo
c
it
y
=
—
M
a
x
Ve
lo
c
it
y
Table
5
. Be
nc
hm
ark
fu
nctio
ns
Fu
n
ct
i
o
n
n
Ran
g
e
PSO
PSO
G
SA
SCA
T
L
BO
MPSO
MT
L
BO
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Po
pu
l
ation b
ase
d op
ti
miz
atio
n alg
or
it
hms i
mp
r
ove
me
nt
…
(
M. M.
H.
Elro
by
)
3267
Table
5
. Be
nc
hm
ark
fu
nctio
ns
(
c
on
ti
nue
)
Fu
n
ct
i
o
n
n
Ran
g
e
PSO
PSO
G
SA
SCA
T
L
BO
MPSO
MT
L
BO
Figure
4
.
C
onve
rg
e
r
at
e c
urve
s for
F1 t
o F3
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
32
6
1
-
32
7
4
3268
Figure
4
.
C
onve
rg
e
r
at
e c
urve
s for
F1 t
o F3
(
con
ti
nue
)
Figure
5
.
Co
nverg
e
r
at
e c
urve
s for
F4 t
o F6
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Po
pu
l
ation b
ase
d op
ti
miz
atio
n alg
or
it
hms i
mp
r
ove
me
nt
…
(
M. M.
H.
Elro
by
)
3269
Figure
5
.
Co
nverg
e
r
at
e c
urve
s for
F4 t
o F6
(
con
ti
nue
)
Figure
6
.
Co
nverg
e
r
at
e c
urve
s for
F7 t
o F9
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
32
6
1
-
32
7
4
3270
Figure
7.
Co
nverg
e
r
at
e c
urve
s for
F10 t
o
F
12
Figure
8
.
Co
nverg
e
r
at
e
c
urve
s for
F13 t
o
F
15
Evaluation Warning : The document was created with Spire.PDF for Python.