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.
4
,
A
ugus
t
2020
,
pp.
44
26
~
44
35
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ij
ece
.
v10
i
4
.
p
p4
426
-
44
35
4426
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Discr
ete
penguin
s search
op
timizati
on alg
or
i
thm to s
olve
flow sho
p schedu
lin
g prob
lem
Ilya
ss
M
z
il
i
1
, M
oha
mmed E
ssa
id
Rif
fi
2
, F
at
ih
a
Be
nz
ak
r
i
3
1
,2
Depa
rtment
of
Com
pute
r
Sci
en
ce
,
Facu
lty
of
Sc
ie
nc
es,
Chou
aï
b
Doukkali
Univ
er
sit
y
,
Moroc
co
3
Depa
rtment of
Mat
hematics,
Fa
cul
t
y
of
Sci
ence, Choua
ïb
Doukkali Unive
rsit
y
,
Morocc
o
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
30
, 2
019
Re
vised
Dec
2
6
,
2019
Accepte
d
Fe
b 24
, 202
0
Flow
shop
sche
duli
ng
prob
lem
is
one
of
t
he
m
ost
cl
assi
ca
l
NP
-
har
d
opti
m
iz
ation
pro
ble
m
.
W
hic
h
aim
s
to
find
the
b
est
pla
nn
ing
th
at
m
ini
m
iz
es
the
m
ake
span
(t
ota
l
completion
ti
m
e)
of
a
set
of
ta
sks
in
a
set
o
f
m
ac
hine
s
with
c
ert
a
in
co
nstrai
nts.
In
this
pape
r
,
we
pro
pose
a
n
ew
na
t
ure
inspir
ed
m
et
ahe
urist
ic
to
solve
the
f
low
shop
sche
dul
in
g
proble
m
(FSSP
),
ca
lle
d
pengui
ns
sea
r
ch
opti
m
iz
a
ti
on
a
lgori
thm
(PeSO
A)
base
d
on
c
oll
abor
at
iv
e
hunti
ng
strateg
y
of
pengui
ns.T
h
e
oper
a
tors
and
par
amete
r
v
al
ue
s
of
PeSOA
red
efi
n
ed
to
sol
ve
th
is
proble
m
.
Th
e
p
erf
orm
a
nce
of
th
e
peng
uins
sea
rch
opti
m
iz
ation
a
lg
orit
hm
is
te
sted
on
a
set
of
benc
hm
ark
s
insta
nces
of
F
SSP
from
OR
-
Li
bra
r
y
,
Th
e
r
esult
s
o
f
the
te
sts
show
that
PeSO
A
is
superior
to
som
e
othe
r
m
eta
heur
isti
cs
al
gor
ithm
s,
in
te
rm
s
of
the
qualit
y
o
f
th
e
soluti
ons
found
and the e
x
ec
ut
ion
t
ime.
Ke
yw
or
d
s
:
Co
m
bin
at
or
ia
l
op
ti
m
iz
ation
Flow
s
hop
sc
hedulin
g
pro
bl
e
m
PeSOA
Ma
kesp
a
n
Me
ta
heu
risti
cs
NP
-
Ha
rd
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
:
Ily
ass Mzi
li
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce, Fa
culty
of Sciences
,
Chouaib
Do
ukkali U
niv
e
rsity
,
El Jadida
, M
orocco
.
Em
a
il
:
dr
.m
zili
.il
ya
ss@g
m
ai
l.
com
1.
INTROD
U
CTION
The
Sc
he
du
li
ng
is
a
branc
h
of
t
his
operati
on
al
researc
h
in
pro
duct
ion
m
anag
em
ent
t
hat
ai
m
s
to
i
m
pr
ove
the
ef
fici
ency
of
a
c
om
pan
y
in
te
r
m
s
of
pro
du
ct
i
on
costs
an
d
de
li
ver
y
tim
es.
Sche
du
li
ng
pr
ob
le
m
s
app
ea
r
in
al
l
areas
of
the
e
conom
y:
co
m
pu
te
rs,
c
on
st
ruct
ion
(
pro
j
ect
m
anag
em
ent),
industry
(
wor
kshops
pro
blem
s,
pr
od
uction
m
anag
e
m
ent),
adm
inist
rati
on
(
sche
du
le
)
.Th
e
flo
w
s
hop
prob
le
m
scheduli
n
g
[
1]
is
on
e
of
t
he
m
os
t
diff
ic
ult
c
om
bin
at
or
ia
l
opti
m
izati
on
prob
le
m
belo
ngin
g
to
NP
-
ha
rd
pro
bl
e
m
[2
]
fam
ily
,
it
is
widely
know
n
in
the
industr
y.
The
so
l
utio
n
to
the
prob
l
e
m
in
find
in
g
an
orde
r
f
or
e
xecu
ti
on
of
ta
sk
s
on
m
achines s
ubj
e
ct
to
m
any constrai
nt
s in
an o
pti
m
al
tim
e.
Ov
e
r
the
la
st
few
deca
des,
na
ture
has
been
a
so
ur
ce
of
in
sp
irat
io
n
f
or
m
any
m
et
aheu
ri
sti
cs,
wh
ic
h
has
bee
n
intr
oduce
d
to
s
olv
e
op
ti
m
iz
a
ti
on
pro
blem
s.
A
s
et
of
these
m
e
ta
heurist
ic
s
has
been
te
ste
d
t
o
so
l
ve
discrete
pr
ob
le
m
s
.
The
resu
lt
s
of
t
hese
te
st
s
are
not
uni
que,
the
qual
it
y
of
s
olu
ti
ons
var
ie
s
acc
ordi
ng
t
o
the
char
act
eri
sti
cs
and
cat
egory
of
m
et
ho
d.
Ge
ner
al
ly
,
m
et
ho
ds
bas
ed
on
par
ti
cl
e
popu
la
ti
on
swar
m
intel
li
gen
ce
al
gorithm
s
pr
ovidin
g
so
l
ution
s
of
good
qual
it
y,
fo
r
e
xam
ple:
bat
a
lgo
rithm
(BA)
[
3,
4],
pa
rtic
le
swar
m
op
ti
m
izati
on
(
PS
O)
[5]
,
arti
fici
al
bee
colo
ny
(
ABC)
[6
]
,
cat
s
war
m
op
ti
m
iz
ation
(
CSO)
[
7,
8],
hunti
ng
search
al
gorit
hm
(H
uS)
[9
]
,
el
ep
han
t
he
rd
i
ng
opti
m
izati
on
(
EH
O
[
10
]
,
swall
ow
swa
rm
op
ti
m
iz
at
ion
(S
S
O)
[11],
golde
n
ba
ll
al
gorithm
(G
BA
)
[12,
13
]
,
c
uc
koo
sea
rch
(C
S)
[
14]
,
chick
en
swa
rm
al
go
rithm
(CSA)
[
15
-
16]
and
flo
we
r
poll
inati
on
al
gorithm
(F
PA)
[17
]
.
In
this
con
te
xt
we
pro
pose
d
a
new
m
et
aheu
risti
c
of
s
wa
rm
intel
l
igence
i
ns
pi
red
by
nat
ur
e
nom
inate
PeS
OA
,
to
s
olv
e
the
f
lo
w
s
hop
sc
he
duli
ng
prob
le
m
on
e
of
the
NP
-
ha
rd
pro
blem
in
c
om
bin
at
or
ia
l
op
ti
m
iz
ation
.
PeSOA
is
an
op
t
im
iz
at
ion
te
chn
iq
ue
in
sp
ire
d
by
the
hu
nting
str
at
egy
of
pe
ngui
ns
,
wh
ic
h
wa
s
devel
oped
t
o
deal
with
opti
m
iz
at
ion
prob
l
e
m
s
in
the
co
nt
inu
ous
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
Discret
e p
e
ngui
ns
se
ar
c
h op
ti
miza
ti
on
algori
thm
t
o
s
olve fl
ow
shop sc
hedul
ing
pro
blem
(
Il
yass
M
zil
i
)
4427
case.
Re
centl
y,
this
al
go
rit
hm
has
been
use
d
to
so
l
ve
di
screte
prob
le
m
s
[18
-
20]
be
cause
of
the
s
at
isfie
d
resu
lt
s i
n
the
c
on
ti
nu
ous case
.
In
this
pa
per,
we
pro
pose
an
adap
ta
ti
on
of
PeSOA
al
gorithm
to
so
lve
FSSP
.
I
n
this
one,
we
ada
pt
the
par
am
et
ers
val
ues
a
nd
operat
or
s
Pe
S
O
A
to
fin
d
good
-
qual
it
y
so
luti
on
s
.
The
pe
rfor
m
ance
of
P
eSO
A
al
gorithm
eval
uated
by
a
te
st
on
a
set
of
be
nch
m
ark
s
of
FSSP
f
r
om
the
OR
-
Lib
rar
y
of
va
rio
us
siz
e
s
and
a
com
par
ison
with
oth
e
r
m
étah
eu
risti
cs.
The
orga
nizat
io
nal
structu
re
of
this
pap
e
r
as
fo
ll
o
w
s:
The
seco
nd
sect
ion
is
con
t
ai
ns
a
sh
ort
de
scriptio
n
of
flo
w
shop
sc
he
duli
ng
pro
blem
.I
n
the
thir
d
sect
ion
PeS
O
A
al
gorithm
are
presente
d.I
n
the
f
ourth
s
ect
ion
,a
da
ptati
on
of
Pe
SOA
to
so
lve
t
he
F
SSP.
T
he
ex
pe
rim
ental
resu
lt
s
and
discuss
i
on
in
the
fifth
sect
ion. Finall
y, a
co
nc
lusio
n
a
nd p
e
r
sp
ect
ives.
2.
FLOW S
HOP
SCHED
ULI
NG PR
OBL
EM SE
ARCH
2
.
1.
Present
at
i
on
The
FSSP
is
a
n
im
po
rtant
iss
ue
in
pr
oduction
sc
hedulin
g.
This
pro
blem
was
first
pro
posed
i
n
19
54
by
Johnso
n
[
21]
and
cl
assi
fied
as
a
NP
-
hard
prob
le
m
.
The
FSSP
is
descr
i
bed
by
a
set
n
j
obs
J
=
{
1,
.
..,
n}
not
relat
ed,
m
us
t
be
treat
e
d
in
m
m
achines
in
M
=
{
1,...,
m
}.
Each
j
ob
j
c
on
ta
in
s
a
s
equ
e
nce
of
op
erati
on
s
O
=
{O
j
i,
j
∈
{1,
.
..,
n}
a
nd
i
∈
{1,...
,
m
}}
wh
ic
h
ha
ve
be
en
exec
ute
d
in
a
giv
e
n
or
der.
Fo
ll
owin
g
thi
s
orde
r,
each
operati
on
Oij
m
us
t
be
per
f
or
m
ed
on
a
sp
eci
fied
m
achine
k
f
or
a
s
pe
ci
fied
tim
e
Pijk
.
Eac
h
m
achi
ne
can
on
ly
process
one
job
at
a
tim
e,
and
eac
h
job
has
to
go
th
rough
each
m
achine
on
ce
a
nd
only
on
ce.
T
he
FSSP
consi
sts
in
fin
ding
a
schedul
e
to
per
f
or
m
op
erati
ons
on
th
e
m
achines
that
m
ini
m
iz
es
the
C
m
ax
(
m
akesp
a
n)
,
ie
the tim
e req
uire
d
to
p
e
rform
a
ll
j
obs.
2
.
2
.
Formul
at
i
on
FSSP
a
re
of
te
n
desig
nated
by
the
,
n,
π
and
m
ax
sy
m
bo
l
s
wh
e
re
re
presents
the
num
ber
of
j
obs
;
is t
he
num
ber
of
m
achines,
π = (
1
,
2
.
..
)
is a
plan
ning
pe
rm
utati
on
of a
ll
j
obs a
nd
m
ax
is t
he
m
akesp
an
.
Let
t
(
,
)
(
1
≤
≤
,
1
≤
≤
m
)
the
process
ing
ti
m
es
of
jo
b
on
t
he
m
ac
hin
e
,
as
su
m
i
ng
that
th
e
pr
e
par
at
io
n
tim
e
fo
r
each
j
ob
is
zer
o
or
include
d
in
t
he
w
orkin
g
ti
m
e
treatm
ent,
;
π
=
(
1
,
2
,
...
,
n
)
is
a
pl
ann
i
ng
per
m
utati
on
of
al
l
jobs.
Π
is
the
overall
plan
ning
of
pe
rm
utati
on
.
(
,
)
is
the
c
om
pleti
on
ti
m
e
of
the
j
ob
on
t
he
m
achine
.Th
e
c
om
pl
et
ion
tim
e
of
the
flo
w
s
hop
pro
blem
planning
π
=
(
1
,
2
,
...
,
)
is
il
lus
trat
ed
as foll
ows
:
(
1
,
1
)
=
(
1
,
1
)
,
(1)
(
,
1
)
=
(
−
1
,
1
)
+
(
,
1
)
,
=
2,
3,…,
,
(2)
(
1
,
)
=
(
1
,
−
1
)
+
(
1
,
)
,
=
2, 3, …
,
,
(3)
(
,
)
=
ma
x
{
(
−
1
,
)
,
(
,
−
1
)
}
+
(
1
,
)
,
=
2,
3,
,
= 2,
3,
. . . ,
,
(4)
∗
= a
r
g
{
m
a
x (
)
=
(
,
)} →
m
in,
∀
∈
Π,
(5)
w
he
r
e
π
*
is
t
he
m
os
t
app
r
opr
ia
te
arr
an
gem
e
nt
w
hich
is
the
pur
pose
of
t
he
per
m
utati
on
of
fl
ow
sho
p
prob
le
m
to f
i
nd
m
ax
(
π *) as t
he
m
ini
m
al
co
m
po
ne
nt.
We
co
ns
i
der
th
e
flow
sho
p
sc
hedulin
g
pr
ob
l
e
m
:
in
a
car
paint
factor
y.
We
cal
culat
ed
the
durati
on
of
paintin
g
of
two
cars
(
j
obs
),
knowin
g
that
car
paintin
g
is
a
sequ
ence
of
two
oper
at
io
ns
,
de
greasi
ng
and
paintin
g
(m
ac
hin
es
).
T
he
F
igure
1
prese
nt
s
the
tim
e
req
uire
d
to
de
gr
e
ase
and
pa
int
the
two
cars
1
and
2.
The
pro
blem
data
is
tw
o
job
s
,
tw
o
m
achine
a
nd
the
com
p
le
ti
on
ti
m
e
of
eac
h
job
on
eac
h
m
achine.
The
opti
m
al
prob
le
m
so
luti
on is car2
-
car
1,
the co
m
pleti
on
tim
e
(
C
m
ax)
is 1
2.
The
F
ig
ure 2
show
s the
us
e of
a g
a
ntt cha
rt to
calc
ulate
the
c
om
pleti
on
tim
e
an
d
the
p
la
nning t
he
im
ple
m
entat
ion
of the
so
luti
on.
De
greasi
ng
Painti
ng
Ca
r1
5
3
Ca
r2
4
4
Figure
1. The
processi
ng tim
e of the
cars
p
a
int
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
.
4
,
A
ugus
t
2020
:
4426
-
4435
4428
TASK
NA
ME
TI
M
E
1
2
3
4
5
6
7
8
9
10
11
12
Deg
reasing
Car2
Car1
Pain
tin
g
Car2
Car1
Figure
2
.
G
a
ntt cha
rt
3.
PeSO
A ALG
ORI
TH
M
The
PeS
OA
a
lgorit
hm
is
a
n
ew
m
et
aheu
r
ist
ic
[2
2]
natu
re
-
in
sp
ire
d
ba
sed
on
the
co
ll
abo
rati
ve
huntin
g
beh
a
vi
or
of
pe
ngui
ns
.
The
hu
nting
strat
egy
of
pe
nguins
ba
s
e
on
c
ollabor
at
e
their
ef
f
ort
s
an
d
synch
ronize
t
he
ir
di
ves
t
o
optim
iz
e
the
gl
ob
al
e
ne
rg
y
i
n
the
process
of
c
ollec
ti
ve
hu
nting
a
nd
nutr
it
ion
.
Pengui
ns
div
i
de
int
o
gr
oups
to
sta
rt
hunting.
As
peng
uin
s
eat
fis
h,
t
he
y
are
di
ving
to
ha
rv
e
st
f
ood
un
ti
l
ox
y
gen rese
r
ve
s ar
e
de
plete
d.
The
huntin
g
st
rategy
of
pe
nguin
s;
eac
h
gro
up
of
peng
uin
s
sta
rts
sear
chi
ng
in
a
s
pecific
po
sit
io
n
a
nd
ran
dom
le
vels,
the
pe
ngui
n
of
eac
h
gro
up
looks
for
foo
ds
in
ra
ndom
way
an
d
in
divi
du
al
ly
in
it
s
gro
up,
and
a
fter
a
n
a
ppr
ox
im
at
e
nu
m
ber
of
di
ves,
the
pe
nguin
s
get
bac
k
on
t
he
ic
e
to
sh
a
r
e
with
it
s
aff
il
ia
te
s.
I
f
the
am
ou
nt o
f
f
ood
i
n
a
posit
ion
is
no
t
en
ough
f
or
the g
r
oup,
p
art
of
the
group
m
igrates
to
ano
t
her
p
osi
ti
on.
The gr
oup t
hat
at
e the m
os
t fish gives
u
s
the l
ocati
on of t
he r
ic
h
f
ood re
pr
es
ented by t
he
hole
an
d t
he
le
vel
.
The
PeS
O
A
al
gorithm
is
per
fo
rm
ed
thr
ough
a
set
of
pe
ngui
n
popula
ti
on
(
rand
om
init
ia
l
so
luti
ons
).
In
t
he
nex
t
st
ep,
t
he
am
ou
nt
of
fis
h
c
onsu
m
ed
is
the
obj
ect
ive
f
unct
ion
relat
e
d
to
eac
h
m
e
m
b
er
of
the
popula
ti
on.
The
op
ti
m
al
value
of
the
obj
ect
iv
e
f
un
ct
i
on
pr
e
sente
d
the
locat
io
n
w
her
e
th
ere
is
a
la
rg
e
a
m
ou
nt
of
f
ood,
w
hich
wa
s
gr
a
nted
b
y
the
gr
ou
ps
.
T
o
find
a
locat
io
n
of
en
ough
f
ood
(b
e
st
so
luti
ons
)
,
peng
uin
s
m
ov
e to a ne
w
locat
i
on ide
ntifie
d.
This m
ov
em
ent ex
presse
d
t
his for
m
ula f
ollo
ws
;
=
+
×
|
−
|
(6)
w
he
re
ra
nd
is
a
rand
om
nu
m
ber
for
distrib
ut
ion
;
an
d
we
ha
ve
th
ree
s
olu
t
ion
,
X
bestlocal
the
best
local
s
olu
ti
on,
X
id
the last
s
ol
ution an
d
t
he X
new
is a
new s
olu
ti
on.T
he ps
eudo
c
ode
of the Pe
SOA al
gorithm
:
Algori
thm 1
:
P
seudo
-
code
o
f
P
eSOA
1.
Generate random population P of the initial solutions
(penguins);
2.
Initialize the probability of existence of fish in the
holes
-
levels;
3.
While (the stop condition is not satisfied) do
4.
For ea
ch i individual of P do
5.
While (RO2> 0) do
6.
-
Take a random step
7.
-
Improve the penguin position using Eq (6);
8.
-
Update quantities of fish eaten for this penguin.
9.
End while
10.
End for
11.
Update quantities of eaten fish in the holes
-
levels.
12.
Update the best group
.
13.
Recalculate the probability of existence of fish in the
holes
-
levels
14.
Update best
-
solution
15.
End while
16.
Return best solution.
4.
APPL
Y
PeS
O
A
TO
FLOW
SHOP
S
CHE
DU
LI
NG PR
OBL
EM
We
a
pp
li
e
d
P
eSO
A
on
FS
S
P
to
im
pr
ove
the
res
ults
a
nd
to
s
how
it
s
perform
ance
against
othe
r
m
et
aheu
risti
cs.
The
PeS
OA
as
al
l
m
e
ta
heurist
ic
s
introdu
c
ed
init
ia
ll
y
fo
r
con
ti
nu
ou
s
optim
iz
at
ion
,
PeSOA
sta
nd
a
rd
c
on
ti
nuous
co
ding
schem
e
cann
ot
be
us
ed
di
re
ct
ly
to
so
lve
discrete
pro
bl
e
m
s
li
ke
flow
sh
op
pro
blem
,
we
adap
t
t
he
al
go
rithm
par
am
et
ers
an
d
the
FS
SP
so
luti
on
form
i
n
the
sp
a
ce
a
nd
the
tra
ns
it
ion
from
the curre
nt s
ol
ution t
o
a
nothe
r on
e
.
4
.
1.
FSSP solu
tion
The
posit
io
n
of
peng
uin
prese
nts
a
so
luti
on. Th
ere
f
or
e, PeSOA
c
onside
red
a
pengu
i
n
in
a
popula
ti
on
as
a
s
olu
ti
on
i
n
the
searc
h
s
pa
ce.
Co
ns
ide
ri
ng
an
instanc
e
of
FSS
P,
for
e
xam
ple
T
able
1
that
c
onta
ins
4
job
s
and 3 m
achines,
a
nd each
jo
b o
per
at
io
n
is
as
so
ci
at
ed wit
h
it
s m
achine and
processi
ng tim
e.
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
Discret
e p
e
ngui
ns
se
ar
c
h op
ti
miza
ti
on
algori
thm
t
o
s
olve fl
ow
shop sc
hedul
ing
pro
blem
(
Il
yass
M
zil
i
)
4429
Table
1.
FSS
P
instance
(4
×
3)
M1
M2
M3
J1
6
1
4
J2
3
6
2
J3
1
2
1
A
so
l
ution
of
FSSP
insta
nce
is
a
per
m
utati
o
n
of
job
s
that
can
be
prese
nt
ed
by
a
series
of
n
intege
rs
(Jobs
)
S
=
{
1,
2,
3...
n},
eac
h
intege
r
i
in
th
e
series
is
desi
gn
at
e
d
th
e
in
de
x
of
jo
b
a
nd
the
order
of
i
i
n
S
is
the
order
of
processin
g
of
the
se
j
obs
in
each
m
achine.
In
our
e
xam
ple,
con
side
r
an
init
ia
l
so
luti
on
of
in
sta
nce
(4.3)
2
-
1
-
4
-
3
wh
ic
h
re
pr
ese
nts
the
orde
r
of
processi
ng
of
the
4
jo
bs
on
eac
h
m
a
chine.
For
ex
a
m
ple,
the
m
achines
sta
rts
processi
ng
jo
bs
2
an
d
e
nd
s
with
job
s
3.
C
on
si
der
i
ng
Table
1
a
nd
the
init
ia
l
so
luti
on,
w
e
can
desig
n
the
Ga
ntt
char
t
a
s
show
n
i
n
Fi
gure
3
of
t
he
i
niti
al
so
luti
on
(2
-
1
-
4
-
3)
t
o
ca
lc
ulate
the
ob
je
ct
ive
functi
on (
Cm
ax) o
f pro
po
se
d solutio
n. The
m
akesp
a
n of
t
he
instance
so
l
ut
ion
is
21.
Machin
es
Ti
m
e
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
M1
2
1
4
3
M2
2
1
4
3
M3
2
1
4
3
Figure
3. Ga
ntt cha
rt of initi
al
so
l
ution 2
-
1
-
4
-
3 (S=
(2
-
1
-
4
-
3))
4
.
2
.
Positi
on
u
pd
ati
ng o
f
Pe
SO
A
The
tran
sit
ion
from
o
ne
so
luti
on
to
a
no
t
her
i
n
the searc
h
spa
ce is an o
per
a
ti
on
b
ase
d
on the notion
of
le
ng
th
a
nd
to
polo
gy.
T
he
le
ngth
is
desig
nat
ed
as
the
c
os
t
of
s
olu
ti
on,
the
topolo
gy
of
pa
ssage
is
a
te
c
hn
i
qu
e
perform
ed
by t
he op
e
rato
rs of
(
7
)
=
⊕
⊗
(
−
)
(7)
w
he
re
ra
nd
is
a
ran
dom
nu
m
ber
f
or
distrib
ution
;
an
d
we
hav
e
th
ree
so
l
ution,
the
best
local
so
luti
on
(S
best
),
the curre
nt s
ol
ution (
S
id
)
a
nd
the n
e
w sol
utio
n (S
new
).
The o
per
at
io
ns i
n
(7)
def
i
ned
as foll
ows:
a)
Substr
uction o
per
at
i
on
−
An
ope
rati
on
be
tween
t
wo
s
olu
ti
ons(sc
he
du
li
ng)
S
1
an
d
S
2
,
wh
ic
h
giv
es
a
disp
la
cem
ent
ve
ct
or
Q.
in
this way,
ex
tra
ct
s the
per
m
utati
on
s a
ppli
ed
t
o
the
tw
o
s
olu
t
ion
s
S
2
to
obtai
n
S
1
,
for e
xam
ple:
Wh
et
her
S
1
=
{j
1
, j
2,
j
3,
j
4
…
j
n
-
1,
j
n
}
a
nd S
2
=
{j
2
,
j
3,
j
1,
j
4
…
j
n,
j
n
-
1
}
Q= S
1
-
S
2
t
he
n Q =
{(
j
1
,
j
2
),
(
j
2
,
j
3
),
…, (
j
n
,
j
n
-
1
)}
In o
t
her
w
ords S
1
-
S
2=
Q
S
1
⊕
Q=S
2
b)
Additi
on ope
ra
ti
on
⊕
An
operati
on
betwee
n
tw
o
di
ff
ere
nt
var
ia
bl
es,
a
s
olu
ti
on
S
2
an
d
a
vecto
r
Q
,
giv
e
a
sol
ution
S
new
.
Is
a
n
a
p
plica
ti
on
of
the
pe
r
m
uta
ti
on
s
of
t
he
vecto
r
Q
t
o
the
so
l
utio
n
S
2
f
or
obta
in
a
ne
w
s
olu
ti
on
S
new
,
for
e
xam
ple:
Wh
et
her
S
2
=
{j
1
, j
2,
j
3,
j
4
, j
5
…
,
j
n
}
a
nd
Q
= {(j
3
,j
1
)
,(
j
4
,j
3
)
,(
j
2
,j
5
)}
S
new
=S
2
⊕
Q
th
en
S
new
=
{j
4
, j
5,
j
1,
j
3
, j2,…,
j
n
}
c)
Mult
ipli
cat
ion
op
e
r
at
io
n
⊗
An
op
e
rati
on
betwee
n
a
real
num
ber
r
(
r
[
0
,
1
]
)
and
disp
la
cem
ent
vecto
r
Q.
The
m
ai
n
ro
le
of
this
op
e
rati
on
is
to
reduce
the
nu
m
ber
of
per
m
utati
on
s
of
the
vecto
r
Q
acc
or
ding
to
t
he
val
ue
of
r
.
We
co
ns
ide
r
a d
is
placem
ent
v
ect
or
Q of n
c
ouple
of p
e
rm
utati
on
Q
=
{
(c
1
, c
2),
(c
3,
c
4)
, (c
5,
c
6)
…
,
(c
2n
-
1
,x
2n)
}
an
d r
eal
n
um
ber
0≤
r
≤
1
.
Q’
=
r×Q
T
hen
m
=r ×
n ≤
n,
Q
’
=
{(c
1
,
c
2),
(c
3,
c
4)
…
,
(c
2m
-
1
, c
2m
)
}
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
.
4
,
A
ugus
t
2020
:
4426
-
4435
4430
The
Fi
gure
4
i
ll
us
trat
es
the
use
of
the
oper
at
or
s
of
(
7
)
to
m
ov
e
it
from
the
c
urren
t
sol
ution
S
Id
t
o
a n
e
w
s
olu
ti
on
S
new
.
Figure
4. Illust
rati
on of P
e
SOA
op
e
rato
rs funct
ion.
4
.
3
.
The PeSO
A
_ FSSP
Algori
t
hm
As
any
al
gorithm
,
the
first
st
ep
in
the
PeS
OA
al
gorithm
is
to
init
ia
l
iz
e
so
m
e
necessary
par
am
et
ers
for
good
an
d
e
ff
ic
ie
nt
operati
on
of
the
al
go
r
it
h
m
.
In
the
pr
opos
e
d
al
gorit
hm
,
a
few
par
a
m
et
ers
are
to
be
set.
Steps
of the
pr
opos
e
d
al
gorithm
(
PeSOA)
a
re
pr
ese
nted
b
e
low.
Step
1:
I
niti
al
i
zat
ion
-
I
niti
al
iz
e a po
pu
la
ti
on size
p
-
G
ene
rate ra
ndom
p
op
ulati
on P
of t
he
init
ia
l solutio
ns
S
(
sche
du
le
s
).
-
I
niti
al
iz
e Ro2 an
d
the
g
l
obal
so
l
ution S
gbest
Step
2
: C
al
cula
te
co
m
pleti
on
tim
e (C
m
ax)
fo
r
al
l sol
utio
ns
i
n
P
Step
3:
Ca
lc
ula
te
S
best
the
best sch
e
dule
in p
Step
4
: F
or eac
h
s
olu
ti
on
(sc
he
du
le
s
),
c
al
c
ul
at
e h
er
n
e
w ver
s
ion usi
ng
(
7).
Step
5
: Im
pr
ov
ing
of
popula
ti
on so
l
utio
ns
usi
ng
desce
nt m
e
thod.
Step
6
: U
pdat
e S
best
an
d
S
gbest
as foll
ows:
If
(Cm
ax(
Sid)
< Cm
ax
(S
best
)) The
n
S
best
=Ji
d.
If
(Cm
ax
(S
best
)
< Cm
ax
(S
gbes
t
))
T
he
n
S
gbest
= S
best
;
Step
7
: Sto
p
it
e
ra
ti
on
s
if st
op
conditi
on is sat
isfed.
Re
turn to St
ep 4, othe
r
wise.
The
s
olu
ti
on
of
the
pro
blem
is
the last
S
gbest
. Fi
gure
5
s
hows
the
flo
wch
a
rt
of the
pro
pose
d
al
go
rithm
.
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
Discret
e p
e
ngui
ns
se
ar
c
h op
ti
miza
ti
on
algori
thm
t
o
s
olve fl
ow
shop sc
hedul
ing
pro
blem
(
Il
yass
M
zil
i
)
4431
Figure
5. Flo
w
char
t
of the
pro
po
s
ed
alg
or
it
hm
(
PeSO
A
_F
S
SP)
5.
E
X
PERI
MEN
TAL RES
UL
TS A
ND DIS
CUSSIO
N
In
t
his
sect
ion,
we
eval
uated
the
PeS
OA
by
31
ben
c
hm
a
rk
s
i
ns
ta
nces
f
ro
m
the
OR
-
L
ibrar
y
[
23
],
us
in
g
c
la
nga
ge
.
T
he
res
ults
of
this
e
xperi
m
ent
are
extra
ct
ed
from
a
com
pu
te
r
with
In
te
l
(R)
Core
(TM)
2
process
or
D
ual
CPU
2.00G
Hz
M370@
2.4
0G
HZ
2.40
G
Hz
and
4.0
0
GB
of
RAM.
Eac
h
i
ns
ta
nces
is
te
st
ed
10
tim
es
with
the
program
.
Experim
ental
resu
lt
s
and
a
naly
sis
are
div
i
ded
i
nt
o
th
ree
par
ts.
First,
in
sect
io
n
1,
we
e
valuate
t
he
e
ff
ect
of
im
po
rtant
pa
ra
m
et
ers
on
the
pro
posed
al
gorithm
.
In
sec
ti
on
2
,
we
present
the co
m
pu
ta
ti
on
al
r
esults o
f
the PeS
OA
alg
or
it
hm
w
hen
a
pp
li
ed
to s
om
e
instances of th
e O
R
-
li
br
a
ry and
fina
l
sect
ion
;
we
c
om
par
e
our
re
su
lt
s
with
othe
r
m
et
aheu
risti
cs
to
stu
dy
the
perform
ance
an
d
the
e
ff
i
ci
ency
of PeS
OA.
5
.
1.
Par
amet
ers
of
P
eSO
A
_FS
S
P
The
al
gorithm
perform
ance
app
ea
r
in
the
good
set
ti
ng
of
the
pa
ram
et
e
rs.
S
o,
at
first
we
m
ake
a
regulat
ion
of
the
value
s
of
the
key
par
am
eter
s
f
or
t
he
obta
ining
a
n
ef
fecti
ve
ada
ptati
on.
The
F
ig
ure
6
s
how
s
the
res
ults
of
us
e
Hel
1
(
100
x
10
)
inst
an
ce
of
OR
-
li
br
a
ry
to
te
st
the
eff
ect
of
par
am
et
ers
PeSOA.
This
insta
nce
con
ta
in
s
10
0
job
s
f
or
10
-
m
achine.
We
i
ncrea
se
the
po
pula
ti
on
siz
e
of
t
he
al
gorithm
ran
gi
ng
from
10
to
60
,
an
d
m
ark
ed
the
qual
it
y
of
the
so
luti
on
found.
T
he
op
ti
m
al
so
luti
on
a
pp
ea
re
d
in
p
oi
nt
40.
Ther
e
f
or
e,
we
set
the
popula
ti
on size at
40 fo
r
this e
xperim
e
nt.
In
F
ig
ur
es
7
an
d
8
w
e
use
d
tw
o
insta
nce
Hel
1(1
00
x
10)
a
nd
Re
C7
(
20
x
10),
t
o
fin
d
the
good
value
of
par
am
et
er
RO2
.
W
e
va
r
y
the
value
s
of
RO
2
f
r
om
0
to
30
a
nd
m
ark
s
t
he
dev
ia
ti
on
of
th
e
le
ng
t
h
of
the
so
luti
on
f
ound
ove
r
the
le
ng
t
h
of
the
be
st
so
luti
on,
and
ti
m
e
execut
ion
s.
At
point
10,
we
f
ound
good
so
luti
ons
qual
it
y i
n
reas
onable
tim
e. So
, t
he v
al
ue
of R
o2 m
us
t
be
fi
xed it
10 for t
his e
xp
erience.
Cost
=
le
ngt
h
of
th
e
sol
ution
foun
d
−
le
ngt
h
of
th
e
op
t
ima
l
sol
ution
(8)
The val
ues
of t
he param
et
ers
us
e
d
in
this e
xperim
ent are
s
how
n
in
Ta
ble
2.
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
.
4
,
A
ugus
t
2020
:
4426
-
4435
4432
Figure
6.
T
he
im
pact o
f
inc
re
asi
n
g t
he PeS
O
A pop
ulati
on
s
iz
e (M)
Table
2.
T
he
P
aram
et
ers
of
P
eSO
A
Para
m
eters
Meanin
g
Valu
e
Gen
e
Maxi
m
u
m
nu
m
b
er
of
iter
atio
n
s
1000
M
Po
p
u
latio
n
size
40
RO2
o
x
y
g
en
r
eserve
10
5
.
2.
C
omp
u
t
ati
on
al re
sults
The
nu
m
erical
resu
lt
s
of
t
hi
s
ada
ptati
on
a
re
prese
nted
i
n
T
a
ble
3.
T
he
fi
rst
col
umn
c
on
ta
ins
the
nam
e
of
th
e
instance
‘
I
nst
ance’,
the
col
um
n
‘n
×m
’
re
pr
ese
nts
n
job
for
m
m
ac
hin
e.
T
he
c
olu
m
n
‘
BKS
’
sh
ow
the
best
resu
lt
pro
pos
ed
by
the
ot
he
r
al
gorithm
s.
The
col
um
n
‘Best
’
and
‘
Wo
rst
’
presents
the
best
resu
lt
s
a
nd
th
e
worst
res
ults
ob
ta
ine
d
by
th
e
app
li
cat
io
n
of
th
e
PeS
OA
m
et
ho
d
f
or
the
sel
ect
ed
in
sta
nce
.
The
col
um
ns
‘Aver
a
ge
’
co
nt
ai
n
inform
at
i
on
on
a
ver
a
ge
,
the
colum
n
‘P
Da
v(%)
’
de
no
te
s
the
per
c
entage
dev
ia
ti
on
of
th
e
aver
a
ge
of
le
ng
t
h
of
so
l
utio
ns
f
ound
ove
r
the
le
ng
t
h
of
the
opti
m
a
l
so
luti
on
i
n
10
r
un
s
an
d
the last
co
l
um
n
‘Tim
e’ s
hows
the av
e
ra
ge
e
xe
cution t
im
e in
seco
nds
for
th
e 10
runs .
PDav
(
%
)
=
BKS
−
A
v
e
ra
g
e
BKS
×
100%
(9)
Table
3
prese
nts
the
com
puta
ti
on
al
resu
lt
s
of
PeS
OA
al
gorithm
on
31
instances
of
OR
-
li
brary.
The
col
um
n
‘P
Dav(%
)’
w
hic
h
ta
kes
the
value
0.0
0
ndic
at
ed
in
bold
wh
e
n
al
l
the
so
luti
on
s
fou
nd
i
n
10
te
st
s
are
e
qu
al
with
the
le
ngth
of
t
he
best
known
so
luti
on,
an
d
t
he
value
le
ss
t
han
0.0
0
in
bold
-
blu
e
,
if
the
a
ver
a
ge
of
s
olu
ti
ons
found
on
al
l
the
t
est
s
is
le
ss
than
the
le
ng
t
h
of
the
best
know
n
s
olu
ti
on,
in
t
his
case
the
s
ol
utions
fou
nd
a
re
bette
r
than
t
he
best
known
s
olu
ti
on.
the
r
est
;
26%
of
the
value
s,
are
le
ss
tha
n
0.
5%
,
w
hich
m
eans
that
the b
est
s
ol
ution
foun
d,
on
the
10
te
sts,
ap
pr
ox
im
at
es
l
ess
tha
n
0.5% o
f
t
he
best
-
known
so
l
utio
n
.t
he
16%
of
t
he
val
ues,
are
m
or
e
0.5%
,
w
hich
m
eans
that
the
PeS
OA
does
not
r
each
the
best
-
know
n
so
l
utio
n
or
nea
r
so
luti
on
to
this
so
luti
on
i
n
the
10
te
st.
The
num
erical
values
in
T
a
ble
3
s
how
t
hat
PeS
O
A
is
of
gr
eat
a
bili
ty
to
pro
vid
e
high
qual
it
y solutio
ns i
n reaso
na
ble t
i
m
e.
Figure
7. T
he
im
pact o
f
cha
ngin
g
the
v
al
ue
of R
O2
for
s
olut
ion
s
Hel1
inst
ance
Figure
8. T
he
im
pact o
f
cha
ngin
g
the
v
al
ue
of RO
2
f
or s
ol
utions Re
C7
in
sta
nce
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
Discret
e p
e
ngui
ns
se
ar
c
h op
ti
miza
ti
on
algori
thm
t
o
s
olve fl
ow
shop sc
hedul
ing
pro
blem
(
Il
yass
M
zil
i
)
4433
Table
3.
N
um
e
rical
s r
es
ults o
btain
Ins
tan
ce
m
x
n
BKS
Bes
t
W
o
rst
Av
erage
PDav
(%)
Ti
m
e
(S)
Car1
1
1
x 5
7038
7038
7038
7038
0
.00
0
.11
Car2
1
3
x4
7166
7166
7166
7166
0
.00
0
.07
Car3
1
2
x 5
7312
7312
7312
7312
0
.00
0
.08
Car4
1
4
x 4
8003
8003
8003
8003
0
.00
0
.08
Car5
1
0
x 6
7720
7720
7720
7720
0
.00
0
.07
Car6
8
x 9
8505
8505
8505
8505
0
.00
0
.07
Car7
7
x 7
6590
6590
6590
6590
0
.00
0
.04
Car8
8
x 8
8366
8366
8366
8366
0
.00
0
.06
Hel1
1
0
0
x 1
0
516
515
516
5
1
5
,5
-
0
.09
4
0
.58
Hel2
2
0
x 1
0
136
135
136
1
3
5
,5
-
0
.36
7
.21
ReC
1
2
0
x 5
1247
1247
1247
1247
0
.00
0
.64
ReC
3
2
0
x
5
1109
1109
1109
1109
0
.00
1
.17
ReC
5
2
0
x 5
1242
1245
1245
1245
0
.24
0
.14
ReC
7
2
0
x 1
0
1566
1566
1566
1566
0
.00
2
.26
ReC
9
2
0
x 1
0
1537
1537
1537
1537
0
.00
2
.19
ReC
1
1
2
0
x 1
0
1431
1431
1431
1431
0
.00
1
.97
ReC
1
3
2
0
x 1
5
1930
1930
1936
1933
0
.15
9
.45
Re
C1
5
2
0
x 1
5
1950
1950
1950
1950
0
.00
7
1
.58
ReC
1
7
3
0
x 1
5
1902
1902
1902
1902
0
.00
5
.17
ReC
1
9
3
0
x 1
0
2093
2099
2099
2099
0
.29
1
7
.89
ReC
2
1
3
0
x 1
0
2017
2046
2046
2046
1
.44
1
5
.45
ReC
2
3
3
0
x 1
0
2011
2020
2020
2020
0
.45
8
8
.46
ReC
2
5
3
0
x 1
5
2513
2525
2530
2
5
2
7
,5
0
.57
3
2
5
.23
ReC
2
7
3
0
x 1
5
2373
2379
2384
2
3
8
1
,5
0
.36
8
0
.30
ReC
2
9
3
0
x 1
5
2287
2298
2298
2298
0
.48
4
3
9
.38
ReC
3
1
5
0
x 1
0
3045
3053
3061
3057
0
.39
8
6
.88
ReC
3
3
5
0
x 1
0
3114
3118
3118
3118
0
.13
1
4
0
.56
ReC
3
5
5
0
x 1
0
3277
3277
3277
3277
0
.00
0
4
.58
Re
C3
7
7
5
x 2
0
4951
5062
5080
5071
2
.42
4
6
8
.70
ReC
3
9
7
5
x 2
0
5087
5180
5184
5182
1
.86
1
5
9
.33
ReC
4
1
7
5
x 2
0
4960
5079
5118
5
0
9
8
.5
2
.79
2
3
6
.54
5
.
3
.
Co
m
pa
ri
s
on
w
ith other
me
ta
h
euri
stic
To
furthe
r
pr
e
sent
the
pe
rfo
rm
ance
of
t
he
pro
posed
Pe
SOA,
a
com
par
iso
n
was
m
ade
betwee
n
the
pr
opos
e
d
al
gorithm
with
res
ults
of
E
M,
an
d
GA
li
ste
d
in
the
w
ork
of
Y
ua
n,
Hen
e
quin,
W
a
ng,
a
nd
Gao
(20
06
)
[24]
an
d
PS
O
-
E
DA
i
n
the
work
of
H
ongc
he
ng
Liu
et
al
.
(20
11)
[
25
]
a
nd
GB
A
in
the
wor
k
F.
Say
oti, M.E.
Ri
ff
i )
(
2016)
[
26
]
,
u
si
ng
di
ff
e
ren
t i
ns
ta
nces
of O
R
-
li
br
a
ry.
The res
ults are
li
ste
d
in T
a
ble
4.
Table
4.
C
om
par
iso
n of PeS
O
A wit
h diff
e
re
nt alg
or
it
hm
s: GA, EM,
G
B
A
and PSO
-
E
D
A
Pro
b
lem
PDav (%
)
Ins
tan
ce
n
×
m
BKS
PeSOA
EM
GA
PSO
-
E
DA
GBA
Car1
1
1
x 5
7038
0
0
0
0
0
Car2
1
3
x4
7166
0
0
0
0
0
Car3
1
2
x 5
7312
0
0
1
.19
0
0
Car4
1
4
x 4
8003
0
0
0
0
0
Car5
1
0
x 6
7720
0
0
0
0
0
Car6
8
x 9
8505
0
0
0
0
0
Car7
7
x 7
6590
0
0
0
0
0
Car8
8
x 8
8366
0
0
0
0
0
Hel1
1
0
0
x 1
0
516
-
0
.09
-
-
-
2
.92
6
3
Hel2
2
0
x 1
0
136
-
0
.36
-
-
-
1
.76
4
7
R
ec1
2
0
x 5
1247
0
4
.41
6
.96
0
.09
6
0
.17
6
4
Rec3
2
0
x 5
1109
0
1
.98
4
.45
0
.03
6
0
.25
2
Rec5
2
0
x 5
1242
0
.24
2
.01
3
.82
0
.24
2
0
.30
5
9
Rec7
2
0
x
1
0
1566
0
3
.70
5
.31
0
0
.88
1
2
Rec9
2
0
x
1
0
1537
0
3
.97
4
.73
0
.20
2
2
.01
0
4
Rec1
1
2
0
x
1
0
1431
0
4
.05
7
.39
0
.12
6
2
.06
1
4
Rec
13
2
0
x1
5
1930
0
.15
4
.87
5
.97
0
.22
3
2
.03
6
2
Rec1
5
2
0
x 1
5
1950
0
3
.79
4
.29
0.
303
2
.31
2
8
Rec1
7
3
0
x 1
5
1902
0
5
.57
6
.08
0
.28
9
2
.35
5
4
Rec1
9
3
0
x 1
0
2093
0
.29
5
.97
6
.07
0
.61
2
2
.88
5
8
Rec2
1
3
0
x 1
0
2017
1
.44
3
.22
6
.07
1
.40
8
2
.64
2
5
Rec2
3
3
0
x 1
0
2011
0
.45
5.
97
7
.46
0
.59
7
3
.02
8
3
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
.
4
,
A
ugus
t
2020
:
4426
-
4435
4434
This
com
par
is
on
is
base
d
on
the
per
c
enta
ge
dev
ia
ti
on
,
i.e
.
the
qual
it
y
of
so
luti
on
f
ound
com
par
e
d
to
be
st
-
kn
own
so
l
utio
n
acc
or
ding
in
OR
-
li
brary.
W
hi
ch
is
confirm
ed
by
Figure
9
i
nd
ic
at
ing
t
he
lo
wer
cu
rv
e
associat
ed
wit
h
PeS
OA.
He
n
ce,
PeS
O
A
is
m
or
e
appropri
at
e
to
be
adap
te
d
to
so
l
ve
FSSP
a
nd
it
can
give
good
resu
lt
s.
In
T
a
ble
4
,
the
exp
erim
ental
resu
lt
s
of
PeS
OA
al
gorithm
are
com
par
ed
with
the
res
ults
of
the
EM,
GA,
PS
O
-
ED
A
an
d
GBA
m
et
hods
.
t
hese
r
esults
cl
early
sh
ow
that
Pe
S
OA
out
perf
orm
s
the
oth
er
E
M,
GA,
PSO
-
E
DA
a
nd
GB
A
al
gori
thm
s
in
reso
l
ut
ion
of
al
l
twenty
-
tw
o
te
ste
d
instance
s
.T
he
pro
pose
d
P
eSO
A
al
gorithm
gets
17
22
Be
st
s
olu
ti
ons
w
hile
EM
/
G
A
/
PS
O
-
ED
A
/
GBA
on
ly
ge
ts
8
20
/
7
20
/
9
20
/
8
22
of
the
bes
t
so
luti
ons
a
m
on
g
22
or
20
insta
nces.
F
urt
her
m
or
e,
we
fin
d
that
the
a
ve
rag
e
of
PD
a
v
(%)
for
the
E
M
/
GA
/
PS
O
-
EDA
/
GBA
al
gorith
m
s
is
equ
al
to
2,4
75
/
3,489
/0,
206
/
1,1
65,
w
hile
our
a
lgorit
hm
PeSOA
is
e
qual
to
0,096.
Figure
9
sho
ws
the
st
ud
y
of
t
he
P
Dav
(
%)
va
riat
ion
bet
wee
n
the
P
eSO
A
al
gorith
m
and
the
E
M,
G
A
,
PSO
-
E
DA
an
d
GBA
al
gorith
m
s
fo
r
the
tw
el
ve
in
sta
nces
of
diff
e
ren
t
s
iz
es.
I
n
F
i
gure
9,
the
l
ow
e
r
curve
associat
ed
with
the
PeS
O
A
al
gorithm
is
b
et
te
r,
in
te
rm
s
of
the
qual
it
y
of
the
s
olu
ti
on.
T
his
can
e
xpla
ine
d
the
perform
ance
and
ef
fici
ency
of
FS
S
J
reso
luti
on
PeSOA
al
gori
thm
in
te
r
m
s
of
the
qual
it
y
of
the s
olu
ti
ons
f
ound.
Figure
9. The
per
ce
ntage
d
e
vi
at
ion
for t
he
P
eSO
A
a
nd o
t
he
r
m
et
aheu
risti
cs
6.
CONCL
US
I
O
N
In
t
his
pa
pe
r,
we
pr
opos
e
d
a
swar
m
-
intel
li
gen
ce
al
g
or
it
hm
,
so
-
cal
le
d
P
eSO
A
to
s
olve
flo
w
s
hop
sche
du
li
ng
pr
oble
m
.
This
adap
ta
ti
on
c
on
sist
s
of
a
dd
i
ng
a
set
of
cha
nges
to
the
sta
nd
a
r
d
ve
rsion
of
P
eSO
A
reg
a
rd
i
ng
po
sit
ion
re
pr
ese
ntat
ion
an
d
it
s
upda
te
equ
at
ion,
as
well
as
the
op
erat
or
s
of
this
equ
at
io
n.
Be
side
s,
a
local
search
m
echan
ism
was
us
e
d
to
im
pr
ov
e
resear
ch
st
rategy.
T
he
Pe
SOA
al
gorith
m
was
app
li
ed
on
a
set
of
ben
c
hm
ark
instances
of
F
SSP
f
ro
m
the
OR
-
Li
br
a
ry.
It
has
bee
n
s
hown
t
hat
the
effe
ct
iveness
as
well
as
perform
ance
of
PeS
OA,
ha
s
reache
d
that
of
t
he
othe
r
te
ch
niques
s
uch
a
s
GA,
EM,
GB
A
an
d
P
SO
-
E
DA
i
n
te
rm
s
of
the
qu
al
it
y
of
so
luti
ons
an
d
th
e
com
pu
ti
ng
ti
m
e.
F
uture
w
ork
is
to
exte
nd
t
he
P
eSO
A
ap
plica
ti
on
for
oth
e
r
com
bin
at
or
ia
l
optim
iz
ation
pr
ob
le
m
s
s
uch
as o
the
r
scheduli
ng
m
od
el
s,
log
ist
ic
network
m
od
el
s,
veh
ic
l
e
routin
g
m
od
el
s
,
et
c.
We
will
al
so
try
t
o
im
pr
ove
t
he
PeS
O
A
al
gorithm
in
orde
r
t
o
obta
in
e
ve
n
bette
r
r
esults
by h
y
br
i
dizat
ion
with
oth
e
r
m
et
hods
or b
y m
od
i
ficat
ion
s
of
the p
a
ram
et
ers
or the al
gorith
m
o
per
at
ors
.
REFERE
NCE
S
[1]
N.
Gupta
,
et
al
.
,
"F
lowshop
sch
edul
ing
r
ese
ar
ch
aft
er
five
d
ecad
es,
"
European
J
ournal
of
Operational
Re
search
,
vol.
169
,
no
.
3
,
p
p.
699
-
711
,
200
6.
[2]
M.
R.
Gar
e
y
,
et
al
.,
"Th
e
compl
exi
t
y
of
flowshop
and
jobshop
s
che
dul
ing,
"
Mat
hemati
cs
of
ope
rations
research
,
vol.
1
,
no
.
2
,
pp
.
117
-
129,
1976
.
[3]
Y.
Saji,
et
al
.,
"
Discre
te
ba
t
-
ins
pire
d
al
gori
thm
for
tra
v
el
l
ing
sa
l
esm
an
proble
m
,
"
In
Se
cond
Wor
ld
Conf
ere
nce
on
Complex
S
yste
m
s,
2014.
WCCS
2
014,
IE
EE
,
pp
.
2
8
-
31
,
2014
.
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
Discret
e p
e
ngui
ns
se
ar
c
h op
ti
miza
ti
on
algori
thm
t
o
s
olve fl
ow
shop sc
hedul
ing
pro
blem
(
Il
yass
M
zil
i
)
4435
[4]
M.
E.
R
iffi
,
et
a
l
.,
"Incorporatin
g
a
m
odifi
ed
u
niform
cro
ss
over
an
d
2
-
ex
cha
n
ge
nei
ghborhoo
d
m
ec
hani
sm
i
n
a
discre
t
e
bat
a
lgori
thm
to
solve
the
quadr
at
i
c
assignm
ent
proble
m
,
"
Egy
pt
ia
n
Informatic
s
Journal
,
vol.
18
,
pp
.
221
–
232
,
20
17.
[5]
X.
H.
Shi
,
et
a
l
.,
"P
artic
le
sw
a
rm
opti
m
iz
at
ion
-
base
d
al
gor
it
h
m
s
for
TS
P
and
gene
ra
li
z
ed
T
SP
,
"
Information
proce
ss
ing
lette
r
s
,
vol. 103, no. 5
,
pp
.
169
-
176
,
2
007.
[6]
D.
Kara
boga
an
d
B.
Gorkem
li
,
"
A
combinat
oria
l
artifi
ci
a
l
bee
col
on
y
al
gor
it
h
m
for
tra
vel
ing
sale
sm
an
problem
,
"
In
Inte
rnat
ional
Symposium on
I
nnovat
ions
in
Int
el
li
g
ent Sy
stems
and
Applications,
2011
.
IE
EE
,
pp
.
50
-
53
,
2011
.
[7]
A.
Bouzi
di,
and
M.
E.
Riffi
,
"
Im
prove
d
CS
O
to
Solve
the
TSP
,
"
In
Inte
rnational
Confe
re
nce
on
Adv
anc
e
d
Inte
lligen
t
Syst
e
ms
for Su
stainab
le
De
velopment
,
2018.
Springer
,
Cham
,
pp.
252
-
2
60
,
2018
.
[8]
A.
Bouzi
d
i,
et
a
l
.,
"
Cat
sw
arm
opti
m
iz
ation
for
solving
th
e
op
en
shop
sche
duli
ng
proble
m
,
"
Journal
of
Industria
l
Engi
ne
ering
Int
e
rnational
,
vol. 1
5,
no
.
2
,
pp
.
367
-
378,
2019
.
[9]
A.
Agharghor,
et
a
l
.,
"
Im
prove
d
hunti
ng
se
arch
al
gori
thm
for
the
qu
adr
a
ti
c
a
ss
ignment
proble
m
,
"
Indone
sia
n
Journal
of
Elec
t
rical
Engi
ne
erin
g
and
Computer
Sci
en
ce
(
IJE
ECS
)
,
vol.
14
,
no
.
1
,
pp.
143
-
154
,
20
19.
[10]
A.
Hos
sam
,
et
al
.,
"
E
le
ph
ant
s
he
rding
opti
m
iz
a
tion
for
solving
t
he
tra
v
el
li
ng
sa
l
esm
an
proble
m
,
"
In
Inte
rnation
al
Confe
renc
e
on
A
dvanc
ed
Int
el
l
i
g
ent
S
yste
ms
for S
ustainabl
e
Dev
elopm
ent
,
Spring
e
r,
Cham
,
pp
.
122
-
130
,
2018
.
[11]
S.
Bouzi
di
,
and
M.
E.
Riff
i,
"D
i
scre
te
sw
allow
sw
arm
opti
m
iz
at
i
on
al
gorit
hm
f
or
tra
ve
ll
ing
sa
le
s
m
an
proble
m
,
"
In
Proceedi
ngs
of
th
e
2017
Inte
r
nati
onal
Con
fe
re
nce
on
Smar
t
Di
git
al
Environm
ent
,
ACM
,
pp.
80
-
84
,
2017
.
[12]
F.
Sa
y
o
ti
,
et
al
.,
"O
pti
m
iz
at
ion
o
f
m
ake
span
in
jo
b
shop
sche
duli
n
g
proble
m
b
y
go
lde
n
ba
ll
al
gori
t
hm
,
"
Indone
sian
Journal
of
Elec
t
rical
Engi
ne
erin
g
and
Computer
Sci
en
ce
(
IJE
ECS
)
,
vol.
4
,
no
.
3
,
p
p.
542
-
547
,
201
6.
[13]
K.
Rutt
a
n
at
e
erawichi
en
,
et
al
.,
"
An
improved
golde
n
ba
ll
a
lgori
t
hm
for
the
ca
pa
ci
t
at
ed
v
ehi
c
le
r
outi
ng
proble
m
,
"
In
Bio
-
Inspired
Computing
-
Theories
and
App
li
ca
ti
ons.
Spring
er,
Be
rlin
,
He
ide
lb
e
rg
,
pp.
341
-
356,
2014.
[14]
A.
Ouaa
rab
,
e
t
a
l
.
,
"
Discr
ete
cu
c
koo
sea
rch
al
gor
it
hm
for
the
tra
v
el
li
ng
sale
sm
an
Problem,
"
Neural
Computing
an
d
Appl
ic
a
ti
ons
,
vo
l.
24
,
no
.
7
-
8
,
pp
.
1659
-
1669
,
20
14.
[15]
S.C.
B.
Sem
la
l
i,
et
al
.,
"P
ar
al
l
el
h
y
brid
ch
ic
ken
s
warm
opti
m
iz
ati
on
for
solving
th
e
quadr
at
i
c
assignm
ent
proble
m
,
"
Inte
rnational
Jo
urnal
o
f El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE)
,
vol
.
9
,
no
.
3,
pp.
2064
-
207
4
,
2019
.
[16]
H.
Ye,
et
al
.
,
"A
n
Im
prove
d
Chic
ken
Sw
arm
Optimiza
t
ion
for
T
SP
,
"
in
Inte
rnati
onal
Confe
ren
ce
on
Appl
ic
a
ti
ons
and
Techni
qu
es
in
Cyb
er
Se
curity
and
In
telli
g
ence.
Spring
er,
Cha
m
,
pp.
211
-
220
,
2019
.
[17]
D.
P.
Sam
dea
n,
et
al
.
,
"F
lower
Polli
na
ti
on
Al
gorit
hm
(FP
A)
to
Solve
Quadr
at
i
c
As
signm
ent
Problem
(QA
P
)
,"
Conte
mpor
ary
Mathe
mat
ic
s an
d
Applications
,
v
ol.
1
,
no
.
2
,
pp
.
5
8
-
67,
2019
.
[18]
I.
Mzi
li
and
M.
E
.
Riff
i,
"D
iscr
ete
pengui
ns
se
arc
h
op
ti
m
izati
on
al
g
orit
hm
to
solv
e
t
he
tr
ave
l
ing
sal
e
sm
an
proble
m
,
"
Journal
of
Theoretical
&
Appl
ie
d
Information
Te
c
hnology
,
vol
.
72
,
no.
3,
2015.
[19]
I.
Mzi
li
,
e
t
a
l
.,
"H
y
brid
Pengu
ins
Sear
ch
Opti
m
iz
at
ion
Algori
thm
and
Gene
tic
Algori
thm
Solv
ing
Tr
aveling
Sale
sm
an
Problem,
"
In
Inte
rna
ti
onal
Conf
ere
n
ce
on
Ad
vanced
Information
Te
chnol
ogy
,
Serv
i
ce
s
and
Syst
ems
,
Springer,
Cham
,
pp.
461
-
473
,
20
17
.
[20]
I.
Mzil
i
,
et
al
.,
"
Penguins
sea
rch
opti
m
iz
ation
al
g
orit
hm
to
solv
e
quadr
atic
assign
m
ent
proble
m
,
"
In
Proce
edi
ngs
of
the
2nd
i
nt
ernational
Conf
ere
nce on
B
ig
Data
,
C
l
oud
and
App
li
ca
ti
ons,
2017
,
BDCA'17.
ACM
,
pp
.
1
-
6
,
2017
.
[21]
S.
M.
Johns
on,
"O
pti
m
al
two‐a
nd
thre
e‐st
age
p
roduc
ti
on
sch
ed
ule
s
with
setup
ti
m
es
inc
lud
ed,
"
Naval
research
logi
stic
s q
uarter
ly
,
vol
.
1
,
pp
.
61
-
68,
1954
.
[22]
Y.
Ghera
ibia,
an
d
A.
Moussaoui
,
"P
engui
ns
sea
rch
o
pti
m
iz
ation
a
lgori
thm
(PeSOA),
"
In
Inte
rnati
onal
Con
fe
renc
e
on
Industrial,
Engi
nee
ring
and
Other
Appl
ic
a
tions
of
Appl
ie
d
Inte
lligen
t
System
s
,
2013
,
Berl
i
n
2013
,
Springer
Heide
lb
erg
,
pp.
222
-
231
,
2013
.
[23]
J.
E.
Bea
sl
e
y
,
"O
R
-
Li
bra
r
y
:
dis
tri
buti
ng
t
est
pr
oble
m
s
b
y
elec
t
ronic
m
ail,
"
Jou
rnal
of
the
ope
rational
resea
rc
h
socie
t
y
,
vol. 41,
pp.
1069
-
1072
,
1990.
[24]
K.Yuan
,
e
t
a
l
.,
"A
new
heur
isti
c
-
EM
for
p
ermut
at
ion
f
lowshop
sche
dul
ing,
"
IF
A
C
Proce
ed
ings
Vol
umes
,
vol
.
3
9,
pp.
33
-
38
,
2006
.
[25]
H.
Li
u
,
et
al
.,
"A
h
y
brid
par
t
i
cl
e
sw
arm
optim
iz
at
ion
with
e
stim
at
ion
of
dis
tri
buti
on
a
lgori
t
hm
for
solving
per
m
uta
ti
on
flo
ws
hop
sche
duli
n
g
proble
m
,
"
Ex
p
ert
Syst
ems wi
th
Appl
ic
a
ti
ons,
vo
l.
38
,
pp
.
4348
-
4
360,
2011
.
[26]
F.
Sa
y
ot
i,
and
M.
E.
Riff
i,
"G
ol
den
ball
al
gor
it
h
m
for
solving
flow
shop
s
che
duli
ng
probl
em,
"
The
Inte
rnation
al
Journal
of
Int
erac
tive
Mu
lt
im
edia and
Art
if
i
ci
al
I
nte
lligen
ce
,
vol
.
4,
no
.
1
,
pp
.
15
-
18,
2016
.
Evaluation Warning : The document was created with Spire.PDF for Python.