Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
C
om
put
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
11
,
No.
1
,
Febr
uar
y
2021
, pp.
471
~
480
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v11
i
1
.
pp471
-
480
471
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
A
n
ovel
popu
lation
-
bas
ed
l
ocal
s
ea
rc
h
for
n
urs
e
rosteri
ng p
roblem
An
m
ar
Ab
u
h
amdah
1
,
W
ad
i
i B
ou
li
la
2
,
G
h
aith M
. Jar
adat
3
,
A
na
s
M.
Qut
ei
sh
at
4
,
Mutase
m
K.
Alsma
di
5
,
Ibr
ah
im
A.
Alm
ar
as
hde
h
6
1
MIS
Depa
rtment,
Fa
cul
t
y
of
Bus
ine
ss
Adm
ini
stra
ti
on,
Taiba
h
Uni
ver
sit
y
,
Saudi
Arabi
a
2
IS Depa
rtment
,
Coll
ege of
Com
pute
r
sc
ie
n
ce
an
d
Engi
n
ee
ring
,
T
ai
bah
Univer
sit
y,
Saudi
Arabi
a
2
RIAD
I
La
bora
t
or
y
,
Nat
ional
Sc
hool
of
Com
puter Sci
en
ce,
Mano
ub
a
Univer
si
t
y
,
Tuni
sia
3
Depa
rtment of
Com
pute
r
Scie
n
ce
,
Facu
lty
of
C
om
pute
r
Scie
n
ce a
nd
In
form
at
ion
Technol
og
y
,
Jer
ash
Univer
sit
y
,
J
orda
n
4
Depa
rtment of
Com
pute
r
and
Network
Eng
ineer
ing,
Fa
cul
t
y
of E
ngine
er
ing
T
ec
h
nolog
y
,
Al
-
Bal
qa
a
Ap
plied
Univer
si
t
y
,
Jordan
4
El
e
ct
ri
ca
l
and
C
om
pute
r
Engi
n
e
eri
ng,
Facu
lty
of
Engi
n
ee
ring
,
So
har
Univer
si
t
y
,
Om
an
5
,6
MIS
Depa
rtment,
Co
ll
eg
e
of
A
ppli
ed
Studie
s
a
nd
Com
m
unity
Servic
e
,
Im
am Abdulra
hm
an
Bin
Faisa
l U
nive
rsit
y
,
Saud
i
Arabi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
y
17
, 20
20
Re
vised
Jun
3
0
,
20
20
Accepte
d
Aug 5
, 2
0
20
Populat
ion
-
bas
e
d
appr
oa
che
s
re
gula
rl
y
are
bett
er
th
an
singl
e
base
d
(lo
ca
l
sea
rch
)
appr
oaches
in
expl
or
ing
the
sea
rch
spa
c
e.
How
eve
r
,
th
e
dra
wbac
k
of
popula
ti
on
-
b
ase
d
appr
oac
h
es
is
in
expl
oiting
the
sea
rch
spac
e.
Se
ver
al
h
y
bri
d
appr
oac
h
es
hav
e
prove
n
the
ir
eff
icien
c
y
thr
ough
diffe
r
ent
dom
ai
ns
of
opti
m
iz
ation
pr
oble
m
s
by
in
c
orpora
ti
ng
and
int
egr
a
ti
ng
the
strengt
h
of
popula
ti
on
and
loc
a
l
sea
rch
app
roa
che
s.
Me
anwhil
e
,
h
y
br
id
m
et
hods
have
a
dra
wbac
k
of
i
ncr
ea
sing
th
e
pa
ramet
er
tun
ing.
Rec
en
tly
,
popu
l
at
ion
-
b
ase
d
loc
a
l
sea
r
ch
was
proposed
for
a
unive
rsit
y
cour
s
e
-
ti
m
etabli
ng
pr
oble
m
with
fewe
r
par
amete
r
s
tha
n
ex
isti
ng
a
pproa
che
s,
the
p
roposed
appr
oa
c
h
prove
s
it
s
eff
ective
n
ess.
T
he
proposed
app
roa
ch
emplo
y
s
t
wo
oper
at
ors
to
i
nte
nsif
y
and
dive
rsif
y
the
se
a
rch
spac
e
.
The
f
irst
oper
at
or
is
appl
ie
d
to
a
singl
e
soluti
on
,
while
th
e
sec
on
d
is
appl
i
ed
for
al
l
solut
ions.
Th
is
pape
r
a
ims
to
inve
stig
ate
the
per
fo
rm
anc
e
of
population
-
base
d
local
se
a
rch
for
th
e
nur
se
roster
ing
proble
m
.
Th
e
IN
RC2010
databa
s
e
with
a
d
ataset
compos
e
d
of
69
insta
nc
es
is
used
to
te
st
the
per
form
anc
e
of
PB
-
LS.
A
com
par
ison
was
m
a
de
bet
wee
n
the
p
erf
orm
anc
e
of
PB
-
LS
and
othe
r
exi
sting
a
pproa
che
s
in
th
e
l
it
er
at
ur
e
.
Result
s
show
good
per
form
ances
of
proposed
appr
oac
h
compa
red
to
oth
er
appr
oac
h
es,
where
population
-
base
d
loc
a
l
sea
r
ch
provide
d
bes
t
result
s
in
55
ca
ses
over
69
insta
n
ce
s used
i
n
exp
er
iments
.
Ke
yw
or
d
s
:
Dive
rsificat
ions
Gr
a
vitat
ion
al
e
m
ula
ti
on
Local
s
ea
rch
a
ppr
oach
es
Nurse
r
ost
erin
g
p
r
oble
m
Popu
la
ti
on
-
b
as
ed
a
ppr
oach
es
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
BY
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
An
m
ar F
a
khri
Abu
ham
dah
,
MIS
Dep
a
rtm
e
nt,
Fac
ulty
of
Business
Adm
i
nistrati
on
,
Tai
bah Unive
r
sit
y
,
Jana
dah Bi
n U
m
ay
ya
h
Road
Tay
ba,
Me
din
a
4235
3,
Saudi
Ar
a
bia
.
Em
a
il
:
aabu
ha
m
dah
@tai
bahu
.edu.sa
;
anm
ar_
81
@hotm
ail.co
m
1.
INTROD
U
CTION
Nurse
r
os
te
ri
ng
pro
blem
(N
RP)
invol
ves
assigni
ng
a
se
t
of
nurse
s
to
diff
e
ren
t
s
hift
s
su
bject
to
a
var
ie
ty
of
c
onstrai
nts
unti
l
a
co
m
plete
ro
ste
r
is
const
ru
ct
ed
[
1
,
2].
Co
nst
raints
in
the
NRP
pr
ob
le
m
can
be
cat
egorized
as
hard
a
nd
s
of
t
[3
-
5].
T
he
m
ai
n
goal
,
w
he
n
al
locat
in
g
nurses
to
va
rio
us
s
hifts
is
t
o
fu
l
fill
the
ha
rd
c
onst
rains
(i.e.
feasible
r
os
te
rs
)
an
d
try
to
m
ini
m
iz
e
the
soft
co
ns
trai
nts
as
m
uch
a
s
possi
ble
with
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
870
8
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
47
1
-
480
472
pen
al
iz
at
io
n
(in
orde
r
to
pro
duce
a
high
-
qu
a
li
ty
nu
rses
s
hif
ts).
The
sm
al
ler
value
in
dicat
es
a
bette
r
qu
a
li
ty
of
nurse
r
os
te
ri
ng
sh
ifti
ng.
NRP
is
an
NP
-
hard
pr
oble
m
du
e
to
the
diff
ic
ulty
of
fin
ding
optim
al
sh
ifts
[3
,
4].
Howe
ver, f
i
nd
i
ng a
ro
ste
rin
g wit
h
a
good
nu
rse qu
al
it
y reli
es on t
he
a
pproach
us
e
d durin
g
the
searc
h [
3
-
5]
.
Re
centl
y,
seve
ral
m
et
ho
ds
ha
ve
bee
n
ap
pl
ie
d
to
so
lv
e
NRP
[
6
-
18]
.
Jarad
at
et
al
.
[
6]
propose
d
the
hybri
d
el
it
ist
-
ant
syst
em
a
lgorit
hm
.
The
pro
po
se
d
al
go
r
it
h
m
aim
s
to
increase
t
he
div
ersit
y
of
the
searc
h
sp
ace i
n
the
el
it
ist
-
ant syst
em
al
gorithm
b
y i
nteg
r
at
ing i
t wi
th ex
te
rn
al
m
em
or
y.
Ra
j
eswa
ri
et
al
.
[7
]
us
e
d
a
m
od
i
fied
ne
lder
-
m
ead
m
et
ho
d
i
n
the
bee
col
ony
optim
iz
ation
al
gorithm
for
m
ulti
-
ob
j
ec
ti
ve
m
a
the
m
at
i
cal
program
m
i
ng.
T
he
m
et
ho
do
l
og
y
is
a
c
om
bin
at
ion
of
s
pecific
local
s
earch
,
honeybee
deci
sion
m
aki
ng,
and
m
ulti
-
age
nt
par
ti
cl
e
en
vi
ronm
ent
syst
e
m
.
Aw
adall
ah
et
al
.
[8
]
pr
opose
d
a h
ybrid
-
har
m
on
y sea
rc
h
al
gorithm
w
it
h
hil
l cl
i
m
b
ing
f
or increasi
ng
the e
xp
l
oitat
ion
m
e
chan
ism
. Th
e authors
m
od
ifie
d
the
m
e
m
or
y
to
ac
cel
erate
the
c
onve
r
gen
ce
rate
by
us
in
g
t
he
global
s
war
m
op
ti
m
iz
ation
c
on
ce
pt.
Santos
et
al
.
[9
]
us
e
d
int
eger
-
pr
ogram
m
ing
te
chn
i
ques
of
the
c
om
pact
and
m
onolit
hic
form
ulati
on.
The
pro
posed
te
chn
iq
ue
is
di
stribu
te
d
into
two
po
ols,
th
e
first
one
im
pro
ves
the
du
al
bounds
f
or
rapi
d
so
luti
on
pro
duct
ion
to
l
ow
e
r
the
opti
m
a
l
so
luti
on,
w
hile
the
sec
ond
is
us
e
d
to
s
pee
d
-
up
the
pro
duct
ion
near
-
opti
m
al
.
Aw
a
dalla
h
et
al
.
[
10
]
re
plac
ed
the
be
e
op
erator
i
n
the
arti
fici
al
bee
colo
ny
al
gorithm
b
y
the
hill
-
cl
i
m
bin
g
optim
iz
er
to
pro
po
se
a h
y
bri
dizat
ion
t
hat ex
pl
oit
s
the
se
arch
s
pace f
or g
oo
d
s
olu
ti
on q
ualit
y.
Burke
a
nd
Cu
rtois
[11],
pro
po
s
ed
ne
w
a
ppr
oac
hes
for
branc
h
a
nd
pr
ic
e
al
gorithm
and
a
n
e
j
ect
io
n
chain
,
wh
e
re
they
int
egr
at
e
d
both
al
gorithm
s
by
us
ing
the
dynam
i
c
program
m
ing
m
et
ho
d.
L
ü
a
nd
Ha
o
[
5]
pro
po
s
e
d
an
a
d
aptive
ne
ighbor
hood
s
earch
by
us
in
g
tw
o
dif
fer
e
nt
neig
hborh
oo
d
m
ov
es
a
nd
adap
ti
vely
al
te
rati
on
s
betwee
n
thre
e
intensific
at
io
n
an
d
di
ver
s
ific
at
ion
searc
h
strat
egies
r
efer
rin
g
to
the
searc
h
hi
story
.
Bi
lgin
et
al
.
[
12]
prese
nted
a
gen
e
ral
hype
r
-
heurist
ic
ap
pro
ach
t
hat
can
w
ork
i
nto
t
wo
di
ff
ere
nt
ti
m
et
a
blin
g
healt
hcar
e
pro
blem
s (
i.e. p
at
i
ent ad
m
issi
on
an
d nurse roste
rin
g)
and pr
ov
i
ded
a set o
f
a low
-
le
vel h
e
ur
i
sti
c fo
r
each
prob
le
m
.
Valo
ux
is
et
al
.
[13]
pr
opos
e
d
a
strat
egy
bas
ed
on
phases
t
o
produce
a
good
qual
it
y
so
luti
on.
The
first
phase
deals
with
eac
h
nurse
an
d
ea
ch
day
of
the
week
a
nd
the
s
econd
deals
wi
th
a
s
pecific
al
locat
ed
daily
sh
ift.
N
onobe
[
14
]
us
e
d
m
et
aheu
risti
c
base
d
on
th
e
co
ns
trai
nt
-
op
tim
iz
at
ion
pro
b
le
m
to
find
s
uitable
values
to
va
riables
that
ha
ve
const
raint
vio
l
at
ion
s’
weig
ht.
The
a
uthor
a
dopted
the
ta
bu
sea
rc
h
to
im
pro
ve
the
pe
rfo
rm
ance
of
c
on
tr
olli
ng
dynam
ic
al
l
y
the
ta
bu
te
nu
re
.
He
us
e
d
c
ons
trai
n
weig
hts
to
e
xam
ine
so
luti
ons
durin
g
t
he neig
hborh
ood sea
rc
h.
Most
of
these
appr
oach
es
a
re
hybr
i
dizat
ion
betwee
n
popul
at
ion
-
base
d
an
d
local
-
base
d
appr
oach
es
.
In
ge
ner
al
,
the
m
a
in
ad
van
ta
ge
of
popula
ti
on
-
based
ap
proach
e
s
is
their
abili
ty
to
exp
l
or
e
t
he
researc
h
s
pace
,
wh
e
reas
l
ocal
-
searc
h
a
ppr
oa
ches
a
re
m
o
re
ada
pte
d
to
ex
plo
it
the
r
esearch
spa
ce
[19
-
35
]
.
T
he
refor
e
,
com
bin
ing
the
popula
ti
on
-
ba
sed
an
d
local
-
searc
h
ap
pro
aches
will
hel
p
ta
king
ad
va
ntage
of
thes
e
tw
o
appr
oach
es
a
nd
pro
du
ce
s
a
good
ap
proach
fo
r
s
olv
in
g
NRP.
H
oweve
r,
com
bin
in
g
these
two
a
ppr
oach
e
s
involves
c
on
si
der
i
ng
a
n
inc
reasin
g
num
ber
of
par
am
eter
s.
T
his
cons
ti
tutes
a
new
chall
eng
e
f
or
m
os
t
hybri
dizat
ion
appr
oach
es
.
H
ence,
m
os
t
re
searche
s
m
ov
e
towa
rd
pro
posing
oth
e
r
a
ppro
ac
hes
with
f
ew
e
r
par
am
et
ers’
tu
ning
f
or
NRP
.
The
m
ai
n
co
ntributi
on
of
t
his
pa
pe
r
is
to
us
e
popula
ti
on
-
base
d
local
search
(P
B
-
L
S)
[
36
]
,
wh
ic
h
is
ap
pl
ie
d
in
an
oth
e
r
opti
m
iz
at
ion
prob
le
m
(i.e.
Course
ti
m
e
t
abling)
to
t
he
nurse
ro
ste
rin
g
pro
bl
e
m
.
PB
-
LS
ha
s
pro
ved
it
s
e
ff
ic
ie
ncy
with
few
e
r
pa
ram
et
ers;
hen
ce
,
it
can
be
ap
pli
ed
to
the
NR
P, w
hich has
dif
fer
e
nt
neig
hborh
ood
structu
res.
Our
w
ork
ai
m
s
to
stud
y
the
pe
rfor
m
ance
of
PB
-
LS
ov
e
r
th
e
nurse
r
os
te
ri
ng
pro
blem
.
T
he
pro
pos
e
d
appr
oach
will
help
to
obta
in
a
good
qua
li
ty
so
luti
on
f
or
nurs
e
r
os
te
rin
g
s
hiftin
g.
In
orde
r
t
o
e
valuate
the
e
ff
ic
ie
ncy
of
PB
-
LS,
we
com
par
e
the
pe
rfor
m
ance
of
PB
-
LS
with
oth
er
a
ppro
ac
hes
app
li
ed
to
N
RP
in
the li
te
ratur
e
. Si
xty
-
nin
e
d
at
as
et
s w
ere
used
a
nd r
es
ults in
dic
at
e that PB
-
L
S
can
pro
du
ce
go
od r
es
ults.
Fo
ll
owin
g
the
introd
uction,
the
pa
per
is
str
uctu
red
as
f
ollow
s
.
Sect
io
n
2
pr
ese
nts
a
de
scriptio
n
of
the
dataset
us
e
d
in
this
pap
e
r
.
Sect
ion
3
de
ta
il
s
the
pr
opose
d
m
et
ho
do
l
ogy.
Sect
ion
4
pr
ese
nts
the
re
su
lt
s
ob
ta
ine
d
by
ap
plyi
ng
PB
-
LS
to
the
co
ns
ide
r
ed
dataset
an
d
discuss
it
s
fi
nding
s
.
Finall
y,
Sect
ion
5
co
nc
l
ud
e
s
the p
a
pe
r
a
nd di
scusses pe
rsp
e
ct
ives for
the
pro
posed
wo
rk.
2.
DA
T
AS
ET
D
ESCRIPT
IO
N
A
69
sta
nd
a
r
d
NRP
ben
c
hma
rk
dataset
s
ar
e
us
e
d
to
te
st
t
he
pe
rfo
rm
ance
of
t
he
propo
sed
m
et
ho
d.
Dataset
s
we
re
ob
ta
ine
d
f
r
om
PA
TAT
2010,
i
nter
national
nurse
r
os
te
rin
g
c
omplet
ion
(INR
C201
0,
https:/
/ww
w.k
uleu
ven
-
ku
la
k.be/n
rp
c
om
petition
/i
ns
ta
nces
)
[
37
]
a
nd
a
nothe
r
ver
si
on
intr
oduce
d
i
n
[
38
].
NRC2
010
dataset
s
are
cl
assif
ie
d
into
four
gro
ups,
toy;
spri
nt;
m
edium
an
d
lo
ng
i
ns
ta
nc
es.
Toy
insta
nc
es
are
pro
vid
e
d
f
or
te
sti
ng
pur
po
se
s,
w
he
re
the
ot
her
t
hr
ee
s
pri
nt;
m
ediu
m
and
lo
ng
insta
nc
es
are
pro
vid
e
d
f
or
com
petit
ion
pu
rpose, an
d use
d by resea
rc
hers
to e
valuate t
he
ir appr
oac
hes.
Each
dataset
of
the
t
hr
ee
co
m
pet
it
ion
data
set
s
co
ntains
t
hr
ee
gro
ups
of
instances
,
the
early
,
la
te
,
and
hi
dd
e
n
i
ns
t
ance g
r
oups.
L
at
er
on,
an
up
da
te
with
e
xtra
gro
up
a
dde
d
to
them
na
m
ed
as
hin
t g
rou
p.
T
able 1
su
m
m
arizes t
he
dataset
s th
at
are
us
ed
in
t
his
p
a
per.
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
A novel
popula
ti
on
-
based
loc
al se
ar
ch
for
nu
rse r
os
te
ri
ng
pr
oble
m
(
An
m
ar
A
buha
mda
h)
473
Table
1.
IN
RC
2010
dataset
s s
umm
arizat
ion
Ins
tan
ces
Ear
l
y
ins
tan
ces
Hid
d
en
ins
tan
ces
Hin
t ins
tan
ces
Late
in
stan
ces
Sp
rint run
10
10
3
10
Mediu
m
run
5
5
3
5
Lon
g
r
u
n
5
5
3
5
3.
METHO
DOL
OGY
In
this
pa
per,
a
popula
ti
on
-
ba
sed
l
ocal
sea
rc
h
a
ppr
oach
(
P
B
-
LS)
is
pro
posed
[
36
]
for
nu
rse
ro
ste
rin
g
pro
blem
s
(N
RP).
PB
-
LS
is
assem
bled
of
gr
avita
ti
on
al
em
ulati
on
local
s
earch
(
GE
LS),
based
on
the
natu
ral
pr
i
nciple
of
gr
avita
ti
on
al
at
tract
ion
[
39
]
.
P
B
-
LS
is
pro
po
sed
to
ov
e
rc
om
e
so
m
e
of
the
po
pu
la
ti
on
base
d
appr
oach
es
li
m
it
at
ion
s su
c
h as
[
36
]:
-
Using
a
popula
ti
on
of
so
luti
on
s
in
local
se
arch
to
inc
reas
e
the
capab
il
it
y
of
the
intens
ific
at
ion
proce
ss
and
to
ov
e
rc
om
e
the
weak
ne
ss
of
the
intensific
at
io
n
process
,
w
hich
le
ads
to
non
-
si
gn
ific
a
nt
i
m
pr
ovem
ents.
-
Using
Me
m
or
y
Gu
ide
with
el
it
e
so
luti
on
s
fo
r
us
e
fu
l
in
f
or
m
at
ion
(in
de
scen
ding
ord
er)
to
ov
e
rc
om
e
the
lim
i
ta
ti
on
of
the
a
ppro
a
c
hes
that
did
not
e
m
plo
y
m
e
m
or
y
gu
idanc
e
fo
r
the
sea
r
ch.
I
n
ad
diti
on,
this
al
lows
overco
m
ing
the
d
epe
ndency
on
rand
om
iz
ati
on
that
re
qu
i
r
es
a
rep
ai
r
m
echan
ism
to
be
eff
ic
ie
nt i
n
c
onstrai
ned pr
ob
le
m
s an
d
overc
om
ing
the iss
ue of
r
a
ndom
ly
u
pd
at
in
g
t
he p
opulati
on.
-
Less
par
am
et
erized and
syst
em
ic
al
l
y updati
ng the
popula
ti
on.
PB
-
LS
pro
ves
it
s
eff
ic
ie
ncy
over
a
n
NP
-
ha
r
d
op
ti
m
iz
a
ti
on
pro
blem
of
un
i
ver
sit
y
co
urse
tim
e
ta
bling
pro
blem
[
36
]
.
This
m
otivated
us
to
us
e
th
is
m
et
ho
d
f
or
NRP
by
us
i
ng
diff
e
ren
t
nei
ghbo
rho
od
str
uc
tures.
PB
-
LS
sta
rts
with
an
init
ia
l
popula
ti
on
a
nd
trie
s
to
m
ini
m
iz
e
so
ft
con
s
trai
nts
it
erati
v
el
y
by
exp
lo
ring
t
hei
r
neig
hbor
s
olu
t
ion
s
.
T
hese
s
olu
ti
ons
a
re
obta
ined
by
usi
ng
one
or
m
or
e
nei
ghbor
hood
struct
ur
e
s
over
the
current
so
l
ution.
For
m
or
e
detai
ls
abo
ut
the
con
strai
nts
Hint,
Hard
a
nd
S
of
t
prese
nt
ed
in
the
IN
R
C201
0
dataset
s
an
d
th
ei
r
m
a
the
m
a
ti
c
al
fo
rm
ulati
on
s
,
rea
der
s
ca
n
re
fer
to
[
5].
As
i
n
[
6],
f
our
gro
up
s
of
nei
ghbo
rho
od
structu
res
a
re
use
d:
-
Sing
le
sh
ift
pe
r
d
ay
neig
hborh
ood
st
ru
ct
ur
e
a
sso
ci
at
ed wit
h har
d
c
onstrai
nt
.
-
Week
e
nd,
pe
rs
on
al
re
qu
est
,
a
lt
ern
at
ive
qu
al
ific
at
ion
s,
over
tim
e,
and
unde
r
ti
m
e
are
the
m
os
t
vio
la
te
d
const
raints as
s
ociat
ed wit
h
s
oft
constrai
nts.
-
Shuffle
, gree
dy
sh
uffle
, c
or
e
sh
uffle
to
s
wa
p l
arg
e
secti
ons
of p
e
rs
on
al
sc
he
du
le
s
.
-
Sh
a
ke
a
sh
ift
,
week
e
nd, a
nd t
wo p
e
ople
for
so
luti
on s
ha
king
.
PB
-
LS
cal
c
ula
te
s
the
gra
vitat
ion
al
f
orce
value
(
F,
as
su
m
ing
m
ini
m
iz
at
ion
prob
le
m
)
by
cal
culat
ing
the
diff
e
re
nce
betwee
n
two
obj
ect
s;
the
tria
l
so
luti
on
s
obj
ect
iv
e
functi
on
values
(i.e
.
Ts)
and
the
c
urren
t
so
luti
on
obj
ect
ive fu
nction va
lue (
i.e
. Cs) as
pr
ese
nted
in
(
1
)
.
F= Cs
-
Ts
(1)
Fig
ure
1
s
how
s
the
ps
e
udo
-
c
od
e
f
or
the
PB
-
LS
a
ppr
oac
h
for
NRP.
Be
lo
w
the
de
scri
ption
of
s
om
e
te
rm
s that are u
se
d
in
Fig
ure
1.
-
In
it
ia
l
so
luti
on
:
Si
;
an
d
t
he q
ua
li
ty
o
f
Si
deno
te
d
by
f(
Si)
-
Be
st ob
ta
in
ed
s
olu
ti
on: S
b
a
nd the
qu
al
it
y o
f Si
d
e
no
te
d by f
(S
b)
-
The
m
axi
m
u
m
n
um
ber
of it
er
at
ion
s
de
no
te
d by N
.m
ax
-
Nu
m
ber
of it
er
at
ion
s t
o
re
set
the s
olu
ti
on
dir
ect
ion
s t
o upda
te
d
en
oted
b
y
R.it
er
-
Fo
r
ce
value fo
r
gravity
d
e
note
d by
force
-
N
th
vel
ocity
v
e
ct
or
m
e
m
or
y t
o pro
vid
e
direc
ti
on
value de
note
d by
VV
n
-
Nu
m
ber
of s
ha
king
neig
hborh
ood
de
no
te
d b
y
Nsn
The
PB
-
LS
al
gorithm
sta
rts
with
a
zero
di
rec
ti
on
val
ue
(i.e.
zero
velocit
y)
that
init
ia
li
zes
t
he
searc
h,
and
the
n
this
value
is
upda
te
d
thr
oughout
the
searc
h
process
.
For
bette
r
sh
ifti
ng,
the
search
s
pace
i
s
adm
inist
e
red
by
the
force
value
(
us
in
g
(
1
)
),
an
d
the
n
the
search
s
pace
is
intens
ifie
d
us
in
g
an
y
local
search
al
gorithm
.
PB
-
LS
use
s
the
MPC
A
-
AR
DA
al
go
rithm
[
36
]
as
a
local
search
du
e
t
o
it
s
abili
ty
of
com
ple
m
entary
exp
lo
rati
on
and
e
xploit
at
io
n.
MPC
A
-
AR
DA
is
pro
po
s
e
d
in
[
40
]
as
hy
br
idiza
ti
on
be
tween
m
ul
ti
-
neighb
orhoo
d partic
le
c
olli
sion
al
gorithm
an
d
a
dap
ti
ve ran
dom
iz
ed
d
esce
nt alg
or
it
hm
.
In
ste
p
1
(i.e.
In
it
ia
li
zat
ion
phase)
,
we
init
ia
li
ze
al
l
the
req
ui
red
par
am
eter
s
m
entioned
in
Table
2.
The
n,
t
he
init
ia
l
so
luti
on
for
the
m
e
m
or
y
of
the
ve
loci
ty
vector
vv
i
s
ge
ner
at
e
d
by
app
ly
in
g
s
ha
king
neig
hborh
oods.
I
niti
al
ly
,
so
lu
ti
on
s
(vv
1
,
…,
vv
N
)
ar
e
produ
ced
base
d
on
the
nu
m
ber
of
neig
hborh
oods
a
nd
their
directi
ons
are
set
t
o
ze
ro
in the
vv
m
e
m
o
ry.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
870
8
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
47
1
-
480
474
Figure
1. Pse
udoc
ode
for
PB
-
LS a
ppro
ac
h [
36
]
over
nurse
ro
ste
rin
g prob
l
e
m
Table
2
.
Param
et
ers
use
d i
n P
B
-
LS
for NRP
Para
m
eter
Valu
e
N.ma
x
Ter
m
in
atio
n
Criter
ia (
m
ax
i
m
u
m
ite
rat
io
n
s) equ
al to 1
0
0
,
0
0
0
R.
iter
Iter
atio
n
s n
u
m
b
er
t
o
r
eset o
r
to
retun
e
the
d
irection
s eq
u
al to1
0
Nsn
Sh
ak
in
g
n
eig
h
b
o
rh
o
o
d
stru
ctu
res
Lo
ca
l Sea
rch
MPCA
-
ARDA
(ite
ration
s n
u
m
b
er
eq
u
al to1
0
)
In
ste
p
2
(i.e.
I
m
pr
ovem
ent
ph
ase
),
vv
m
e
m
or
y
is
reo
r
der
e
d
in
dec
re
asi
ng
order
ba
sed
on
their
directi
on
value
s.
T
he
s
olu
ti
on
s ar
e c
om
par
ed
b
ase
d on their
directi
on
value
s,
an
d
t
he
s
olu
t
ion
with t
he high
e
st
directi
on
value
is sele
ct
ed fo
r furt
he
r
im
pr
ove
m
ent p
ri
or
t
o othe
r
s
olu
ti
ons
in
vv
.
a.
In
s
te
p
2.1
,
if
al
l
directi
on
values
are
dif
f
eren
t,
we
sel
e
ct
the
highest
so
luti
on
direct
ion
val
ue
a
nd
set
S
i
=vv
1
,
the
n
M
PCA
-
ARD
A
is
app
li
ed
to
ge
ner
at
e
S
i
*
unti
l
the
stoppin
g
crit
eria
are
m
e
t.
MPC
A
-
AR
DA
us
es
dif
fer
e
nt
s
hak
i
ng
nei
ghbo
r
hood
.
Lat
er,
the
f
or
ce
val
ue
is
cal
culat
ed
usi
ng
(
1
)
(i.e.
Cs
=
S
i
an
d
Ts=
S
i
*
)
and
is
a
dded
(
neg
at
ive
or
po
sit
ive)
to
the
s
tore
d
directi
on
value
of
vv
1
.
The
best
so
l
ut
ion
(i.e
.
Sb
)
i
s
updated
with
S
i
*
in
case
of
f(
S
i
*
)
(i.e.
the
qu
al
it
y
of
S
i
*
)
is
bette
r
than
f(
Sb
)
(i.e.
the
qu
al
it
y
of
S
b).
If
the
qu
al
it
y
of
S
i
*
is
bette
r
than
S
i
,
we
r
eplace
vv
1
with
S
i
*
.
Otherwi
se,
the
unim
pr
ov
e
d
c
ounter
i
s
increase
d
(
UnImpr
ove
1
)
by
on
e
for
the
s
el
ect
ed
so
luti
on.
I
n
case
U
nImprove
1
is
equ
al
to
pr
e
-
s
et
un
im
pr
ove
d
it
erati
on
s
(i.e.
R
.it
er
),
the
directi
on
of
vv
n
is
returne
d
to
zer
o
a
nd
vv
1
so
l
ution
is
change
d
wi
t
h
the
best
neig
hbo
r
s
olu
ti
on
ge
ner
at
e
d
f
ro
m
Sb
ra
ndom
neig
hbo
r
s
(i.e
.
S
b*
)
by
s
hak
i
ng
th
e
neig
hborh
oo
d
s
.
This
pr
act
ic
abi
li
ty
tr
ie
s to
escape
from
local op
ti
m
a and
att
e
m
pts to
e
xp
a
nd the
searc
h sp
ace.
b.
In
ste
p
2.2,
i
f
the
dir
ect
ion
va
lues
are
sim
ilar
f
or
al
l
vv
s
olu
ti
ons,
ste
p
2.1.a
is
e
xec
uted
to
disti
ngui
s
h
the s
olu
ti
ons
of sim
i
la
r
direct
ion
value
s.
T
hi
s tries to
preser
ve
a
set
of
div
e
rse
s
olu
ti
ons.
4.
RESU
LT
S
The
pro
pose
d
appr
oach
is
ex
per
im
ented
for
25
r
uns
(as
s
ugge
ste
d
in
[
6])
throu
gh
69
in
sta
nces
that
wer
e
a
nnounc
ed
in
I
NRC2
010
(
https:/
/w
ww.kuleu
ve
n
-
ku
la
k.be/n
rp
c
om
pet
it
ion
/i
ns
ta
nces)
for
10
0,000
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
A novel
popula
ti
on
-
based
loc
al se
ar
ch
for
nu
rse r
os
te
ri
ng
pr
oble
m
(
An
m
ar
A
buha
mda
h)
475
it
erati
on
s.
T
he
m
achi
ne
us
e
d
is
an
I
ntel
Co
re
i5
3.2
G
Hz
proce
ss
or,
8
G
B
RAM,
an
d
t
he
co
de
im
plem
ented
us
in
g
j
ava
la
ng
uag
e
over
Net
Be
ans
I
DE
v
8.2.
As
sta
te
d
in I
NRC2
01
0,
the
instances
re
quire
a
so
l
utio
n
withi
n
appr
ox
im
at
ely
10
seco
nds
for
the
s
pr
int
i
ns
t
ances,
10
m
inu
te
s
f
or
the
m
edi
um
instance
s
,
an
d
10
ho
urs
f
or
the lo
ng insta
nc
es.
Howe
ver,
dif
fe
ren
t
fact
or
s
a
ffec
t
the
per
f
or
m
ance
of
the
m
achine
su
c
h
as
m
e
m
or
y,
clo
ck
s
pee
d
an
d
the
op
e
rati
ng
s
yst
e
m
.
Ther
e
f
or
e
,
the
res
ults
of
the
r
uns
a
r
e
obta
ine
d
at
diff
e
re
nt
ti
m
es
.
A
s
dep
ic
te
d
i
n
[
5],
si
m
ulati
on
s
in
our
a
ppro
ac
h
a
re
pe
rfor
m
ed
unde
r
rela
xing
t
i
m
eou
t
conditi
on
s
,
a
nd
we
e
m
plo
y
10
00
se
conds
for
t
he
s
pr
i
nt
instance
s,
50
00
sec
onds
f
or
the
m
edium
instance
s
a
nd
20
hours
f
or
the
l
ong
inst
ances.
As
s
how
n
i
n
T
able 2, PB
-
LS
e
m
plo
ys f
our p
aram
e
te
rs:
a.
The
te
rm
inatio
n
crit
eria
(
N.
ma
x
),
w
hich
i
s
pr
e
sente
d
in
[6
]
a
nd
ta
kes
into
c
onsider
at
ion
the
rel
a
x
tim
eou
t co
nd
it
ion.
b.
Iterati
on
num
ber
to
r
eset
(or r
et
un
e
)
the
d
i
re
ct
ion
s
(
R.it
er
)
, whic
h
is
pr
e
se
nted
i
n [
2
].
c.
Sh
a
king
-
nei
gh
bo
r
hood
str
uct
ur
es
(
Ns
n
)
, whi
ch
is
pr
e
sente
d i
n [6
]
;
d.
Iterati
on
num
ber
f
or
t
he
loc
al
search
(in
t
his
case,
MP
CA
-
ARD
A
use
d
as
a
l
ocal
search
with
10
it
erati
on
s)
, whi
ch
is
pr
e
sente
d i
n [
36
]
.
In
ord
er
to
e
va
luate
the
pe
rfo
rm
ance
of
PB
-
LS,
a
c
om
par
ison
is
m
ade
bet
ween
PB
-
LS
pe
rfor
m
ance
and
re
su
lt
s
of
si
m
il
ar
m
et
ho
ds
based
on
thei
r
publishe
d
w
orks
.
Table
3
shows
a
com
par
ison
base
d
on
s
pr
i
nt
instances, Ta
bl
e 4
show
s a c
om
par
ison
b
ase
d
on m
ediu
m
i
ns
ta
nces a
nd Tab
le
5
sho
ws
a com
par
ison
ba
sed
on
long in
sta
nces.
In
Tables
3
-
5,
the
be
st
res
ult
ob
ta
ine
d
by
P
B
-
LS
(d
e
note
d
as
Best
)
a
nd
the
rank
of
PB
-
LS
(
de
nted
a
s
Ra
nk
)
are
il
lust
rated
an
d
com
par
e
d
to
the
ot
her
a
ppro
ac
hes
.
The
best
res
ul
ts
so
far
are d
e
picte
d
by
bold co
lo
r.
Me
anwhil
e,
in
T
ables
3
-
5,
th
e
e
m
pty
cel
ls
with
a
sym
bo
l
(
-
)
de
no
te
tha
t
the
al
go
rithm
is
no
t
ap
plied
to
the
corres
pondin
g
instance,
f
or
exam
ple,
R4,
R6,
R7,
R9
a
nd
R10
did
no
t
a
pp
ly
their
al
go
rithm
s
to
the
9
hin
t
instances,
and
R6,
R
8
a
nd R1
0 did n
ot apply
their al
gorith
m
s to
the 20
hid
de
n
i
ns
ta
nces
.
Table
3.
C
om
par
iso
n betwee
n PB
-
LS
a
nd sim
il
ar
m
et
ho
ds
in the li
te
ratu
re
u
si
ng sprint i
nst
ances
Dataset
PB
-
LS
R1
R2
R3
R4
R5
R6
R7
R8
R9
R1
0
Ran
k
Bes
t
sp
rint_
early
_
0
1
8
57
57
56
58
56
56
56
56
57
56
56
sp
rint_
early
_
0
2
Sa
m
e
58
59
59
64
58
58
58
58
59
58
58
sp
rint_
early
_
0
3
Sa
m
e
51
51
51
59
51
51
51
51
51
51
51
sp
rint_
early
_
0
4
Sa
m
e
59
59
59
67
59
59
59
59
60
59
59
sp
rint_
early
_
0
5
Sa
m
e
58
58
58
63
58
58
58
58
58
58
58
sp
rint_
early
_
0
6
2
54
54
53
58
54
54
54
54
54
54
54
sp
rint_
early
_
0
7
Sa
m
e
56
56
56
61
56
56
56
56
56
56
56
sp
rint_
early
_
0
8
Sa
m
e
56
56
56
58
56
56
56
56
56
56
56
sp
rint_
early
_
0
9
Sa
m
e
55
55
55
61
55
55
55
55
55
55
55
sp
rint_
early
_
1
0
Sa
m
e
52
52
52
58
52
52
52
52
52
52
52
sp
rint_
h
id
d
en
_
0
1
Sa
m
e
32
32
32
46
32
32
-
32
-
33
-
sp
rint_
h
id
d
en
_
0
2
Sa
m
e
32
32
32
44
32
32
-
32
-
32
-
sp
rint_
h
id
d
en
_
0
3
Sa
m
e
62
62
62
78
62
62
-
62
-
62
-
sp
rint_
h
id
d
en
_
0
4
Sa
m
e
66
66
66
78
66
66
-
66
-
67
-
sp
rint_
h
id
d
en
_
0
5
Sa
m
e
59
59
59
69
59
59
-
59
-
59
-
sp
rint_
h
id
d
en
_
0
6
Sa
m
e
130
130
134
169
130
130
-
130
-
134
-
sp
rint_
h
id
d
en
_
0
7
Sa
m
e
153
153
153
187
153
153
-
153
-
153
-
sp
rint_
h
id
d
en
_
0
8
Sa
m
e
204
204
204
240
204
204
-
204
-
209
-
sp
rint_
h
id
d
en
_
0
9
Sa
m
e
338
338
338
372
338
338
-
338
-
338
-
sp
rint_
h
id
d
en
_
1
0
Sa
m
e
306
306
306
322
306
306
-
306
-
306
-
sp
rint_
h
in
t
_
0
1
2
75
78
73
90
-
75
-
-
78
-
-
sp
rint_
h
in
t
_
0
2
2
46
47
43
56
-
46
-
-
47
-
-
sp
rint_
h
in
t
_
0
3
2
50
50
49
69
-
50
-
-
57
-
-
sp
rint_
late_
0
1
Sa
m
e
37
37
37
52
37
37
37
37
40
37
37
sp
rint_
late_
0
2
2
42
42
41
56
42
42
42
42
44
42
42
sp
rint_
late_
0
3
2
48
48
45
60
48
48
48
48
50
48
48
sp
rint_
late_
0
4
2
73
73
71
95
73
73
75
73
81
75
76
sp
rint_
late_
0
5
Sa
m
e
44
44
46
57
44
44
44
44
45
44
45
sp
rint_
late_
0
6
Sa
m
e
42
42
42
52
42
42
42
42
42
42
42
sp
rint_
late_
0
7
Sa
m
e
42
42
44
55
42
44
42
42
46
42
43
sp
rint_
late_
0
8
Sa
m
e
17
17
17
19
17
17
17
17
17
17
17
sp
rint_
late_
0
9
Sa
m
e
17
17
17
17
17
17
17
17
17
17
17
sp
rint_
late_
1
0
Sa
m
e
43
43
43
54
43
43
43
43
46
43
44
No
te:
R1
:
Hy
b
rid
Elitist
-
An
t
S
y
ste
m
[
6
],
R2
:
Directe
d
Bee
Co
lo
n
y
Op
ti
m
izatio
n
Alg
o
rith
m
[
7
],
R3
:
Hy
b
ridizatio
n
o
f
h
ar
m
o
n
y
se
arc
h
with
h
ill
-
cli
m
b
in
g
[
8
],
R4
:
Integ
er
p
rog
ra
m
m
in
g
tech
n
iq
u
es
[9],
R5
:
A
h
y
b
rid
artif
icial
b
ee
c
o
lo
n
y
[
1
0
],
R6
:
New
ap
p
roach
f
o
r
b
ranch
an
d
p
rice
alg
o
rith
m
an
d
an
ejecti
o
n
ch
ain
[11
],
R7
:
Ad
ap
tiv
e
n
eig
h
b
o
rhood
search
[
5
],
R8
:
Hy
p
er
-
h
eu
ristic
ap
p
roach
[
1
2
],
R9
: A
syste
m
atic
t
wo
-
p
h
ase app
roach
[
1
3
]
an
d
R1
0
: A
g
en
eral
co
n
strain
t
o
p
ti
m
iz
atio
n
[
1
4
].
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
870
8
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
47
1
-
480
476
Table
4.
C
om
par
iso
n betwee
n PB
-
LS
a
nd
sim
il
ar
m
et
ho
ds
in the li
te
ratu
re
u
si
ng m
ediu
m
instances
Dataset
PB
-
LS
R1
R2
R3
R4
R5
R6
R7
R8
R9
R1
0
Ran
k
Bes
t
m
e
d
iu
m
_
earl
y
_
0
1
Sa
m
e
240
241
245
280
240
245
244
240
242
240
241
m
e
d
iu
m
_
earl
y
_
0
2
Sa
m
e
240
241
243
281
240
245
241
240
241
240
240
m
e
d
iu
m
_
earl
y
_
0
3
Sa
m
e
236
236
239
287
236
242
238
236
238
236
236
m
e
d
iu
m
_
earl
y
_
0
4
Sa
m
e
237
237
245
278
237
240
240
237
238
237
239
m
e
d
iu
m
_
earl
y
_
0
5
Sa
m
e
303
303
310
330
303
308
308
303
304
303
304
m
e
d
iu
m
_
h
id
d
en
_
0
1
Sa
m
e
111
111
143
410
111
155
-
117
-
130
-
m
e
d
iu
m
_
h
id
d
en
_
0
2
Sa
m
e
220
220
230
412
221
254
-
220
-
221
-
m
e
d
iu
m
_
h
id
d
en
_
0
3
Sa
m
e
34
34
53
182
34
54
-
35
-
36
-
m
e
d
iu
m
_
h
id
d
en
_
0
4
Sa
m
e
78
78
85
168
78
94
-
79
-
81
-
m
e
d
iu
m
_
h
id
d
en
_
0
5
Sa
m
e
119
119
182
520
119
177
-
119
-
122
-
m
e
d
iu
m
_
h
in
t_
0
1
2
42
42
42
64
-
48
-
-
40
-
-
m
e
d
iu
m
_
h
in
t_
0
2
Sa
m
e
91
91
91
133
-
94
-
-
91
-
-
m
e
d
iu
m
_
h
in
t_
0
3
2
140
140
135
187
-
140
-
-
144
-
-
m
e
d
iu
m
_
late_
0
1
2
161
161
176
234
157
174
187
164
163
158
176
m
e
d
iu
m
_
late_
0
2
Sa
m
e
18
18
30
49
18
31
22
20
21
18
19
m
e
d
iu
m
_
late_
0
3
Sa
m
e
29
29
35
59
29
38
46
30
32
29
30
m
e
d
iu
m
_
late_
0
4
Sa
m
e
35
35
42
71
35
48
49
36
38
35
37
m
e
d
iu
m
_
late_
0
5
Sa
m
e
107
107
129
272
107
137
161
117
122
107
125
Table
5.
C
om
par
iso
n betwee
n PB
-
LS
a
nd sim
il
ar
m
et
ho
ds
in the lit
eratu
re
u
si
ng lo
ng
inst
ances
Dataset
PB
-
LS
R1
R2
R3
R4
R5
R6
R7
R8
R9
R1
0
Ran
k
Bes
t
lo
n
g
_
early
_
0
1
2
197
198
194
339
197
197
198
197
197
197
197
lo
n
g
_
early
_
0
2
Sa
m
e
219
220
228
399
219
229
223
222
220
219
219
lo
n
g
_
early
_
0
3
Sa
m
e
240
240
240
349
240
240
242
240
240
240
240
lo
n
g
_
early
_
0
4
Sa
m
e
303
303
303
411
303
303
305
303
303
303
303
lo
n
g
_
early
_
0
5
Sa
m
e
284
284
284
383
284
284
286
284
284
284
284
lo
n
g
_
h
id
d
en
_
0
1
Sa
m
e
346
346
389
4466
346
400
-
346
-
363
-
lo
n
g
_
h
id
d
en
_
0
2
Sa
m
e
89
89
108
1071
89
117
-
89
-
90
-
lo
n
g
_
h
id
d
en
_
0
3
Sa
m
e
38
38
48
163
38
51
-
38
-
38
-
lo
n
g
_
h
id
d
en
_
0
4
Sa
m
e
22
22
27
113
22
29
-
22
-
22
-
lo
n
g
_
h
id
d
en
_
0
5
Sa
m
e
41
41
55
139
41
56
-
45
-
41
-
lo
n
g
_
h
in
t_
0
1
2
40
40
40
126
-
42
-
-
33
-
-
lo
n
g
_
h
in
t_
0
2
2
28
28
29
122
-
30
-
-
17
-
-
lo
n
g
_
h
in
t_
0
3
Sa
m
e
55
55
79
278
-
83
-
-
55
-
-
lo
n
g
_
late_
0
1
Sa
m
e
235
235
249
588
235
257
286
237
241
235
235
lo
n
g
_
late_
0
2
Sa
m
e
229
229
261
577
229
263
290
229
245
229
229
lo
n
g
_
late_
0
3
Sa
m
e
220
220
259
567
220
262
290
222
233
220
220
lo
n
g
_
late_
0
4
Sa
m
e
221
221
257
604
222
261
280
227
246
221
221
lo
n
g
_
late_
0
5
Sa
m
e
83
83
92
329
83
102
110
83
87
83
83
As
s
how
n
in
Table
s
3
-
5,
resu
lt
s
in
dicat
e
that
PB
-
LS
pro
du
ces
ve
r
y
good
qual
it
y
so
luti
ons.
The
pro
pose
d
appr
oach
rea
c
hes
55
opti
m
a
l
so
luti
on
s
ou
t
of
69
instanc
es.
Wh
il
e
in
oth
er
instan
ces
PB
-
L
S
ob
ta
ine
d
t
he
se
cond ra
nk ex
ce
pt in o
ne
in
sta
nc
e (i.e.
sprint
_e
arly
_01).
Desp
it
e
that
PB
-
LS
has
obta
ined
seco
nd
place
i
n
s
om
e
instances,
it
ou
tpe
rfo
r
m
ed
the
sam
e
appr
oach
es
in
oth
e
r
dif
fer
e
nt
instances.
F
or
exam
ple,
PB
-
LS
ob
ta
ine
d
the
ei
gh
th
rank
in
sp
ri
nt_
ea
rly
_01
an
d
ou
t
perform
ed
R2,
R4,
R5
,
R6,
R7,
R
9
and
R10
.
I
n
ad
diti
on
,
P
B
-
LS
ou
tperfo
rm
ed
R2
in
sp
rint
_ear
l
y_02,
R4
in
m
ediu
m
_h
i
dd
e
n_02,
R
5
in
s
pr
int
_late
_07,
R
6
in
m
edium
_ear
ly
_01,
R7
i
n
m
edi
um
_h
idd
e
n_01
,
R9
i
n
sp
ri
nt_hid
den_
01,
R1
0
in
sprint_lat
e_
05.
More
ov
e
r,
PB
-
LS
outpe
rfo
r
m
ed
R1
in
sprint_ea
rly
_02,
R3
in
sp
ri
nt_
ea
rly
_02,
a
nd
R8
in
s
pri
nt_
ea
rly
_02. Re
su
lt
s
ind
ic
at
e
that
PB
-
L
S outpe
rfor
m
ed
al
l
appro
ac
hes
in
so
m
e
instances
a
nd
ob
ta
ine
d
sim
ilar
pe
rfor
m
anc
es
to
ot
her
a
ppr
oac
hes
in
s
om
e
cases.
This
confirm
s
that
PB
-
LS
can b
e
c
onside
red
as
a
go
od
a
ppr
oach
f
or
N
RP.
To
bette
r
e
val
uate
t
he
perform
ance
of
th
e
pro
posed
a
ppro
ac
h,
Fig
ure
2
de
picts
a
com
par
is
on
betwee
n
PB
-
L
S
an
d
oth
e
r
a
ppr
oac
hes
ref
e
r
eei
ng
t
o
the
nu
m
ber
of
best
r
esults
(optim
a
l solution
s
) o
btained
ov
e
r
t
he 69 i
nst
ances.
Figure
2
sho
w
s
the
num
ber
of
best
resu
lt
s
obta
ined
ov
e
r
69
dataset
s
for
t
he
pro
posed
a
ppr
oach
an
d
the
consi
dered
existi
ng
w
ork
s.
The
pr
opos
e
d
ap
proac
h
obta
ined
55
be
st
op
ti
m
al
so
luti
on
s
over
69
in
sta
nces
.
R4
com
es
in
t
he
seco
nd
rank
with
52
be
s
t
op
tim
al
so
luti
on
s.
R3
c
ome
s
in
the
la
st
rank
with
on
l
y
on
e
o
ptim
al
so
luti
on
.
Figure
3
de
picts
furthe
r
a
nal
ysi
s
by
com
par
in
g
sim
i
la
r,
worse
a
nd
bette
r
res
ults
of
t
he
pro
po
s
e
d
appr
oach
ove
r
the
69
instan
ce
s
to
t
he
oth
e
r
c
on
si
der
e
d
ap
proach
e
s.
In
the
le
gend,
the
blue
col
or
de
no
te
s
that
the
pro
pose
d
a
ppr
oach
has
t
he
sam
e
nu
m
be
r
of
bette
r
res
ults
than
the
e
xisti
ng
a
ppr
oa
ch.
T
he
or
a
ng
e
colo
r
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
A novel
popula
ti
on
-
based
loc
al se
ar
ch
for
nu
rse r
os
te
ri
ng
pr
oble
m
(
An
m
ar
A
buha
mda
h)
477
represe
nts
the
nu
m
ber
of
tim
es
that
the
pr
opose
d
ap
proac
h
has
the
best
resu
lt
s.
The
gray
color
re
pr
esents
the
num
ber
of
tim
es
that
the
existi
ng
ap
pro
ach
has
t
he
be
st
res
ults.
F
or
exam
ple,
the
pro
posed
ap
pr
oa
ch
a
nd
R1
ob
ta
in
ed
62
si
m
il
ar
nu
m
ber
of
best
sol
ution
s
.
Howe
ver,
the
pro
posed
ap
proac
h
ou
t
perform
ed
R1
in
7
cases
.
I
n
a
ddit
ion
,
in
al
l
cases,
PB
-
LS
ob
ta
ine
d
se
ve
r
al
bette
r
res
ul
ts
gr
eat
er
tha
n
the
ot
her
exi
sti
ng
appr
oach
es
(
orang
e
ba
r
agai
nst
gr
ay
ba
r).
T
able
6
s
umm
ar
iz
es
exp
e
rim
en
ta
l
resu
lt
s
achi
eved
by
PB
-
L
S
over
sp
ri
nt insta
nce
s,
Ta
ble 7 f
or the m
edium
instances,
a
nd Ta
ble 8 f
or the l
ong i
ns
ta
nces.
Figure
2. Com
par
is
on of t
he nu
m
ber
of b
e
st res
ults
obta
ine
d ov
e
r
t
he 69 i
ns
ta
nces
Figure
3. The
nu
m
ber
of sim
i
la
r,
worse a
nd
bette
r
re
su
lt
s c
om
par
ed
t
o
the
prop
os
ed
app
r
oach
Table
6.
E
xper
i
m
ental
r
esults
for our a
ppr
oa
ch on I
NRC2
010 o
ve
r
s
pr
i
nt instances
Dataset
Bes
t
av
g
σ
Ti
m
e
sp
rint_
early
_
0
1
57
5
8
.7
1
.8
4*
sp
rint_
early
_
0
2
58
5
8
.6
1
.0
3*
sp
rint_
early
_
0
3
51
5
1
.8
0
.7
7*
sp
rint_
early
_
0
4
59
6
0
.1
0
.8
4*
sp
rint_
early
_
0
5
58
5
8
.2
0
.4
6*
sp
rint_
early
_
0
6
54
5
4
.2
0
.4
3*
sp
rint_
early
_
0
7
56
5
6
.6
0
.5
3*
sp
rint_
early
_
0
8
56
5
6
.4
0
.4
4*
sp
rint_
early
_
0
9
55
5
5
.5
0
.7
4*
sp
rint_
early
_
1
0
52
5
2
.5
0
.4
6*
sp
rint_
h
id
d
en
_
0
1
32
3
3
.9
1
.2
68
sp
rint_
h
id
d
en
_
0
2
32
3
3
.5
1
.1
86
sp
rint_
h
id
d
en
_
0
3
62
6
3
.5
1
.9
6*
sp
rint_
h
id
d
en
_
0
4
66
6
7
.2
0
.7
24
sp
rint_
h
id
d
en
_
0
5
59
5
9
.6
0
.6
8*
sp
rint_
h
id
d
en
_
0
6
130
1
3
3
.4
2
.4
48
sp
rint_
h
id
d
en
_
0
7
153
1
5
6
.1
3
.7
5*
sp
rint_
h
id
d
en
_
0
8
204
2
0
5
.8
1
.6
71
sp
rint_
h
id
d
en
_
0
9
338
3
4
0
.2
2
.8
34
sp
rint_
h
id
d
en
_
1
0
306
3
0
6
.6
1
.9
7*
sp
rint_
h
in
t
_
0
1
75
7
7
.2
5
.4
44
sp
rint_
h
in
t
_
0
2
46
4
9
.2
2
.9
26
sp
rint_
h
in
t
_
0
3
50
5
4
.7
6
.7
35
sp
rint_
late_
0
1
37
3
8
.3
1
.1
23
sp
rint_
late_
0
2
42
4
3
.7
1
.5
18
sp
rint_
late_
0
3
48
5
0
.2
0
.9
8*
sp
rint_
late_
0
4
73
7
6
.6
2
.4
28
sp
rint_
late_
0
5
44
4
5
.1
0
.8
6*
sp
rint_
late_
0
6
42
4
2
.5
0
.6
9*
sp
rint_
late_
0
7
42
4
3
.9
0
.9
8*
sp
rint_
late_
0
8
17
17
0
5*
sp
rint_
late_
0
9
17
17
0
4*
sp
rint_
late_
1
0
43
4
5
.2
1
.2
10*
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
870
8
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
47
1
-
480
478
Table
7.
E
xper
i
m
ental
r
esults
for our a
ppr
oa
ch on I
NRC2
010 o
ve
r
m
edium
instances
Dataset
Bes
t
av
g
σ
Ti
m
e
m
e
d
iu
m
_
earl
y
_
0
1
240
2
4
1
.6
1
.7
957
m
e
d
iu
m
_
earl
y
_
0
2
240
2
4
1
.5
1
.1
567*
m
e
d
iu
m
_
earl
y
_
0
3
236
2
3
8
.6
1
.2
921
m
e
d
iu
m
_
earl
y
_
0
4
237
2
3
8
.7
1
.4
249*
m
e
d
iu
m
_
earl
y
_
0
5
303
3
0
5
.3
1
.8
353*
m
e
d
iu
m
_
h
id
d
en
_
0
1
111
1
2
0
.8
3
.8
2153
m
e
d
iu
m
_
h
id
d
en
_
0
2
220
2
3
3
.7
8
.2
4370
m
e
d
iu
m
_
h
id
d
en
_
0
3
34
3
5
.9
1
.9
705
m
e
d
iu
m
_
h
id
d
en
_
0
4
78
8
2
.4
2
.5
2081
m
e
d
iu
m
_
h
id
d
en
_
0
5
119
1
2
9
.3
5
.4
4429
m
e
d
iu
m
_
h
in
t_
0
1
42
5
1
.8
4
.6
1505
m
e
d
iu
m
_
h
in
t_
0
2
91
1
0
5
.3
7
.6
2453
m
e
d
iu
m
_
h
in
t_
0
3
140
1
5
1
.1
8
.7
2691
m
e
d
iu
m
_
late_
0
1
161
1
7
2
.6
6
.3
3553
m
e
d
iu
m
_
late_
0
2
18
2
1
.8
2
.3
2417
m
e
d
iu
m
_
late_
0
3
29
3
2
.9
2
.2
583*
m
e
d
iu
m
_
late_
0
4
35
3
6
.8
0
.7
1108
m
e
d
iu
m
_
late_
0
5
107
1
1
6
.7
6
.8
4230
Table
8.
E
xper
i
m
ental
r
esults
for our
a
ppr
oa
ch on I
NRC2
010 o
ve
r
lo
ng in
sta
nces
Dataset
Bes
t
av
g
σ
Ti
m
e
lo
n
g
_
early
_
0
1
197
2
0
1
.4
3
.7
1
1
1
1
4
*
lo
n
g
_
early
_
0
2
219
2
3
1
.3
6
.3
1
2
0
2
1
*
lo
n
g
_
early
_
0
3
240
240
0
3
0
4
1
*
lo
n
g
_
early
_
0
4
303
3
0
4
.9
0
.7
5
1
1
2
*
lo
n
g
_
early
_
0
5
284
2
8
5
.6
1
.2
8
3
9
2
*
lo
n
g
_
h
id
d
en
_
0
1
346
3
5
5
.9
5
.4
2
1
2
9
7
*
lo
n
g
_
h
id
d
en
_
0
2
89
9
1
.6
2
.8
1
8
8
4
4
*
lo
n
g
_
h
id
d
en
_
0
3
38
4
0
.1
1
.2
1
4
9
4
7
*
lo
n
g
_
h
id
d
en
_
0
4
22
2
4
.9
1
.5
2
4
6
6
3
*
lo
n
g
_
h
id
d
en
_
0
5
41
4
5
.2
2
.7
2
6
6
9
0
*
lo
n
g
_
h
in
t_
0
1
40
4
2
.8
1
.8
1
5
0
6
0
*
lo
n
g
_
h
in
t_
0
2
28
3
0
.2
1
.1
1
4
2
5
1
*
lo
n
g
_
h
in
t_
0
3
55
5
7
.5
1
.5
3
1
6
9
4
*
lo
n
g
_
late_
0
1
235
2
3
9
.2
2
.7
2
4
8
1
1
*
lo
n
g
_
late_
0
2
229
2
3
0
.8
1
.2
1
6
9
6
6
*
lo
n
g
_
late_
0
3
220
2
2
2
.1
1
.0
2
4
0
3
5
*
lo
n
g
_
late_
0
4
221
2
2
4
.3
2
.4
1
5
4
0
6
*
lo
n
g
_
late_
0
5
83
8
4
.9
0
.7
2
0
4
3
4
*
In
Ta
ble
s
6
-
8,
the
best
r
esult
ob
ta
ine
d
over
25
r
un
s
is
de
note
d
by
Best
,
t
he
ave
ra
ge
res
ult
is
denoted
by
av
g
,
the
sta
nd
a
r
d
dev
ia
ti
on
is
de
no
te
d
by
σ
,
an
d
t
he
c
om
pu
ta
ti
on
al
t
i
m
e
of
the
be
s
t
resu
lt
is
de
note
d
by
Time
(
wh
e
re
*
is
the
ti
m
e
lim
i
t
in
IN
RC
20
10).
Fo
r
e
xam
ple,
in
Table
6,
the
PB
-
LS
achieve
s
a
value
of
57
as
the
best
re
s
ult
(over
25
r
uns)
for
s
pr
i
nt_e
arly
_01
insta
nce
in
4
seco
nds,
with
an
a
ve
rag
e
of
58.7
a
nd
a
sta
nd
a
r
d
de
viati
on
of
1.8
.
Ta
bles
6,
7,
a
nd
8
il
lustrate
that
P
B
-
LS
obta
ins
43
out
of
69
instance
s
within t
he
ti
m
e
lim
i
t (which
m
ark
ed
as
*) of INRC
20
10.
Re
su
lt
s
in
Tabl
e
s
3
-
5,
a
nd
a
na
ly
sis
in
Fig
ure
2
a
nd
Fig
ure
3
co
nf
irm
that
the
pro
po
se
d
a
ppr
oac
h
ca
n
pro
du
ce
go
od
qual
it
y
so
luti
on
s
f
or
the
NRP
c
om
par
ed
to
ot
her
ex
ist
ing
a
ppro
a
c
hes
i
n
the
li
te
ratur
e
.
In
a
ddit
i
on
,
re
su
lt
s
in
Ta
bles
6
-
8
s
how
t
hat
the
pro
pose
d
app
r
oach
r
un
s
in
a
go
od
m
ann
er
for
al
l
inst
ances
with
instance
s
of
d
ive
rsity
.
A
s
a
perpecti
ve
to
this
w
ork
wil
l
be
to
ap
ply
the
propose
d
ap
proac
h
in
the
c
onte
xt
of
rem
ote
sens
ing
big
data
[
41
-
4
4
]
,
t
o
e
xp
lore
t
he
c
on
te
xt
of
ca
se
-
bas
ed
reasonin
g
usi
ng
‘
hype
r
-
he
ur
ist
ic
’
[45,
4
6
]
, a
nd to
ev
al
uate
the e
f
fect o
f un
ce
rtai
nty i
n
the
pr
oc
ess of sea
rc
h
f
or nur
s
e r
os
te
ri
ng pr
ob
le
m
[
4
7
-
49
]
.
5.
CONCL
US
I
O
N
In
this
pa
per,
we
pro
po
se
d
a
popula
ti
on
-
bas
ed
l
ocal
searc
h
ap
proac
h
(P
B
-
LS)
f
or
the
nur
se
r
os
te
r
i
ng
pro
blem
.
The
popula
ti
on
-
bas
ed
is
m
otivate
d
by
a
gr
a
vitat
ion
al
em
ulati
o
n
local
search
al
gorithm
to
intensify
the
searc
h
spa
ce
and
by
a
n
MPC
A
-
AR
DA
t
o
di
ver
si
fy
the
s
earc
h.
A
com
par
iso
n
is
m
ade
betwee
n
the
pe
rfo
rm
ance
of
the prop
ose
d
a
pproach
a
nd p
er
form
ances
of
ot
her
exiti
ng
ap
proac
hes
in
the
li
te
ratu
re
over
69
dataset
s.
I
n
this
pa
pe
r,
te
n
existi
ng
ap
proac
hes
are
consi
der
e
d
f
or
com
par
ison.
Re
su
lt
s
ind
ic
at
e
that
the pr
opos
e
d
a
ppr
oach p
rod
uc
es a
good
qu
a
li
ty
so
luti
on
c
om
par
ed
to t
he e
xisti
ng appr
oa
ches.
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
A novel
popula
ti
on
-
based
loc
al se
ar
ch
for
nu
rse r
os
te
ri
ng
pr
oble
m
(
An
m
ar
A
buha
mda
h)
479
Our
a
ppr
oach
ob
ta
ine
d
55
optim
a
l
so
luti
ons
over
69
cases
,
an
d
it
is
ra
nk
e
d
fi
rst
w
hile
c
om
par
ing
it
to
oth
e
r
ap
proach
e
s
accor
di
ng
to
the
nu
m
ber
of
op
ti
m
al
so
luti
on
s.
Additi
on
al
ly
,
resu
lt
s
ind
ic
at
e
that
the
pro
posed
appr
oach
ou
t
pe
rfor
m
ed
al
l
t
he
existi
ng
ap
proac
hes
w
hile
com
par
ing
t
he
num
ber
of
bette
r
so
luti
ons
to
t
he
nu
m
ber
of
worse
s
olu
ti
on
s.
Thes
e
res
ults
confirm
the
good
pe
rfo
rm
a
nce
of
the
pro
po
s
ed
appr
oach f
or
s
olv
in
g
t
he pr
ob
lem
o
f
nurse ro
ste
ring
.
As
f
utu
re
wor
k,
we
pr
opos
e
to
us
e
m
or
e
case
stud
ie
s
to
te
st
the
per
f
or
m
ance
of
th
e
pr
op
os
e
d
appr
oach
a
nd
t
o
ap
ply
it
in
con
te
xt
of
rem
ote
sensing
big
data
.
A
dd
it
io
na
ll
y,
in
this
pap
er
,
we
us
e
a
sing
le
heurist
ic
ap
pro
ach
wit
h
di
ff
e
r
ent
nei
ghbor
hoods
over
a
popula
ti
on
of
s
olu
ti
ons;
w
hic
h
is
ch
os
e
n
ba
sed
on
gr
a
vity
form
ul
a.
A
no
t
her
c
ha
ll
eng
in
g
to
pic
to
be
e
xplore
d
i
s
the
stu
dy
of
t
he
case
-
base
d
reasonin
g
m
eth
od
to
al
low
us
in
g
of
‘h
y
per
-
he
uri
sti
c’.
T
his
will
he
lp
to
deter
m
i
ne
the
bette
r
he
ur
ist
ic
f
or
a
gi
ven
popula
ti
on,
a
nd
hen
ce
avoidi
ng
so
l
ving
prob
le
m
s f
ro
m
scr
at
ch.
REFERE
NCE
S
[1]
P.
Brucke
r
,
et
a
l
.
,
“
Personnel
sc
hedul
ing:
m
ode
l
s
and
complexi
t
y
,
”
Eur
opean
J
ournal
of
Oper
ati
onal
R
es
earc
h,
vol.
210
,
no
.
3
,
p
p.
467
-
473
,
201
1.
[2]
M.
Dorigo
and
T.
Stütz
l
e,
“
Ant
col
on
y
op
ti
m
iz
a
t
ion:
over
vie
w
an
d
rec
ent
adv
anc
e
s,
”
Handbook
of
Meta
heur
isti
cs
,
Springer,
pp
.
22
7
-
263,
2010
.
[3]
I
.
I
.
I.
B
art
hold
i,
“
A gua
ran
te
ed
-
a
cc
ura
c
y
round
-
o
ff
al
gorit
hm
for c
y
cl
ic
sche
du
li
n
g
and
set
cove
r
in
g,
”
Oper
erati
on
s
Res
earc
h
,
vol
.
2
9
,
no
.
3
,
pp
.
501
-
510,
1981
.
[4]
H.
H.
Mill
ar
an
d
M
.
Kira
gu,
“
C
y
c
li
c
and
non
-
c
y
clic
sche
dul
in
g
of
12
h
shift
nurses
by
ne
twork
progra
m
m
ing
,
”
Eur
opean
J
ourn
al
of
Oper
ati
ona
l
R
es
earc
h
,
vol
.
104
,
no
.
3
,
pp
.
5
82
-
592,
1998
.
[5]
Z.
Lü
and
H.
K.
Hao,
“
Adapt
ive
n
ei
ghborhoo
d
sea
rch
for
nu
rse
roster
ing
,
”
Eur
opean
J
ourn
al
of
Oper
at
ion
al
Res
earc
h
,
vol
.
2
18
,
no
.
3
,
pp
.
86
5
-
876,
2012
.
[6]
G.
M.
Jara
dat
,
e
t
al
.
,
“
H
y
brid
E
li
t
ist
-
Ant
S
y
stem
for
Nurs
e
-
Rosteri
ng
Problem,
”
Journal
of
King
Sa
ud
Univ
ersity
–
Computer
and
I
nformation
Sc
ience
s
,
vo
l. 31
,
no.
3
,
pp.
378
-
384
2019.
[7]
M.
Raj
eswari
,
e
t
al
.
,
“
Dire
ct
ed
Bee
Colon
y
Op
ti
m
iz
a
t
ion
Algorit
hm
to
Solve
the
Nurs
e
Rosteri
ng
Proble
m,
”
Comput
ati
onal
I
nte
ll
ige
n
ce and
Neurosci
ence
,
v
ol.
2017
,
pp
.
1
-
2
6,
2017
.
[8]
M.
A.
Aw
ada
ll
a
h,
et al.
,
“
H
y
brid
iz
a
ti
on
of
har
m
o
n
y
sea
r
ch
with
h
il
l
c
li
m
bing
for
highly
const
ra
in
ed
nurse
roster
in
g
proble
m
,
”
Neura
l
Comput
ing
and
Appl
ic
at
ions
,
vo
l.
28
,
no
.
3
,
pp
.
4
63
-
482,
2017
.
[9]
H.
G.
Santos,
e
t
al.
,
“
Inte
ger
p
rogra
m
m
ing
te
c
hnique
s
for
th
e
nurse
roster
ing
proble
m
,
”
Ann
a
ls
of
Oper
ati
ons
Res
earc
h
,
vol
.
2
39
,
no
.
1
,
pp
.
22
5
-
251,
2016
.
[10]
M.
A.
Aw
ada
ll
a
h,
et
al.
,
“
A
hy
b
rid
art
ifici
al
bee
col
on
y
for
a
nurse
roster
ing
proble
m
,
”
Appl
ied
Soft
Comput
ing
,
vol.
35
,
pp
.
726
-
739,
2015
.
[11]
E.
K.
Burke
an
d
T.
Curtoi
s,
“
New
appr
oac
he
s
to
nurse
roster
ing
benc
hm
ark
insta
nce
s,
”
Eur
opean
J
ournal
of
Oper
ati
onal
Res
earc
h
,
vo
l. 237
,
no.
1
,
pp
.
71
-
81
,
2014.
[12]
B.
Bil
gin,
e
t
al
.
,
“
One
hy
pe
r
-
heur
isti
c
appr
oa
ch
to
two
ti
m
et
abl
ing
proble
m
s
in
hea
lt
h
ca
re
,
”
J
ourn
al
of
Heuristic
s
,
vol.
18
,
no
.
3
,
pp
.
401
-
434
,
2012
.
[13]
C.
Valouxi
s,
et
al
.
,
“
A
sy
st
emat
ic
two
phase
ap
proa
ch
for
th
e
nurse
roster
ing
pr
oble
m
,
”
Eur
o
pean
J
ournal
of
Oper
ati
onal
Res
earc
h
,
vo
l. 219
,
no.
2
,
pp
.
425
-
4
33,
2012
.
[14]
K.
Nonobe,
“
INRC2010:
An
appr
oa
ch
usi
ng
a
gen
eral
constra
in
t
op
ti
m
iz
ation
solv
er,
”
The
F
irst
Inte
rnatio
nal
N
urs
e
Roste
ring
Competi
ti
on
(
PA
TAT
-
2010)
,
2010
.
[Online
]
.
Avai
la
bl
e:
ht
t
p:/
/www
.
kule
uv
en
-
kortri
jk
.
be/
nrp
co
m
pet
it
ion
.
[15]
J.
F.
Ba
rd
and
H
.
W
.
Purnom
o,
“
Prefe
ren
c
e
sch
e
duli
ng
for
nurse
s
using
col
um
n
gene
ra
ti
on,
”
European
Journal
o
f
Operational
Re
s
earc
h
,
vo
l. 164
,
no.
2
,
pp
.
510
-
5
34,
2005
.
[16]
H.
G.
Santos,
e
t
al.
,
“
Inte
ger
p
rogra
m
m
ing
te
c
hnique
s
for
th
e
nurse
roster
ing
proble
m
,
”
Anna
ls
of
Operations
Re
search
,
vol
.
2
39
,
no
.
1
,
pp
.
22
5
-
251,
2016
.
[17]
S.
Ceschia,
et
a
l.
,
“
Solving
the
INRC
-
II
nurse
roster
ing
prob
l
emb
y
sim
ula
t
ed
anneal
ing
b
ase
d
on
la
rg
e
-
sca
le
nei
ghborhoods,
”
i
n
Proce
edi
ngs
of
the
12th
In
t
ernati
onal
Conf
enf
ere
n
ce
on
Pr
act
i
ce
and
Theo
ry
of
Aut
omat
ed
Timetabl
ing
(
PA
TAT
-
2018)
,
2018.
[18]
L.
Antoine,
et
a
l.
,
“
A
rota
ti
on
-
b
ase
d
bra
nch
-
and
-
pric
e
app
roa
ch
for
the
nurse
sch
edul
ing
probl
em,
”
Math
emati
ca
l
Program
ming
Computati
on
,
pp.
1
-
34,
2019
.
[19]
C.
Blum
and
A.
Roli
,
“
Met
ahe
ur
isti
cs
in
combinat
ori
al
opt
imizat
ion:
over
v
ie
w
a
nd
conc
ep
tual
c
om
par
ison,
”
AC
M
Comput
ing
Surv
eys
,
vo
l. 35
,
no.
3,
pp
.
268
-
308
,
2003.
[20]
D.
y
v
az,
e
t
a
l.
,
“
Perform
anc
e
ev
a
lua
ti
on
of
evo
lutionar
y
heu
ristics
in
d
y
namic
en
vi
ronm
ent
s,
”
App
l
ied
Int
el
l
ige
n
ce
,
vol.
37
,
no
.
1
,
pp
.
130
-
144
,
2012
.
[21]
H.
R.
Cheshm
ehga
z,
et
a
l.
,
“
Eff
ective
loc
a
l
evol
uti
on
ar
y
se
arc
hes
distr
ibuted
on
an
isla
n
d
m
odel
solving
bi
-
objecti
v
e
opt
i
m
iz
at
ion
problem
s,
”
Appl
i
ed
In
t
el
l
ig
ence
,
vol
.
3
8
,
no
.
3
,
pp
.
331
-
356,
2013
.
[22]
M.
K.
Alsm
adi
,
“
Quer
y
-
sensit
i
ve
sim
il
ari
t
y
m
ea
sure
for
con
t
ent
-
base
d
imag
e
ret
ri
eval
using
m
et
a
-
heur
ist
ic
al
gorit
hm
,
”
Jou
rnal
of
King
Sa
ud
Unive
rs
it
y
-
C
omputer
and
Inf
orm
ati
on
Sci
en
c
es
,
vol.
30,
no
.
3,
pp.
373
-
381
,
2018.
[23]
M.
K.
Alsm
adi,
“
Conte
nt
-
Base
d
Im
age
R
et
ri
e
val
Us
ing
Co
lo
r,
Shape
and
T
e
xture
Descri
p
t
ors
and
Fe
at
ure
s,
”
Arabian
Journal
for Scie
n
ce
a
nd
Engi
ne
ering
,
vol
.
45
,
pp
.
3317
-
3
330
,
2020
.
[24]
M.
K.
Alsm
adi
,
et
a
l.
,
“
Robust
fea
tur
e
ex
tra
c
ti
on
m
et
hods
f
or
ge
ner
al
f
ish
class
ifi
cation,
”
I
nt
ernati
onal
Journal
o
f
El
e
ct
rica
l
&
Co
mputer
Engi
n
ee
r
ing
,
vo
l. 9,
no.
6,
pp.
5192
-
5204,
2019.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
870
8
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
47
1
-
480
480
[25]
M.
K.
Alsm
adi
,
“
Hy
b
rid
Gene
ti
c
Algori
thm
with
Ta
bu
Sear
ch
with
Bac
k
-
Propaga
ti
on
Al
gorit
hm
for
Fis
h
Cla
ss
ifi
c
at
ion
:
D
et
ermining
the
Appropria
te
Fea
ture
Se
t,
”
Int
e
rnational
Journal
of
Applied
Enginee
ring
Re
searc
h
,
vol.
14
,
no
.
23
,
p
p.
4387
-
4396
,
2
019.
[26]
M.
K.
Als
m
adi
,
“
An
eff
ic
ie
n
t
sim
il
ari
t
y
m
ea
sur
e
for
cont
ent
b
ase
d
image
ret
ri
eva
l
us
ing
m
em
et
i
c
al
gori
thm,
”
Egy
pti
an
Journa
l
of
Basic and A
ppli
ed
Scienc
es
,
vol.
4
,
no
.
2
,
pp
.
112
-
122,
2017
.
[27]
M.
Alsm
adi
,
“F
ac
i
al
Re
cogni
t
io
n
under
Expre
ss
ion
Vari
a
ti
ons
,”
The
Inte
rnation
al
Arab
Journal
of
Information
Technol
ogy
,
vo
l. 13,
no.
1A,
pp
.
133
-
141,
2016
.
[28]
M.
Alsm
adi
,
et
al
.
,
“
A
hy
br
id
m
emeti
c
a
lgori
th
m
with
bac
k
-
propa
gation
cl
assifi
er
for
fish
cl
assific
a
ti
on
base
d
o
n
robust
feature
s
ext
ra
ct
ion
from
PLGF
and
shape
m
ea
surem
e
nts,
”
In
formatio
n
Technol
og
y
J
ournal
,
vol
.
10
,
no.
5
,
pp
.
944
-
9
54,
2011
.
[29]
M.
Alsm
adi
,
et
al
.
,
“
Fis
h
Cla
ss
i
fic
a
ti
on:
Fis
h
Cla
ss
ifi
c
at
ion
Us
ing
Mem
et
ic
Algori
thms
with
Bac
k
Propaga
t
io
n
Cla
ss
ifi
er
,”
LAP L
ambert
Aca
d
e
m
ic
Publishing,
2012.
[30]
K.
Yous
ef
,
et
al
.
,
“
Appl
y
ing
the
big
bang
-
b
i
g
cru
nch
m
et
a
heur
isti
c
to
la
r
ge
-
size
d
oper
at
i
onal
probl
ems
,
”
The
Inte
rnat
iona
l
Journal
of
Elec
tric
al
and
Comp
ute
r E
ng
ine
ering
(
IJE
CE)
,
v
ol
.
1
0,
no
.
3
,
pp
.
284
8
-
2502,
2020
.
[31]
A.
Abuham
dah,
“
Adapti
ve
e
li
t
ist
-
ant
s
y
s
te
m
for
m
edi
ca
l
cl
ust
ering
proble
m
,
”
The
Journal
of
K
in
g
Saud
Univ
ersit
y
-
Computer
and
Information
Sc
ience
s
,
v
ol
.
32,
pp.
709
-
717
,
2020
.
[32]
A.
Abuham
dah,
“
Adapti
ve
Acc
ept
a
n
ce
Criterio
n
(AA
C)
al
gorit
hm
for
Optimiza
ti
on
Problems
,
”
T
he
Journal
of
Computer
Scien
ce
,
v
ol
.
11
,
no
.
4
,
pp
.
675
-
691
,
2
015.
[33]
A.
Abuham
dah,
et
al.
,
“
Adapti
v
e
Grea
t
De
luge
(AG
D)
for
Medic
a
l
Clust
eri
ng
Problem
,
”
T
he
I
nt.
J.
Eme
rg.
Sc
i
.
(
IJE
S)
, v
ol
.
4
,
n
o.
1
,
pp
.
1
-
13
,
2
014.
[34]
A.
Abuham
dah,
et
al
.
,
“
H
y
brid
iz
a
ti
on
Bet
we
en
Ite
rative
Sim
ula
te
d
Annea
li
ng
and
Modifie
d
Grea
t
Delug
e
for
Medic
a
l
Cluste
r
i
ng
Problems
,
”
World
of
Computer
Scienc
e
and
Information
Tec
hnology
Journal
(
WCSIT
)
,
vol.
2
,
n
o.
4
,
pp
.
131
-
1
36,
2012
.
[35]
A.
Abuham
dah
and
M.
A
y
ob,
“
Adapti
ve
Rando
m
iz
ed
Desce
nt
Algorit
hm
using
Round
Robin
f
or
Solv
ing
Course
Ti
m
et
ab
li
ng
Pro
ble
m
s
,
”
Int
ernat
ional
Conf
ere
nc
e
on
Int
ellige
n
t
Syste
ms
Design
and
Appl
i
cat
ion
s
(
ISDA
)
IEE
E
,
pp.
1201
-
1206
,
2010
.
[36]
A.
Abuham
dah,
et
al
.
,
“
Popula
ti
on
base
d
loca
l
sea
r
ch
for
un
ive
rsi
t
y
cour
se
ti
m
et
abling
pro
ble
m
s,
”
App
l
ied
Inte
ll
ige
n
ce
,
vol
.
40
,
no.
1,
pp.
44
-
53,
2013
.
[37]
S.
Haspesla
gh,
et
al.
,
“
The
fir
st
int
ern
ational
nurse
roster
ing
competi
ti
on
,
”
A
nnals
of
Opera
ti
ons
Re
search
,
vol.
218
,
no
.
1
,
p
p.
221
-
236
,
201
4.
[38]
S.
Ceschia,
et
al.
,
“
The
se
cond
in
te
rna
ti
ona
l
nurse
roster
ing
compe
ti
ti
on
,
”
Annal
s
o
f
Operations
R
ese
arch
,
vo
l.
274
,
pp.
171
-
186
,
20
1
9
.
[39]
B
.
L
.
W
ebster,
“
Solving
combinat
ori
al
opti
m
i
za
t
ion
probl
ems
using
a
new
a
lgori
thm
bas
ed
on
g
rav
i
ta
t
iona
l
at
tr
ac
t
ion,
”
PhD
the
sis,
Co
ll
eg
e
o
f
Engi
n
ee
ring
a
t F
lori
da
Inst
it
ut
e of
Technol
og
y
,
2004.
[40]
A.
Abuham
dah
and
M.
A
y
ob,
“
MP
CA
-
AR
DA
f
or
solvi
ng
cour
s
e
ti
m
etabli
ng
pr
oble
m
s,
”
i
n
3rd
conf
ere
n
ce
on
d
ata
mining
and
op
tim
izati
on
(
DMO
)
,
New York
,
pp
.
171
-
177,
2011
.
[41]
I.
Chebbi
,
e
t
al.
,
“
Big
dat
a:
C
once
pts,
ch
al
l
en
ges
and
appl
ica
ti
ons
,
”
i
n
Computati
onal
co
ll
e
ctive
intelligen
ce
,
Springer,
Ch
am,
pp.
638
-
647,
20
15.
[42]
I.
Chebbi
,
e
t
al.
“
A
compari
son
of
big
remote
sensing
dat
a
proc
e
ss
ing
with
Hado
op
MapReduc
e
and
Spark
,
”
in
4
th
Inte
rnational
C
onfe
renc
e
on
A
dvanc
ed
Te
chno
logi
es
for
Signa
l
and
Image
Pr
oce
ss
ing
(
ATSIP)
,
IEE
E
,
pp.
1
-
4
,
2018.
[43]
W
.
Bouli
l
a,
et
al
.
,
“
A
novel
d
ec
ision
support
s
y
stem
for
th
e
i
nte
rpre
ta
t
ion
of
remote
sensing
big
data
,”
Eart
h
Sci
en
ce Inf
orm
at
ic
s
,
vo
l. 11
,
no.
1,
pp
.
31
-
45
,
20
18.
[44]
W
.
Bouli
la,
“
A
top
-
down
appr
oac
h
for
sem
an
ti
c
segm
entati
on
of
big
remote
sensing
images
,”
Earth
S
ci
en
ce
Informatic
s
,
vo
l.
12
,
no.
3,
pp.
29
5
-
306,
2019
.
[45]
I.
R.
Fara
h
,
e
t
al
.
,
“
Inte
rpre
tati
on
of
m
ult
i
-
sensor
remote
sensing
images:
m
ult
i
-
app
roa
ch
fusi
on
of
unce
rt
ai
n
informati
on,
”
I
E
EE
Tr
ansacti
ons
on
Geosci
ence and R
emot
e
S
en
sing
,
vol
.
46
,
no
.
12,
pp.
4142
-
41
52,
2008
.
[46]
I.
R.
Fara
h
,
e
t
a
l.
,
“
Multi
appr
oa
ch
s
y
stem
base
d
on
fusion
of
mul
ti
spec
tral
imag
es
for
la
nd
-
cov
e
r
cl
assifi
ca
t
ion,”
IEE
E
Tr
ansacti
o
ns on
Geosci
ence
and
Re
mot
e
S
e
nsing
,
vol
.
46
,
n
o.
12
,
pp
.
4153
-
4161,
2018
.
[47]
A.
Ferc
hi
chi,
et
al
.
,
“
Reducing
u
nce
rt
ai
nt
ie
s
in
l
a
nd
cove
r
ch
ange
m
odel
s
using
s
ensit
ivit
y
ana
l
y
s
is,”
Know
le
dge
and
Information Syste
ms
,
vol
.
55
,
no
.
3,
pp.
719
-
7
40,
2018
.
[48]
W
.
Bouli
l
a,
al.,
“
Com
bini
ng
Dec
ision
Fus
ion
and
Unce
r
ta
in
t
y
Propaga
ti
on
to
Im
prove
La
nd
Cover
Chang
e
Predic
ti
on
in
Sa
te
llite
Im
age
Da
ta
base
s,
”
The
Journal
of
Mult
imedi
a
Proce
ss
ing
and
Technol
ogie
s
,
vol.
2,
no.
3,
pp.
127
-
139
,
20
11.
[49]
W
.
Bouli
la,
et
al.,
“
Sensiti
vi
t
y
an
aly
s
is a
pproa
ch
t
o
m
odel
epi
stemic
and
a
leator
y
i
m
per
fec
ti
on
:
Applicati
on
to La
n
d
Cover
Chang
e
p
red
iction
m
odel,
”
Journal
of
com
putat
ional sci
en
ce
,
vol
.
23
,
pp
.
5
8
-
70,
2017
.
Evaluation Warning : The document was created with Spire.PDF for Python.