TELKOM
NIKA
, Vol. 11, No. 5, May 2013, pp. 2508 ~
2515
ISSN: 2302-4
046
2508
Re
cei
v
ed
De
cem
ber 1
2
, 2012; Re
vi
sed
March 11, 20
13; Accepted
March 20, 20
13
Automated Guide Vehicles Dynamic Scheduling
Based on Annealin
g Genetic Algorithm
Zou Gan*
1
,
Li Tao
1,
2
,
Qi
n Ying
1,2
1
Mechan
ical
a
nd Electric
al E
ngi
neer
in
g Col
l
ege, K
unm
in
g Univers
i
t
y
of Scienc
e an
d T
e
chno
log
y
Kunmi
ng, 65
01
18, PR Chi
na.
2
Chin
a Ship
bu
il
din
g
T
r
ading (Kunmi
ng) Co., Ltd,650
05
1, PR Chin
a.
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: zouga
n@h
o
t
m
ail.com*
1
, zxs
t
@126.com
2
Abstrac
t
Dispatc
h
in
g au
tomat
ed g
u
id
e
d
veh
i
cles (AG
V
s)
is the co
mmo
n
a
ppro
a
ch
for AGVs schedu
lin
g
in practic
e
, the informatio
n
ab
out loa
d
arriva
l
s
in
adva
n
ce w
a
s not used to
opti
m
i
z
e th
e pe
rforma
nce of th
e
auto
m
ate
d
gu
i
ded ve
hicl
es system (AGVsS). Accord
ing
to the chara
c
teristics of the AGVsS, th
e
math
e
m
atic
al
mo
de
l of AGVs schedu
li
ng w
a
s establ
ish
ed.
A heuristic a
l
g
o
rith
m cal
l
ed A
nne
ali
ng Ge
ne
tic
Algorit
h
m
(AGA) w
a
s prese
n
t
ed to d
eal w
i
t
h
the AG
Vs s
c
hed
uli
ng
prob
le
m, an
d a
ppl
i
ed the
al
gorith
m
dyna
mical
l
y by
usi
ng
it re
peat
edly
un
der
a c
o
mbi
ned
ro
l
l
i
n
g o
p
ti
mi
z
a
ti
on
strategy. the
p
e
rformanc
e of
the
prop
osed a
ppr
oach for AGVs schedu
lin
g w
a
s comp
are
d
w
i
th the dispatch
ing rul
e
s by si
mu
lati
on. Resu
lt
s
show
ed th
at the
appr
oac
h p
e
rforms
sig
n
ifi
c
antly b
e
tter t
han
the
disp
at
chin
g ru
les
an
d prov
ed
that
it is
really
effective for AGVsS.
Key
w
ords
:
a
u
tomated
gu
i
de ve
hicl
es (
A
GVs), dyna
mic
sc
hed
ul
in
g, rolli
ng
opti
m
i
z
at
io
n strategy,
ann
eal
in
g ge
n
e
tic alg
o
rith
m (
A
GA)
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Automated
G
u
ided V
ehi
cl
es
(AGVs)
a
r
e u
n
man
n
e
d
mea
n
s of
inplant tran
sportation
whi
c
h are commonly used in facilities
such as
warehouses, manuf
acturi
ng plants, terminal
s and
distrib
u
tion
centers. AGV
s
in
crea
se
e
fficienc
y
and
red
u
ce
co
sts by
helpi
ng
to auto
m
ate
a
manufa
c
turi
n
g
facility o
r
wareh
o
u
s
e [1].
Appli
c
ation
o
f
AGVs
ha
s b
een
bro
ade
n
ed
sin
c
e th
e l
a
te
20th centu
r
y. In the past few decade
s, much rese
a
r
ch has be
en d
e
voted to the improveme
n
t of
AGVs syste
m
(AGVsS).
Basically, the releva
nt
issue
s
of AGVsS can b
e
divided into
the
followin
g
mai
n
cate
go
rie
s
: guide
-path
desi
gn, e
s
tim
a
ting the
req
u
ired
num
ber of AGVs, id
le
AGVs p
o
sitio
n
ing, AGV
s
sche
duling,
ba
ttery m
ana
ge
ment an
d
co
nflict re
sol
u
tion. The
s
e
issues
relate
to diffe
rent level
s
of
the d
e
ci
sio
n
makin
g
pro
c
e
s
s [2]. As on
e
of the
en
abli
ng te
chn
o
logi
es,
sched
uling of
AGVs have
attracte
d co
n
s
ide
r
abl
e atte
ntion. The A
G
Vs
sch
eduli
ng is
re
spo
n
sible
for managi
ng
AGVs efficiently to guarantee se
rve
j
obs a
s
qui
ck as possibl
e. Many rules
or
algorith
m
s for the sch
eduli
ng of AGVs h
a
ve been p
r
o
posed [2-5].
In gene
ral,
AGVsS can
be divide
d
into two
main
catego
ries:
ce
ntrali
zed
and
dec
ent
rali
ze
d
co
nt
rol
sy
st
e
m
s.
Th
e cent
ralized syste
m
u
s
es
info
rm
ation avail
a
ble at th
e
ce
ntral
controlle
r whi
l
e the decentralize
d
syste
m
dispat
ch
e
s
A
G
Vs ba
se
d o
n
local info
rm
ation availabl
e
at the d
e
ci
sio
n
mom
ent [3]
.
In practi
ce,
due to
thei
r e
fficiency,
cen
t
ralize
d
cont
rol sy
stem
s a
r
e
more
po
pula
r
. Since
di
spat
chin
g rule
s a
r
e quite
st
raig
htforwa
r
d
an
d ea
sy to
use
.
A wid
e
vari
e
t
y
of dispatchin
g rule
s in
clu
d
ing
nea
re
st-workstation
-first
(NWF
), n
eare
s
t ve
hicl
e first (NVF)
and
modified firs
t
c
o
me firs
t s
e
rved (MOD-FCFS) ar
e use
d
for centralized cont
rol AGVsS [4]. Ba
sed
on the
ways
in which tran
spo
r
tation
re
que
sts
ar
e a
ssi
gne
d, Egb
e
lu h
a
ve divi
ded
dispatchi
ng
rule
s into two
catego
rie
s
: load-i
n
itiated
disp
atch
i
ng rules in
whi
c
h
jobs at wo
rkstation have t
h
e
prio
rity to cl
aim vehi
cle
s
and
vehi
cle
-
initiat
ed
dispatchi
ng
rule
s in
which v
ehicl
es have
the
prio
rity to clai
m jobs [5]. Di
spat
chin
g is
related to imm
ediate d
e
ci
si
ons
su
ch
as
whe
r
e a ve
hi
cle
sho
u
ld be se
nt to at a
spe
c
ific mome
nt. Howeve
r,
if
advan
ce load
arrival information is kn
o
w
n,
they are outp
e
rform
ed by sche
duling a
p
p
roa
c
h
e
s [2].
In the literat
ure, vehi
cle
routing
pro
b
l
e
ms
with time wind
ows
(VRPTW) hav
e been
studie
d
exten
s
ively.
B.L. Golden
et al.
provid
e a
re
view of v
ehi
cle ro
uting
with time wi
ndo
ws
inclu
d
ing
the Pick-up and Delie
ry
Probl
em
with
Tim
e
Wind
ow
(PDPTW), multi
-
Travel Sale
sman
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2508– 251
5
2509
Problem
with
Time Wind
o
w
(m
-TSPT
W) and solutio
n
s app
roa
c
h
e
s
[6].They me
ntion two ma
in
types of optimization al
go
rithms for V
R
PT
W: dyna
mic pro
g
ram
m
ing and b
r
anch-a
n
d
-
bo
und.
Both method
s a
r
e very t
i
me co
nsumi
ng an
d
can
not
solve practical
probl
ems within an
accepta
b
le time limit. Xu
et al. propose the co
lumn
-gen
eration a
l
gorithm to solve VRPTW.
In
orde
r to d
eal
with practi
cal
-
size proble
m
s. They u
s
e
several h
e
u
r
ist
i
cs to
gen
erat
e col
u
mn
s wit
h
negative re
d
u
ce
d co
sts and elimin
ate unattra
c
tive column
s by sop
h
i
s
ticate
d col
u
mn
manag
eme
n
t scheme
s
[7].
Wenfe
ng
Wa
ng et al.
pr
e
s
ent an i
m
proved ge
netic al
gorithm
to sol
v
e
the VRPTW [8]. Corde
au e
t
al. summari
zed
several
o
f
the most important an
d mode
rn heu
ri
stics
for vehicle ro
uting pro
b
le
m [9], Victor
Pillac et
al. provide a surv
ey on solutio
n
approa
che
s
for
dynamic vehi
cle ro
uting problem
s [10].Two main
ap
proache
s inclu
de an ada
pt
a
t
ion of the static
solutio
n
and
an implem
ent
ation of stat
ic algorithm
s u
nder a
rolling
hori
z
on.
After studying
the literature on
the vehicl
e sched
uling,
we fi
nd t
hat
most
st
u
d
ie
s con
c
e
r
n
external tran
spo
r
t sy
stem
s. Fe
w aut
h
o
rs u
s
e m
e
thod
s succe
s
sfully depl
oyed for
external
transpo
rt pro
b
lems for i
n
te
rnal t
r
an
spo
r
t
like
th
e AGV
s
S. The
probl
em of dyn
a
mi
c
scheduli
ng
o
f
AGVs h
a
s
no
t attracted
m
any re
se
arch
ers [2]. Since
the sch
eduli
ng ap
proa
ch
has
bee
n u
s
ed
widely a
nd ef
ficiently for e
x
ternal tra
n
sp
ort. We
devel
op such sch
e
duling
strate
g
y
for AGVsS
in
this pa
per. T
he main
purpose is to in
vestigat
e the
potential co
ntribution
of the sche
duli
n
g
approa
ch for
AGVs. AGVs sch
eduli
ng p
r
oble
m
is si
m
ilar to the vehicle
sched
uling pro
b
lem f
o
r
external tra
n
s
po
rt. Ho
we
ver, it also has
some
di
fferences
su
ch a
s
the o
b
jective
s
, tra
v
e
l
distan
ce
s an
d load a
rrival
rate. There i
s
so
me
litera
t
ure on
sche
duling met
h
o
d
s for AGV
s
[3
]
[11]. Several obje
c
tives are also u
s
e
d
, inclu
d
ing min
i
mizing the a
v
erage lo
ad
waiting time
and
minimisi
ng e
m
pty vehicle
travel dista
n
ces, but
they focu
s on
static problem
s. An AGVsS is a
system
with
a high degree of
uncertai
nty, short
travel times and, often, high
vehicle utilization
rates. In AGVsS, normally,
we only kno
w
a limit
ed numbe
r of jobs in advan
ce.
The sched
ule of
vehicle
s
sho
u
ld be u
pdat
ed wh
en n
e
w
tran
sp
ortat
i
on re
que
st informatio
n arrives. Thi
s
p
aper
focu
se
s mainl
y
on dynamic sch
edulin
g p
r
oble
m
of
AGVs, adapts
su
ccessful dyna
mic sche
dulin
g
approa
ch for AGVsS, and comp
are
s
its perfo
rma
n
ce
s with co
mmon dispat
chin
g rule
s by
s
i
mulation.
The remai
n
d
e
r of thi
s
pa
per i
s
o
r
ga
ni
zed
as follo
ws: In
se
ctio
n 2, we de
scrib
e
the
mathemati
c
al
model, Ann
e
a
ling G
eneti
c
Algorithm
(A
GA) an
d the
rolling o
p
timization st
rategy
o
f
AGVs
sched
uling
pro
b
lem
.
Re
sults an
d
discu
s
sion
a
r
e p
r
e
s
e
n
ted
in sectio
n 3.
we
co
ncl
ude
the
pape
r in Secti
on 4.
2. Rese
arch
Metho
d
2.1.
Mathem
atical Model
of AGV
s
Sch
e
duling
In AGVsS, AGVs picks-up load
s at
some l
o
catio
n
s an
d deliv
ers th
em to their
destin
a
tion
s satisfying ce
rtain
time-win
dows.
the A
G
Vs
scheduli
ng proble
m
can be fo
rmul
ated
as
a V
R
PTW. m-TSPT
W i
s
a
spe
c
ial
case
of th
e V
R
PTW in
whi
c
h th
e ve
hicl
e capa
citie
s
are
infinite [6]. We formul
ate the sch
eduli
n
g pro
b
le
m fo
r AGVs a
s
a
m-TSPTW
by proje
c
ting ti
me-
wind
ows at delivery stait
ons to the
corre
s
p
ondin
g
pick-u
p st
ations a
nd a
pick-up a
n
d
a
corre
s
p
ondin
g
delivery job
is co
nsi
dere
d
logica
lly as a singl
e job-node. We ma
ke the follo
wi
ng
assumptio
n
s
for the studyi
ng AGVsS:
All AGVs have unit-loa
d
ca
pacity.
AGVs ope
rat
e
contin
uou
sl
y without bre
a
kd
own.
There are n
o
traffic pro
b
le
ms.
AGVs ch
oo
se
the sho
r
test
path to pick u
p
and delive
r
load
s.
AGV loading
and unl
oadin
g
times are fixed and con
s
idere
d
in trav
el times.
AGVs ca
n al
ways p
a
rk at their drop-off locatio
n
s.
Only one de
p
o
t in AGVsS.
the following notations will
help in the descripti
on of formulation of AG
Vs scheduling problem.
D
— AGVs
depot.
N
— set of jobs.
K
— s
e
t of AGVs
.
ijk
x
— take the value 1 if job
i
and job
j
are served by AG
V
k
Conse
c
uti
v
ely, and 0 otherwise.
i
e
— relea
s
e time of job
i
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Auto
m
a
ted
Guide Vehicles Dynam
ic Scheduling Based on Annealing Genetic …
(
Z
ou Gan)
2510
i
l
— the latest pick-u
p time allowed of job
i
.
i
s
— pick-up ti
me of job
i
.
ij
t
— T
h
e
travel
time from j
o
b
-
nod
e
i
to
j
, eq
uals the
trave
l
time from th
e o
r
igin
of jo
b
i
to the
destin
a
tion of
job
i
, plus the
travel time from the destin
a
tion of job
i
to the origin of
job
j
.
Dk
s
— the s
t
art time of AGV
k
at the AGVs depot.
kD
s
— the arrival
time of AGV
k
at the AGVs depot.
B
— a sufficien
tly large positi
v
e numbe
r.
Normally, mi
nimizin
g
the
averag
e job
waiting
tim
e
i
s
the m
o
st i
m
porta
nt obj
ective of
AGVs sche
du
ling pro
b
lem.
The AGVs
scheduli
ng prob
lem can b
e
formulate
d
as f
o
llows:
N
i
i
i
e
s
N
mize
i
min
1
(1)
s
ubjec
t to:
N
i
,
x
K
kN
j
ijk
1
(2)
K
k
,
N
i
,
x
x
N
D
j
jik
N
D
j
ijk
(3)
K
x
x
K
kN
i
Dik
K
kN
j
jDk
(4)
K
k
,
N
j
,
x
B
s
t
s
Djk
j
Dj
Dk
1
(5)
K
k
,
N
j
,
i
,
x
B
s
t
s
ijk
j
ij
i
1
(6)
K
k
,
N
i
,
x
B
s
t
s
iDk
kD
iD
i
1
(7)
N
D
i
,
l
s
e
i
i
i
(8)
Con
s
trai
nts (2)-(4
) form a
multi-commo
dity flow fo
rm
ulation. The
constraint
(5
)
sho
w
s that if an
AGV
k
serve
s
job
j
after job
i
, the const
r
ain
t
j
ij
i
s
t
s
must be satisfied. Con
s
traints (5
)-(7
)
ensure fea
s
ib
ility of
the sch
edule. Equat
i
ons (9) i
s
time-wi
ndo
w for
job
i
.
For the stati
c
AGVs sched
uling problem
, we
assume
that all AGVs start at the
AGVs
depot.
Ho
we
ver, du
ring
the p
r
o
c
e
s
s
of optimizat
io
n, AGVs m
a
y start at
an
y load’
s d
r
op
-off
station. The
r
efore, we ne
ed to modify
the formul
ation to refle
c
t this by re
pla
c
in
g (4
) and
(6
)
by
con
s
trai
nts (9
) and (10)
re
spectively, wh
ere
k
D
is the virtual startin
g
d
epot of AGV
k
.
K
x
x
K
kN
j
jDk
K
kN
i
ik
D
k
(9)
K
k
,
N
j
,
x
B
s
t
s
jk
D
j
j
D
k
D
k
k
k
1
(10
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2508– 251
5
2511
2.2. Anneali
ng Gene
tic Algorithm (AGA)
Thoug
h opti
m
al solution
s to AGVs
sch
edulin
g probl
em can b
e
o
b
tained
usi
n
g
exact
method
s, the
comput
ation
a
l time requi
red to so
lve
this pro
b
le
m to optimality is prohibit
i
ve
.
Heu
r
isti
c met
hod
s su
ch a
s
Gen
e
tic Algorithm
s (GA
)
,Simulated
Annealin
g (S
A) often prod
uce
optimal
o
r
ne
ar optimal sol
u
tions
in a re
aso
nabl
e am
ount of
co
mp
uter time.
He
uristi
c m
e
tho
d
s
have been p
r
oved useful in a variet
y of search a
nd
optimizatio
n
probl
em
s over the years. But
most of si
ngl
e heu
risti
c
al
gorithm
s h
a
ve their o
w
n
shortcomin
gs i
n
the optimi
z
ation [12]. In many
probl
em
s, GA may have a tenden
cy to conve
r
g
e
towa
rd
s lo
cal
optima rath
er than the glo
bal
optimum
of t
he p
r
o
b
lem [
13]. The
exe
c
ution
time
of SA is sen
s
itive to the
scal
e of th
e p
r
o
b
l
e
m.
In dealin
g wit
h
larg
e-scale
issue,
the
co
mputational ti
me of SA for
getting a
sati
sfacto
ry solut
i
on
is not acce
ptable [14]. Accordin
g to the cha
r
a
c
teri
stics of AGVsS, we propo
se a
hybrid algo
rithm
calle
d Ann
e
a
ling
Geneti
c
Algo
rithm
(AGA) to
op
timize AGV
s
sche
duling
pro
b
lem. A
G
A
facilitates the exhaustive and para
ll
el treatment of the problem
an
d to increase t
he probability of
finding glob
al
minimums in
a reasona
ble
computatio
n time.
The propo
se
d AGA is describ
ed a
s
follows:
Step1:
Initial
i
ze th
e pa
ra
meters, i.e., pop
ulation
size
p
opsize
, Maximum g
ene
ra
tions
of
evolution
gen
MAX
,crossover probability
c
p
,mutation rate
m
p
, initial tem
perature
0
T
,
temperature cooli
ng
coefficient
q
, final te
mperature
end
T
.
Step2:
ge
nerate i
n
itial popul
ation
0
P
, compute indi
viduals’ fitness val
u
e
popsize
,
,
,
i
f
i
2
1
.
Step 3:
set cycle co
unt variable
0
gen
.
Step 4:
ope
ration of sel
e
ction, cro
s
so
ver and m
u
ta
tion, com
p
u
t
e the fitness value of n
e
w
individual
'
i
f
, if
i
'
i
f
f
,new in
dividua
l is acce
pted,
otherwise
, n
e
w individ
ual
is a
c
cepted
with a
probability
)
T
)
f
f
exp((
p
i
'
i
.
Step 5:
if
MAXgen
gen
, then
1
gen
gen
, go to step 4, otherwi
se ,go to step 6.
Step 6:
if
end
i
T
T
,
Termin
ate th
e alg
o
rithm
a
nd p
r
int the
solution, oth
e
rwise,
i
i
qT
T
1
,
go
to step 3.
The fram
ewo
r
k of the AGA
is given in Figure 1.
In this
pap
er,j
obs have
be
e
n
num
be
red
according
to i
n
crea
sing
job
s
’ relea
s
e
tim
e
, a
solutio
n
is e
n
co
ded a
s
a
chromo
som
e
formed
of multiple se
gments.
kj
x
rep
r
esents AGV
k
serv
e
s
jo
b
j
,a ch
romo
som
e
of the A
G
VsS sch
e
d
u
ling p
r
obl
e
m
ca
n be
e
x
presse
d a
s
(
,
0
,
,
,
,
,
0
,
,
,
i
0
2
22
21
1
12
11
t
s
i
i
i
i
i
,
,
0
,
,
,
,
,
0
,
2
1
mw
m
m
i
i
i
)
with
0
repre
s
ent
s the
AG
Vs de
pot. Fo
r
example, the
chromo
som
e
0
,
8
,
7
,
0
,
6
,
5
,
4
,
0
,
3
,
2
,
1
,
0
expre
ss
3 AGVs se
rve
8 jobs, It indicate
s that
AGV 1 s
e
rves
0
,
3
,
2
,
1
,
0
in se
quen
ce, AGV 2 serves
0
,
6
,
5
,
4
,
0
in se
que
nce a
nd AGV
3 serve
s
0
,
8
,
7
,
0
in seq
uen
ce
respe
c
tively. Roul
ette-whe
el schem
e is
applie
d in the
sele
ction p
r
o
c
ed
ure.
The metho
d
of gene
ration
of the initial popul
ation,
the crossove
r o
perato
r
an
d
mutation op
erator
in this pap
er
are o
r
igin
ated
from literature [8].
2.3. Rolling Optimiza
tion
Strate
g
y
for
AGVs Sche
duling
In AGVsS, we may kno
w
informatio
n a
bout
job
s
’ arrival in advance. Base
d on
this
informatio
n, we can u
s
e rolling optimi
z
ation stra
te
gy to sche
dule
AGVs. Whe
n
an AGV start
s
to
serve
a job, it
has to fini
sh
it. Cancellatio
n
of
job
s
is n
o
t allowe
d. Normally, there
are two rolli
n
g
optimizatio
n strategie
s
, i.e. rolling
by time and
ro
lli
ng
by the num
be
r of job
s
[2]. F
o
r the
rollin
g
by
time, we sch
edule all
kn
o
w
n job
s
d
u
rin
g
a time pe
ri
od
H
, Dependi
ng on diffe
re
nt con
d
itions,
the numbe
r o
f
sche
duled j
obs can diffe
r signifi
cantly
for the time
hori
z
on
H
. Howev
e
r, AGVs
only follow th
e re
sultin
g
schedul
e du
rin
g
a time
peri
o
d
1
H
h
After the time period
h
the
system invo
kes the
sche
duling al
go
rithm agai
n
to
sched
ule al
l kno
w
n jo
bs in the pe
ri
od
H
h
h
,
. The pro
c
e
s
s stop
s when
all jobs have
been
finishe
d
. Rolling o
p
timization
stra
tegy by
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Auto
m
a
ted
Guide Vehicles Dynam
ic Scheduling Based on Annealing Genetic …
(
Z
ou Gan)
2512
the num
ber
of jobs i
s
simil
a
r to the one rolling by
time, the differences
are rol
ling time peri
od
H
,
h
repla
c
in
g b
y
the rolling n
u
mbe
r
of the jobs
M
,
m
respe
c
ti
vely (sho
wn i
n
Figure 2).
Figure 1. The
framework of
t
he anneali
n
g geneti
c
alg
o
rithm (AGA
)
i
t
1
i
t
im
m
1
i
Figure 2. Roll
ing optimization strat
egies illustrati
on (by time - left a
nd by the number of jobs-
right)
As sh
own in
Figure 2, in rolling by time
po
licy, the n
u
mbe
r
of sch
edule
d
job
s
a
t
each
step ca
n differ signifi
cantly.
Wh
en
too m
any
job
s
are t
a
ke
n into
a
c
count, the
ru
n
n
ing tim
e
of t
he
sched
uling al
gorithm may
increa
se sig
n
ificantly
an
d
may not m
eet the real
time sche
duli
n
g
requi
rem
ents.
Whe
n
we
use rollin
g opti
m
ization
stra
tegy by the nu
mber
of jobs,
sometim
e
s, t
he
numbe
r of j
o
bs
we
kn
ow i
n
advan
ce
m
a
y be n
o
t en
ough. In thi
s
pape
r, we p
r
opo
se a
com
b
ined
rolling optimi
z
ation st
rategy
.When the
numb
er of jobs known in
adv
ance is suffici
ent
M
i
, we
apply the rolling optimi
z
ati
on strategy b
y
the numbe
r of jobs, othe
rwi
s
e
rolling
by time policy is
use
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2508– 251
5
2513
3. Results a
nd Analy
s
is
In this
study, the layout
we
have
sele
cte
d
is
sh
own in
Figure 3.
We
model th
e A
G
VsS
in Plant 9 si
mulation
software. In our
experim
en
t, the job
s
’ inter-arrival times for the loadi
ng
station
s
a
r
e expon
entially distribute
d
with inter arrival
times equal
80
70
60
50
40
5
4
3
2
1
,
,
,
,
,
,
,
,
seconds. T
h
e probability that
a load is
sent f
r
om a
loadin
g
statio
n
i
to a unlo
ading
station
j
are
ij
p
(
n
p
ij
/
1
,
n
is th
e numb
e
r of
unloadi
ng
station
s
).Fo
r
the numbe
r
of AGVs, 3 levels have b
een u
s
ed
6
5
4
,
,
. Informatio
n a
bout job
s
’
arrival
s
is
kn
own in a
d
van
c
e of 240
se
con
d
s.
the sp
eed of AGVs is 2 m/s. The para
m
eters for
the combin
e
d
rolli
ng
strategy and
the AGA:
250
H
,
25
M
,
8
.
0
,
30
popsize
,
200
gen
MAX
,
6
.
0
c
p
,
05
.
0
m
p
,
1000
0
T
,
9
.
0
q
,
1
end
T
.
The m
a
in
pe
rforma
nce
cri
t
erion
is min
i
mizing
the
averag
e jo
b
waiting
time. The
se
con
dary
ob
jective is mini
mizing
the m
a
ximum
job
waiting time. Result
s of the
experim
ents
are
sho
w
n in Ta
b
l
e 1.
Figure 3. The
experime
n
tal
AGVsS layout
Table 1. Re
sults of differe
nt sch
eduli
n
g
strategi
e
s
(±
rep
r
e
s
ent
s the 95% confid
ence interval
)
No.AGVs
4 5 6
Method
MOD-
FCFS
NVF
ROLL
-
AG
A
MOD-
FCFS
NVF
ROLL
-
AG
A
MOD-
FCFS
NVF
ROLL
-
AG
A
A
v
er-wai
t
(s
)
192.8
(±5.6)
136.5
(±3.8)
65.3
(±2.2)
132.5
(±3.1)
87.9
(±2.5)
34.8
(±1.8)
93.6
(±2.7)
68.4
(±2.3)
17.2
(±1.1)
Max
-
wait
(s
)
958.4
1267.2
792.3
407.6
563.4
354.7
187.7
239.3
130.6
No.
AGVs: nu
mber of
AGVs;
A
v
er
-wait:averag
e
job waiti
n
g
t
i
me
;
Ma
x
-
w
a
i
t
:
ma
x
i
mu
m j
o
b
w
a
i
t
i
n
g
t
i
me
;
MO
D
-
FC
F
S
:
m
odified first com
e
first served; NVF:
nearest ve
hicle first;
RO
LL-AG
A: Rolli
ng optim
i
z
a
t
ion strat
egy based on annealing
genetic algorithm.
Table 1 a
nd
Figure 4 sh
o
w
that the Mo
dified First Come First Served (MO
D
-F
CFS) i
s
the wo
rst stra
tegy for the A
G
VsS in
all th
ree
ca
se
s. Th
e Ne
arest Ve
hicle
First (NVF) pe
rform
s
a
bit better tha
n
the
MO
D-F
C
FS. Th
e av
erag
e j
ob
wa
iting time
s a
nd the
maxi
mum jo
b
wai
t
ing
times obtai
ni
ng by Rolli
ng
optimizatio
n strategy
b
a
se
d on Anne
ali
ng Gen
e
tic A
l
gorithm
(RO
L
L
-
AGA) are si
g
n
ificantly sm
a
ller than th
e
MOD-F
C
FS a
nd the
NVF strategy, whi
c
h are
amon
g
th
e
best di
spat
ch
ing rul
e
s in li
teratures. Th
e ROL
L
-AGA
sho
w
s
goo
d
performan
ce
for the AGVs
sched
uling a
nd prove to b
e
robu
st to different ope
rati
ng co
ndition
s.
In general, we have found
that the dispat
chin
g rule
s such as MO
D-F
C
FS an
d NVF
don’t perfo
rm
well for the
AGVsS. The disp
atchi
ng rules di
sp
atch
AGVs only base
d
on di
sta
n
ce
betwe
en the l
oad
s and th
e
AGVs or th
e
jobs’
waiting
time, the information ab
out
jobs a
rrival
s
in
advan
ce i
s
n
o
t co
nsi
dere
d
an
d in
crea
se th
e ave
r
a
ge a
nd m
a
ximum jo
b
wai
t
ing times.
T
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Auto
m
a
ted
Guide Vehicles Dynam
ic Scheduling Based on Annealing Genetic …
(
Z
ou Gan)
2514
AGVsS is a
system of u
n
certainty, ba
se
d on th
e info
rmation a
bout
job
s
a
rrival
s
in advan
ce, t
he
prop
osed RO
LL-AGA, whi
c
h sha
r
e
s
th
e
advant
a
g
e
s
of both SA and GA, redu
ce
s the ave
r
age
and maximu
m job waiting
times greatly
and imp
r
ove the syste
m
pe
rforma
nce.
4 A
G
V
s
5 A
G
V
s
6 A
G
V
s
0
20
40
60
80
10
0
12
0
14
0
16
0
18
0
20
0
A
v
er
a
g
e
j
o
b w
a
i
t
i
n
g t
i
m
e
s
M
O
D
-
FC
FS
NV
F
RO
L
L
-
A
G
A
4 A
G
V
s
5 A
G
V
s
6 A
G
V
s
0
200
400
600
800
10
00
12
00
14
00
Ma
x
i
mu
m j
o
b
w
a
i
t
in
g
t
i
me
s
M
O
D
-
FC
FS
NV
F
RO
L
L
-
A
G
A
Figure 4. The
performan
ce
of the three sche
duling
strategie
s
4. Conclusio
n
AGVsS is b
e
comi
ng p
o
p
u
lar in
auto
m
atic mate
ri
als h
andlin
g
system
s, flexible
manufa
c
turi
n
g
sy
stem
s a
n
d
even
contai
ner-ha
ndling
appli
c
ation
s
.
AGVs sched
uling
i
s
a cru
c
ial
probl
em to improve the e
fficiency of the AGVsS.
In this pape
r, we have di
scussed the A
G
Vs
sched
uling issue
s
in AGVsS. Usi
ng si
m
u
lation, we p
r
ove that the
MOD-F
C
FS
rule and the NVF
rule, both
a
m
ong the
be
st dispatchin
g rule
s in lit
eratu
r
e, pe
rform
s
not so
well in
simul
a
tion
environ
ment. They dispatching AGVs
o
n
ly base
d
on
distan
ce b
e
twee
n the loa
d
s an
d the A
G
Vs
or the j
o
b
s
’ waiting time, th
e inform
ation
about lo
ad a
rrival
s
in a
d
vance is
not u
s
ed to
optimi
z
e
the pe
rforma
nce
of the
A
G
VsS. Lite
ra
ture
on
external transpo
rt ha
s
sho
w
n
that
sche
duli
n
g
vehicle
s
may
lead to bette
r pe
rform
a
n
c
e than di
spat
chin
g them. In this pa
pe
r, we Apply thi
s
to
the AGVsS, f
o
rmul
ate the
AGVsS
sche
duling
proble
m
a
s
a
m-T
SPTW.
AGV
s
S is
an
s
y
s
t
em
with
a hig
h
d
egre
e
of
un
certainty, We
prop
ose AGA
to deal
with t
he AGVs
sch
edulin
g probl
em,
and a
pply thi
s
he
uri
s
tic
dynamically by
usin
g it
re
pea
tedly unde
r a
combi
ned
roll
ing optimi
z
ati
on
strategy. We
comp
are the
performan
ce
of the pr
opo
sed a
pproa
ch with the di
spat
chin
g rul
e
s.
Re
sults
sho
w
that the approa
ch
pe
rforms sig
n
ifica
n
t
ly better t
han the dispat
ching rul
e
s. From
pra
c
tical
p
o
in
t of view, it
make
s the
p
r
opo
se
d
sch
edulin
g a
pproach m
o
re
a
ttractive fo
r t
h
e
AGVs
S.
Ackn
o
w
l
e
dg
ement
This wo
rk wa
s
supp
orted by
the
intern
ational
coo
p
e
r
ation p
r
oj
ect
s
of China
Mi
nistry
of scie
n
ce an
d techn
o
logy
(201
0DFB70
700).
Referen
ces
[1]
Mantel
RJ a
n
d
La
nde
w
e
er
d
HRA. Des
i
g
n
and
op
erati
o
n
a
l co
ntrol
of a
n
AGV s
y
stem
.
Internatio
na
l
Journ
a
l of Prod
uction Eco
n
o
m
ics
. 1995; 4
1
(3
): 257-26
6.
[2]
Le-An
h T
.
Intelli
ge
nt Contro
l of Vehic
l
e-B
a
s
ed Inter
nal
T
r
ansport S
y
stems. PhD the
s
is, Erasmu
s
Univers
i
t
y
R
o
tterdam. 20
05.
[3]
T
a
lbot L. Desi
gn an
d perfor
m
ance a
n
a
l
ysi
s
of
multi station aut
omated
gui
ded
v
ehic
l
e
s
y
stems. Ph.D
T
hesis, Univer
site Catho
liq
ue
de Louv
ai
n. 2003.
[4]
Le-An
h T and
De Koster, R(M) BM. A
revie
w
of desi
gn
and co
ntrol of
automated g
u
i
de
d vehic
l
e
s
y
stems.
Euro
pea
n Jour
nal o
f
Operationa
l R
e
searc
h
. 200
6; 171(1): 1–
23.
[5]
Egbe
lu PJ
a
n
d
T
anchoc
o J
M
A. Char
acter
i
zati
o
n
of
auto
m
ated
gui
de
d
veh
i
cle
dis
p
a
t
ching
rul
e
s.
Internatio
na
l Journ
a
l of Prod
uction R
e
se
arch
. 1984; 2
2
(3): 359-
374.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2508– 251
5
2515
[6]
BL Gol
d
e
n
a
n
d
AA Ass
ad.
Vehic
l
e
Ro
utin
g: Metho
d
s
an
d Stud
ies.
Nor
t
h-Holl
an
d,
Els
e
vier
Scienc
e
Publ
ishers. 19
88; 65-8
4
.
[7]
Xu H, C
hen Z
L
, Raja
gop
al
S and Arun
ap
uram S. Solvin
g a practica
l p
i
ckup a
nd d
e
li
ver
y
pr
ob
lem.
T
r
ansportati
on Scienc
e
. 200
3; 37(3): 347-
36
4.
[8]
W
enfeng
W
a
n
g
,
Z
untong W
ang, F
e
i Qi
ao
. An Impr
ove
d
Gen
e
tic A
l
g
o
rithm for
Ve
hicle
R
outi
n
g
Probl
em w
i
t
h
time w
i
nd
o
w
.
Internatio
na
l S
y
mp
osi
u
m o
n
Co
mp
ut
er S
c
ienc
e a
n
d
C
o
mputati
o
n
a
l
T
e
chno
logy
,
S
han
gh
ai
. 200
8;
189-1
94.
[9]
Cord
eau, JF
Gendre
au, M
Lap
orte, G Po
tvin, JY Seme
t F
.
A guide
to veh
i
cle
routi
ng h
eur
istics.
Journ
a
l of the
Operatio
nal R
e
search Soc
i
ety
. 2002; 53(
5): 512-5
22.
[10]
Victor Pill
ac, Michel Ge
ndr
e
au, Christe
l
l
e
G
uéret, André
s
L. Medag
lia.
A revie
w
of d
y
n
a
mic ve
hicl
e
routin
g pro
b
le
ms.
Europea
n Journ
a
l of Ope
r
ation
a
l Res
ear
ch
. 2012; 2
25(
1): 1-11.
[11]
Vis IFA, De K
o
ster, R(M) BM
and Savelsber
gh MWP. Minimum vehi
cle fleet siz
e
under
time-
w
indo
w
constrai
nts at a contain
e
r terminal.
T
r
ans
port
a
tion Sci
enc
e
. 200
5; 39(2): 24
9–2
60.
[12]
KC T
an, LH Lee, QL Z
hu, K Ou, Heuristic met
hods for v
ehicl
e routi
ng
prob
lem
w
i
th ti
me
w
i
n
d
o
w
s.
Artificial Intel
lig
ence i
n
Eng
i
ne
erin
g
. 200
1; 15
(3): 281-2
95.
[13]
Yi Z
han
g,
Xi
a
opi
ng
Li, Qia
n
W
ang. H
y
b
r
id
gen
etic a
l
g
o
ri
thm for perm
u
tation fl
o
w
sh
o
p
sche
d
u
lin
g
problems w
i
th
total
fl
o
w
time
minimiz
a
tion.
Europ
e
a
n
Jour
nal of Operati
o
nal R
e
searc
h
. 200
9; 196(
3):
869
–8
76.
[14]
Cha
ndra Sek
h
ar Pedam
all
u
, Linet Ozdam
ar. In
vestigatin
g a h
y
brid si
mulate
d ann
e
a
lin
g an
d loca
l
search a
l
g
o
rit
h
m for constrain
ed o
p
timiza
tion.
Euro
pea
n
Journa
l of Operati
ona
l Res
earch
. 2
008
;
185(
3): 123
0–1
245.
Evaluation Warning : The document was created with Spire.PDF for Python.