TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 15, No. 2, August 201
5, pp. 390 ~
396
DOI: 10.115
9
1
/telkomni
ka.
v
15i2.830
1
390
Re
cei
v
ed Ap
ril 23, 2015; Revi
sed
Jul
y
5 2015; Accept
ed Jul
y
20, 2
015
Heuristic Approaches t
o
Solve Travelin
g Salesman
Problem
Malik Muneeb Abid
1
*, Iqbal Muhamma
d
2
1
School of T
r
ansportati
on a
n
d
Log
istics, South
w
est
Jia
o
T
ong Un
iversit
y
, Che
ngd
u, Chi
n
a, 6100
31
2
School of Infor
m
ation Sci
enc
es and T
e
chno
log
y
,
South
w
e
s
t JiaoT
ong Uni
v
ersit
y
, Ch
en
g
du, Chi
n
a
.
*Corres
p
o
n
id
n
g
author, e-ma
i
l
: malik.mun
ee
b.abi
d@h
o
tmai
l.com
A
b
st
r
a
ct
T
h
is pa
per
pro
v
ides th
e surv
ey of the
he
uri
s
ti
cs soluti
on
a
ppro
a
ches
for
the travel
in
g s
a
les
m
an
prob
le
m (T
SP). T
SP is easy to understa
nd, h
o
w
e
ver, it is
very difficult to solve. Due to co
mp
lexity i
n
volv
e
d
w
i
th exact sol
u
tion a
ppr
oach
e
s
it is h
a
rd to s
o
lve T
SP w
i
thi
n
feasi
b
l
e
ti
me
. T
hat
’
s
w
h
y
di
fferent he
uristi
cs
are g
e
n
e
ral
l
y a
ppli
e
d
to so
lve
T
SP. Heuristic
s
to so
lv
e
T
SP are
pr
ese
n
ted here
w
i
th deta
i
l
ed alg
o
rith
ms. A
t
the end, co
mp
ariso
n
a
m
on
g selecte
d
ap
pro
a
ches sh
ow
s the efficie
n
cy of appro
a
ches i
n
terms of soluti
o
n
qua
lity an
d time consu
m
ed to
solve T
SP.
Ke
y
w
ords
: T
SP, heuristics, survey, NP-har
d
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
The com
puta
t
ional compl
e
xity
theory make
s
it
po
ssible
to valid
ate the
co
ncepts
of
“ea
s
y” a
nd “h
ard
”
problem
s an
d the di
stinction
a
m
on
g them. Probl
ems
can
be f
o
rmally
classi
fied
based on th
ei
r co
mplexity and if a pr
obl
em stri
ctly be
long
s to the fa
mily of NP-h
ard o
r
compl
e
te
probl
em
s, we kno
w
in
a
d
vance that
there i
s
little
ch
an
ce
of finding
an
efficient
and
ex
act
algorith
m
to
solve it. The
algorith
m
fo
r su
ch
a p
r
o
b
lem ha
s a
n
executio
n time bu
rstin
g
fo
r
increa
sing problem si
ze
s,
and
in
majo
rity ca
se
s is
not feasi
b
le f
o
r mo
st p
r
a
c
tical pu
rp
ose
s
.
Comp
uters a
r
e playing eve
r
y effective ro
le in so
lving
different com
p
lex probl
em
s but the mat
t
er
of fact is th
at some
proble
m
s a
r
e fun
d
a
m
entally
ha
rd
er to
solve. A
l
though, fo
r some p
r
obl
em
s it
is po
ssi
ble to
develop inte
lligent algo
rithms that
sol
v
e the probl
e
m
s expe
ditio
u
sly, howeve
r
, it
seem
s
sub
s
tantially hard
even in som
e
cases im
poss
ible to come up with any solution [24].
Daven
d
ra [13
]
defined TSP as, “Given
a set of
citie
s
of different distan
ce
s a
w
ay from
each othe
r, a
nd the obj
ecti
ve of TSP is to find t
he sh
ortest p
a
th fo
r a sale
spe
r
son to visit every
city exactly o
n
ce
an
d
return ba
ck to
the
ori
g
in
city”.
TSP is
an i
m
portant
ap
plied p
r
obl
em
with
many fasci
na
ting variants;
like theoretical math
e
m
a
t
ics, co
mpute
r
sci
en
ce, NP hard probl
em,
combi
nato
r
ial
optimizatio
n
and o
peration re
se
ar
ch
[25]. TSP is cl
assified
as
symmet
r
ic,
asymmet
r
ic a
nd multiple T
SP based on
the distan
ce
and directio
n betwe
en two
cities in a gra
p
h
(Figu
r
e 1
)
. If distan
ce be
tween two ci
ties is
same
in each di
rection it is
symmetri
c
wi
th
undirecte
d
na
ture othe
rwi
s
e it is asymm
e
tric.
Figure 1. Cla
ssifi
cation of
TSP
TSP ca
n b
e
solved
by u
s
i
ng o
p
timal al
gorit
hm
s
(e.g
., IP formulati
on),
however, these
solutio
n
s are comp
utationa
lly
infeasibl
e
and
it
i
s
alm
o
st impo
ssible
to gen
erate
optimal
soluti
on
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Heu
r
isti
c App
r
oa
che
s
to Solve T
r
a
v
elin
g Salesm
an
Problem
(Mal
ik Mune
eb Ab
id)
391
within
re
aso
n
able tim
e
. Fo
r thi
s
rea
s
o
n
heuri
s
tics are
used
to
solv
e TSP. Thi
s
pape
r
provid
es
a
survey of different ap
proa
che
s
u
s
ed to
solve the TS
P heuri
s
ticall
y.
1.1. Illustration of TSP
The
co
re o
b
j
e
ctive of TSP
is to
minimi
ze the
total t
r
a
v
eling cost
of obje
c
t a
r
ou
n
d
tours.
In ord
e
r to
u
nderstan
d TS
P, let us expl
ore th
e
given
belo
w
exam
ple. Figu
re
2
sho
w
s the
ro
ad
distan
ce b
e
twee
n the three town
s i.e. ABC and a
d
d
i
tionally assu
me that a sal
e
sp
erson, wh
ose
busi
n
e
ss i
s
t
o
sell
s l
ubri
c
ant items to
different
com
panie
s
lo
cate
d in the
s
e th
ree
citie
s
, m
u
st
travel (startin
g from
home
)
. Here the d
e
c
imal
valu
es
near the lin
e
edge
s in th
e
diagram a
r
e t
h
e
driving dis
t
anc
e
s
between the c
i
ties
.
In this
exa
m
ple, we
are assumi
ng
that we h
a
ve a
symmetri
c
T
SP i.e.
the cost in going f
r
om A to B is
the same a
s
the cost in g
o
ing from B to
A
but in asymm
e
tric TSP the co
st coul
d no
t be the same
betwee
n
the two point
s.
In the given situation, one
que
stion ari
s
es
in mind th
at how many
TSP tours co
uld be?
The
gen
eral
answe
r to
thi
s
q
u
e
s
tion
is,
for th
e
com
p
lete g
r
a
ph
with
n
vertice
s
, the
num
be
r of
different TSP route
s
wo
uld
be Equation
(1).
!
(
1
)
Figure 2. A 4 city TSP prob
lem
Let us calcula
t
e the chea
pe
st t
our by wo
rking o
n
abov
e Figure 2
.
HABCH
8
2
+
1
40
+9
0+
5
2
=
3
64
HACB
H
8
2
+
1
02
+9
0+
6
9
=
3
43
HCAB
H
52
+10
2
+140
+6
9=3
6
3
Thus, g
o
ing from H to A, th
en to C, then
to B and then
back H could
be the best choice.
1.2. Related
Wor
k
In literature
TSP is used
in two forms:
i) combi
n
atorial optimi
z
ation versio
n and ii)
deci
s
io
n version. In first v
e
rsi
on it i
s
u
s
ed to
find a
minimum
Hamiltonian
cy
cle a
nd in l
a
ter
versio
n to ch
eck the existe
nce of smalle
r gra
ph.
Theo
retical compute
r
scie
nce
and
op
e
r
ation
s
resea
r
ch,
both fiel
ds of
co
mbi
natorial
optimizatio
n
contai
ns TSP
.
In this probl
em, set of
cities a
r
e give
n
with their
dist
ances to fin
d
the
sho
r
test rout
e to each
city without visiting a ci
ty twice. In 1930,
it was first formul
ated a
s
a
mathemati
c
al
model a
nd a
pplied to
so
many area
s
to find their
op
timal solutio
n
s
e.g., Clu
s
te
ring
of array of data [1], Hand
ling of a warehou
se m
a
te
rials [2] a
nd
cry
s
tal struct
ure a
nalysi
s
[3].
Re
sou
r
ce co
nstrai
ned
sch
edulin
g pro
b
lem with agg
regate dea
dli
ne also
solv
ed with TSP [4].
Re
sea
r
che
s
[5, 6] took o
r
i
enteeri
ng a
n
d
pri
z
e
colle
ction pro
b
lem
s
as
spe
c
ial
case
s of resou
r
ce
con
s
trai
ned
TSP. One of the best
kno
w
n an
d mo
re
compl
e
x co
mbinato
r
ial p
r
oble
m
is Ve
hicle
routing
p
r
obl
e
m
, to dete
r
mi
ne the
o
r
de
r
of vehicl
e fo
r
cu
stome
r
s se
rving from fle
e
t of vehi
cle
s
, is
being solved by
TSP
[7].
TSP is ea
sy to unde
rsta
n
d
but it’s real
ly diffi
cult to solve it. In 1
972, Ri
ch
ard
M. Karp
sho
w
e
d
that
it is
NP-com
plete to
solv
e Ham
iltonia
n
cy
cle p
r
o
b
l
e
m, whi
c
h
e
x
plains t
he
NP-
hard
n
e
ss
of TSP. This ca
n be ju
dge
d that ho
w mu
ch co
mputatio
nal difficulty i
s
attached to
find
the optim
al ro
ute [9]. With
the p
a
ssa
ge
o
f
time TSP i
s
getting m
o
re
and
more
so
phisti
c
ated
a
nd
solving in
stan
ce
s are la
rg
er now. A brief view of TSP mileston
es i
s
given belo
w
in Table 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 390 –
396
392
Table 1. Sum
m
ary of the Mileston
es a
c
hi
eved in TSP (12)
Year
Research Tea
m
Size of Ins
t
anc
e
1954
G. Dantzig, R.
F
u
lkerson,
and S.
Johnson [14]
49 cities
1971
M. Held and R.M
.
Karp [15]
64 cities
1975
P.M. Camerini, L
.
Fratta,
and F. M
a
ffioli [16]
67 cities
1977
M. Grötschel [17]
120 cities
1980
H. Cro
w
der a
nd
M.W. Padberg [1
8]
318 cities
1987
M. Padberg a
nd
G. Rinaldi [19]
532 cities
1987
M. Grötschel and
O. Holland [20]
666 cities
1987
M. Padberg a
nd
G. Rinaldi [19]
2,392 cities
1994
D. Applegate, R.
Bixb
y
,
V. C
h
vátal, and W. Cook [2
1]
7,397 cities
1998
D. Applegate, R.
Bixb
y
,
V. C
h
vátal, and W. Cook [2
2]
13,509 cities
2001
D. Applegate, R.
Bixb
y
,
V. C
h
vátal, and W. Cook [2
3]
15,112 cities
2006
D. Applegate, R.
Bixb
y
,
V. C
h
vátal, W. Cook,
and K. Helsgaun
[23]
24,978 cities
This pap
er i
s
o
r
ga
nized
as foll
ows:
Next
sectio
n
provid
es th
e alg
o
rithm
details of
sele
cted heu
ristic sol
u
tion
s
for
TSP. After
that all of these te
chni
que
s a
r
e
analyzed a
nd
comp
ared for solution q
uali
t
y and time consumed fo
r
different probl
em data.
At last co
ncl
u
si
o
n
s
and future di
rection
s
are prese
n
ted.
2. Heuris
tic Solution Tec
hniques
There a
r
e va
riou
s h
euri
s
ti
cs and
ap
prox
imate solut
i
on ap
proa
ch
es,
which h
a
d
bee
n
devise
d
duri
n
g last de
ca
d
e
s, to find so
lution wi
thin
reasona
ble time and
with 2-3 % optima
lity
gap [8]. T
h
e
r
e a
r
e
num
e
r
ou
s a
p
p
r
oa
che
s
used t
o
solve TSP
sh
own in
li
terature. So
me
importa
nt an
d mo
stly used a
pproa
ch
es
are
enli
s
t
ed h
e
re
with
their
appli
c
ation al
gorith
m
s.
Variabl
es for
these al
gorithms are illustrated in Table
2.
Table 2. Illust
ration of Vari
able
s
Descrip
tion
Variable
Initial solution pr
ovided b
y
the us
er
S
Objective function value
z
The best solution
S*
Cities in TSP (Nodes in
transporta
tion net
w
o
rk)
i, j
Population P
Gene
rations G
Gene
ration count
er
N
g
en
Initial temperatur
e
T
Cooling factor
r
Number of
times the tempera
t
ure
T
is decreased
ITEM
P
Maximum numbe
r of ne
w
solution
s to be accepted at each temper
at
ure
NLI
M
IT
Maximum numbe
r of solutions evaluated at each te
mperatu
r
e
NO
VER
Gain par
ameter
L
2.1.
Brute Forc
e
Metho
d
Algorithm for
TSP using Brute-force me
t
hod contain
s
the followin
g
step
s:
Algorithm 1: TSP using Brute Force Me
thod
Step 1: calcul
ate the total numbe
r of tours (w
he
re citie
s
rep
r
e
s
e
n
t the numbe
r of node
s).
Step 2: draw
and list all the
possible tou
r
s.
Step 3: calcul
ate the distan
ce of ea
ch to
ur.
Step 4: choo
se the sho
r
test
tour; this is the optimal so
lution.
2.2. Greedy
Appr
oach
Gree
dy app
ro
ach
solve
s
TSP by using t
he five steps,
given in Algorithm 2.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Heu
r
isti
c App
r
oa
che
s
to Solve T
r
a
v
elin
g Salesm
an
Problem
(Mal
ik Mune
eb Ab
id)
393
Algorithm 2: TSP using G
r
eedy Approa
ch
Step1: Look
at all the arcs with minimu
m distan
ce.
Step 2: Choo
se the
n
chea
pest arcs
Step 3: List the distan
ce of
arcs
starting f
r
om the mini
mum dista
n
ce to maximum distan
ce.
Step 4: Dra
w
and che
ck if it forms a Ham
iltonian cy
cle.
Step 5: If ste
p
4 forms a
Hamiltonian
cycle than
we h
a
ve an optim
al solutio
n
; write down the
tour of the op
timal solution
and calculate
their distan
ce.
2.3.
Near
es
t Neig
hbor He
uristic
Algorithm 3 shows the met
hodol
ogy to solv
e TSP by
Nea
r
e
s
t Neig
hbor
Heu
r
i
s
tic.
Algorithm 3: TSP using Neare
s
t Nei
g
h
bor Heu
r
isti
c
Step 1: Pick any starting n
ode.
Step 2: Look
at all the arcs coming o
u
t of
the starting
node that hav
e not been vi
sited an
d
cho
o
se the n
e
xt close
s
t no
de.
Step 3: Repe
at the pro
c
ess until all t
he node
s have b
een visited at
least on
ce.
Step 4: Check and
see if a
ll node
s are v
i
sited. If so
re
turn to the sta
r
ting point wh
ich give
s us a
tour.
Step 5: Dra
w
and write do
wn the tour, a
nd cal
c
ul
ate the dista
n
ce o
f
the tour.
2.4.
Bran
ch and
Bound M
e
th
od
Bran
ch
and
b
ound
techniq
ue i
s
comm
o
n
ly
used
opti
m
ization
tech
nique. F
o
rm
u
l
ation of
TSP by using
bran
ch an
d b
ound te
chni
q
ue is given in
Algorithm 4.
Algorithm 4: TSP using Branch and Bo
und Metho
d
Step 1: Choo
se a sta
r
t nod
e.
Step 2: Set b
ound to a very large value,
let’s say infin
i
ty.
Step 3: Choo
se the chea
p
e
st
arc bet
we
en the cu
rren
t and unvis
ite
d
node a
nd a
dd the dista
n
c
e
to the current
distan
ce an
d
repeat while
the curre
n
t distan
ce is le
ss than the bou
nd.
Step 4: If current distan
ce i
s
less than b
ound, then
we are do
ne
Step 5: Add up the distan
ce and bo
und
w
ill be eq
ual to the cu
rre
nt distan
ce.
Step 6: Repe
at step 5 until
all the arcs h
a
ve been
cov
e
red.
2.5.
2-Op
t Algori
t
hm
For n no
de
s in the TSP proble
m
, the 2-Opt algo
rith
m con
s
ist
s
of the steps
sh
own in
Algorithm 5.
Algorithm 5: TSP using 2
-
Opt Algorithm
Step 1:
Let S be the initial solution
provide
d
by the user
and z its
objec
tive func
tion value. Set S*=
s
,
z*=z, i=1 and
j=i+1=2.
Step 2:
Con
s
id
er the
excha
nge results in a sol
u
tion S that
has objective fun
c
tion value
z’
<z*,
set z*
=z’
and S*=S’. If j<n re
peat ste
p
2; otherwi
se set i=i
+
1 an
d j=i+1. If i<n, repeat
step 2
;
otherwi
se g
o
to step 3.
Step 3:
If S
≠
S*, s
e
t S
=
S*, z
=
z
*
, i=1, j=
i+
1=
2 and go to s
t
ep
2. Otherwise, output S* as the best solutio
n
and termi
nate
the pro
c
ess.
2.6.
Greedy
2-Op
t Algorithm
The g
r
ee
dy 2-Opt
algo
rithm is
a vari
ant of
2-opt
algorith
m
. It con
s
i
s
ts of t
he ste
p
s
s
h
ow
n
in
Algo
r
i
th
m 6
.
Algorithm 6: TSP using
G
r
eedy 2-O
p
t Algorithm
Step 1:
Let S be the initial solution
provide
d
by the user
and z its
objec
tive func
tion value. Set S*=
s
,
z*=z, i=1 and
j=i+1=2.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 390 –
396
394
Step 2:
Tran
sp
ose Node i and
No
de j, i<j. Com
pare the
re
su
l
t
ing obje
c
tive function valu
e z with z*. If
z
≥
z* a
nd j<n, set j=j+1 a
nd repeat st
e
p
2; Otherwise go
to step 3.
Step 3:
If z
<
z
*
, set S*=
S
, z
*
=z
, i=
1,
j=
i+
1=
2 and
go to s
t
ep 2. If z
≥
z* and
j=n, set i=i
+
1,
j=j+1 a
nd re
p
eat step 2. Otherwi
se, outp
u
t S* as the best sol
u
tion a
nd termin
ate
the pro
c
e
ss.
2.7.
Gene
tic Alg
o
rithm
This al
go
rith
m is ba
se
d
on ge
netics. The Ge
net
ic Algo
rithm
works a
s
shown in
Algorithm 7.
Algorithm 7: TSP using
G
enetic Algo
rit
h
m
Step 0:
Obtain the m
a
ximum num
ber of individ
uals in the p
o
pulation P an
d the maximu
m numbe
r of
gene
ration
s
G from the user, gene
rate
P solution
s for the first gen
eration’
s po
p
u
lation ra
ndo
mly,
and re
presen
t each solutio
n
as a st
ring.
Set generatio
n cou
n
ter N
g
en
=1.
Step 1:
Determine th
e fitness of e
a
ch
solutio
n
in
the cu
rre
nt gene
ration’
s
popul
ation an
d record the
string that ha
s t
he be
st fitness.
Step 2:
Gene
rate sol
u
tions for the
next generation’s p
opulati
on as follo
ws:
1. Retain 0.1
P
of the solutions
with t
he best fitness in
the previou
s
popul
ation.
2. Gene
rate 0
.
89P solution
s via mating, and
3. Select 0.01
P solution
s from the previo
us po
pulatio
n
rando
mly an
d mutate the
m
.
Step 3:
Update N
g
en
= N
g
en
+
1
. If N
g
en
≤
G, go to Step 1. Otherwise sto
p
.
2.8.
Simulated Annealing (SA)
Statistical m
e
ch
ani
cs i
s
t
he ba
si
c id
e
a
of si
mulate
d ann
ealin
g
(SA). Analo
g
y
of the
behavio
r of physi
cal sy
stems in
the
heat bath i
s
main motivation of SA. Solution sta
t
e is
rep
r
e
s
ente
d
by the tem
p
e
r
ature. Initiating
with
a
n
in
itial tempe
r
at
ure,
algo
rith
m move
s to
the
next temperat
ure until it rea
c
he
s a fro
z
en
state.
Algorithm 8: TSP using Si
mulated Ann
ealing
Step 0: Set
S
= initial feasi
b
le sol
u
tion
Step 1: Repe
at step 2
NOVER
times or until the number of succe
s
sful ne
w sol
u
tions i
s
equ
al to
NLIMIT
Step 2: Pick a pair of ma
chine
s
ran
dom
ly and excha
nge their p
o
si
tions.
Step 3: Set
T =
rT
an
d
ITEMP =
ITEMP +
1
. If
ITEMP
<= 1
00, go to
step 1; otherwise stop.
2.9. Neur
al
Net
w
ork
In
th
is
pr
oc
ess
a
t
ea
ch
pr
oc
es
s
i
n
g
s
t
e
p
a
c
i
ty
is surv
eyed. On
co
mplete ite
r
ation visit
s
all cities in a
pre
s
et order.
First of all a node
i
s
sele
cted from ran
d
o
mly from the network an
d a
gain pa
ramet
e
r is u
s
ed to
attain the result and qua
lit
y if solution.
Algorithm 9 shows the wo
rking
mech
ani
sm o
f
neural net
work.
Algorithm 9: TSP using ne
ural net
wo
rk
Step 1. Find the nod
e j
c
which is
closed t
o
city i.
Step 2. Move node j
c
and its neig
hbo
rs o
n
the ring toward
s city i.
The dista
n
ce each nod
e wil
l
move is determin
ed by
a function f (L,
n), whe
r
e n i
s
the distan
ce
measured alo
ng the loop b
e
twee
n nod
e
s
j and j
c
.
n =inf (j- j
c
(m
od N), j
c
-
j
(mod N
))
Every node j is moved fro
m
current po
siti
on to a new o
ne.
The fun
c
tion f is defined to
be
f(L,n)=sqrt(1/2)*exp
(-n
2
/G
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Heu
r
isti
c App
r
oa
che
s
to Solve T
r
a
v
elin
g Salesm
an
Problem
(Mal
ik Mune
eb Ab
id)
395
3. Discussio
n
s
Previou
s
se
ction demo
n
st
rates th
e different metho
dologi
es to
solve NP
-ha
r
d TSP
probl
em app
roximately. Although, different te
chniq
u
e
s had b
een
devised p
r
e
v
iously, but,
all
available te
chniqu
es a
r
e
not efficient in term
s
of solution time
and qu
ality.
Comp
ari
s
o
n
by
Mare
dia
[10] sho
w
s
that Neare
s
t Ne
ig
h
bor
heu
risti
c
works
well
bu
t it is not
su
re that it will
g
i
ve
us
solution
s
good a
s
b
r
ut
e force. Moreover, g
r
eed
y appro
a
ch i
s
not a g
o
o
d
app
roximat
i
on
techni
que
for TSP. Co
mp
arison
of Bra
n
ch
an
d
bou
nd te
chni
que
with
brute f
o
rce m
e
thod
is
presented in
Figure 3 (data extrac
ted from Maredia [
10]). SA algorit
hm obtains has ability to find
the goo
d qua
lity final solutions
by mech
anism
of
gra
dually goin
g
from o
ne
solu
tion to the ne
xt.
The mai
n
differen
c
e
of SA from the
2-o
p
t is that
the l
o
cal
optimiza
t
ion algo
rithm
is often
re
stri
ct
their
sea
r
ch f
o
r the
optim
al sol
u
tion in
a
downhill
di
re
ction which m
ean th
at the i
n
itial solution
is
cha
nge
d only
if it re
sults in
a d
e
crea
se
i
n
the
obje
c
tive fun
c
tion val
u
e
.
However
,
it is
s
h
own by
Kim et al. [11] that 2-opt al
gor
ithm
wo
rks well
wh
en the problem
si
ze is l
e
ss tha
n
50 citie
s
. F
o
r
comp
ari
s
o
n
o
f
the app
roa
c
hes
we
are referri
ng to d
a
t
a Kim et al. [11]. Data
extracted f
r
om Ki
m
et al. [11] is plotted in Figu
re 4 an
d 5, a
c
cordi
ng to result
s it is ob
vious that the
solution of T
S
P
usin
g the
neu
ral n
e
two
r
k o
u
tperfo
rms th
an all
other a
ppro
a
che
s
fo
r all
pro
b
lem
sizes.
Ho
wev
e
r,
if we
compa
r
e the
time
co
nsum
ed
to
so
lve the
pro
b
le
ms, it i
s
ea
sy
to re
alize that
neu
ral
net
wo
rk
approa
ch i
s
takin
g
mu
ch
more time
a
s
comp
ared
to
others. Ge
net
ic alg
o
rithm
h
a
s b
een
used
for
many optimi
z
ation p
r
oble
m
s; ho
weve
r, the u
s
e
of
ge
netic al
go
rith
m for TSP h
a
s
di
sadva
n
ta
ges
of prematu
r
e
conve
r
ge
nce and p
oor
local
sea
r
ch
capa
bility. These di
sad
v
antage
s ca
n be
overcome by
adaptation
of other effici
ent wo
rk
in
g
algorith
m
s e.
g., artificial immune
syst
ems
[13].
Figure 3. Co
mpari
s
o
n
of Bran
ch an
d Bound Te
ch
niq
ue with Brute
Force Metho
d
Figure 4. Len
gth Comp
ari
s
on of TSP
Heu
r
isti
cs
Figure 5.
Time Comparis
on of TSP Heuris
tics
Advantage
s
of these h
e
u
r
istics
app
ro
ach
e
s
are th
at they are
simple
an
d
easy to
unde
rsta
nd. These req
u
ire
le
ss
p
r
og
ra
mming and
storage
requi
rements a
nd
prod
uce m
u
ltiple
s
o
lutions wit
h
in less
time [
26]. Howe
ver, he
uri
s
tics lo
cal
imp
r
o
v
ements in
h
euri
s
tics
can
be
sou
r
ce of lack in glob
al prosp
e
ctive of obje
c
tive function.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 390 –
396
396
4. Conclusio
n
and Re
co
mmendation
s
This p
ape
r p
r
ovide
s
the
survey of dif
f
erent
he
uri
s
tics u
s
e
d
to solve TSP. First, an
example of T
SP is pre
s
ent
ed to give the idea of
TS
P. This su
rve
y
is limited to some
sele
ct
ed
approa
che
s
becau
se it is not possibl
e
to cove
r
all approa
che
s
here. So, onl
y most relev
ant
approa
che
s
are di
scusse
d here. Altho
ugh a num
be
r of heuri
s
tics had b
een
devise
d
however;
some
are effi
cient with
re
spect to time and some a
r
e outpe
rformi
ng in sol
u
tion
quality. Future
work will consider
th
e formulation
of heuri
s
tics to
solve t
he TSP
in a
reasonable time
with bench
mark pro
b
lem
data of mile stone
s prese
n
ted in Table
1.
Referen
ces
[1]
La
w
l
er EL,
Le
nstra JK, Rin
n
o
o
y
AHG, Sh
mo
y
s
DB. T
he T
r
aveling S
a
l
e
sman Pr
obl
e
m
. Chich
e
ster:
John W
i
l
e
y
.
1
9
85.
[2]
Ratliff HD,
Ro
sentha
l AS.
O
r
der-Picki
ng
in
a
Recta
ngu
la
r W
a
reh
ouse:
A So
lva
b
le
Case
for th
e
T
r
avelin
g Sal
e
sma
n
Prob
le
m
.
PDRC Re
port Series N
o
. 81-
10. 198
1.
[3]
Bland RE, Shallcross DF.
L
a
rge T
r
av
el
ing
Sales
m
an
Probl
e
m
Arisi
ng
from Ex
peri
m
e
n
ts in X-r
a
y
Crystallo
gra
p
h
y
: a Prelimin
ar
y Report on C
o
mp
utatio
n.
T
e
chnic
a
l Re
port No. 730. 1
987.
[4]
Miller, Pek
n
y
J. Exact Soluti
on of Larg
e
A
s
y
mmetric T
r
avelin
g Sal
e
sm
an Prob
lems.
Scienc
e 25
1.
199
1: 754-
761.
[5]
Balas E. T
he Prize Col
l
ecti
ng
T
r
aveling Sal
e
sman Prob
lem.
Netw
orks 19.
198
9: 621-
636.
[6]
Golde
n
BL, Le
v
y
L, Vohr
a R. T
he Orienteeri
ng Prob
lem.
N
a
val R
e
searc
h
Log
istics 34.
1
987: 30
7-3
18.
[7]
Christofi
des N.
Vehicl
e Ro
utin
g, in T
he T
r
ave
lin
g Sa
lesma
n
Probl
em. In: La
w
l
er, Le
nstra, Rino
o
y
Ka
n
and Shm
o
ys.
Editors
. John W
i
le
y
.
198
5: 431-
448.
[8]
Reg
o
C, Gamboa D, Glover F
,
Osterman C.
T
r
av
eling sal
e
s
m
an pro
b
lem h
euristics: le
adi
ng metho
d
s
,
implem
entati
o
n
s
and lat
e
st ad
vances.
Euro
p
ean Jo
urn
a
l of
Operatio
nal
Re
search
. 20
11;
211(
3): 427-
441.
[9]
Cook W
.
History of the T
SP.
The T
r
aveli
ng S
a
lesm
an Prob
l
e
m. Georgia T
e
ch. 200
9.
[10] Maredi
a
A. His
tor
y
,
A
nal
ys
is, and
Implem
ent
atio
n
of T
r
aveli
ng S
a
lesm
an
Probl
em (T
SP) an
d R
e
l
a
ted
Probl
ems. T
h
esis. Dep
a
rtment of
Comp
u
t
er and Math
ematica
l
Scie
nces Un
iversit
y
of Ho
uston
-
Do
w
n
t
o
w
n
. 20
10.
[11]
Kim IB, Shim IJ, Z
hang M. C
o
mparis
on
of T
SP Algor
ithms, Project for Mode
ls in F
a
ci
lities Pl
an
nin
g
and Mater
i
als
Han
d
li
ng. 19
98
.
[12] http://
w
w
w
.
mat
h
.u
w
a
ter
l
oo.c
a
/tsp/histor
y
/
mi
le
stone.html (ass
essed o
n
08-
0
6
-20
15).
[13]
Dave
ndra D.
T
r
avelin
g Salesm
an Prob
le
m, T
heor
y
and Ap
plic
atio
ns. IN
T
E
CH ope
n acce
s
s
pub
lish
e
rs. 201
0.
[14]
Dantzi
g G, F
u
l
k
erson
R, Joh
n
s
on S. So
luti
on
of a l
a
rg
e-scal
e
travel
in
g-sal
e
sman pr
ob
lem.
Operati
ons
Research 2.
19
54: 393-
41
0.
[15]
Held M, Karp RM.
T
he traveling-salesm
an problem and minimum spanning trees: part II.
Ma
th
em
a
t
i
c
al
Progra
m
mi
ng 1.
1971: 6-
25.
[16]
Cameri
ni PM, F
r
atta L, Maffi
oli F
.
On impro
v
i
ng re
la
xati
on
methods b
y
m
odifi
ed gra
d
i
e
n
t
techniqu
es.
Mathe
m
atic
al Progra
m
mi
ng Study
3.
197
5:
26-3
4
.
[17]
Grötschel M, P
adb
erg M. O
n
the s
y
mmetric
traveli
ng s
a
les
m
an
prob
lem: t
heor
y
an
d c
o
m
putatio
n. In:
R. He
nn, B. K
o
rte, W Oettli.
Editors
.
Lectu
re N
o
tes i
n
Ec
onom
ics a
n
d
Mathematic
al
S
y
stems
15
7
.
Sprin
ger, Berli
n
. 1977: 1
05-1
15.
[18]
Cro
w
d
e
r H, Pa
dber
g MW
. Sol
v
ing l
a
rg
e-scal
e
s
y
mm
etric travell
i
n
g
sal
e
s
m
an pr
obl
ems
to optima
lit
y.
Mana
ge
me
nt Scienc
e 26.
19
8
0
: 495-5
09.
[19]
Padb
erg M, Ri
nal
di G. Optimizatio
n
of a 5
3
2
-c
it
y
s
y
mm
etric traveli
ng s
a
l
e
sman
prob
le
m b
y
branc
h
and cut.
Opera
t
ions Res
earch
Letters 6.
198
7: 1-7.
[20]
Grötschel M, Holl
an
d O. A cutting pl
an
e
al
gorithm for mi
nimum p
e
rfect 2-matchin
g
s.
Co
mp
uting 39
.
198
7: 327-
344.
[21]
Appl
egat
e D, Bixb
y R, Chv
á
ta
l V, Cook W
.
F
i
ndin
g
cuts i
n
the T
SP (A
prelim
in
ar
y
r
e
p
o
rt).
DIMAC
S
T
e
chnic
a
l Re
p
o
rt.
1995: 95-
0
5
.
[22]
Appl
egat
e D, Bixb
y R, Co
o
k
W
,
Chvátal
V. On the soluti
on of trav
elin
g sa
lesma
n
pro
b
lems
.
Rhei
nisc
he F
r
i
edric
h-W
ilh
elm
s
-Univers
ität Bonn. 19
98.
[23]
Appl
egat
e D, Bixb
y R, Chvát
a
l V, Cook W
.
T
he
T
r
aveling
Salesm
an Pro
b
lem: A Comp
utation
a
l Stud
y.
Princeto
n, Ne
w
Jerse
y
, USA:
Princeto
n Univ
ersit
y
Press. 2
006.
[24]
F
a
tma A, Karkor
y
A, A Abud
almol
a
. Imple
m
ent
atio
n of Heuristics for
Solvi
ng T
r
avell
i
ng Sa
lesma
n
Probl
em Usin
g
Nearest Ne
ig
hbo
ur an
d Min
i
mum Spa
nni
n
g
T
r
ee Algorith
m
s.
Internation
a
l Jour
nal of
Mathe
m
atic
al, Co
mp
utation
a
l,
Natural a
nd P
h
ysical E
ngi
ne
erin
g.
201
3; 7(10).
[25] http://en.
w
i
k
i
p
e
d
ia.or
g
/
w
iki/T
r
avelli
ng
_sal
esm
an_
prob
lem (a
ssessed o
n
08-
06-2
015).
[26]
T
u
rban E. Decision su
pp
ort and e
x
p
e
rt s
y
ste
m
. Macmillan s
e
ries i
n
inform
ation s
y
stem. 1
990.
Evaluation Warning : The document was created with Spire.PDF for Python.