Journal of mathematics and computer Science 12 (2014) 27 - 38

An Efficient Algorithm for Solving a Stochastic Location-Routing Problem

H.A. Hassan-Poura, M. Mosadegh-Khaha, M. Zareei1

a
Department of Industrial Engineering, Faculty of Engineering, Imam Hossein University, Tehran, Iran
[email protected]
[email protected]
[email protected]

Article history:
Received February 2014
Accepted June 2014
Available online July 2014

Abstract
The purpose of this paper is to determine the location of facilities and routes of vehicles, in which
the facilities and routes are available within probability interval (0, 1). Hence, this study is coherent the
stochastic aspects of the location problem and the vehicle routing problem (VRP). The location
problem is solved by optimization software. Because of the computational complexity of the stochastic
vehicle routing problem (SVRP), it is solved by a meta-heuristic algorithm based on simulated
annealing (SA). This hybrid algorithm uses genetic operators in order to improve the quality of the
obtained solutions. Our proposed hybrid SA is more efficient than the original SA algorithm. The
associated results are compared with the results obtained by SA and optimization software.

Keywords: Location; Vehicle routing problem; Stochastic; simulated annealing; Genetic operators.

1. Introduction
The location routing problem (LRP) can be divided to the facility location problem (FLP) and the
vehicle routing problem (VRP). In the first problem, all customers are directly linked to a depot. In the
second problem, it can be supposed that the locations of depots are predetermined. Min et al. [1] and
Nagy and Salhi [2], classified the LRP based on a number of aspects, such as facility layers,
hierarchical levels, number of facilities, size of vehicle fleets, vehicle capacity, facility capacity, nature
of demand/supply, number of objective functions, types of model data, planning horizon, time
windows, solution space and solving procedure. In most previous studies in LRP, facilities are
completely closed or opened to serve customers. In addition, routes are completely available to transit
the vehicles. However, Lee and Chang [3] assumed that the facilities would not provide services for

1
Corresponding author

27
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

some reasons, such as breakdown or shutdown of unknown causes. Hwang [4, 5] considered that
facilities are available in a known probability. Hassan-Pour et al. [6] developed two new aspects of the
LRP considering stochastic availability of facilities and routes. Wu et al. [7] divided a multi-depot
location-routing problem into a LRP and a VRP. Doerner et al. [8] solved a multi-criteria tour
planning. Hwang [4] presented a supply-chain logistics system with a service level. He divided the
main problem into a stochastic set covering problem (SSCP) and a VRP. Hwang [5] presented a model
for a stochastic set covering problem. Zhou et al. [9] designed a bi-criteria allocation problem and
proposed a genetic algorithm to solve the given problem. Chu et al. [10] extended insertion and two-
phase heuristics for a periodic capacitated arc routing problem (ARP). Alumur and Kara [11]
considered a location-routing problem minimizing the transportation risk and cost. Erkut and Alp [12]
presented a hazardous material shipment problem to minimize the total transportation risk. Caballero
et al. [13] developed a tabu search (TS) method for a multi-objective LRP. Erdogan and Esin [14]
explained a routing table updating by the use of the self cloning ant colony that is based on one of the
meta-heuristics. Liu and Chung [15] proposed a heuristic method, called variable neighborhood tabu
search, adopting six neighborhood searching approaches, such as insertion and 2-opt. This method
finds the optimal solution for the VRP with backhauls and inventory.
In this paper, we assume that facilities may be partially destroyed because of crisis conditions or
unknown reasons. Also, in catastrophic conditions, the routes are not completely available. Hence,
facilities and routes are in the available state with the known probability between (0, 1). Then, we
solve the location problem by an SSCP approach and the multi-objective multi-depot stochastic
vehicle routing problem (MO-MDSVRP) by a hybrid meta-heuristic method based on SA.

2. The model of the SSCP

According to the literature survey, we solve the location problem using an SSCP approach. The main
assumptions are as follows:
 Facilities are uncapacitated.
 The establishment cost of facilities is determined.
 Every customer can be covered by multi depots simultaneously.
 The facilities are available within the probability interval (0, 1).
Hassan-Pour et al. [6, 16] presented the mathematical model of the SSCP that minimizes the
establishment cost of facilities, subject to the coverage minimum of customers by some of the
facilities. This model is solved by the optimization software. The illustrative results are provided and
discussed in this paper.

3. Computational results of SSCP

To solve the SSCP, we use data generated at random, as shown in Table 1.

Table 1 Data generated to solve the SSCP

Entity Parameter Value/ Data
Customer No. of customers 50,75 and 100
No. of depots 10 and 20
Covering coefficient of customer i
U(0, 1)
by facility j
Depot Availability of facility j U(0, 1)
Covering probability of customer
i U(0.9, 1)
by some of the available facilities

28
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

By considering these data generated, six problems are solved by the optimization software. This
software is used to find the optimal solution in the case where the average CPU time for the SSCP is
always less than one second. Computational results of facilities location are shown in Table 2.

Table 2 Computational results of SSCP

Problem No. of
No. of No. of
name OFV depots
customer depots
used
P1 50 10 93061 8
P2 50 20 146968 14
P3 75 10 104509 9
P4 75 20 161625 14
P5 100 10 115415 10
P6 100 20 175189 16

4. Mathematical formulation of the MO-MDSVRP

In the MO-MDSVRP module, we consider a VRP with two objective functions and multiple depots.
Also, the decision variables should be found under a stochastic environment, namely, facilities and
routes are available within the probability interval (0, 1). We develop the MO-MDSVRP model under
assumptions as follows:
 The vehicle fleet is homogeneous.
 The depots are multiple.
 The capacity of each vehicle is limited.
 The total demand served by each vehicle cannot exceed its capacity.
 The working time of each vehicle is limited.
 The service time in each node is given.
 Each node is visited only once by a single vehicle.
 The availability of each route is probabilistic.
 We formulate the MO-MDSVRP as follows:
Minimize Z =
1   
i 1,...,N j 1,....N v 1,...,NV
C ij X ijv (1)

Maximize Z =
2  ( 
v 1,...,NV i 1,...,N j 1,...,N
Pij X ijv ) (2)

s.t.
 
i 1,..., N v 1,..., NV
X ijv  1 , j  M  1,..., N (3)


j 1,...,N v 1,...,NV
 X ijv  1 , i  2,..., N (4)


i 1,...,N
X ipv  
j 1,...,N
X pjv  0 , v  1,..., NV ; p  M  1,..., N (5)


i 1,...,N
di ( 
j 1,..., N
X ijv )  Cv , v  1,..., NV (6)


i 1,...,N
t iv 
j 1,...,N
X ijv   
i 1,..., N j 1,..., N
t ijv X ijv  Tv , v  1,..., NV (7)

 
i 1,....M j M 1,...,N
X ijv  1 , v  1,..., NV (8)

29
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

  X
v 1,...,NV i S j S
ijv  S  r (S ) , S  A  {1} , S   (9)

X ijv 0,1 , i , j ,v (10)

Notations:
M Number of facilities.
N Number of demand nodes or facilities (depot is at the node i = 1).
NV Number of available vehicles.
i, j Index of demand nodes or facilities (1≤ i, j ≤ N).
V Node set in graph G(V,A) that is equal to {1,…,i ,…,N).
A Arc set in graph G(V,A) that is equal to {(i, j): i,jV, ij}.
S Arbitrary subset of set v.
r(S) Minimum number of vehicles needed to serve set S.
Cv Capacity of vehicles.
di Demand at node i.
tiv Service time (time for rendering service) to node i by vehicle v.
tijv Travel time between arc (i, j) by vehicle v.
Tv Maximum service time and travel time for the vehicle v.
Cij Travel cost between arc (i, j).
Pij Availability of path between arc (i, j).
Qij Unavailability of path between arc (i, j), where Qij = 1- Pij.
Xijv 1, if arc i, j is traversed by vehicle v; 0, otherwise.

This model determines vehicle routings and transportation schedules for the facilities, which are
determined by the SSCP model. The objective function (1) minimizes the transportation cost, and the
objective function (2) maximizes the probability of delivery to customers. Constraints (3) and (4)
cause each customer is served by one vehicle. Constraint (5) defined for equivalence in input and
output of each node. Constraint (6) indicates the maximum capacity of vehicles. Constraint (7) shows
the maximum working time for each vehicle. Constraint (8) causes that each vehicle travels only from
one depot to customers. Constraint (9) ensures that sub-tours do not establish. Constraint (10) defines
the type of decision variables.
We interpret how the objective function (2) can be changed from multiplication to summation, and so
from the minimum to the maximum (Hassan-Pour et al. [6]). The objective function Z2 is in the
multiplication form. To prevent that the value of Z2 become zero, a fixed value added to it. Then, to
convert it to a linear function, the logarithm of the obtained function is maximized. Finally, in order to
change it to maximization, the objective function is multiplied by -1. According to this interpretation,
the objective function (2) is changed by:

Z‫׳‬2 =Min   
i 1,...,N j 1,...,N v 1,...,NV
Qij X ijv (11)

A multi-objective solution technique should be adopted for the MO-MDSVRP. Rahimi-Vahed et al.
[17] classified five methods for multi-objective optimization problems, such as scalar methods,
interactive methods, fuzzy methods, meta-heuristic methods, and decision-aided methods. We employ
a linear composition objective function for the ease of applications as good as Alumur and Kara [11],
Erkut and Alp [12], and Caballero et al. [13].

Based on a convex combination of transportation cost (Z1) and unavailability of routes (Z2), the
combined objective function obtained as follows (Hassan-Pour et al. 2009):

30
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

Z1 Z 2‫׳‬
Min Z opt   ( )  (1   )( ) (12)
Z 1max Z 2‫ ׳‬max

In this equation, Z1max and Z2‫׳‬ max

are the maximum cost and the maximum unavailability of routes,
respectively.

5. SA methodology
The SA methodology is a meta-heuristic method that uses a stochastic approach to direct the search.
To crystallize a solid in the annealing operation, it is heated to a high temperature and gradually cooled
to low. This heating process allows the atoms to move randomly in order to reach a minimum energy
state. This analogy can be used in solving any combinatorial optimization. It makes the search to
proceed to a neighboring state even if the move causes the value of the objective function to become
worse (Tavakkoli-Mogaddam et al. [18]).

5.1. Hybrid SA algorithm

We generate an initial solution by a heuristic method and improve the quality of the solution by the
genetic operators. Hence, we hybridize the SA algorithm with the genetic operators. The flowchart for
our hybrid SA algorithm is shown in Fig 1.

31
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

Start

Notations:
Define the initial temperature (T0) and Set Tk= T0. Ln Epoch length, the number of
Define the rate of temperature decrease (). accepted solutions in each temperature.
Define the maximum number of consecutive Kn Maximum number of
temperature trails (Kn) and the number of accepted consecutive temperature trails.
solutions in each temperature (Ln). T0 Initial temperature.
Generate an initial solution (Z0) and Set ZBest= Z0. α Rate of the current temperature
decrease (i.e., cooling schedule).
Z Feasible solution.
Set K= 0 and L=0. F(z) Objective function value for z.
L Counter for the number of
accepted solutions in each temperature;
Select 1-opt or 2-opt operators randomly, and K Counter for the number of
generate the neighbor solution (Zl) and set ZNew=Zl. consecutive temperature trails, where Tk is
equal to temperature in iteration k

ΔF = F(ZNew) - F(ZBest).

Accept the new solution as

Yes the best solution (ZBest= ZNew)
Is ΔF < 0?
and so the solution of stage L
(ZL= ZNew).
No

Is R (0, 1) < Accept the new solution as the

Exp (-ΔF/ Tk)? solution of stage L (ZL= ZNew).
Yes

No
Set L=L+1.

Yes
Is L < Ln?

No

Set K=K+1.

Set Tk = Yes Is K< Kn , Tk>0?

Tk-1 -  Tk-1.

No

The best solution is obtained and stops.

32
Fig. 1 Flowchart for the hybrid SA algorithm
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

5.2. Initial solutions

An initial solution is generated by a heuristic method based on the capacitated VRP model. In this
model, each vehicle serves to one node. The homogeneous fleet is used. First, an un-served node is
selected at random for each route, and then a vehicle is allocated to that. Until the capacity of the
vehicle is violated, another un-served node is searched at random continually. Also, the capacity of the
allocated vehicle is updated at each route.

5.3.Neighborhood solution
To obtain a neighborhood solution, the genetic operators, namely mutation (i.e., 1-opt) and crossover
(i.e., 2-opt), are embedded in the SA algorithm. Hence, these operators are selected randomly in order
to improve the solution in each iteration of the algorithm. Tables 3 and 4 show the representation of
some of the solutions in the genetic operators. In mutation operator, two routes of two vehicles are
randomly chosen out of the existing feasible solutions, and then a customer is deleted from a route and
added to the other route (e. g. customer C3 in table 3). In crossover operator, two routes belonging to
two vehicles are randomly chosen from the existing feasible solutions, and then two nodes out of two
routes are exchanged with each other (e. g. customer C2 and customer C3 in table 4).

Table 3 The representation of solution in the mutation operator.

Customers C1 C2 C3 C4 C5 Tout planning
Previous Route 1, Vehicle 1 1 0 1 1 0 C1- C3- C4
solution Route 2, Vehicle 2 0 1 0 0 1 C2- C5

Next Route 1, Vehicle 1 1 0 0 1 0 C1- C4

solution Route 2, Vehicle 2 0 1 1 0 1 C2- C3- C5

Table 4 The representation of solution in the crossover operator.

Customers C1 C2 C3 C4 C5 Tout planning
Previous Route 1, Vehicle 1 1 0 1 1 0 C1- C3- C4
solution Route 2, Vehicle 2 0 1 0 0 1 C2- C5

Next Route 1, Vehicle 1 1 1 0 1 0 C1- C2- C4

solution Route 2, Vehicle 2 0 0 1 0 1 C3- C5

5.4. Assessment of the hybrid SA efficiency

The illustrative results are provided to the model verification and large-sized problems. Our proposed
hybrid SA is programmed in the Microsoft Visual Basic 6.0 and executed on a Pentium 4 CPU 2.4GHz
with 256MB RAM.

5.5.Parameter settings
To solve the MO-MDSVRP, we use data generated at random, as shown in Table 5. We solve this
problem with the various values of  between 0 and 1 in increments of 0.1. As the decision maker is
concerned about the cost and probability, based on trade-off curve, it is expected to select a solution
generated by  between 0.4 and 0.6. Hence for to simplicity, the coefficient of the objective function is
set to 0.5.

33
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

Table 5 Data required to solve the MO-MDSVRP

Entity Parameter Value/ Data
No. of depots (M) Output of SSCP (Optimization software)
Covering coefficients Output of SSCP (Optimization software)
Depot Travel cost (Cij) U(500, 1000) cost unit
Travel time (tijv) U(0.3, 1.8) Hour
Availability of arcs i and j (pij) U(0, 1)
6 to 10 Customers for small size and 20, 25, 35,
No. of customers (N)
50,75,100 Customers for large size problems
Demand (di) U( 0.1, 1.1) Ton
Customer Travel cost (Cij) U(500, 1000) cost unit
Travel time (tijv) U(0.3, 1.8) hours
Service Time (tiv) U(0.3, 0.8) hour
Availability of arcs i and j (pij) U(0, 1)
3,4 Vehicles for small size and 5 to 7, 30, 40 and 50
No. of vehicles (NV)
vehicles for large size problems
Vehicle
Vehicles capacity (Cv) 2 Ton
Work time (Tv) 8 Hours
No. of accepted
100
Solution in each temperature (Ln)
Maximum no. of consecutive(Kn)
Program 100
temperature trials
control
Initial temperature (T0) 5
Final temperature (Tk) 0
Rate of decreasing temp. (α) 0.95
Objective
coefficient
 0.5 (0.1 to 0.9 in trade-off curve)

5.6.Model verification for small-sized problem

Seven small-sized test problems are solved by the optimization software and the hybrid SA algorithm.
Table 6 reports the results obtained of solving a number of problems with 3 and 4 vehicles and 6 to 10
customers. For each solution method, this table presents the objective function value (OFV), the
computation time expressed in seconds, and the gap (in percent) between the optimal solution values
obtained from two methods. It also shows that our hybrid SA algorithm is as good as mathematical
programming in term of the solution quality. This approach solves such a hard problem in the
maximum gap of 1.3%. In addition, the average percentage gap of solutions obtained by the
optimization software and the hybrid SA is 0.5% showing the verification of our hybrid SA.

Table 6 Computational results for small-sized problems

No. of Exact algorithm Hybrid SA
Problem Depots/ Gap
name Customers/ Time Time (%)
OFV OFV
Vehicles (sec.) (sec.)
P7 1/ 6/ 4 5.5 32 5.5 1.12 0
P8 1/ 7/ 3 5.32 58 5.32 0.98 0
P9 1/ 7/ 4 5.88 305 5.88 1.25 0
P10 1/ 8/ 4 6.34 2612 6.34 1.39 0
P11 1/ 9/ 3 5.93 1496 6.01 0.86 1.3
P12 1/ 9/ 4 6.54 7451 6.62 1.31 1.2
P13 1/ 10/ 3 6.56 15751 6.62 1.23 0.9
Average gap 0.5

34
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

Fig. 2 indicates the average CPU time related to the hybrid SA and exact method solutions in respect
to 3 and 4 vehicles and 6 to 10 customers given in Table 6. This figure shows the solution time for the
mathematical approach rises drastically when the problem size increases.

15000
Exact algorithm
12000 Hybrid SA

Time (Sec.) 9000

6000

3000

0
1/6/4 1/7/3 1/7/4 1/8/4 1/9/3 1/9/4 1/10/3
Problem (Depot/Customers/Vehicles)

Fig.Fig.
2 CPU time
2 CPU timefor small-sized
in small-sized problems
problem

5.7. Model efficiency for large-sized problems

Since the optimization software packages are unable to solve large-sized problems, we can simplify
the mathematical model by relaxing some of the constraints and obtain the lower bound (LB)
solutions. Then, the results of hybrid SA is compared with lower bound (LB) solutions in five test
problems. Even by relaxation constraints, the optimization software cannot obtain the LB solution at
reasonable time in more than one depot, seven vehicles, and thirty five customers, as shown in Table 7.
Fig. 3 illustrates the average CPU time related to the hybrid SA and LB solutions in respect to 5 to 7
vehicles and 20, 25 and 35 customers as given in Table 7. This figure indicates the solution time for
the mathematical approach rises drastically when the problem size increases. In addition, as shown in
Table 7, we solve large-sized problems by the SA algorithm with 8, 9, 10, 14, and 16 depots; 50, 75,
and 100 customers; and 30, 40, and 50 vehicles that are located in the SSCP. Although, usually the
solution of exact method are better than the hybrid SA, the average gap between the hybrid SA and LB
solution is equal to 21%, which is an acceptable outcome for large dimensions.
To demonstrate the efficiency of the hybrid SA, we solve large-sized problems by the SA algorithm
without 1-opt and 2-opt operators. As shown in Table 7, for large sizes, the average gap between
solutions obtained by the SA (without operators) and hybrid SA is equal to 16%, which demonstrates
the hybrid SA algorithm is more efficient than the SA algorithm. However, the solution time for the
hybrid SA is better than SA in most large-sized problems. Fig. 4 illustrates the average CPU time
related to the SA and hybrid SA solutions in large-sized problem as given in Table 7.

Table 7 Computational results for large-sized problems

Strategy ІІ:
No. of Strategy І:
LB SA (Without LB & І І & ІІ
Problem depots/ Hybrid SA
operators)
name customers/
Time Time OFV Time Gap Gap
vehicles OFV OFV
(sec.) (sec.) (sec.) (%) (%)
P14 1/ 20/ 5 8.05 178 9.52 7.17 - - 18 -
P15 1/ 20/ 6 7.62 478 8.85 8.3 - - 16 -
P16 1/ 35/ 6 12.95 3127 16.35 19.2 - - 26 -
P17 1/ 35/ 5 12.63 4484 15.93 14.4 - - 26 -
P18 1/ 25/ 7 9.96 6922 11.90 13.7 - - 19 -

35
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

P19 8/ 50/ 30 - - 32.71 703 39.23 850 - 20

P20 9/ 75/ 30 - - 49.19 1282 53.28 1647 - 8.3
P21 9/ 75/ 40 - - 53.05 1911 61.11 2126 - 15.2
P23 14/ 50/ 30 - - 33.22 1134 41.57 1184 - 25
P24 14/ 50/ 40 - - 34.20 1782 41.98 1535 - 22.7
P25 14/ 75/ 40 - - 51.14 3204 57.67 2694 - 12.8
P22 10/ 100/ 70.69 6315 -
- - 61.63 5881 14.7
40
P26 16/ 100/ 75.60 8230 -
- - 68.93 7509 9.7
50
Average gap 21 16

7000

6000 Exact algorithm (LB)

Hybrid SA
5000
Time (Sec.)

4000

3000

2000

1000

0
1/20/5 1/20/6 1/35/6 1/35/5 1/25/7
Problem (Depots/Customer/Vehicles)

Fig.
Fig.33CPU
CPUtime
time for
for the lower
lower bound
bound

7500 Hybrid SA
SA
6000
Time (Sec.)

4500

3000

1500

0
8/50 /30

9/75 /30

9/75 /40

14/5 0/30

14/5 0/40

14/7 5/40

10/1 00/40

16/1 00/50

Problem (Depots/Customers/Vehicles)

Fig.4 4CPU
Fig. CPUtime
time
in for large-sized
large-sized problems
problem

6. Conclusions
In this paper, we have developed an efficient hybrid meta-heuristic method for the special type of the
location-routing problem (LRP) that serviceability of facilities and availability of routes are under

36
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

uncertainty. To solve this problem, we have used a hybrid SA-GA meta-heuristic. In this hybrid
method, we have used two genetic operators, namely mutation and crossover that embedded in the SA
algorithm. The SA algorithm allows the search to escape from local optimum, and GA operators are
used to obtained the neighborhood solution and improve the quality of solutions. To show the
performance of our hybrid SA algorithm, the average CPU times and the quality of solutions have
been compared with the SA algorithm and optimization software. The hybrid SA works better than SA
to obtain the solution. Also, the genetic operators makes the SA become more efficient in terms of the
average CPU time in the most large-sized problems.
We suggest some future studies, such as other meta-heuristics can be developed to solve the presented
model. Also, Or-opt and λ-opt operators can be used to generate the various initial solutions. The
model can be further enhanced by including pick-up and delivery distribution processes and
considering parameters as fuzzy and probabilistic.

References
[1] H. Min, V. Jayaraman, R. Srivastava, Combined location-routing problem: A synthesis and future
research directions, European Journal of Operational Research. 108 (1998) 1-15.
[2] G. Nagy, S. Salhi, Location- routing: Issues, models and methods, European Journal of Operational
Research.176 (2007) 649-672.
[3] S.D. Lee, W.T. Chang, On solving the discrete location problems when the facilities are prone to
failure, Applied Mathematical Modeling. 31 (2007) 817-831.
[4] H.S. Hwang, Design of supply-chain logistics system considering service level, Computers &
Industrial Engineering. 43 (2002) 283-297.
[5] H.S. Hwang, A stochastic set-covering location model for both ameliorating and deteriorating
items”, Computers & Industrial Engineering. 46 (2004) 313-319.
[6] H.A. Hassan-Pour, M. Mosadegh-Khah, R. Tavakkoli-Moghaddam, Solving a multi-objective multi-
depot stochastic location-routing problem by a hybrid simulated annealing, Proceeding of the
institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture. 223 (2009) 1045-
1054.
[7] T.H. Wu, C. Low, J.W. Bai, Heuristic solution to multi-depot location-routing problems, Computers
& Operations Research. 29 (2002) 1393-1415.
[8] K. Doerner, A. Focke, W.J. Gujahr, Multicriteria tour planning for mobile healthcare facilities in a
developing country, European Journal of Operational Research. 179 (2007) 1078-1096.
[9] G. Zhou, H. Min, M. Gen, A genetic algorithm approach to the bi-criteria allocation of customers to
warehouses, International Journal of production Economics. 86 (2003) 34- 45.
[10] F. Chu, N. Labadi, C. Prins, Heuristics for the periodic capacitated arc routing problem, Journal of
Intelligent Manufacturing.16(2) (2005) 243-251.
[11] S. Alumur, B.Y. Kara, A new model for the hazardous waste location-routing problem, Computers
& operations Research. 34 (2007) 1406-1423.
[12] E. Erkut, O. Alp, Designing a road network for hazardous material shipment, Computers &
Operations Research. 34 (2007) 1389-1405.
[13] R. Caballero, M. Gonzalez, F.M. Guerrero, J. Molina, C. Paralera, Solving a multiobjective location
routing problem with a metaheuristic based on tabu search: Application to a real case in Andalusia,
European Journal of Operational Research. 176 (2007) 1751-1763.
[14] S.Z. Erdogan, E.M. Esin, Routing table updating by using intelligent agents, Journal of Intelligent
Manufacturing. 20(2) (2009) 187-193.
[15] S.C. Liu, C.H. Chung, A heuristic method for the vehicle routing problem with backhauls and
inventory, Journal of Intelligent Manufacturing. 20(1) (2009) 29-42.

37
H.A. Hassan-Pour, M. Mosadegh-Khah, M. Zareei / J. Math. Computer Sci. 12 (2014) 27 - 38

[15] S.C. Liu, C.H. Chung, A heuristic method for the vehicle routing problem with backhauls and
inventory, Journal of Intelligent Manufacturing. 20(1) 29-42.
[16] H.A. Hassan-Pour, M. Mosadegh-Khah, R. Tavakkoli-Moghaddam, An integrated location-routing
problem in a stochastic environment by a hybrid simulated annealing, Proceedings of the 6th
International Symposium on Intelligent & Manufacturing Systems, Sakarya University, Turkey. (2008)
552-558.
[17] A.R. Rahimi-Vahed, M. Rabbani, R. Tavakkoli-Moghaddam, S.A. Torabi, F. Jolai, A multi-objective
scatter search for mixed-model assembly line sequencing problem, Advanced Engineering
Informatics. 21 (2007) 85-99.
[18] R. Tavakkoli-Moghaddam, N. Saffaei, Y. Gholipour, A hybrid simulated annealing for capacitated
vehicle routing problems with the independent route length, Applied Mathematics and
Computation.176 (2006) 445-454.

38