Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5464
An Improvement of Genetic Algorithm for Location-routing Problem
Apichat Buakla1 and Sirichai Tanratanawong2
School of Logistics and Supply Chain, Naresuan University, Phisanulok, Thailand.
Abstract
This paper proposed an improvement of genetic algorithm for
solving the location-routing problem (LRP). In this study,
tournament selection technique and k-mean clustering are
applied to create customer routing in each depot. The
algorithm is tested on benchmark problems set by data and
compared with former algorithm. The computation results
illustrate that the proposed algorithm can produce optimal
solutions better than comparative method. The findings prove
that the proposed algorithm can solve LRP effectively which
lead to minimize the total cost of network distribution in real
world business.
INTRODUCTION
Nowadays, the logistics costs consume a major proportion of
business operational costs. In Thailand, the average logistics
cost is 14.2% of GDP and 52.8% of total logistics cost is
transportation cost [1] which impacts on goods price increase
[2] [3] [4] and household expenses [5]. These costs can be
condensed extremely by designing an efficient distribution
network of the business supply chain. For example, in the
present the popular distribution network is transportation of
goods from factory to distribution centers or central depots.
After that, some goods which must be transported long haul use
full truckload (FTL) to carry and wait in regional depot. When
transporting goods from regional depot to each customer, the
customer demand should be less than truckload (LTL).
Therefore, the last distribution section must ship around to
many customers or milk run, which collects many goods of
demand in one truck, transports them from regional depot to
fulfil each demand and come back to the same depot as shown
in figure 1.
The regional depot location and suitable vehicle routing for
serving all customers are two important factors, which create
the potential for this distribution network. In the past, designers
found depot locations first and vehicle routing later, not solving
two problems at the same time. This procedure led to higher
distribution costs and management problems in routing.
Because when designers select depot location, first priority has
to constraint and effect in routing design later. To improve the
efficiency in supply chain distribution network, researchers
develop the mathematical model which combines and solves
two questions, depot location and vehicle routing at the same
time, calling it location-routing problem (LRP)
The LRP is one type of geographical problem similar to vehicle
routing problem [6] which consist of location data and their
attributes to both supply side (depot and vehicle data) and
demand side (customer data). In 1989, Salhi and Rand [7]
introduced LRP benefits when compared with former
procedures, after that LRP has been widely studied among
researchers. The high complexity and significance of the
problem have been attracting researchers to study this
discipline extensively. In real world problems, there are many
variants of the LRP incorporating constraints and conditions as
indicated by many review research works [8] [9] [10] e.g. the
Capacitated LRP (CLRP) that the depots and vehicles have
capacity constraints [11] [12] [13] and Two-echelon LRP
(LRP-2E) which compose of three layers (factories,
warehouses/depots and customers) [14] [15][16].
Presently researchers bring many methodologies to solve LRP.
In the earliest period, they suggested the exact method [17] [18]
to solve the problem only in small sized instances. After that,
heuristics [19] [20][21], meta-heuristics [22] [23] [24]and
hybrid methods [25] [26] [27] have been developed to work out
the disadvantages of the existing method. They can solve the
problem efficiently by using less computation time. The hybrid
method is the most recent approach for solving LRP. Due to the
significance of this problem, the researches around the world
have attracted to continually construct new methodologies to
solve this problem. Therefore, this study aims to develop a new
effective methodology for solving LRP which is necessary for
distribution network design in supply chain.
This paper proposes an improvement of genetic algorithm for
location-routing problem. The tournament selection and k-
mean clustering are applied to generate an initialization of the
population. If initial population is sufficiently diversified then it
is possible to choose the best solution for recombination and
reduction of the computational time.
The following contents include: section 2, describes the
location-routing problem (LRP). Section 3, presents the
proposed algorithm in details. The experimental results are
demonstrated in section 4. Finally, the conclusion is illustrated
and discussed in section 5.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5465
Potential regional depot
Opened regional depotRegional / Central DC
CustomerMilk Run 1
Regional DCFTL
Milk Run 2 FTL
Milk Run 3
FTLCentral DC
FTL
>>
>
>
Figure 1: Modern distribution network
PROBLEM DESCRIPTION
The CLRP is the basic variants of LRP [28] which can be
described as a combination of location and vehicle routing
problem with capacity constraints on depots and vehicles.
Demand side and supply side data of network are given as
follows: the number, location and demand of customers, the
number, location, and capacity of all potential depots, the
vehicle type and size. The problem must be overcome
constraints as follow: each customer demand must be served by
one single vehicle; each route must begin and end at the same
depot and not exceed vehicle capacity and the total load of the
routes assigned to a depot must not exceed depot capacity.
Prins et.al.[29] defines CLRP as follows: it is a weighted and
directed graph G = (V,A,C). V is a set of nodes which
combines with a subset I of m (possible depot locations) and a
subset J of n (customers). Every arc a = (i,j) in the arc set A
have own transportation costs Ca. Each depot site i ∈ I have
capacity Wi and an opening cost Oi. Customer demand dj for
each customer j ∈ J is available. A vehicle set K of capacity Q
is given. When network is operated, each vehicle is used only
on one single route and has a fixed cost F. The number of total
vehicles used or routes performed is a decision variable.
The problem objective is to minimize the total cost of network
by discovering which depots should be opened and which
routes should be arranged. The depot cost is an opening cost Oi
while the total route cost combines with the vehicle fixed cost F
and the arc costs Ca. The mathematical model is defined below:
Mathematical model:
. . . ,i i a ak aki I a Ak K k K a I
z O y C x F x
Subject to:
1 ,akk K a j
x j J
(2)
. , ,j akj J a j
d x Q k K i V
(3)
0 , , ak aka i a i
x x k K i V
(4)
1 ,aka I
x k K
(5)
1 ***** , aka L S
x S S J k K (6)
1 , , ak ak ija i J a j
x x f i I j J k K
(7)
. . ,j ij i ij J
d f W y i I (8)
0,1 , akx a A k K (9)
0,1 iy i I (10)
0,1 , ijf i I j V (11)
The objective functions (1) comprises of all costs described
before. Each customer belongs to one route only and has only
one predecessor which is guaranteed by constraints (2). Vehicle
and depot capacity constraints are specified to inequalities (3)
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5466
and (8). Constraints (4) and (5) confirm each route
continuousness and the origin depot of return. Constraint (6) is
sub-tour elimination constraint. Constraint (7) guarantees that if
a linking route among potential depots and customers is
opened, each customer can be assigned to only one depot.
Finally, all variables are Boolean which be indicated to
constraints (9), (10) and (11).
GENETIC ALGORITHM
Genetic algorithm (GA) is a heuristic search algorithm based
on the evolution. The GA concept is derived from genetic
transformation from older to next generation which consists of
selections, crossovers and mutations. Nowadays, GA is used to
solve optimization which represents an intelligent exploitation
of a random search problem. The original GA consists of the
following 5 components:
Step 1: generate an initialization of the populations NP vector
solutions from randomization.
Step 2: the fitness function/objective functions of the
populations are evaluated, after the initialization of
populations.
Step 3: the procedure for generating next generation of
population includes the following steps:
a) Apply proportional fitness selection (roulette
wheel), select two current populations, P1 and P2.
b) Apply one-point crossover operator to p1 and p2
with crossover rate (Pc) to obtain a child
chromosome C1 and C2.
c) Apply mutation operator to C1 and C2 with
mutation rate (Pm) to produce C1’ and C2’.
d) Add C’ to the successor population.
Step 4: replace the source population with the successor
population.
Step 5: if stopping criteria has not been met, return to Step 2.
THE PROPOSED ALGORITHM
This paper intends to improve the algorithm of the initialization
of the population by using tournament selection. The use of
former genetic algorithm to find the problem solution after
generating population step. The procedure of the proposed
algorithm includes the following steps:
Step 1: Initialization of the population; the structure of initial
population consists of the following components:
1. Generate feasible solutions for opened depot by equation
(12) then, sort the ascending order cost of each solution.
!
0k ! !
q jj j j
n nSolution for k nk n k
(12)
where 1,2,3, ,j n
n the maximum number of depots.
customerj depot
j
weightk
w
The example of general feasible solution, for instance A1
shown in figure 2.
Figure 2: The feasible solution for open depot of instance A1.
2. K-Means algorithm is applied to customer cluster in to the
depot. The number of clusters (k) is the same to the number of
depots in the solution. The k-mean algorithm consists of the
following components: (i) use the coordinates of each depots in
the solution as the central point of cluster. (ii) after that,
grouped customers by the k-means algorithm. The result of k-
means algorithm shown in figure 3.
Figure 3: The result of cluster using k-mean algorithm
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5467
3. The tournament selection is applied to create routing
customer in each depot of solution.
Figure 3: The example of a chromosome from initial
population step.
After generating population fitness, the objective functions of
the populations are evaluated.
Step 2: The selection step, using roulette wheel for random
selectin of two current individuals, P1 and P2.
Step 3: The crossover operation is performed in this step. The
two-point crossover is applied for P1 and P2. After that, the
depot part of the child consists of the left and the right part of
first and second parent.
Figure 4: The example of chromosomes crossover.
Step 4: The mutation applied for the permutation vector VP
is to insert a customer, selected at first position 1rM , at a
new position 2rM , also chosen at random. The new
chromosome is then obtained by swapping customer at
position 1rM and 2rM .
Figure 5: The example of chromosomes mutation.
Following that, the clockwise search process is applied to
move the customer in each vehicle and repeat until all
customers are moved. When clockwise search process is
terminated, all novel solutions which show better results than
the previous solution will be transferred to the next stage. The
figure 6 shows the clockwise process in 31car of depot 3.
Figure 6: Shows clockwise search process for 31car .
Step 5: Replace the source population with the successor
population.
Step 6: If stopping criteria has not been met, return to Step 2.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5468
Figure 7: The flowchart of the proposed algorithm.
EXPERIMENTAL RESULTS
The performance of the proposed algorithm is implemented in
Matlab language and tested on the 12 instances from Prins et al.
benchmark data set. [30] for the problem on vehicle and depot
capacities. The description of instances which can be classified by
number of depot and vehicles are shown in Table 1.
Table 1: The description of instance
Instance #
Depot
Capacity of
Depot
Capacity of
Vehicle
#
Customer
A1 5 140 70 20
A2 5 300 150 20
A3 5 70,140 150 20
A4 5 70,140 70 20
A5 5 350,420 70 50
A6 5 350,420 150 50
A7 5 350 70 50
A8 5 350 150 50
A9 5 300 150 50
A10 5 350 70 50
A11 5 350,420 70 50
A12 5 350,420 150 50
The computational results of the proposed model are
summarized in comparison with GRASP algorithm results from
Prins [30] who created this benchmark dataset in Table 2 and
shown thoroughly in Table 3. The comparisons demonstrate
percentage of average different result which separated in total
distribution network cost, number of opened depot and used
vehicle, the setup costs of depot (Cd) and the routing costs (Cr).
Table 2: The result comparison with former algorithm
Instance Cost #dep #veh Cd Cr
A1-A4 Average diff.(%) 0.05 - - - 0.09
Better - - - - -
A5-A12 Average diff.(%) -0.57 - - -1.60 -0.51
Better 7/8 - - 1/8 6/8
Tested instances Average diff.(%) -0.36 - - -1.06 -0.31
Better 7/12 - - 1/12 6/12
The minus numbers indicate that the proposed algorithm
produces less number of vehicle used, more effective decisions
in depot location and routing design especially in large sized
instance. Number of instances which are founded better by
results proposed by method are compared with number of
tested instances. The proposed model can reduce total network
cost by 0.36% in 7 instances form 12 tested instances. Figure 8-
11 illustrates the solution for the instance A1, A3, A6 and A9
respectively. The results demonstrate that the proposed
algorithm provides better depot location and vehicle routing
compared with the former algorithm as shown in the table.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5469
Table 3: The detail result from former algorithm and proposed algorithm
Instance GRASP Purposed % Difference
Cost #dep #veh Cd Cr Cost #dep #veh Cd Cr Cost #dep #veh Cd Cr
A1 55021 3 5 25549 29472 55131 3 5 25549 29582 0.20 - - - 0.37
A2 39104 2 3 15497 23607 39104 2 3 15497 23607 - - - - -
A3 48908 3 5 24196 24712 48908 3 5 24196 24712 - - - - -
A4 37542 2 3 13911 23631 37542 2 3 13911 23631 - - - - -
A5 90632 3 12 29319 59467 90160 3 12 25442 64718 -0.52 - - -13.22 8.83
A6 64761 2 6 15385 49376 63256 2 6 15385 47871 -2.32 - - - -3.05
A7 88786 3 12 29319 64270 88715 3 12 29319 59396 -0.08 - - - -7.58
A8 68042 3 6 29319 38723 67893 3 6 29319 38574 -0.22 - - - -0.38
A9 84055 3 12 19785 64270 84181 3 12 19875 64396 0.15 - - 0.45 0.20
A10 52059 3 6 18763 33296 51992 3 6 18763 33229 -0.13 - - - -0.20
A11 87380 2 12 18961 68419 86203 2 12 18961 67242 -1.35 - - - -1.72
A12 61890 2 6 18961 42929 61830 2 6 18961 42869 -0.10 - - - -0.14
Figure 8: Illustration of the solution for the problem instance
A1 with 20 customers.
Figure 9: Illustration of the solution for the problem instance
A3 with 20 customers.
Figure 10: Illustration of the solution for the problem instance
A6 with 50 customers.
Figure 11: Illustration of the solution for the problem instance
A9 with 50 customers.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5470
CONCLUSION
This paper proposed an improvement of genetic algorithm for
location routing problem (LRP). In order to improve the
initialized population of genetic algorithm, tournament
selection technique and k-mean custering are applied to create
customer routing in each depot. The testing on 12 instances
shows that the proposed algorithm provides the better results
when comparing with former methodology. The proposed
algorithm can reduce total distribution network cost when
comparing with the former algorithm. The experimental results
confirm that the proposed algorithm is the new alternative
methodology for solving the LRP which can minimize total
cost of distribution network in real world business.
REFERENCES
[1] Office of the National Economic and Social
Development Board.(2016). Thailand’s logistics report 2016. Bangkok, ISSN 1906-4373
[2] Golden, B. L., & Wasil, E. A. (1987). OR Practice—
Computerized Vehicle Routing in the Soft Drink Industry.
Operations research, 35(1), 6-17.
[3] De Backer, B., Furnon, V., Prosser, P., Kilby, P., &
Shaw, P. (1997, November). Local search in constraint
programming: Application to the vehicle routing
problem. In Proc. CP-97 Workshop Indust. Constraint-Directed Scheduling (pp. 1-15). Austria: Schloss
Hagenberg.
[4] Bräysy, O., & Gendreau, M. (2005). Vehicle routing
problem with time windows, Part I: Route construction
and local search algorithms. Transportation science, 39(1), 104-118.
[5] Rodrigue, J. P., Comtois, C., & Slack, B. (2009). The geography of transport systems. Routledge.
[6] Kantawong, K. (2017). Multi-phase Method for
Geographical Transportation. International Journal of Geoinformatics, 13(3).
[7] Salhi, S., & Rand, G. K. (1989). The effect of ignoring
routes when locating depots. European journal of operational research, 39(2), 150-156.
[8] Nagy, G., & Salhi, S. (2007). Location-routing: Issues,
models and methods. European journal of operational research, 177(2), 649-672.
[9] Prodhon, C., & Prins, C. (2014). A survey of recent
research on location-routing problems. European Journal of Operational Research, 238(1), 1-17.
[10] Drexl, M., & Schneider, M. (2015). A survey of
variants and extensions of the location-routing problem.
European Journal of Operational Research, 241(2),
283-308.
[11] Ghiani, G., & Laporte, G. (2001). Location-arc routing
problems. Opsearch, 38(2), 151-159.
[12] Duhamel, C., Lacomme, P., Prins, C., & Prodhon, C.
(2008, October). A memetic approach for the
capacitated location routing problem. In Proceedings of the 9th EU/Meeting on Metaheuristics for Logistics and Vehicle Routing, Troyes, France.
[13] Schneider, M., & Löffler, M. (2017). Large composite
neighborhoods for the capacitated location-routing
problem. Transportation Science.
[14] Perboli, G., Tadei, R., & Vigo, D. (2011). The two-
echelon capacitated vehicle routing problem: models
and math-based heuristics. Transportation Science, 45(3), 364-380.
[15] Nguyen, V. P., Prins, C., & Prodhon, C. (2012). A
multi-start iterated local search with tabu list and path
relinking for the two-echelon location-routing problem.
Engineering Applications of Artificial Intelligence, 25(1), 56-71.
[16] Abedinzadeh, S., Ghoroghi, A., Afshar, S., &
Barkhordari, M. (2017). A Two-Echelon Green Supply
Chain with Simultaneous Pickup and Delivery.
International Journal of Transportation Engineering and Technology, 3(2), 12.
[17] Laporte, G., & Nobert, Y. (1981). An exact algorithm
for minimizing routing and operating costs in depot
location. European Journal of Operational Research, 6(2), 224-226.
[18] Albareda-Sambola, M., Dı́az, J. A., & Fernández, E.
(2005). A compact model and tight bounds for a
combined location-routing problem. Computers & Operations Research, 32(3), 407-428.
[19] Barreto, S., Ferreira, C., Paixao, J., & Santos, B. S.
(2007). Using clustering analysis in a capacitated
location-routing problem. European Journal of Operational Research, 179(3), 968-977.
[20] Lopes, R. B., Barreto, S., Ferreira, C., & Santos, B. S.
(2008). A decision-support tool for a capacitated
location-routing problem. Decision Support Systems, 46(1), 366-375.
[21] Boudahri, F., Aggoune-Mtalaa, W., Bennekrouf, M., &
Sari, Z. (2013). Application of a clustering based
location-routing model to a real agri-food supply chain
redesign. In Advanced methods for computational collective intelligence (pp. 323-331). Springer, Berlin,
Heidelberg.
[22] Jabal-Ameli, M. S., Aryanezhad, M. B., & Ghaffari-
Nasab, N. (2011). A variable neighborhood descent
based heuristic to solve the capacitated location-routing
problem. International Journal of Industrial Engineering Computations, 2(1), 141-154.
[23] Derbel, H., Jarboui, B., Hanafi, S., & Chabchoub, H.
(2012). Genetic algorithm with iterated local search for
solving a location-routing problem. Expert Systems with Applications, 39(3), 2865-2871.
[24] Alvim, A. C., & Taillard, E. D. (2013). POPMUSIC for
the world location-routing problem. EURO Journal on
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 13, Number 7 (2018) pp. 5464-5471
© Research India Publications. http://www.ripublication.com
5471
Transportation and Logistics, 2(3), 231-254.
[25] Hemmelmayr, V. C., Cordeau, J. F., & Crainic, T. G.
(2012). An adaptive large neighborhood search
heuristic for two-echelon vehicle routing problems
arising in city logistics. Computers & operations research, 39(12), 3215-3228.
[26] Ting, C. J., & Chen, C. H. (2013). A multiple ant
colony optimization algorithm for the capacitated
location routing problem. International Journal of Production Economics, 141(1), 34-44.
[27] Villegas, J. G., Prins, C., Prodhon, C., Medaglia, A. L.,
& Velasco, N. (2013). A matheuristic for the truck and
trailer routing problem. European Journal of Operational Research, 230(2), 231-244.
[28] Jokar, A., & Sahraeian, R. (2012). A heuristic based
approach to solve a capacitated location-routing
problem. Journal of Management and Sustainability, 2(2), 219.
[29] Prins, C., Prodhon, C., & Calvo, R. W. (2006, April). A
memetic algorithm with population management (MA|
PM) for the capacitated location-routing problem. In
European Conference on Evolutionary Computation in Combinatorial Optimization (pp. 183-194). Springer,
Berlin, Heidelberg.
[30] Prins, C., Prodhon, C., & Calvo, R. W. (2006). Solving
the capacitated location-routing problem by a GRASP
complemented by a learning process and a path
relinking. 4OR, 4(3), 221-238.