See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/285959002

A multi-size compartment vehicle routing problem for

multi-product distribution: Models and solution procedures

Article  in  International Journal of Artificial Intelligence · January 2013

CITATIONS READS

9 371

3 authors, including:

Krisada Asawarungsaengkul Tarit Rattanamanee

King Mongkut's University of Technology North Bangkok King Mongkut's University of Technology North Bangkok
10 PUBLICATIONS   20 CITATIONS    6 PUBLICATIONS   10 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

ENG-NEW-60-15 View project

ENG RA View project

All content following this page was uploaded by Tarit Rattanamanee on 16 January 2019.

The user has requested enhancement of the downloaded file.

International Journal of Artificial Intelligence,
ISSN 0974-0635; Int. J. Artif. Intell.
Autumn (October) 2013, Volume 11, Number A13
Copyright © 2013 by IJAI (CESER PUBLICATIONS)

A Multi-Size Compartment Vehicle Routing Problem for

Multi-Product Distribution: Models and Solution
Procedures

Krisada Asawarungsaengkul11 , Tarit Rattanamanee2 , and Teeradej Wuttipornpun3

Department of Industrial Engineering

Faculty of Engineering, King Mongkuts University of Technology North Bangkok
Bangkok, 10800, Thailand
1
Email : [email protected], 2 Email : [email protected],
3
Email : [email protected]

ABSTRACT
This paper presents a Multi-Size Compartment Vehicle Routing Problem with Split Pattern
(MC-VRP-SP). This MC-VRP-SP focuses on the delivery of multiple types of liquid products
from depot to customers. The main objective is to minimize the total traveling cost. Multiple
types of fluid product are loaded to the vehicles having several capacities with non-identical
capacity of compartments. Then, two mathematical models are proposed to represent
this MC-VRP-SP. Since the order quantity may be over the capacity of compartment, the
splitting of order is needed. The original customers demand is split according to the pre-
determined patterns so that it can be loaded to compartments appropriately. However, the
split demands must be delivered by the same truck and one compartment can support only
a single split demand. The optimization approach is utilized to solve these mathematical
models within the limited computational time. A homogeneous and heterogeneous fleet of
multi-compartment trucks are considered in the numerical experiment. The results from
numerical examples show that the optimization approach can yield us the optimal solution
only in small-scale problem. For large-scale problem, the optimization approach can de-
termine only the feasible solutions. Thus, the local search so called 2-opt is employed to
make an improvement on solution of the large-scale problem. Moreover, a saving algorithm
is utilized to separate the customers into clusters. Then each cluster is solved separately
by CPLEX. The results show that this clustering can significantly improve the solutions on
the large-scale problem.
Keywords: Multi-size compartment vehicle routing problem, Multi-product, Homogeneous
and heterogeneous fleet, Split pattern, Clustering, Local search.
2000 Mathematics Subject Classification: 90-08, 90B06, 90C11, 90C27
Computing Classification System: G.1.6, I.2.8

1 INTRODUCTION

The vehicle routing problem (VRP) is widely popular problem in logistics and supply chain.
The original VRP first introduced by Dantzig and Ramser (1959) which classify as the classi-
cal capacitated vehicle routing problem (CVRP). The CVRP is concerned with delivery goods

www.ceser.in/ijai.html
www.ceserp.com/cp-jour
www.ceserpublications.com
 International Journal of Artificial Intelligence

to a set of customers that have certainly demands. The objective of problem is to determine
shorted distance routing for serve customers. The typical constraints of CVRP are the truck
capacity, the time limit for traveling. The literatures on the CVRP have received much atten-
tion in Laporte et al (2000) and Toth and Vigo (2002). The vehicle with several compartments
or multi-compartment vehicle routing problem (MC-VRP) is a variant of the capacitated VRP.
The MC-VRP consists of designing set of routes to serve the demand of multi-product of a
set of customers. In many real world problems, the MC-VRP can be applied in situation that
the goods are different in types, for example milk (cows and goats) and fluid product (gasoline
and diesel), different in quality, for instance expired dates, different in treatment conditions, for
instance level of storage temperatures, and so on. To save transportation costs, vehicles with
several compartments are employed in order to allow transporting the heterogeneous goods
together on the same vehicle, but in different compartments.

The MC-VRP has widely practical applicability, however the MC-VRP has received not much
attention in the literature. The MC-VRP is a variant of the capacitated vehicle routing prob-
lems (CVRP). A review on VRP can be found in Cordeau et al (2007). They discussed on
the formulations of VRP, exact algorithm, heuristic algorithms, and metaheutistics. A vehicle
with multiple compartments in the distribution of foods to convenient stores can be found in
Chajakis and Guignard (2003). This paper tries to minimize both delivery and cooling costs.
The cooling costs are the costs of keeping non-ambient temperature items at suitable tem-
peratures during deliveries. The mathematical models which are 0-1 programming and mixed
integer programming were formulated. The Lagrangean Relaxations was utilized to solve this
problem. Derigs et al (2011) presented a good review and summary on the vehicle routing
with compartments in terms of applications, modeling, and heuristics. This paper presented a
model formulation, an integer programming model of the MC-VRP. The demand of customer
order cannot be split. Many heuristics and metaheuristics are reviewed and the comparisons
on solutions of the benchmark problems are presented as well. Other literatures on the mod-
eling, optimizations, heuristics and evolutionary optimizations can also be found in Raza and
Vidyarthi (2009), Farahani et al (2012), and Purcaru et al (2013).

The deliveries of liquid fuels to petrol station using vehicle with multiple compartments was
presented by Abdelaziz et al (2002). The adjustment of order quantities was employed in order
to fit vehicle compartments before starting dispatch. This paper proposed an integer linear
programming and applied the variable neighborhood search heuristics to determine the solu-
tion. Another research on the petrol delivery was done by Surjandari et al (2011). This paper
studied on the petrol delivery assignment with multi-product, multi-depot, split deliveries, and
time windows. The purpose of this paper is to determine the fleet assignment and to design
the delivery routes to petrol station using the available trucks with minimum total traveling cost.
Each compartment has equal capacity and is not equipped with a flow meter. The demand
quantities must be equal or an integer multiple of compartments capacity. The fluid product
in each compartment has to be entirely unloaded into a storage tank. Each petrol station re-
quires the delivery within the time window. Each truck can be assigned for several trips and
each petrol station can have multiple visits. The Tabu search algorithm has been used to solve
this problem. Avella et al (2004) also proposed solving a fuel delivery problem by heuristic and

238
 International Journal of Artificial Intelligence

exact approaches. The customer order is an integer multiple of one thousand litres and may
require one or more compartments. Each compartment must be completely emptied after re-
plenishment. All orders of a customer have to be delivered by one truck. The branch-and-price
is employed to improve solution obtained from heuristic which is the packing/routing heuristic.
The improvement on logistic of multi-compartment chemical tankers was also proposed by
Jetlund and Karimi (2004). Cornillier et al (2008) also proposed the petrol station replenish-
ment problem (PSRP). The objective of their work is to deliver petroleum products to petrol sta-
tions by an unlimited heterogeneous fleet of compartmented tank trucks. The solution consists
of determining the quantities to deliver within a given interval, allocating products to compart-
ments and designing the routes to stations. The possibly standard ILP algorithm was proposed
to solve the loading problem. The routing problem is handled using a matching approach and
a column generation scheme. They proposed a formulation of a mixed-integer linear program-
ming (MILP) and a heuristic algorithm to maximize the total profit for a single ship. The petrol
station replenishment problem (PSRP) with time window can be found in Corniller et al (2009).
Their research consider the quantity of petroleum products to be delivered to petrol stations
by a limited heterogeneous fleet of tank-trucks, the assignment of products to compartments,
truck routing and scheduling. A mathematical formulation and two heuristics were presented
to determine the appropriate solution. A paper of a generalized version of the trip packing
problem was proposed by Boctor et al (2011). The solution procedure is to assign a number
of trips to a fleet of a limited number of non-identical tank-trucks. Working time of vehicles
is limited and the net revenue of each trip depends on the truck used. This paper provides
a mathematical formulation and proposes some construction, improvement and neighborhood
search.
The split demand of the order quantity is typical found in the split delivery vehicle routing prob-
lem (SDVRP). Most papers proposed the splitting of customers demand among vehicles on
the different routes rather than the splitting of an order quantity and assigning the split order
quantities to compartments in one vehicle. The split delivery in VRP (SDVRP) and the multi-
depot VRP (MDVRP) can be found in Gulczynski et al (2011), which the demand of customer
can be split into two or more trucks and multi-trip must be enable. A customer can be visited
more than one truck. An integer programming-based heuristic was utilized to search for the
high-quality solution. Muyldermans and Pang (2010) presented a study on a MC-VRP with
demands of different products. The initial solution is obtained by utilizing the Clarke and Wright
saving algorithm (Clarke and Wright, 1964). Then the local search including 2-opt, cross, ex-
change, relocate are used to improve the solution. The neighbor lists and marking and the
guided local search are also combine with the local search to determine the better solution.
Asawarungsaengkul et al (2012) presented a MC-VRP with splitting of the order quantity in de-
livery of the multiple types of fluid products to petrol stations. They propose two mathematical
models with the fixed-split pattern and float-split pattern. LINGO12 is used to determine the
solutions.
This paper focuses on the delivery of multiple types of fluid products by multi-size compartment
vehicles to customers. The capacity of each compartment for storing the fluid product cannot
be adjusted and is different in capacity. If the customers order of a fluid product is over than

239
 International Journal of Artificial Intelligence

the compartments capacity, that demand order will be split so that the order of a customer can
be loaded to exactly one vehicle. Thus this paper presents the procedure of splitting the order
quantity and assigning the split order to multi-compartment trucks in the delivery of multiple
types of fluid products from a supplier to petrol retailers. The problem can be formulated as the
mixed integer programming (MIP). The optimization software namely CPLEX12.2 are utilized to
solve this specific MC-VRP with two types of splitting patterns. The numerical experiment with
various sizes of customers will be provided and both homogenous and heterogeneous fleets
of multi-compartment trucks are in our interest. The 2-opt local improvement will be utilized to
improve the solution obtained from optimization software. Moreover, the clustering technique
that divides the large-scale problem into the small-scale problems will also be presented in the
last session as well.

2 PROBLEM DESCRIPTIONS OF THE PROPOSED MC-VRP-SP

This paper proposes a MC-VPR-SP for the distribution of the multiple types of fluid products
to customers. The order quantity for each type of fluid products is less than trucks capacity
and therefore, the orders from several customers will be loaded to the same truck on delivery
routes. The objective of this paper is to minimize the total traveling cost. The main activities of
this problem is to split the order quantity and load each split demand to the compartments with
various capacities and then determine the optimal routes.
The characteristics of the customer order and the truck routing are subject to the following as-
sumptions and constraints:
1. each customer must be visited only once a day,
2. the orders are known and the planning period is made for one day delivery,
3. each truck must not be loaded over its capacity for a single route,
4. one compartment can contain only a single customers demand,
5. there are many types of fluid product and several compartment capacities in the homoge-
neous and heterogeneous fleet of trucks,
6. the unload time at customer is assumed to be constant,
7. each order for each fluid product type must be a multiple of 2 kilolitres, and
8. the vehicle is not equipped with a flow meter which means that all compartments will be
emptied after replenishment.

In MC-VRP-SP for fluid product delivery, each type of fluid product is incompatible. Each order
can easily be split and delivered in independent compartments. Let Yjpq be a predetermined
integer number. The split pattern Yjpq is used to divide the original order quantity of customer
j of product p (denoted by Djp ) by split pattern q where q = 1, ..., Q. After splitting, quantity of
 Djp
each split order will be identical as following equation: Djpq = .
Yjpq
However, it is not allowed to mix between two customers orders; even if they are the same type
of the fluid product. At the customer, driver just replenishes the fluid products to the big tanks.

240
 International Journal of Artificial Intelligence

3 MATHEMATICAL FORMULATION OF THE MC-VRP-SP

The mathematical models for this MC-VRP were proposed by Rattanamanee and Asawarungs-
aengkul (2011). This paper makes a revision on those mathematical models. The integer non-
linear program which is hard to solve is modified. The time constraint is added to the models.
To deal with the MC-VRP-SP for fluid product delivery, two mathematical models are pro-
posed and optimization software CPLEX is employed to provide us the practical solution for
the planner of this problem. CPLEX determines the solution for MIP by using branch and cut
technique.

We consider the splitting of the order quantity by integer number as described in pervious
section and design of truck routes. The decision variables are the integer number used for
splitting the order quantity, the selected compartments for each split-order quantity, and the
truck routes. The constraints of this MC-VRP are the trucks capacity, the compartments ca-
pacity, and the working hours per day. The multi-product, multi-compartment, different capacity
among compartments, homogeneous fleet of trucks, and heterogeneous fleet of trucks are the
main characteristics taken into account in this research. There is only one depot in this MC-
VRP. Next, the proposed mathematical models are grouped into two categories based on the
splitting pattern available. Model 1 is the representation for the single-split pattern while model
2 is considered as the multi-split pattern. In the single-split pattern, there is an integer number
available for splitting the order quantity. For the multi-split pattern, each order quantity can be
divided by an integer number selected from the available set which depends upon the order
size.

3.1 The MC-VRP-SP for Single-Split Pattern: Model 1

Model 1 is formulated as the mixed integer linear programming (MILP). The model formulation
is based on the following notations.

TC total cost of traveling

N number of customers including one depot
dij the shortest path distance from i to j; i, j = 1, 2, 3, ..., N ; where 1 is depot
P number of products p = 1, 2, 3, ..., P
K number of available trucks
Ck variable cost of truck k (bath/km), k = 1, 2, 3, ..., K
Wk number of compartments for truck k, w = 1, 2, 3, ..., Wk
Djp demand of customer j on product p (kl)
Capk capacity of truck k
Capckw capacity of compartment w of truck k; w = 1, 2, 3,..., Wk
Ui arbitrary real numbers which specify the position of node i
Yjp integer number of customer j on product p

Djp split-order quantity which equals to Djp /Yjp
Un unload time which is assumed a constant value
Ve the average velocity of a truck

241
 International Journal of Artificial Intelligence

T max Time limit for one trip

The decision variables are:

p
Zkwj 1 if split order quantity of product p for customer j is loaded to compartment w of truck k
0 otherwise
Xijk 1 if truck k travels from node i to node j
0 otherwise
Ejk 1 if truck k travels to node j
0 otherwise

The mathematical model can be written as following equalities and inequalities.


N 
N 
K
Minimize T C = dij Xijk Ck (1)
i=1 j=1 k=1

subject to

(N − 3)Xjik + (N − 1)Xijk + Ui ≤ Uj + (N − 2) for i, j = 2, 3, ..., N ; i = j; ∀k (2)

N 
N
Xihk − Xhjk = 0 for h = 2, 3, ..., N ; ∀k (3)
i=1 j=1
N
Xijk = Ejk for j = 2, 3, ..., N ; i = j; ∀k (4)
i=1

K
Ejk = 1 for j = 2, 3, ..., N (5)
k=1
N
X1jk ≤1 ∀k (6)
j=2

K Wk p
Zkwj = Yjp ∀j, p (7)
k=1 w=1
N  P
p
Zkwj ≤1 ∀k, w (8)
j=1 p=1
p
Ejk ≥ Zkwj ∀k, w, j, p (9)

N W k 
P
p 
Zkwj Djp ≤ Capk ∀k (10)
j=1 w=1 p=1
p 
Zkwj Djp ≤ Capckw ∀k, w, j, p (11)
N  N
Xijk dij N  N
Ve + Xijk U n ≤ T max ∀k (12)
i=1 j=1 i=1 j=2
p
Xijk ∈ {0, 1}, Ejk ∈ {0, 1}, Zkwj ∈ {0, 1} ∀k, w, j, p (13)

In this mathematical model, the objective function is to minimize the total cost of fuel distribution
(Equation (1)). Constraint (2) represents a sub-tour elimination adopting from Desrochers
and Laporte (1991). Constraint (3) states that a truck that visits node j must leave node j.
Constraint (4) and (5) forces each customer to have only one vehicle visited. Constraint (6)
states that each vehicle has only one route. Constraint (7) ensures that the amount of product
p (after splitting) of customer j must be delivered. One compartment can contain only a single
split order quantity is stated in Constraint (8). Constraint (9) means that each compartment
can contain the split demand if and only if the customer j is visited by truck k. Constraint
(10) ensures all trucks must not be loaded over theirs capacity for a single route. Constraint

242
 International Journal of Artificial Intelligence

(11) compels each compartment not to be loaded over its capacity. Constraint (12) is used to
make sure that the traveling time for each route does not exceed the time limit. Constraint (13)
defines decision variable to be binary.

3.2 MC-VRP-SP for Multi-Split Patterns (MILP): Model 2

Some variables are needed to be defined for model 2. The additional notations are used in the
model 2 are listed below.

Yjpq integer number q of customer j on product p

Q number of patterns; Q ∈ {1, ..., Qmax }
Qmax maximum number of split pattern available

Djpq split demand order equals Djp /Yjpq

The customers order can be split by dividing the order quantity with an available integer num-
ber selected from a set of predetermined integer number Yjpq .

The original of the mathematical model 2 proposed by Rattanamanee and Asawarungsaengkul

(2011) is classified as mixed integer nonlinear programming (MINLP). In this paper, mathemat-
ical model 2 is modified to be a mixed integer linear programming (MIP). A decision variable
pq pq
Skwj ∈ {0, 1} is added in to model 2. The variable Skwj is used to make linear relation between
pq
variable Zkwj and Vjpq . The linear relation is shown in equation (27) and (28). The model 2 in
a mixed integer programming (MIP) form can be solved by CPLEX more effectively.

The mathematical model for the multi-split patterns MC-VRP can also be written as following
equalities and inequalities.

N 
N 
K
Minimize T C = dij Xijk Ck (14)
i=1 j=1 k=1

subject to

(N − 3)Xjik + (N − 1)Xijk + Ui ≤ Uj + (N − 2) for i, j = 2, 3, ..., N ; i = j; ∀k (15)

N 
N
Xihk − Xhjk = 0 for h = 2, 3, ..., N ; ∀k (16)
i=1 j=1
N
Xijk = Ejk for j = 2, 3, ..., N ; i = j; ∀k (17)
i=1

K
Ejk = 1 for j = 2, 3, ..., N (18)
k=1
N
X1jk ≤1 ∀k (19)
j=2

K Wk pq
Zkwj = Yjpq Vjpq ∀j, p, q (20)
k=1 w=1
N  P Q
pq
Zkwj ≤1 ∀k, w (21)
j=1 p=1 q=1
pq
Ejk ≥ Zkwj ∀k, w, j, p, q (22)
N W k P  Q
pq 
Skwj Djpq ≤ Capk ∀k (23)
j=1 w=1 p=1 q=1
pq 
Skwj Djpq ≤ Capckw ∀k, w, j, p, q (24)

243
 International Journal of Artificial Intelligence


N 
N
Xijk dij 
N 
N
Ve + Xijk U n ≤ T max ∀k (25)
i=1 j=1 i=1 j=2

Q
Vjpq = 1 ∀j, p (26)
q=1
pq pq
Zkwj + Vjpq − Skwj −1≤0 ∀k, w, j, p, q (27)
pq pq
−Zkwj − Vjpq + 2Skwj ≤ 0 ∀k, w, j, p, q (28)
pq pq
Xijk ∈ {0, 1}, Ejk ∈ {0, 1}, Zkwj ∈ {0, 1}, Vjpq ∈ {0, 1}, Skwj ∈ {0, 1} ∀k, w, j, p, q (29)

In model 2, the objective function is minimizing the total traveling cost of fuel distribution. Most
constraints of model 2 are not different from those of model 1 except constraint (20), (23), (24),
and (26). These constraints are utilized to ensure that one of an integer number Yjpq will be
selected to split the order quantity and all split demands will be loaded appropriately.

4 NUMERICAL EXAMPLES

This section discusses the numerical example and the solution procedure by utilizing these
two mathematical models in section 3. The delivery of fuel oils to petrol stations is used to
demonstrate the solution procedures for MC-VRP-SP. The hypothetical problems which vary
the number of customer range from 5 to 50 customers are generated with 2 replicates of
each problem size. The optimization software namely IBM ILOG CPLEX v.12.2.2 is used to
find solution. The constant variables for these problems are defined and will be used for all
problems. The maximum order quantity is 16 kiloliters, the unload time is assumed a constant
value of 20 minutes per customer, the average velocity of a truck is 40 km/hour, and time limit
for one trip is 8 hours. For single-split pattern, the denominators Yjp are as following:
⎧
⎨ 1; if Djp = 8, 12, 16
⎪
Yjp = 2; if Djp = 10, 12, 14, 16
⎪
⎩
0; if Djp = 0

For multi-split pattern model 2, we set Qmax = 3 and the denominators Yjpq are depended upon
the order quantity Djp . Parameter Q is defined as
⎧
⎨ 3; if Djp = 8, 12, 16
⎪
Q= 2; if Djp = 4, 6
⎪
⎩
1; if Djp = 2, 10, 14
⎧
⎨ 2, 4, 8; if Djp = 16
⎪
when Q = 3 denominators can be Yjpq = 2, 3, 6; if Djp = 12 ,
⎪
⎩
1, 2, 4; if Djp = 8

1, 2; if Djp = 4
when Q = 2 denominators can be Yjpq = ,
1, 3; if Djp = 6
⎧
⎨ 2; if Djp = 4
⎪
and when Q = 1, denominator can beYjpq = 1; if Djp = 6 .
⎪
⎩
0; if Djp = 0

244
 International Journal of Artificial Intelligence

Several numerical examples are tested by utilizing mathematical models in section 3. The
customer positions locate in x-y coordinates which have the boundary between -50 to 50 kilo-
meters away from the depot where the depot is located at [0, 0] and the order quantity is
an integer of multiple of 2,000 liters. The numerical examples are separated into two groups
which are the MC-VRP-SP with the homogeneous fleet of trucks and the MC-VRP-SP with the
heterogeneous fleet of trucks. The maximum computational time is 7200 seconds.

4.1 MC-VRP-SP with a homogeneous fleet of trucks

In this section, the single-split pattern and multi-split patterns MC-VRP will be examined with
the homogeneous fleet of trucks. The information of customer including locations, demand
order, available vehicles, and results are displayed in Table 1, 2, 3, and 4, respectively.

Table 1: X-Y coordinates for 10 customers with a homogeneous fleet of trucks

Coordinate Customer
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10
X -22.25 -4.75 -50.00 -15.00 -31.75 45.50 -17.00 2.50 36.50 34.75
Y -45.00 -23.00 -15.00 25.50 4.25 5.50 -37.50 30.25 -39.25 -20.00

Table 2: Demands (kl) of each customer for 10 customers with a homogeneous fleet of trucks
Product Customer
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10
1 6 0 4 0 0 0 0 0 6 0
2 0 0 0 0 2 4 0 0 4 0
3 0 8 0 16 6 0 0 0 0 12
4 0 0 0 4 0 0 2 0 0 4
5 2 0 0 0 0 4 0 8 0 0

Table 3: Data of available homogeneous fleet of trucks

Type Amount Capacity (kl) Compartment Capacity (kl) Cost (Bath/km)
1 6 40 88664422 19

Table 4: Result for a problem of 10 customers with a homogeneous fleet of trucks

Truck Routes - Model 1 Truck Routes - Model 2
1 D→C9→C10→D 1 D→C8→C4→D
2 D→C2→C7→C1→C3→C5→C8→D 2 D→C9→C10→C6→D
3 D→C4→C6→D 3 D→C2→C7→C1→C3→C5→D
Total Cost 8,846 Baht Total Cost 7,135 Baht
State optimal State optimal
Time (sec.) 28.39 Time (sec.) 116.80
Note: D = depot

245
 International Journal of Artificial Intelligence

Other problem sets ranging from 5 to 50 customers are also examined where each problem
set consists of 2 subproblems. The upper bound (UB), lower bound (LB), and calculation time
of all problems with a heterogeneous fleet of trucks can be found in Table 5 - 8.

Table 5: Problem 1 of 5, 10, 15 customers with a homogeneous fleet of trucks

Model Total cost (Baht) for each problem
5 Time 10 Tme 15 Time
1 5,517 UB 1.17 7,572 UB 17.70 8,349 UB 7200
5,517 LB 7,572 LB 7,898 LB
2 4,960 UB 1.20 7,572 UB 138.8 8,349 UB 7200
4,960 LB 7,572 LB 6,609 LB

Table 6: Problem 2 of 5, 10, 15 customers with a homogeneous fleet of trucks

Model Total cost (Baht) for each problem
5 Time 10 Tme 15 Time
1 4,226 UB 1.20 8,446 UB 28.39 10,161 UB 7200
4,226 LB 8,446 LB 9,112 LB
2 4,075 UB 0.56 7,135 UB 116.8 9,613 UB 7200
4,075 LB 7,135 LB 7,468 LB

Table 7: Problem 1 of 20, 30, 40 and 50 customers with a homogeneous fleet of trucks
Model Total cost (Baht) for each problem
20 30 40 50
1 12,285 UB 18,217 UB ∗ 22,816 UB ∗ 40,439 UB
8,179 LB 9,317 LB 7,585 LB 8,427 LB
2 11,853 UB 17,767 UB - UB - UB
5,048 LB 6,820 LB 7,820 LB 8,239 LB

Table 8: Problem 2 of 20, 30, 40 and 50 customers with a homogeneous fleet of trucks
Model Total cost (Baht) for each problem
20 30 40 50
1 12,512 UB 18,104 UB ∗ 23,701 UB ∗ 39,739 UB
8,260 LB 7,749 LB 7,949 LB 8,308 LB
2 12,007 UB 18,686 UB - UB - UB
4,952 LB 7,628 LB 7,533 LB 8,239 LB
Note: ∗ CPLEX was terminated because of out of memory. The computational time limit was reduced to handle this problem.

CPLEX can provide us the global optimal solutions up to the problem of 10 customers. Mostly,
CPLEX is able to yield us the best solution when utilizing the multi-split patterns model (model
2). For the large-scale problem (40 and 50 customers), solving the single-split pattern model
enable us to have the best solution. We found that CPLEX is unable to find the feasible solution
within 2 hours when solving the problem of 50 customers and utilizing the multi-split pattern
(model 2). It is noticed that the multi-split pattern models are more difficult to find the feasible
solution for the large-scale problem.

246
 International Journal of Artificial Intelligence

4.2 MC-VRP-SP with a heterogeneous fleet of trucks

In this section, the single-split and multi-split pattern MCVRP will be examined with a fixed
heterogeneous fleet of trucks. A problem of 10 customers is solved and the information of
customer locations, demands of each customer, available vehicles, and results are displayed
in Table 9, 10, 11, and 12, respectively.

Table 9: X-Y coordinates for 10 customers with a heterogeneous fleet of trucks

Coordinate Customer
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10
X 39.25 7 -20.75 -22 -5.5 22.25 -3 -22.75 -18 7.5
Y -25.25 -33.75 17.5 -1.5 45.5 21.5 11.5 13.5 -28.25 6.25

Table 10: Demands (kl) of each customer for 10 customers with a homogeneous fleet of trucks
Product Customer
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10
1 8 0 14 4 0 12 4 0 16 0
2 0 0 0 0 0 0 0 4 0 8
3 0 0 0 0 6 0 6 2 0 0
4 0 0 0 8 0 2 0 0 0 0
5 0 4 0 0 0 0 0 0 0 0

Table 11: Data of available heterogeneous trucks

Type Amount Capacity (kl) Compartment Capacity (kl) Cost (Bath/km)
1 1 16 442222 15
2 1 16 4444 15
3 1 40 88888 19
4 1 40 88664422 19
5 1 32 666644 17
6 1 32 884444 17

Table 12: Result for a problem of 10 customers with a heterogeneous fleet of trucks
Truck-Type Routing - Model 1 Truck-Type Routing - Model 2
1-3 D→C10→C1→C2→C9→D 1-1 D→C10→D
2-4 D→C4→D 2-4 D→C3→C8→C4→D
3-5 D→C6→C5→C7→D 3-5 D→C6→ C5→ C7→ D
4-6 D→ C8→C3→ D 4-6 D→ C9→ C2→ C1→D
Total Cost 6,548 Baht Total Cost 5,894 Baht
State optimal State optimal
Time (sec.) 4.63 Time (sec.) 171.6

Other problem sets ranging from 5 to 50 customers are also examined where each problem
set consists of 2 problems. The computational time limit is 7200 seconds. The UB and LB of
all problems with a heterogeneous fleet of trucks are in Table 13 - 16.

247
 International Journal of Artificial Intelligence

Table 13: Problem 1 of 5, 10, 15 customers with a heterogeneous fleet of trucks

Model Total cost (Baht) for each problem
5 Time 10 Tme 15 Time
1 4,300 UB 1.08 6,548 UB 4.63 8,642 UB 525
4,300 LB 6,548 LB 8,642 LB
2 4,211 UB 1.03 5,894 UB 171.6 8,642 UB 7200
4,211 LB 5,894 LB 4,825 LB

Table 14: Problem 2 of 5, 10, 15 customers with a heterogeneous fleet of trucks

Model Total cost (Baht) for each problem
5 Time 10 Tme 15 Time
1 2,004 UB 1.37 5,891 UB 10.94 9,090 UB 1684
2,004 LB 5,891 LB 9,090 LB
2 2,004 UB 5.5 5,666 UB 268.4 8,853 UB 7200
2,004 LB 5,666 LB 4,992 LB

Table 15: Problem 1 of 20, 30, 40 and 50 customers with a heterogeneous fleet of trucks
Model Total cost (Baht) for each problem
20 30 40 50
1 12,798 UB 16,198 UB ∗ 23,044 UB ∗ 29,695 UB
6,195 LB 6,808 LB 7,585 LB 7,550 LB
2 12,985 UB 16,572 UB ∗ 28,092 UB - UB
5,029 LB 5,900 LB 6,995 LB 7,394 LB

Table 16: Problem 2 of 20, 30, 40 and 50 customers with a heterogeneous fleet of trucks
Model Total cost (Baht) for each problem
20 30 40 50
1 10,666 UB 14,354 UB ∗ 26,213 UB ∗ 30,351 UB
5,417 LB 6,802 LB 7,281 LB 7,797 LB
2 10,895 UB 14,704 UB ∗ 24,004 UB - UB
4,793 LB 5,932 LB 6,575 LB 6,914 LB

4.3 Clustering Method and 2-opt Algorithm

In large-scale problem, we can observe the big gap between UB and LB. Thus, the Clarke &
Wright’s saving algorithm in Clarke and Wright (1964) is employed to perform the clustering of
customers and then each cluster is separately solved model 2. Equation (30) represents the
saving algorithm. Fig. 1 (a) and 1 (b) show how to save the distance when visit 2 two node
before traveling back to the depot.

Sij = D0i + Dj0 − Dij (30)

where

248
 International Journal of Artificial Intelligence

D0i is the distance from depot 0 to node j

Dij is the distance from node i to node j
Sij is a saving value traveling directly between node i to node j

Figure 1: Saving Algorithm

The clustering technique is used to arbitrarily separate the customers into the smaller groups.
In this paper, each cluster has 10 customers. Then, each cluster is separately solved by
utilizing model 2 for the Problem of 40 and 50 customers. Each cluster will be obtained the
optimal solution.

In the problem with a homogeneous fleet of trucks, the clustering technique reduce the total
cost of traveling from 22,816 baht to 20,525 baht in problem 1 of 40 customers as show in
Table 17 and from 23,701 baht to 22,493 baht in problem 2 of 40 customers (see Table 21).
For the problems of 50 customers, the total cost can be reduced from 40,439 baht to 28,815
baht in problem 1 as shown in Table 18 and from 39,739 baht to 25,644 baht in problem 2 (see
Table 21).

Table 17: Results of clustering method for problem 1 of 40 customers with a homogeneous
fleet of trucks
Group Customer Cost (Baht) Time (sec.)
1 8, 37, 10, 5, 23, 29, 25, 38, 15, 13 5,232 1753.0
2 20, 16, 31, 4, 34, 2, 11, 28, 18, 26 6,915 155.8
3 9, 32, 40, 1, 12, 24, 27, 30, 3, 22 5,530 1406.6
4 14, 6, 36, 39, 21,33, 7, 17, 35, 19 2,848 241.3
Total 20,525 3556.7

Table 18: Results of clustering method for problem 1 of 50 customers with a homogeneous
fleet of trucks
Group Customer Cost (Baht) Time (sec.)
1 10, 39, 25, 32, 8, 28, 17, 44, 36, 33 7,430 1130.3
2 38, 49, 7, 21, 14, 42, 29, 26, 48, 9 6,369 2608.6
3 4, 31, 37, 2, 47, 23, 19, 46, 34, 30 5,272 1094.6
4 50, 11, 5, 24, 3, 22, 12, 41, 15, 1 5,146 1750.0
5 6, 13, 16, 18, 20, 27, 35, 40, 43, 45 4,598 75.7
Total 28,815 6659.2

249
 International Journal of Artificial Intelligence

In the heterogeneous fleet of trucks, the total cost of traveling reduces from 23,044 baht to
21,802 baht in problem 1 of 40 customers as shown in Table 19 and from 24,004 baht to
19,783 baht in problem 2 of 40 customers (see Table 22). For problem of 50 customers, the
total cost of traveling is reduced from 29,695 baht to 25,000 baht in problem 1 as shown in
Table 20 and from 30,351 baht to 28,312 baht in problem 2 (see Table 22).

Table 19: Results of clustering method for problem 1 of 40 Customers with a heterogeneous
fleet of trucks
Group Customer Cost (Baht) Time (sec.)
1 17, 37, 4, 7, 20, 6, 18, 28, 30, 2 7,019 923
2 10, 23, 33, 25, 9, 31, 16, 14, 29, 34 5,990 1267.2
3 8, 19, 12, 27, 13, 5, 36, 35, 21, 40 5,516 376.5
4 15, 22, 24, 39, 38, 3, 32, 11, 1, 26 3,277 34.9
Total 21,802 2601.6

Table 20: Results of clustering method for problem 1 of 50 customers with a heterogeneous
fleet of trucks
Group Customer Cost (Baht) Time (sec.)
1 47, 36, 12, 30, 22, 23, 9, 5, 18, 31 7,316 341.3
2 21, 26, 25, 17, 11, 41, 46, 27, 14, 39 7,077 1178.2
3 34, 2, 50, 8, 6, 40, 4, 33, 28, 20 5,031 1290.1
4 13, 29, 7, 10, 48, 16, 19, 43, 44, 24 2,518 3229.3
5 1, 3, 15, 49, 45, 32, 35, 37, 38, 42 3,059 56.5
Total 25,000 6095.4

The 2-opt procedure works on improving single route by systematically exchange the route
direction between two pairs of consecutive customers in the route and evaluate whether the
routing cost of the route is improved or not. The exchange mechanism is illustrated in Fig. 2,
in which the route direction between customers i –(i + 1) and j –(j + 1) are interchanged. If
the routing cost of the modified route is better than the routing cost of original one, the route
is updated with the modified one. This paper employs the 2-opt to improve the best solution
of the problem of 40 and 50 customers which are obtained from CPLEX. The improvement in
percent by implementing 2-opt are in the Table 21 and 22. The results show that 2-opt can
reduce the cost of traveling for 7 out of 8 problems.

Figure 2: The 2-opt Algorithm

250
 International Journal of Artificial Intelligence

The comparisons on the several solution approaches are shown the Table 21 and 22. Fig. 3
to Fig. 6 are also used to illustrated the performance of each solution procedure. For small-
scale problem (number of customers are less than or equal to 15), the solutions received from
the CPLEX in solving 2 mathematical models is not significant difference. However, Model 2
can yield us the lowest cost. In the most cases of large-scale problem, using Model 1 can
find the lowest traveling cost. The solutions in the problems of a homogeneous fleet of and
heterogeneous trucks are in the same trend.

When using 2-opt and clustering technique, it is noticed that the clustering technique using the
saving algorithm can make a significant improvement on the traveling cost of both problems in
the large-scale problem.

Table 21: Comparisons of the clustering method, the best UB, and 2-opt on a homogeneous
fleet of trucks
Size Problem Best UB Best UB 2-opt Clustering ∗ % Improvement
(Baht) obtained from (Baht) solution (Baht) 2-opt Clustering
40 1 22,816 Model 1 21,914 20,525 3.95 10.04
2 23,701 Model 1 22,924 22,493 3.28 5.10
50 1 40,439 Model 1 38,636 28,815 4.46 28.74
2 39,739 Model 1 39,624 25,644 0.29 35.47
Note: ∗ % Improvement compared to the Best UB

Table 22: Comparisons of the clustering method, the best UB, and 2-opt on a heterogeneous
fleet of trucks
Size Problem Best UB Best UB 2-opt Clustering ∗ % Improvement
(Baht) obtained from (Baht) solution (Baht) 2-opt Clustering
40 1 23,044 Model 1 23,044 21,802 0.00 5.39
2 24,004 Model 2 23,002 19,783 4.17 17.58
50 1 29,695 Model 1 28,517 25,000 3.97 15.81
2 30,351 Model 1 30,060 28,312 0.96 6.72

251
 International Journal of Artificial Intelligence

Figure 3: Solutions of the homogeneous fleet of trucks - Problem 1

Figure 4: Solutions of the homogeneous fleet of trucks - Problem 2

252
 International Journal of Artificial Intelligence

Figure 5: Solutions of the heterogeneous fleet of trucks - Problem 1

Figure 6: Solutions of the heterogeneous fleet of trucks - Problem 2

253
 International Journal of Artificial Intelligence

5 CONCLUSIONS AND DISCUSSIONS

This paper proposes three mathematical models and the solution procedures including the op-
timization approaches (CPLEX), 2-opt algorithm, and clustering technique. The objective of the
mathematical models is to minimize the total traveling cost. The major constraint is that each
compartment on a truck can contain a single customer’s demand. Since the compartments
are not identical, the predetermined split pattern including single- and multi-split patterns are
proposed to separate the demand so that the split demand will not be over the compartment
capacity. CPLEX can guarantee the optimality in the problem size up to 15 customers. In
additions, using model 2 can potentially obtain the lower cost than the other alternatives. Two
groups problem with a homogeneous fleet and a heterogeneous fleet yield us the solutions
in the same fashion. Next, the 2-opt heuristic and the clustering using the saving algorithm
are employed to improve the solution in the large-scale problem. It is noticed that 2-opt can
improve the solutions received from CPLEX up to 4.6% in the problem of 50 customers with
a homogeneous fleet of trucks. The clustering technique can yield us best solution compared
to other approaches. Using the clustering can make an improvement on a solution of 50 cus-
tomers by 35.47%. In addition, the cumulative computational time for all clusters is still lower
than 7,200 seconds. These results reveal that separating the problem into the small-scale en-
ables us to have the lower traveling cost. It is recommended that, for large-scale problem, the
practical solution procedure is to make the clusters before searching for the optimal routes.

6 ACKNOWLEDGEMENT

This research was supported by Faculty of Engineering, King Mongkuts University of Tech-
nology North Bangkok, Thailand. This support is gratefully acknowledged. The authors also
thank Dr.Suebsak Nanthavanij at Sirindhorn International Institute of Technology, Thammasat
University, Thailand for his careful reading and valuable input.

7 REFERENCES

Abdelaziz, B., Roucairol, C., Bacha C., 2002, Deliveries of Liquid Fuels to SNDP Gas Station
using vehicle with multiple compartment. Proceedings of 2002 IEEE International Conference
on Systems, Man and Cybernetics, Hammamet, Tunisia, 1, 478-483.

Asawarungsaengkul, K., Rattanamanee, T., Nanthavanij S., 2012, A Multi-Compartment Vehi-

cle Routing Problem with Splitting of the Order Quantity for Delivery of the Multiple Types of
Fuel Oils to Gas Stations. Proceedings of The 2nd International Conference on Computer and
Management (CAMAN 2012), Wuhan, China, 3352-3355.

Avella, P., Boccia, M., Sforza, A., 2004, Solving a fuel delivery problem by heuristic and exact
approaches. European Journal of Operation Research. 152(1), 170-179.

Boctor F.F., Renaud J., and Cornillier F., 2011, Trip packing in petrol stations replenishment.
Omega. 39, 8698.

254
 International Journal of Artificial Intelligence

Chajakis, E. D., Guignard, M., 2003, Scheduling Deliveries in Vehicle with Multiple Compart-
ment. J. Global Opt. 26, 43-78.

Clarke, G., Wright, J.W., 1964, Scheduling of Vehicle from Central Depot to a Number of De-
livery Point. Operation Research. 12, 568-581.

Cordeau, J-F., Laporte, G., Savelsbergh, M.W.P., Vigo, D., 2007, Vehicle routing. Handbooks
in operations research and management science: transportation. 14, 367-428.

Cornillier, F. and Boctor, F.F., Laporte, G., and Renaud, J., 2008, An exact algorithm for the
petrol station replenishment problem. Journal of the Operational Research Society. 59(5),
607-615.

Cornillier, F., Laporte, G., Boctor, F.F., Renaud, J., 2009, The Petrol Station Replenishment
Problem with Time Windows. Computer & Operation Research. 36(3), 919-935.

Dantzig, G.B., Ramser, J.H., 1959, The Truck Dispatching Problem. Management Science.
6(1), 80-81.

Derigs, U., Gottlieb, J., Kalkoff, J., Piesche, M., Rothlauf, F., 2011, Vehicle routing with com-
partment: applications, modeling and heuristics. OR SPECTRUM. 33(4), 885-914.

Desrochers, M., Laporte, G., 1991, Improvements and Extension to the Miler-Tucker-Zemlin
Subtour Elimination Constraints. Operations Research Letters. 10, 27-36.

Farahani, S.M, Abshouri, A.A, Nasiri, B., Meybodi, M.R., 2012, Some hybrid models to improve
firefly algorithm performance. International Journal of Artificial Intelligence. 8(12), 97-117.

Gulczynski, D., Golden, B., Wasil, E., 2011, The multi-depot split delivery vehicle routing prob-
lem: An integer programming-based heuristic, new test problems, and computational results.
Computers & Industrial Engineering. 61(3), 794-804.

Jetlund, A.S., Karimi, I.A., 2004, Improving the Logistic of Multi-compartment Chemical Tankers.
Computer & Chemical Engineering. 28(8), 1267-1283.

Laporte, G., Gendreau, M., Potvin, J.Y., Semet, F., 2000, Classical and modern heuristics for
the vehicle routing problem. International Transactions in Operational Research. 7, 285-300.

Purcaru, C., Precup, R.E., Lercan, D., Fedorovici, L.O., David, R.C., Dragan, F., 2013, Optimal
robot path planning using gravity search algorithm. International Journal of Artificial Intelli-
gence. 10(13), 1-20.

Rattanamanee, T., Asawarungsaengkul, K., 2011, An Optimization Approach of the Multi-

Compartment Vehicle Routing Problem with Single & Multiple Split-Pattern for Fuel Oil De-
livery. Proceedings of The 2011 International Conference on Computer and Computational
Intelligence (ICCCI 2011), Bangkok, Thailand, 27-32.

Raza, Z., Vidyarthi, D.P., 2009, A computation grid scheduling model to minimize turnaround
using modified GA. International Journal of Artificial Intelligence. 3(9), 86-106.

Surjandari, I., Rachman, A., Dianawati, F., Wibowo, R.P., 2011, Petrol Delivery Assignment
with Multi-Product, Multi Depot, Split Deliveries and Time Windows. International Journal of

255
 International Journal of Artificial Intelligence

Modeling and Optimization. 1(1), 375-379

Toth, P., Vigo, D., 2002, The vehicle routing problem. SIAM Monographs on Discrete Mathe-
matics and Applications, Philadelphia, PA.

256

View publication stats