4 Int. J. Intelligent Systems Technologies and Applications, Vol. 18, Nos.

1/2, 2019

Mixed integer programming for vehicle routing

problem with time windows

Divya Aggarwal* and Vijay Kumar

Computer Science and Engineering Department,
Thapar University,
Patiala, Punjab, 147001, India
Email: aggarwaldivya2593@gmail.com
Email: vijaykumarchahar@gmail.com
*Corresponding author

Abstract: Being a key element in logistics distribution, vehicle routing

problem becomes an importance research topic in management and
computation science. Vehicle routing problem (VRP) with time windows is a
specialisation of VRP. In this paper, a brief description of VRP is presented.
A mixed integer programming (MIP) is utilised to solve the vehicle routing
problem with time windows (VRPTW). A novel mathematical model of
MIP is formulated and implemented using IBM CPLEX. A novel constraint is
designed to optimise the number of vehicle used. The proposed model
is used to optimise both transportation cost and number of vehicle used
simultaneously. The proposed model is tested on two well-known instances
of Solomon’s benchmark test problem. Experimental results illustrate that the
proposed formulation provides promising solutions in reasonable computation
time. The sensitivity analysis of customer nodes is also studied.

Keywords: VRP; vehicle routing problem; time windows; Solomon’s instance;

CPLEX; MIP; mixed integer programming.

Reference to this paper should be made as follows: Aggarwal, D. and

Kumar, V. (2019) ‘Mixed integer programming for vehicle routing problem
with time windows’, Int. J. Intelligent Systems Technologies and Applications,
Vol. 18, Nos. 1/2, pp.4–19.

Biographical notes: Divya Aggarwal received her BTech in Computer Science

and Engineering from HCTM Technical Campus under Kurukshetra University
in 2015. Since July 2015, she is pursuing MTech in the field of Vehicle
Routing from Thapar University, Patiala. Her main research interests include
design and analysis of algorithms.

Vijay Kumar received his BTech from the M.M. Engineering College, Mullana.
He received his MTech from the Guru Jambheshwer University of Science
and Technology, Hisar. He received his PhD from the National Institute of
Technology, Kurukshetra. He has been an Assistant Professor at the
Department of Computer Science and Engineering, Thapar University, Patiala,
Punjab. He has more than nine years of teaching and research experience.
He has more than 40 research papers in international journals, book chapters
and conference proceedings. He is on the panel of reviewers of Elsevier and
Springer journals. His main research focuses on soft computing, image
processing, data clustering and multi-objective optimisation.

Copyright © 2019 Inderscience Enterprises Ltd.

 Mixed integer programming for vehicle routing problem with time windows 5

This paper is a revised and expanded version of a paper entitled ‘Mixed integer
programming for vehicle routing problem using CPLEX’ presented at
International Conference on Cloud, Big Data and IoT, Coimbatore, India,
29–30 March, 2017.

1 Introduction

Logistics distribution has a keen role in the present logistics management. For the last
few decades, vehicle routing problem (VRP), due to its key role in logistics distribution,
has become a topic of main concern for researchers working in the field of distribution
and computation. It has been observed that distribution costs include half of the total
system cost in most of the businesses such as food or drinks. These might sums up to
70% of the total cost (De Backer and Furnon, 1997; Golden and Wasil, 1987; Halse,
1992). Hence, the planning of vehicles and routes could help to minimise the input
resources. Toth and Vigo (2002) reported in their work that computerised methods
would decrease the transportation cost by 5–20%. The well-known examples of VRP are
supermarket distribution, freight transportation and industrial gases delivery.
The simple routing problem is known as travelling salesman problem (TSP).
It suggests that a salesman visiting number of cities should visit all the cities travelling
the minimum distance and returning to the same city where started. The VRP is the
extension of multiple travelling salesman problem (MTSP) in which each request is
associated with each city and each vehicle has a specific capacity limit. When the time
window is added to each customer location then the problem becomes vehicle routing
problem with time windows (VRPTW). The VRPTW is an expansion of VRP. The aim
of VRP is to promote an arrangement of customers with known requests while
minimising the cost on given set of routes. VRPTW involves the delivery in specific time
window provided by the customers.
The time window constraint can be further classified into two categories such as soft
and hard time windows. The former one involves the violation of time constraint at the
sake of cost. If the vehicle arrives before its earliest time period, then vehicle can violate
time window constraints. A new penalty function is added to vehicle in case of arriving
late or early. The former one does not take into consideration a vehicle to arrive after the
latest time to begin service. Hard time window constraint restricts the service time to fall
between the earliest and the latest time constraints. If the vehicle arrives before its earliest
time, then it has to wait until the time window starts.
The focus of this paper is hard time window constraint. The mixed integer
programming (MIP) has been used to solve the VRPTW. A novel constraint has been
designed to minimise the number of vehicle used. This constraint has been utilised in the
MIP. The proposed model has been tested on two instances of Solomon’s benchmark
problem. The remaining paper is structured as follows. Section 2 presents the basic
concepts of VRP and VRPTW. Section 3 introduces the overview of related work done in
the field of VRPTW. The problem formulation is described in Section 4. The
computational results are presented in Section 5. Finally, the concluding remarks are
drawn in Section 6.
6 D. Aggarwal and V. Kumar

2 Background

This section provides the basic concepts used in VRP and VRPTW.

2.1 Vehicle routing problem

The assignment of designing delivery or pickup service in the area of transportation and
supply chain is termed as VRP. VRP was first introduced by Dantzig and Ramser (1959)
as a generalisation of TSP. The main aim of VRP is to select the optimal routes for
gasoline delivery trucks which provide service between single distribution terminal
and number of service stations. Collection of household garbage, package delivery and
delivery trucks are the main sources of distribution or collection in real world. VRP has a
vital role in distributions and logistics.
VRP is the most studied problem in optimisation literature because of its practical
applicability and its considerable difficulty. VRP refers to the situation where distinct
routes can be determined for each vehicle whose starting and ending journey at one or
more depots. In case of adequate supply of vehicles, it also aims to use a minimum
number of vehicles. Figure 1 shows the illustration of classical VRP having 10 customers
and 3 vehicles. The routes in VRP are discriminated through colour lines.

Figure 1 A classical vehicle routing problem (see online version for colours)

2.2 Vehicle routing problem with time windows

The VRPTW is the specialisation of classical VRP, being very applicable in real world
problems. It involves a fleet of vehicles starting from a depot to serve a number of
customers which are diversely located. Each customer has a fixed demand request.
Owing to the applicability of VRPTW in real life situations, it has been widely
investigated through various researchers. Figure 2 describes the illustration of VRPTW
 Mixed integer programming for vehicle routing problem with time windows 7

having hard time window constraint. The routes in VRPTW are discriminated through
colour lines. Each customer has fixed time constraint window.

Figure 2 A typical vehicle routing problem with time windows (see online version for colours)

The main difference between VRPTW and VRP is that there are time windows imposed
by customers. Each customer can only be visited within the time window. For VRPTW,
the time window is the important factor during the selection of routes. The VRP is only
routing problem. VRPTW is the combination of routing and scheduling problem. Besides
these dissimilarities, both VRP and VRPTW are NP-hard optimisation problems.

3 Literature review

Various researches have been done in the field of VRPTW which can be used to specify
many real world problems. The early works on the VRPTW include case studies such as
Pullen and Webb (1967), Knight and Hofer (1968) and Madsen (1976). Earlier studies on
VRPTW include both optimisation algorithms and heuristic approaches. However, the
focus of current research is heuristic approaches. Various surveys have been done in the
field of VRPTW such as Golden and Assad (1986, 1988), Desrochers et al. (1988) and
Solomon and Desrosiers (1988). Many real-life applications have been proposed for
VRPTW including bank deliveries (Lambert et al., 1993), postal deliveries (Mechti et al.,
2001), industrial refuse collection (Golden et al., 2001), national franchise restaurant
services (Russell, 1995) and industrial gases delivery (Campbell et al., 2001).
VRPTW has been solved using heuristic and exact optimisation approaches. Owing to
the complexity of the VRP family problems, exact methods become useless especially
when large-scale instances are tackled. Kolen et al. (1987) used branch and bound
approach to solve a VRPTW having customer nodes varying between 6 and 15. When the
number of nodes is set to 6, then computation took about 1 min. The branch and bound
8 D. Aggarwal and V. Kumar

approach was unable to process for 12 customer nodes. Sam (1995) worked on reducing
the delivery costs by considering only travel time taken by vehicle while traversing from
one customer to another customer. Moreover, this work involves product delivery by
violating the time windows and compensating it by adding a penalty in the objective
function. Dondo et al. (2003) proposed a mixed integer linear programming (MILP) for
solving multi-depot heterogeneous-fleet VRPTW problem. Dondo and Cerda (2007)
offered three-phase heuristics for m-depot routing problem with time windows and
heterogeneous vehicles. The clustering algorithm was applied on VRPTW which could
solve at most 25 nodes by restructuring MILP.
Cetinkaya et al. (2013) proposed a hybrid approach that combines both MIP and
memetic algorithm (MA) for solving two-stage VRP with arc time windows. He added
time variant to each and every arcs. Suwansuksamran and Ongkunaruk (2013) presented
a MIP for a case study company. In this problem, a binary variable is used to indicate
whether a customer has time window constraint or not. Simbolon and Mawengkang
(2014) suggested a mixed integer approach for VRP with window having a variant of
avoiding routes. When pair of edges occurs frequently, then the sub-route is avoided to
eliminate the heavy traffic. The experimental results revealed that MIP formulation gave
good results for both small- and medium-size instances. However, it was concluded that
MIP formulation is good enough for instances having up to 50 customers.
Besides these, various researchers have applied heuristics techniques for VRPTW.
Holland (1975) proposed a genetic algorithm which initialises the random chromosome
and generates the solution for VRPTW in bit strings. Thereafter, chromosomes were
selected for crossover and mutation process to produce children. The process keeps on
repeating until the evolution has been converged. El-Sherbeny (2001) solved VRPTW
problem using a multi-objective simulated annealing (MOSA) method. Here, three
objectives were defined such as minimising the vehicle used, total time and concerning
the flexible duration of routes. Beatrice et al. (2006) presented a hybrid approach that
comprises of genetic algorithm (GA) and Pareto ranking to solve VRPTW. This method
was applicable on problems with more than 100 customer nodes.
Sabar et al. (2015) suggested an approach that consists of two phases, i.e., math phase
and hyper heuristic phase. This approach is better than the simple hyper heuristics
approach. Hokama et al. (2016) suggested branch-and-cut approach for solving VRP
having loading constraints. It solves the problem in smaller computation time,
but packing strategies and lower bounds could be improved. Ozbaygin et al. (2017)
proposed a branch-and-price algorithm for VRP. This technique used the concept of
roaming delivery locations. The proposed approach provided optimal solution. However,
it requires the prior knowledge of travel time and itinerary of customer.
From the literature, it has been observed that there are two main research issues
associated with VRPTW. First, the identification of route feasibility is an important issue
for VRPTW. Second, the selection of time window during the design of routes is another
research issue.

4 Proposed approach

In this section, we will describe motivation and problem formulation of VRPTW.

Further, the algorithm of VRPTW is presented.
 Mixed integer programming for vehicle routing problem with time windows 9

4.1 Motivation
After the thorough investigation of VRPTW, it has been observed that the high
transportation cost and long scheduling time are the main two areas of concern. The time
window constraint makes the VRP more applicable in real instances. Moreover, it has
been noticed that there is scope to design VRPTW problem as a bi-objective problem that
minimise both the transportation cost and used number of vehicles.

4.2 Problem statement

The problem is to determine the set of routes that minimise transportation cost and
number of vehicles used while satisfying the following constraints:
• the load of a vehicle cannot increase its predefined capacity
• vehicle will not service the customer before the earliest time and not later than the
latest time
• number of vehicle should be minimised
• each vehicle should visit customer only once.
The novel mathematical formulation of MIP proposed to satisfy service time constraint.
Here, the main aim of proposed approach is to minimise the total transportation cost and
number of vehicles used.

4.3 Problem formulation of VRPTW

From a graphical point of view, VRPTW is defined as a graph (C′, E) which is to be
traversed by a fleet of vehicles V. Here, the vehicles are considered homogeneous,
i.e., every vehicle has same capacity.
Nomenclature indices
i, j (i, j) ∈ E
i customer i
Model parameters
C Set of all customers (1, 2, …, n)
C′ Set of all nodes C ′ = C ∪ 0 ∪ n + 1
Q Vehicle capacity
V Number of vehicles
cij Cost of travel from customer i to customer j
tij Travel time from customer i to customer j
ri Demand to service customer i
[ei, li] Time window requested by customer i, where ei indicates the earliest service
starting time and li indicates the latest service starting time
10 D. Aggarwal and V. Kumar

Decision variables
The model includes two types of decision variables X and S. The decision variable Xijk is
defined as follows:
⎧1, if edge from customer i to customer j is traversed by vehicle k
X ijk = ⎨
⎩0, otherwise
The another decision variable Sik represents the time at which vehicle k service
customer i.
Let us assume that S0k = 0 and S(n+1)k denote the time at which vehicles arrives at
returning depot. The objective of model is to design a set of minimal cost routes, one for
each vehicle, such that all customers are serviced exactly once while minimising the
number of vehicles. Therefore, the objective function is considered as bi-objective.
The split deliveries are not allowed. The routes must be feasible with respect to the
capacity of vehicles and time windows of the customers serviced.
The mathematical model of VRPTW is given below:
Minimise ∑ ∑
k ∈C ′ ( i , j )∈E
cij X ijk (1)

Subject to

∑∑ X
k ∈V j∈C ′
ijk = 1, ∀i ∈ C (2)

∑r ∑ X
i∈C
i
j∈C ′
ijk ≤ Q, ∀k ∈ V (3)

∑X
j∈C ′
ojk = 1, ∀k ∈ V (4)

∑X
i∈C ′
ipk − ∑ X hjk = 0, ∀p ∈ C , ∀k ∈ V
j∈C ′
(5)

∑X
i∈C ′
i , n +1, k = 1, ∀k ∈ V (6)

X ijk ( Sik + tij − S jk ) ≤ 0, ∀(i, j ) ∈ E , ∀k ∈ V (7)

ei ≤ Sik ≤ li , ∀i ∈ C ′, ∀k ∈ V (8)

X ijk ∈ {0,1} ∀(i, j ) ∈ E , ∀k ∈ V (9)

The objective function (1) states that the total cost should be minimised. Constraint (2)
suggests that only one vehicle should service one node while constraint (3) ensures that
the capacity constraint is fulfilled. Here, no vehicle can serve the customer demands
beyond its capacity permits. Constraints (4)–(6) fulfil the flow constraint for the path
traversed by vehicle k. Constraint (7) makes sure that vehicle k, k ∈ V should not arrive at
customer j before Sik + tij if travelling from node i to j. Constraint (8) checks for the
satisfaction of time constraint. An unused vehicle is represented by an empty route from
node 0 to n + 1.
 Mixed integer programming for vehicle routing problem with time windows 11

A novel constraint for minimisation of vehicles used is defined below:

∑∑x
k ∈V j∈C ′
0 jk ≤ V , ∀k ∈ V , ∀j ∈ C ′ (10)

Note that constraint (7) may results in non-convex optimisation because it involves
quadratic terms. These terms are rewritten in linear form which is given below.
Sik + tij − M ij (1 − X ijk ) ≤ S jk ∀i, j ∈ C ′, ∀k ∈ V (11)

Here, M ij is described as a large constant that can be replaced by max(li + tij − ei , 0) .

4.4 Constraint workflow

Figure 3 shows the constraint workflow of VRPTW. The constraint workflow consists of
five main stages. The first stage includes the integrity constraint of given problem. It is
satisfied that each customer should be served exactly once by each vehicle. In the second
stage, the capacity constraint of vehicle is maintained. It ensures that the vehicle should
deliver the goods in their predefined capacity. The third stage ensures that each vehicle
should have an edge between an intermediate node and the start node. The fourth stage is
responsible for time window constraint. It ensures that no vehicle is allowed to serve
before the earliest time and later than the latest time for a particular customer. The fifth
stage is responsible for minimising the vehicle used.

Figure 3 Constraint workflow of VRPTW

12 D. Aggarwal and V. Kumar

4.4.1 Algorithm
Step 1. a) Obtain the initial linear relaxation of the mixed integer programming
formulation.
b) Construct the initial feasible solution and designate this solution as upper
bound.
c) Add the root node to tree and set the root node as current node.
Step 2. If termination criterion is satisfied, then
Return the upper bound and Exit
Else
Select a node that has the smallest objective function value
Endif
Step 3. Solve the corresponding linear relaxation and obtain the optimal solution
Step 4. If current node is infeasible or worse than upper bound, then
Remove the current node from the tree and goto Step 2.
Endif
Step 5. If Optimal solution is integer, then
Update the upper bound and remove it from tree and goto Step 2.
Endif
Step 6. If Optimal solution is not integer, then
Construct a feasible solution from existing solution and update the upper bound.
Endif

5 Experimental results and analysis

In this section, the experimentation is designed to evaluate the performance of proposed

approach on two well-known benchmark instances of Solomon (1987). The algorithm
was coded in IBM CPLEX and implemented on a Core i7 Intel Processor under
Windows 7 with 1 GB RAM.

5.1 Solving tool and test instances

IBM CPLEX is one of the tools widely used to solve combinatorial optimisation
problems. It has a concert technology that provides interfaces to C++, C# and
Java languages. It is accessible through independent modelling systems such
as a mathematical programming language (AMPL) and TOMLAB. It is also recognised
as a constraint solving toolkit suitable for solving optimisation models. It uses inbuilt
procedures to solve the MIP in short time. It can be used to solve a variety of different
optimisation problems in a variety of computing environments. IBM CPLEX is an exact
solver that uses MIP to search the desired solutions.
 Mixed integer programming for vehicle routing problem with time windows 13

The proposed approach is tested on two test instances of VRPTW. The used test
instances are taken from Solomon’s VRPTW instances (Solomon, 1983). The test
instances are R101 and C101. The first test instance namely R101 that belongs to class
with randomly generated customers. This class is solvable up to 33 customer nodes. The
second test instance namely C101 that belongs to class having clustered customer
locations. This class is solvable up to 60 customer nodes. The objective of proposed
approach is to minimise the number of vehicles and carrying capacity of each vehicle
should be 200 units.

5.2 Performance evaluation

Table 1 shows the performance evaluation of proposed approach on test instances R101.
The six different cases are generated from R101 instance. The number of customers is
varying from 5 to 33. The computed distance, used number of vehicles and computational
time are taken for performance comparison. As seen from Table 1, the computational
time increases with increase in number of customer node. The optimum distance and
vehicle used is also increase with increase in customer nodes.
The performance evaluation on test instances C101 is presented in Table 2. The seven
different cases are generated for C101. The number of customer nodes is varying
from 5 to 60. It has been observed from Table 2 that the number of vehicle used is one for
number of customer nodes 5 and 10. For number of customer nodes 20 and 25, the
optimum vehicles used are 3. The number of vehicle used, objective function value and
computation time increase with increase in number of customer nodes.

Table 1 Performance evaluation of proposed approach on test instance R101

Number of Objective Minimised number Average time

Instance customer distance of vehicles (in seconds)
R101 5 156.35 2 0.76
R101 10 269.54 4 1.92
R101 15 383.83 3 1.99
R101 20 511.26 6 2.45
R101 25 618.34 6 4.59
R101 33 713.58 11 11.59

Table 2 Performance evaluation of proposed approach on test instance C101

Number of Objective Minimised number Average time

Instance customer distance of vehicles (in seconds)
C101 5 42.42 1 0.69
C101 10 58.34 1 1.20
C101 15 142.17 2 2.16
C101 20 175.39 3 2.89
C101 25 191.83 3 5.22
C101 50 363.28 5 19.49
C101 60 470.13 6 36.92
14 D. Aggarwal and V. Kumar

5.3 Convergence analysis

Figures 4 and 5 show the convergence curves obtained from proposed model using R101
and C101 test instances, respectively. In these figures, the yellow point indicates a node
where an integer value has been found. The green line represents the evolution of best
integer value computed. The red line gives a bound on the final solution. These figures
are drawn between objective function value and time.
It has been observed from figures that the proposed model explore the promising
regions for solutions. Thereafter, it converges towards the best solution. Figure 4(a)
and (b) shows the convergence curve for 10 and 25 customer nodes. As seen from
Figure 4(a), the slope of convergence curve is varying throughout the process whereas the
slope of convergence curve is same throughout the process for 25 customer nodes.
From Figure 5(a) and (b), it has also been observed that the slope of convergence curve is
same throughout the process for both 10 and 25 customer nodes.

Figure 4 Convergence curve obtained from CPLEX (a) 10 customer nodes (b) 25 customer nodes
for test instance R101 (see online version for colours)
 Mixed integer programming for vehicle routing problem with time windows 15

Figure 5 Convergence curve obtained from CPLEX (a) 10 customer nodes (b) 25 customer nodes
for test instance C101 (see online version for colours)

Therefore, mentioned results depict that the proposed model provides better convergence
towards the optimal solution. The performance of proposed model is least affected for
above-mentioned customer nodes.

5.4 Sensitivity analysis

The effect of number of customer has been investigated on computational time and
vehicle used is studied. Figures 6 and 7 show the performance of proposed approach over
R101 and C101 instance, respectively. For R101 test instance, the computational time
greatly increase with increase in number of customers. The number of vehicle used also
16 D. Aggarwal and V. Kumar

increases with increase in the value of customer nodes. The result reveals that the number
of vehicles used is six for both 20 and 25 customer nodes.

Figure 6 Comparison between customer nodes and vehicles used with average time on secondary
axis for R101 instance (see online version for colours)

Figure 7 Comparison between customer nodes and vehicles used with average time on secondary
axis for C101 instance (see online version for colours)

For C101 test instances, the performance of proposed approach is greatly reduced with
increase in number of customers. The results obtained from Figure 6 depict that
the number of vehicles used and average computation time increase with increase in
 Mixed integer programming for vehicle routing problem with time windows 17

customer nodes. The optimum number of vehicle used is one for 5 and 10 customer
nodes. It can also be observed that the minimum number of vehicle used is three for 20
and 25 customer nodes.

6 Conclusion and future scope

This paper presents an approach for solving VRPTW using MIP. The proposed approach
has been used to minimise transportation cost and number of vehicles used
simultaneously. A new constraint has been designed to optimise the number of vehicles
used. The proposed approach has been tested on two well-known test instances
of Solomon’s benchmark problem. The computational results reveal that both cost and
number of vehicles are optimised at reasonable time. The sensitivity analysis for
customer nodes has been investigated on proposed approach, where it was noticed that
the computational time and number of vehicle used increase with increase in number of
customer node.
This work opens several research directions for future studies. First, metaheuristic
techniques can be applied to solve this problem. Second, metaheuristic can be integrated
with MIP to improving their performance. Third, investigation of different constraint for
VRPTW under different environment would be a valuable contribution. Also, it is worth
to investigate and find the best constraint satisfaction technique among the available
techniques.

Acknowledgements

The work on this paper was partially supported by Thapar University, Patiala under
the seed research grant number TU/DORSP/57/2135. This support is gratefully
acknowledged.

References
Beatrice, O., Brian, J.R. and Franklin, H. (2006) ‘Multi-objective genetic algorithms for vehicle
routing problem with time windows’, Applied Intelligence, Vol. 24, No. 1, pp.17–30.
Campbell, A., Clarke, L. and Savelsbergh, M. (2001) ‘Inventory routing in practice’, in Toth, P.
and Vigo, D. (Eds.): The Vehicle Routing Problem, SIAM, USA, pp.309–330.
Cetinkaya, C., Karaoglan, I. and Gokcen, H. (2013) ‘Two-stage vehicle routing problem with arc
time windows: a mixed integer programming formulation and a heuristic approach’, European
Journal of Operational Research, Vol. 230, pp.539–550.
Dantzig, G.B. and Ramser, J.H. (1959) ‘The truck dispatching problem’, Management Science,
Vol. 6, No. 1, pp.80–91.
De Backer, B. and Furnon, V. (1997) ‘Meta-heuristics in constraint programming experiments with
tabu search on the vehicle routing problem’, Second International Conference on
Metaheuristics (MIC’97), Sophia Antipolis, France, pp.1–14.
Desrochers, M., Lenstra, J.K., Savelsbergh, M.W.P. and Soumis, F. (1988) ‘Vehicle routing with
time windows: optimization and approximation’, in Golden, B. and Assad, A. (Eds.): Vehicle
Routing: Methods and Studies, Elsevier Science Publishers, Holland, pp.65–84.
18 D. Aggarwal and V. Kumar

Dondo, R. and Cerda, J. (2007) ‘A cluster-based optimization approach for the multi-depot
heterogeneous fleet vehicle routing problem with time windows’, European Journal of
Operational Research, Vol. 176, pp.1478–1507.
Dondo, R., Mendez, C.A. and Cerda, J. (2003) ‘An optimal approach to the multiple-depot
heterogeneous vehicle routing problem with time window and capacity constraint’, INTEC
(UNL-CONICET) Guemes, Santa Fe, Argentina, pp.3450–3000.
El-Sherbeny, N. (2001) Resolution of a Vehicle Routing Problem with Multiobjective Simulated
Annealing Method, PhD Dissertation, Faculte Polytechnique de Mons, Belgique.
Golden, B.L. and Assad, A.A. (1986) ‘Perspectives on vehicle routing: exciting new
developments’, Operations Research, Vol. 34, pp.803–809.
Golden, B.L. and Assad, A.A. (1988) Vehicle Routing: Methods and Studies, Elsevier Science
Publishers, New York, USA.
Golden, B.L. and Wasil, E.A. (1987) ‘Computerized vehicle routing in the soft drink industry’,
Operations Research, Vol. 35, pp.6–17.
Golden, B.L., Assad, A.A. and Wasil, E.A. (2001) ‘Routing vehicles in the real world: applications
in the solid waste, beverage, food, dairy and newspaper industries’, in Toth, P. and Vigo, D.
(Eds.): The Vehicle Routing Problem, SIAM, Philadelphia, pp.245–286.
Halse, K. (1992) Modeling and Solving Complex Vehicle Routing Problems, PhD Thesis, Institute
of Mathematical Modelling, Technical University of Denmark, Lyngby, Denmark.
Hokama, P., Miyazawa, F.K. and Xavier, E.C. (2016) ‘A branch-and-cut approach for the vehicle
routing problem with loading constraints’, Expert Systems With Applications, Vol. 47,
pp.1–13.
Holland, J.H. (1975) Adaptation in Natural and Artificial Systems, University of Michigan Press,
Ann Arbor, MI.
Knight, K. and Hofer, J. (1968) ‘Vehicle scheduling with timed and connected calls: a case study’,
Operational Research Quarterly, Vol. 19, pp.299–310.
Kolen, A.W.J., Rinnooy Kan, A.H.G. and Trienkens, H.W.J.M. (1987) ‘Vehicle routing and
scheduling with time windows’, Operations Research, Vol. 35, No. 2, pp.266–273.
Lambert, V., Laporte, G. and Louveaux, F. (1993) ‘Designing collection routes through bank
branches’, Computer’s and Operations Research, Vol. 20, pp.783–791.
Madsen, O.B.G. (1976) ‘Optimal scheduling of trucks – a routing problem with tight due times for
delivery’, in Strobel, H., Genser, R. and Etschmaier, M. (Eds.): Optimization Applied to
Transportation Systems, IIASA, International Institute for Applied System Analysis,
Laxenburgh, pp.126–136.
Mechti, R., Poujade, S., Roucairol, C. and Lemari, B. (2001) Global and Local Moves in Tabu
Search: A Real-Life Mail Collecting Application, Manuscript, PriSM Laboratory, University
of Versailles, France.
Ozbaygin, G., Karasan, O.E., Savelsbergh, M. and Yaman, H. (2017) ‘A branch-and-price
algorithm for the vehicle routing problem with roaming delivery locations’, Transposition
Research Part B, Vol. 100, pp.115–137.
Pullen, H. and Webb, M. (1967) ‘A computer application to a transport scheduling problem’,
Computer Journal, Vol. 10, pp.10–13.
Russell, R.A (1995) ‘Hybrid heuristics for the vehicle routing problem with time windows’,
Transportation Science, Vol. 29, pp.156–166.
Sabar, N.R., Zhang, X.J. and Song, A. (2015) ‘A math-hyper-heuristic approach for large-scale
vehicle routing problems with time windows’, Evolutionary Computation (CEC), IEEE
Congress, Sendai, Japan, pp.830–837.
Sam, R.T. (1995) Vehicle Routing with Time Windows using Genetic Algorithms, submitted to the
book on Application Handbook of Genetic Algorithms, New Frontiers, pp.253–277.
 Mixed integer programming for vehicle routing problem with time windows 19

Simbolon, H. and Mawengkang, H. (2014) ‘Mixed integer programming model for the
vehicle routing problem with time windows considering avoiding route’, International
Journal of Advanced Research in Computer Engineering & Technology (IJARCET), Vol. 3,
pp.2741–2744.
Solomon, M.M. and Desrosiers, J. (1988) ‘Time window constrained routing and scheduling
problems’, Transportation Science, Vol. 22, pp.1–13.
Solomon, M.M. (1983) Vehicle Routing and Scheduling with time Windows Constraints: Models
and Algorithms, PhD Dissertation, Department of Decision Sciences, University of
Pennsylvania.
Solomon, M.M. (1987) ‘Algorithms for the vehicle routing and scheduling problems with time
window constraints’, Operations Research, Vol. 35, pp.254–265.
Suwansuksamran, S. and Ongkunaruk, P. (2013) ‘A mixed integer programming for a vehicle
routing problem with time windows: a case study of a Thai seasoning company’, Proceedings
of the 4th International Conference on Engineering, Project, and Production Management,
Bangkok, pp.1214–1221.
Toth, P. and Vigo, D. (2001) The Vehicle Routing Problem, Monographs on Discrete Mathematics
and Applications, SIAM, Philadelphia.