Multi-Objective Vehicle Routing Problem with Time Windows and

Fuel Consumption Minimizing

Seyed Farid Ghannadpour1 and Mohsen Hooshfar2

1Department of Railway Engineering, MAPNA Co., Tehran, Iran
2Department of Railway Engineering, MAPNA Co., Tehran, Iran

Keywords: Vehicle Routing Problem, Fuel Consumption, Customers' Priority, Multi-Objective, Evolutionary
Algorithm.

Abstract: Transportation often represents the most important single element in logistics costs and its reduction and
finding the best routes that a vehicle should follow through a network is an important decision. the energy
cost is a significant part of total transportation cost and it is important to improve the operational efficiency
by decreasing energy consumption. Unlike most of the studies trying to minimize the cost by minimizing
overall travelling distance, the energy minimizing which meets the latest requirements of green logistics, is
considered in this paper. the customers' priority for servicing is considered as well. Besides, the model is
interpreted as multi-objective optimization where, the energy consumed and the total fleet are minimized
and the total satisfaction rates of customers is maximized. A new solution based on the evolutionary
algorithm is proposed and its performance is compared with the CPLEX Solver. Results illustrate the
efficiency and effectiveness of proposed approach.

1 INTRODUCTION consumption in waste collection. Moreover, Bektaş

and Laporte (2011) studied the pollution-routing
Transportation often represents the most important problem (PRP) that in which the amount of pollution
single element in logistics costs and to its reduction emitted by a vehicle is considered in depth.
finding the best routes is an important decision Minimizing the fuel consumption in VRPs is also
problem. One of the most important and widely considered by Gaur & Mudgal (2013) and Kara et al.
studied combinatorial optimization problems in this (2007) with a new cost function and based on the
area is the vehicle routing problem with time results, the fuel consumption could be reduced by
windows (VRPTW). The literature of the VRPTW, 5% on average. In this regards, Zhang et al. (2014)
due to its inherent complexities and usefulness in introduced an environmental vehicle routing
real life is rich in different models and solution problem (EVRP) with the aim of reducing the
approaches (Chiang & Hsu 2014, Blaseiro et al. adverse effect on the environment and by using a
2011, Dhahri et al. 2014, Ghannadpour et al. 2014, hybrid artificial bee colony algorithm.
Lin 2011, Mavrovouniotis & Yang 2015, Tan et al. Besides, the proposed model in this paper is
2006 and Feng & Liao 2014). interpreted as multi-objective optimization problem.
Although there are different forms of VRPTWs, In real-life, for instance, there may be several costs
most of them minimize the cost by minimizing the associated with a single tour. For this reason,
overall traveling distance or the traveling time. In adopting a multi-objective point of view can be
fact, it is the amount of fuel or energy consumed, not advantageous by determining the trade-offs between
the traveled distance that is the greater concern to the objectives. In the multi-objective area, Tan et al.
transportation companies and meet the latest (2006) and Ombuki et al. (2006) proposed a hybrid
requirements of green logistics. Statistics show that multi-objective evolutionary algorithm (MOEA) for
energy cost is a significant part of total solving the multi-objective VRPTW. Tan et al.
transportation cost (Xiao et al. 2012). in this regard, (2007) proposed a similar approach for VRP with
Tavares et al. (2008) took into account the effect of stochastic demand. Ghannadpour et al. (2014) and
both road inclination and vehicle load on energy Ghannadpour & Hooshfar (2015) solved Dynamic

92
Ghannadpour, S. and Hooshfar, M.
Multi-Objective Vehicle Routing Problem with Time Windows and Fuel Consumption Minimizing.
DOI: 10.5220/0005657900920099
In Proceedings of 5th the International Conference on Operations Research and Enterprise Systems (ICORES 2016), pages 92-99
ISBN: 978-989-758-171-7
Copyright c 2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
 Multi-Objective Vehicle Routing Problem with Time Windows and Fuel Consumption Minimizing

VRPTW as a multi-objective problem by GA. Other It should be noted that this cost function is
similar approach could be found in (Sivaram Kumar mainly focused on energy consumption and it can be
et al. 2014, Garcia-Najera & Bullinaria 2011 and calculated based on the work done by a vehicle over
Garcia-Najera et al. 2015). The remainder of this a route (arc) of network. It is assumed the movement
paper is organized as follows. Section 2 defines the of vehicles is considered as an impending motion
model description. The structure of the solution where the force causing the movement is equal to
technique is discussed in Section 3. Section 4 the friction force. So, a new objective function to
describes the computational experiments carried out minimize the work done by vehicles or the energy
to investigate the performance of the proposed used (equivalent to fuel consumed by vehicles) is
method, and finally Section 5 provides the obtained and should be considered instead of
classical cost function as follows:
concluding remarks.
+
∑ ∑ , ∑ ́ × ×
(2)
2 MODEL DESCRIPTION × ×

The problem considered here is energy minimizing Where is the acceleration of gravity (9.81 /
vehicle routing problem with time windows ) and ́ is the coefficient of friction on link ( , ).
(VRPTW) as a multi objective optimization. Moreover, is the load of vehicle upon leaving
VRPTW is given by a special node called depot, a customer as follows: (∀ ∈ \{0})
set of customer = {0, 1, 2, … , } to be visited and ∑ ∑ + × = (3)
,
a directed network connecting the depot and the
customers. Also a set of fleet = {1, 2, … , } These new constraints and objective function are
located at the depot is available. Each vehicle has a non-linear and should be approximated to liner
limited capacity ( ) and each customer has a equation. For this purpose a new variable is
varying demand ( ). A distance and travel time defined instead of which means the load of
are associated with each arc of the network. On vehicle when moves from customer to customer
the other hand, any customer i must be serviced . The linear formulation is described later.
within a pre-defined time interval [ , ]. Each The concept of customers' satisfaction proposed
in our recent research (Ghannadpour & Hooshfar
vehicle k is also supposed to complete its individual
2015) is also considered and developed here for
route within the total route time ( ). The objective
different kinds of customers. In this paper the
of the classical VRPTW is to serve all the customers preference information of customers is represented
such that the total distance traveled by the vehicles is as a fuzzy time windows as Fig.1. In this approach,
minimized. But this paper, unlike most of the work every customers can be assigned by the expert to one
those minimize the cost by minimizing overall of groups (e.g., important customers ( ), casual
traveling distance, tries to minimize the real cost of a ( ) and etc.) where ∪ = \{0}.
vehicle traveling along a route. It has been
recognized that the real cost of a vehicle in a
network depends on many factors like load of
vehicles, fuel consumption per mile, time spent or
distance traveled up to visit a node, depreciation of
vehicles, maintenance, driver costs and etc.
Although energy consumption is largely determined
by distance, other factors such as load also have a
considerable impact on fuel costs. So, if the other
factors are kept constant, the energy consumption
then mainly depends on distance and load.
The classical cost function of VRPTW is as
equation (1) and it should be modified as
( , ) where load
is the weight of the vehicle (tare plus the load of the
vehicle) over each link ( , ).
Figure 1: Conventional and fuzzy time window for each
(1) customer.
∑ ∑ , ∑

93
ICORES 2016 - 5th International Conference on Operations Research and Enterprise Systems

According to Fig. 1, the classical time window is

changed to the triple [ , , ] and [ ́ , , ] for ( − ( + )) ∗ +
important and casual customers. ( ) is the (( + ) − ) ∗ ∀ ∈ \{0} (15)
membership function of customer i and shows the (1 − ) < 0
grade of satisfaction when the start of service time is
ti. The start time of service for each customer i is as
= + where and are arrival and ≤( + )≤ ∀ ∈ (16)
waiting time at customer i. Therefore a new
objective function should be considered as − ≤( + )≤ + ∀ ∈ (17)
∑ ∈ \{ } × ( ) where, is the
importance degree of customer i. ∑ , ∑ −
The mathematical formulation of the proposed ∀ ∈ \{0} (18)
∑ , ∑ =
model is as follows:

=∑ ∑ ∑ ( × ∀ ∈
,
(4) ≤ × ,∀ ≠ ∈ (19)
+ )×
,∀ ∈
=∑ ∑ (5)
∀ , ∈ ,
=∑ × ( ) (6) ∈ {0,1}, ≥0
∀ ∈
S.t:
Formulas (4-6) are the objective functions
∑ ∑ ≤ ∀ =0 (7) Formula (4-5) minimize total energy consumed and
the total number of vehicles and formula (6)
∑ , = ∀ ∈ , maximizes the total satisfaction rates of customers.
(8) Constraint (8) secures maximum size of fleet.
∑ , ≤1 ∀ ∈
Constraints (8) and (9) define that every customer
∀ ∈ node is visited only once by one vehicle. Constraint
∑ ∑ , =1 (9) (11) is the maximum travel time constraint.
\ {0}
Constraints (12-17) define the arrival time, and the
time windows for different kinds of customers.
= = = ( )= =0 (10) Constraints (13-15) compute the satisfaction level of
each customer Constraints (13-15) are non-linear
∀ ∈ /{0}, and they have relaxed to linear constraints.
+ + + − (1 − Constraint (18) indicates the load of vehicle after it
∀ ∈ , (11)
) ≤ visits a customer. Constraint (19) limits the maximal
=0
load carried by the vehicle and force to zero
∀ ∈ \{0}, when = 0.
+ + + − (1 −
∀ ≠ ∈ , (12)
) ≤
∀ ∈
3 SOLUTION METHOD
( )
( )= ∗
∀ ∈ (13) This section designs an efficient evolutionary
( )
(1 − )+ ∗ method for tackling the proposed model that in
which objectives are met and the constraints are
satisfied. The proposed model is based on the
( ) ( ) conventional VRPTW which is NP-hard and should
( )= ∗ be tackled by heuristics. The evolutionary
( )
( ) ( ) ∀ ∈ (14) algorithms like GA have many advantages in finding
(1 − ) + ∗
( ) an easy way of the solution representation and in
implementation for multi objective models and
ability of incorporation with the different operators
that improve the solutions.

94
 Multi-Objective Vehicle Routing Problem with Time Windows and Fuel Consumption Minimizing

3.1 Representation 3.6 Local Search

In this method each chromosome which is a solution The local search (LS) is employed as mutation to the
to the problem, is represented by an integer string of child chromosome with a probability . This
length . This string of customer identifiers paper uses a -interchange mechanism as local
represents the sequence of deliveries that must be search method that moves customers between routes
covered by vehicles during their routes. to generate neighborhood solution for the proposed.
Given a feasible solution for the model represented
3.2 Pareto Ranking Procedure by = { , … , , … , , … , } where is a set
of customer served by vehicle route . A -
The Pareto ranking procedure (Ghannadpour & interchange between a pair of routes and is a
Hooshfar 2015) which tries to rank the solutions to replacement of subset ⊆ of size | | ≤ by
find the non-dominated solutions is used for another subset ⊆ of size | | ≤ , to get the
evaluation of each chromosome. In this approach,
new route sets , and a new neighbouring
chromosomes assigned rank 1 are non-dominated,
and inductively, those of rank i +1 are dominated by solution = { , … , , … , , … , } where =
all chromosomes of ranks 1 through i. ( − )∪ and = ( − ) ∪ . The
neighbouring ( ) of a given solution is the set
3.3 Population & Initialization of all neighbors { } generated by the λ-interchange
method for a given λ. In one version of the algorithm
In this paper the method of PFIH (originally called GB (global best), the whole neighborhood is
proposed by Solomon (1987)) is used to create the explored and the best move with lower rank is
first chromosome. PFIH method defines the relation selected. In another version, FB (first best), the first
of = + + (( /360) ) to find the admissible improving move is selected if exists;
first customer in each new route where; is the otherwise the best admissible move is implemented.
distance from customer to the central depot; is In this paper 1-interchange (FB) or 2-interchange
the latest time and is the polar coordinate angle of (GB) is employed to the child chromosome with the
the customer . Once the first customer is selected special probability.
for the current route, the heuristic selects from the
set of unrouted customers the one customer which
minimizes the total insertion cost between every 4 COMPUTATIONAL ANALYSIS
edge in the current route without violating the time
and capacity constraints. In this section, since there is no any prior work on
the proposed model, a set of complete randomly
3.4 Selection generated instances with different size (N) is
considered as numerical examples. In the first step,
This paper uses a standard k-tournament selection the validity of new mathematical formulation for
where a tournament set of size k is randomly drawn small and medium instances are implemented by
from the population and the chromosome with a CPLEX Solver separately (with a time limit of 2
lower rank is selected and will then be recombined hours) and the results are analyzed. Finally, the
via the recombination operators to create potential quality of proposed evolutionary method is
new population. evaluated. In this step the instances with larger size
are considered and the results obtained by the
3.5 Recombination proposed method and CPLEX Solver are analyzed.
This paper uses the modified best cost-best rout 4.1 Mathematical Modelling
crossover (BCBRC), which selects a best route from
each parent and then for a given parent, the Table 1 presents a summary of results obtained by
customers in the chosen route from the opposite CPLEX Solver when the single objective energy
parent are removed. The final step is to locate the minimizing VRPTW is considered. The column
best possible locations for the removed customers in labeled “with classical cost function” gives the
the corresponding children. findings of VRPTW when it tries to minimize the
total distance travelled by vehicles (distance
oriented); column “with new cost function” gives the

95
ICORES 2016 - 5th International Conference on Operations Research and Enterprise Systems

findings of model when it tries to minimize the total 4.2 Analysis of Proposed Method
energy consumption (fuel oriented). For each
instance, the vehicles’ total traveling distance In this section, the quality of proposed evolutionary
(indicated by Dis.) and the related fuel consumption method is evaluated. In this step the instances with
(indicated by Related FC) are calculated when the larger size are considered and the results obtained by
distance-oriented model is implemented. Moreover, the proposed method and CPLEX Solver are
the fuel consumption (FC) and the related traveling analyzed. The results of Mathematical Model are
distance (Related Dis.) are also obtained by fuel- found by using the weighting method as follows:
oriented model. The times marked with an asterisk
show the time limit of 2 hours for the CPLEX Solver
× + × − (20)
and the solver is interrupted after this time. For some
instances there is no integer solution up to this time ×
limit.
It can be observed from Table 1 that for the Where, is the weight of objective function
small/medium – scale instances, the FC obtained by estimated by DM and∑ = 1 and the objective
fuel oriented model is on average 5.6% lower than functions are calculated according to relations (4-
the obtained by distance oriented model but with a 6). The proposed heuristic is coded and run on a PC
10.6% increase in distance traveled. In other words, with Core 2 Duo CPU (3.00 GHz) and 2.9 GB of
by 10.6% increase in distance traveled, the fuel cost RAM. Moreover, the model is implemented under
which is a significant part of total transportation cost parameters of Population size = 30 - 100, Generation
can be reduced by 5.6%. It should be noted that the number = 500-1000, Crossover rate = 0.80, Mutation
choice of any solutions (fuel & distance oriented) rate = 0.40, Selection rate of improvement operators
depends on the DM’s preference. = 0.5. It must be mentioned that the population size
and the generation number is adopted with the
Table 1: VRPTW with fuel consumption by CPLEX problem size.
Solver. It should be noted that the Repetition of
experiments is 10 runs. Table 2 presents the average
With classical cost function and best values (among the non-dominated
Instance N solutions) of proposed method over 10 runs and to
CPU t.
Dis. Related FC. the finding of CPLEX Solver.
(Sec.)
1 4 115.3760 0.2030 2847.909
2 5 140.1070 0.2180 2427.428 Table 2: Average and best results over 10 experiments.
3 10 226.6523 2.8750 4392.852
4 12 303.2485 13.359 6106.863 h − ave h − best
N
5 15 321.6250 37.765 9817.879 FC. K Sat. FC K Sat
6 20 497.100 7200* 14827.87 10 4424.69 6.0 25.20 4382.37 6 26.00
7 30 -------- 7200* -------- 15 9845.04 6.6 33.20 9308.00 6 36.00
8 40 -------- 7200* -------- 20 14520.1 8.1 49.50 14345.3 8 54.00
30 19150.3 12 78.40 18009.1 12 79.00
Ave. 221.4018 5118.586
40 29308.4 16.8 103.8 25542.8 16 105.0
With new cost function 70 59805.0 15 117.0 50231.1 15 120.0
Instance N CPU t.
Dis. Related FC. 100 83063.7 18.6 260.5 75654.6 18 270.0
(Sec.) Deviation (%)
1 4 2438.131 0.0541 138.152 N
D D D D D D
2 5 2393.459 0.0620 156.523
3 10 4382.368 2.0150 228.777 10 0.960 0.00 3.08 0.0 0 0
4 12 5933.309 17.357 340.049 15 5.450 9.09 7.78 0.9 0 0
5 15 9220.890 69.531 376.166 20 1.200 1.23 8.33 -1.6 0 0
6 20 -------- 7200* -------- 30 5.960 0.00 0.76 --- --- ---
7 30 -------- 7200* -------- 40 12.85 4.76 1.14 --- --- ---
8 40 -------- 7200* -------- 70 16.01 0.00 2.50 --- --- ---
100 8.920 3.23 3.52 --- --- ---
Ave. 4873.631 247.9334
Ave.

7.340 2.62 3.87 -0.2 0 0

Dev. (%) FC dev. : -5.57 / Dis dev. : 10.64

96
 Multi-Objective Vehicle Routing Problem with Time Windows and Fuel Consumption Minimizing

The column labeled "ℎ − " gives the total travelled by vehicles may be increased or decreased
average findings of proposed heuristic over 10 runs and it is related to the geographical location and time
and it is divided into three columns where the each windows of customers [15]. So by increasing the
of them represents the average of each objective number of vehicles the load of vehicles is decreased
function (indicated by ., and ); column and when the distance cost of solution is changed in
"ℎ− " gives the best results of each objective the opposite direction, the total fuel consumed by
function obtained by proposed heuristic over 10 vehicles is decreased. On the other hand, in the
experiments (indicated by , and ). instance 3, although adding a vehicle provides a
Deviation between the average and best results of schedule with a lower load of vehicles for each
proposed heuristic are listed in the columns labeled route, the distance cost is much higher than that of
, and . Moreover, ( = 1,2,3) the basic model. Therefore the total fuel consumed
represents the deviation between the best value of by all fleets is increased.
objective function obtained by proposed heuristic
over 10 runs and the best value found by the CPLEX
Solver. It should be noted that the listed values of
deviations represents the amount of difference
between the best and average results of proposed
method over 10 experiments and could illustrate the
consistency and reliability of results. Moreover the
deviations between the best results of proposed
method and CPLEX Solver represents the quality of
obtained results and the negative value represent the
amount of improvements obtained by the proposed
approach.
According to this table we can see the results
obtained from proposed method are rather consistent
and the average deviations over 10 experiments are
lower than 8%. Moreover, the average difference
between the best values of proposed method and
CPLEX Solver illustrates the improvement of 0.2%
in the first objective for the first three instances and
for the others the CPLEX Solver cannot find any
solution in a reasonable amount of computational
time.
In general, the relationship between these
defined objectives is unknown until the problem is
solved in a proper multi-objective manner. These Figure 2: Population distribution of the 7th instance.
objectives may be positively correlated with each
other or they may be conflicting to each other.
According to the results, the customers' satisfaction 5 CONCLUSION
rate is improved as the total fuel consumed is
deteriorated. Moreover, the waiting time imposed on This paper presented a new model and solution for
vehicles is increased in these instances due to get the the multi-objective vehicle routing and scheduling
better satisfaction rate of customers. These problem with considering the fuel consumption rate.
behaviours for the 7th instance of Table 6 are Moreover, this paper considered the customers'
illustrated in Fig.2. priority according to customer-specific time
The different behaviour is observed for the total windows, which are highly relevant to the
fuel consumption and the required fleets. They are customers’ satisfaction level.
positively correlated with each other in some Besides, the proposed model was interpreted as
instances like instance #3 and they are conflicting to multi-objective optimization problem and a new
each other in others (like instance #2). By adding a solution based on the evolutionary algorithm was
vehicle to the schedule, the load of vehicles could be proposed. the performance on several completely
decreased along a route but the total distance

97
ICORES 2016 - 5th International Conference on Operations Research and Enterprise Systems

random generated instance problems was compared Blaseiro, S.R., Loiseau, I., Ramonet, J., 2011. An ant
with the CPLEX Solver. The results show the colony algorithm hybridized with insertion heuristics
efficiency and effectively of proposed method. for the time dependent vehicle routing problem with
It should be noted that the proposed model is time windows. Computers & Operations Research 38:
954–966.
very compatible with the constraints of reality and it Chiang, T.C., Hsu, W.H., 2014. A knowledge-based
is under implementation for locomotives routing and evolutionary algorithm for the multiobjective vehicle
assignment for railway transportation division of routing problem with time windows. Computers &
MAPNA Group. In this model the trains are Operations Research 45: 25-37.
considered as customers and they are made up at Dhahri, A., Zidi, K., Ghedira, K., 2014. Variable
different stations of network and they need to Neighborhood Search based Set Covering ILP Model
receive locomotive based on the time table of train for the Vehicle Routing Problem with Time Windows.
scheduling. Moreover, the locomotives are located at Procedia Computer Science 29: 844-854.
some central depots and they depart toward the Feng, H.M., Liao, K.L., 2014. Hybrid evolutionary fuzzy
learning scheme in the applications of traveling
trains to move them from their origins to their salesman problems. Information Science 270: 204-
destinations based on the train scheduling plan. One 225.
sample of train scheduling plan is illustrated in Garcia-Najera, A., Bullinaria, J.A., 2011. An improved
Fig.3. In this case, the trains with low priorities are multi-objective evolutionary algorithm for the vehicle
considered to be having the classical time windows. routing problem with time windows. Computers &
Moreover, the trains with highly priority have the Operations Research 38: 287–300.
fuzzy time windows and the desired time is nearest Garcia-Najera, A., Bullinaria, J.A., Gutiérrez-Andrade,
to the earliest dispatching time of each train. M.A., 2015. An evolutionary approach for multi-
objective vehicle routing problems with backhauls.
Computers & Industrial Engineering 81: 90-108.
Gaur, D.R., Mudgal, A., 2013. Singh RR. Routing
vehicles to minimize fuel consumption. Operations
Research Letters 41: 576-580.
Ghannadpour, S.F., Hooshfar, M., 2015. Multi-objective
evolutionary method for dynamic vehicle routing and
scheduling problem with customers' satisfaction level.
4st International Conference on Operations Research
and Enterprise Systems. SCITEPRESS.
Ghannadpour, S.F., Noori, S., Tavakkoli-Moghaddam, R.,
2014. A multi-objective vehicle routing and
Figure 3: Typical train scheduling plan. scheduling problem with uncertainty in customers’
request and priority. Journal of Combinatorial
Moreover, the detailed schedule of each locomotive Optimization 28: 414-446.
including the departure time, trains in its Kara, I., Kara, B.Y., Yetis, M.K., 2007. Energy
commitments, planned routes, waiting times, fuel Minimizing Vehicle Routing Problem. Lecture Notes
in Computer Science 4616: 62-71.
consumption cost and etc is corresponding to the
Lin, C.K.Y., 2011. A vehicle routing problem with pickup
routes found by the proposed VRPTW and they are and delivery time windows, and coordination of
identified for this route. transportable resources. Computers & Operations
Research 38: 1596-1609.
Mavrovouniotis, M., Yang, S., 2015. Ant algorithms with
immigrants schemes for the dynamic vehicle routing
ACKNOWLEDGEMENTS problem. Information Science 294: 456-477.
Ombuki, B., Ross, B., Hanshar, F., 2006. Multi-Objective
The authors would like to thank MAPNA Group for Genetic Algorithm for Vehicle Routing Problem with
its supports and financing this paper. Time Windows. Applied Intelligence 24: 17-30.
Sivaram Kumar, V., Thansekhar, M.R., Sarvanan, R.,
Miruna Joe Amali, S., 2014. Solving multi – objective
vehicle routing problem with time windows by FAGA.
REFERENCES Procedia Engineering 97: 2176-2185.
Solomon, M.M., 1987. Algorithms for the vehicle routing
Bektaş, T., Laporte, G., 2011. The pollution-routing and scheduling problems with time window
problem. Transportation Research Part B: constraints. Operations Research 35: 254–265.
methodological 45: 1232-1250. Tan, K.C. Cheong, C.Y., Goh, C.K., 2007. Solving
multiobjective vehicle routing problem with stochastic

98
 Multi-Objective Vehicle Routing Problem with Time Windows and Fuel Consumption Minimizing

demand via evolutionary computation. European

Journal of Operational Research 177: 813-139.
Tan, K.C., Chew, Y.H., Lee, L.H., 2006. A hybrid
multiobjective evolutionary algorithm for solving
vehicle routing problem with time windows.
Computational Optimization and Applications 34:
115-151.
Tan, L., Lin, F., Wang, H., 2015. Adaptive comprehensive
learning bacterial foraging optimization and its
application on vehicle routing problem with time
windows. Neurocomputing 151: 1208-1215.
Tavares, G., Zsigraiova, Z., Semiao, V., Carvalho, M.G.,
2008. A case study of fuel savings through
optimisation of MSW transportation routes.
Management of Environmental Quality: An
International Journal 19: 444 – 454.
Xiao, Y., Zhao, Q., Kaku, I., Xu, Y., 2012. Development
of fuel consumption optimization model for the
capacitated vehicle routing problem. Computers &
Operations Research 39: 1419-1431.
Zhang, S., Lee, C.K.M., Choy, K.L., Ho, W., Ip, W.H.,
2014. Design and development of a hybrid artificial
bee colony algorithm for the environmental vehicle
routing problem, Transportation Research Part D:
Transport and Enviroment 31: 85-99.

99