Int. J. Contemp. Math. Sciences, Vol. 5, 2010, no.

11, 497 - 507

Fuzzy Vehicle Routing Problem with

Uncertainty in Service Time

Radha Gupta1, Bijendra Singh2 and Dhaneshwar Pandey3

1
Sri Bhagawan Mahaveer College of Engineering, Bangalore, Karnataka, India
2
School of Studies in Mathematics, Vikram University, Ujjain, (M.P.), India
3
Department of Mathematics, Dayalbagh University, Agra, (U.P.), India
[email protected], [email protected]

Abstract

In this paper we have extended the fuzzy multi objective optimization problem
considered by Gupta et al [8] by incorporating another attribute of minimization of
number of vehicles. Reducing the number of vehicles produces substantial savings in
the global transportation costs though some times this goal is achieved with a
compromise of other goal. The problem is solved by enhancing the genetic algorithm
(FVRPTW) developed by Gupta et al [8] in MATLAB. Also the time taken to serve a
customer is considered to be dynamic since practically it is highly impossible to
predict the exact time taken to serve a customer.

Mathematics Subject Classification: 90B36, 90C29

Keywords: Fuzzy logic, Genetic Algorithm, Vehicle routing problems, Dynamic

service time
498 R. Gupta, B. Singh and D. Pandey

1. Introduction:
The Vehicle routing problem calls for the determination of the optimal set of routes to
be performed by a fleet of vehicles to serve a given set of customers, and it is one of
the most important, and studied, combinatorial optimization problems. Dantzig and
Ramser [3] introduced this problem in 1959 where they described a real-world
application concerning the delivery of gasoline to service station and proposed the
first mathematical programming formulation and algorithmic approach. In 1964,
Clarke and Wright [2] proposed an effective greedy heuristic that improved on the
Dantzig-Ramser approach. Following these two seminal papers, hundreds of models
and algorithms were proposed for the optimal and approximate solutions of the
different versions of VRP. The literature on the development of genetic algorithms
for solving the VRP is rather scant. There is an opportunity for the GA to provide
competitive results, given its relative robustness in the presence of complex
constraints. Some very effective implementations have been reported in the literature
for VRPTW (Potvin and Bengio [5], Thangiah[9]). The work done on the CVRP,
including its distance or time-constrained variant, was mostly aimed at evaluating the
impact of different parameters of a GA on the efficiency of the search. Breedam [1]
compares a GA with previously developed simulated annealing and tabu search
heuristics on different types of VRPs. A detailed literature survey on genetic
algorithm based VRP has been given by Gupta et al [7].

Recently application of fuzzy set to VRP to help find optimal (or near optimal)
solution is widely spreading. A fuzzy vehicle routing and scheduling problem with
five attributes was formulated by Lin [4] and they proposed to solve it with a pure
genetic algorithm method. They used the concept given by Gen and Cheng [6] to
replace the time windows by the concept of fuzzy due time since it can describe
customers preference better than fixed time window. We improved this model (refer
Gupta et al [8]) by also incorporating the time taken to serve each customer since it
affects the time window of the next customer. Here we developed a model with three
attributes; maximization of grade satisfaction, distance minimization, waiting time
minimization and the service time considered was deterministic (i.e. the time taken to
serve the customer was assumed based on experience). However, practically it is
difficult to assess the correct service time. Also it was felt that reducing the number
of vehicles produces substantial savings in the global transportation costs even
though some times this goal is achieved with a compromise of other goal. The
decision can be left to the decision maker’s priority. Therefore, in this paper we have
enhanced our model FVRPTW with four attributes, including fourth attribute to be
minimization of number of vehicles, and making the service time to be dynamic.
Fuzzy vehicle routing problem 499

2. Mathematical model of Fuzzy VRP

We briefly describe the model taken by us in the previous work [6] and then discuss
the improvement done in the model. The concept of time windows does not model
the customer’s preference very well. Even though customers are asked to provide a
fixed time window for service, they really hope to be served at a desired time if
possible. Such a desired time is called a fuzzy due time. Fuzzy vehicle routing
problem is formulated based on the concept of fuzzy due time, where the membership
function of fuzzy due time corresponds to the grade of satisfaction of a service time.
It is improved by adding the time taken to serve each customer dynamically thereby
affecting waiting time of the other customers. The objectives considered here are to
maximize the average grade of satisfaction over customers, minimize the number of
vehicles, the total travel distance and total waiting time for vehicles.

2.1 Grade of Satisfaction

The grade of satisfaction of service µi(t) can be defined for any service time (t>0) as
µi(t) = (1)
where (ei, li) represents the earliest and the latest start time of customer i and t is the
time at which the customer is served. In fuzzy set theory, a triangular fuzzy number
(TFN) with respect to the grade of satisfaction for service time can be defined by the
triplet (ei, ui, li). The membership function of the fuzzy due time of the customer i by
µi(ti) can be shown as equation (2). If a customer is served at his/her desired due
time, the grade of satisfaction for him/her is 1 (full satisfaction); otherwise, the grade
of satisfaction gradually decreases with the increase of difference between the service
time and desired due time. The grade of satisfaction will be 0 (no satisfaction) if the
service time falls outside the time interval.

= (2)

The overall degree of service satisfaction of vehicle k to all its service customers
(SC)k is calculated as

Maximize (SC)k = , (3)

500 R. Gupta, B. Singh and D. Pandey

Where µi(ti) is the degree of satisfaction for customer i

ti is the service time for customer i
nc is the number of customers served by the vehicle k

We can find out the maximal service satisfaction of vehicle k to all its service
customers through maximizing the value of (SC)k.

2.2 Waiting time

With the service satisfaction improved, the number of vehicles, total waiting time
occurring under only certain amount of vehicles available for service probably
increases at the same time. The relationship of the service time between two different
customers of a vehicle is described as follows. After customer i is served at the
service time ti the vehicle arrives and starts to serve customer j at tj. The time rij
denotes the required transport time from customer i to customer j and si denotes the
time taken to serve the customer i and is generated randomly. A waiting time exists
while the service time of customer j is earlier than the earliest time wanted. The
waiting time (WT)j can be given by the equation
(4)
Where
Once the waiting time is minimized, the most effective transportation solution can be
reached. With respect to a customer j, which is served by vehicle k and needs waiting
time, the minimum waiting time can be calculated by
Minimize (WT)j = (5)
Where wi(ti) is the waiting time for a vehicle at customer i
ti is the service time for customer i

2.3 Total distance traveled

Another effective way to reduce the cost of transportation is shortening of distance.
Through the ratio of each distance to the maximum calculated (DT)j, the average ratio
for all transportation distance to the maximum can be obtained as
Minimize (DT)j = (6)
Where dij is the distance of direct travel from customer i to j

2.4 Fleet Size

The dispatching strategy always mainly focuses on minimizing the number of
vehicles in use. A sacrifice of increasing the total transportation distance is always
made to achieve this goal but the tradeoff between fleet size and travel cost must be
considered. We apply genetic algorithm to search for the most appropriate
dispatching strategy to transport the commodities with minimum fleet size.
Fuzzy vehicle routing problem 501

n m
Minimize FS = ∑∑ x
j =1 k =1
0 jk (7)

Where

2.5 Objective Function

Fitness is calculated based on the original objective functions. The weighted sum of
objectives for a chromosome is given by:

⎡ (DT ) j ⎤ ⎡ (SC )kj ⎤ ⎡ (WT ) j ⎤ ⎡ FS ⎤

F = ρ1 ⎢ ⎥ + ρ 2 ⎢1 − ⎥ + ρ3 ⎢ ⎥ + ρ4 ⎢ ⎥ (8)
⎣⎢ (DT ) j max ⎦⎥ ⎢⎣ (SC )kj max ⎦⎥ ⎣⎢ (WT ) j max ⎦⎥ ⎣ (FS )max ⎦

Where (SC)k = , (SC)kmax is the maximum satisfaction grade

(WT)j = , (WT)jmax is the maximum waiting time

(DT)j = , (DT)jmax is the maximum distance traveled

n m
FS = ∑∑ x
j =1 k =1
0 jk , (FS)max is the maximum number of vehicles used

3. Fuzzy Vehicle Routing Problem with Time Windows (FVRPTW)

having Uncertainty in Service Time

The fuzzy logic concept was first introduced by L.A. Zadeh in1965. It was proposed
as an extended version of Aristotelian logic considering all values between 0 and 1.
The updated mathematical model of fuzzy vehicle routing problems with time
windows (FVRPTW) is given below:

Constants

n is the number of customers

m is the number of vehicles
dij is the distance of direct travel from customer i to j
wi(ti) is the waiting time for a vehicle at customer i when his service time is ti
si is the time taken to service the customer i
502 R. Gupta, B. Singh and D. Pandey

[ai,bi] is the time window of customer i

µi(ti) is the degree of satisfaction of customer i

Variables

ti is the service time for customer i

Formulation of Mathematical Model

n m
Minimize ∑∑ x
j =1 k =1
0 jk , (3.1)

Maximize (3.2)

Minimize (3.3)

Minimize (3.4)

Subject to,

(3.5)

(3.6)

(3.7)

(3.8)

(3.9)
Fuzzy vehicle routing problem 503

(3.10)

(3.11)

(3.12)

(3.13)

where the subscript 0 stands for the central depot of the fleet of the vehicles.
Objective (3.1) minimizes fleet size, objective (3.2) maximizes the average grade of
satisfaction, objective (3.3) minimizes total travel distance, and objective (3.4)
minimizes total waiting time for vehicles. Constraint (3.5) ensures that the service
time for each customer is within a tolerable interval of time. Constraint (3.6) ensures
that each customer is serviced by one and only one vehicle. Constraint (3.7) and (3.8)
ensures that for each customer, there are only two customers directly connected with
him, one directly reaches him, and another he directly travels to by a vehicle.
Constraint (3.9) and (3.10) guarantee schedule feasibility with respect to time
considerations.

3.1 Fitness Function

Fitness is calculated based on the objective function as shown in equation (8). The
four objectives considered are:
(i) minimizing the fleet size
(ii) minimizing the total distance traveled,
(iii) maximizing the average grade of customer satisfaction, and
(iv) minimizing total waiting time over vehicles.
There are tradeoffs between objectives. Thus the solution obtained is a compromised
solution (nearly optimal) for considering different objectives at the same time. It is
easier for a dispatcher to acquire a best compromise solution by minimizing the
overall cost of objectives. He can satisfy the bigger costs of the objectives but
sacrifice the smaller to minimize the overall cost.
The application of this improved model in a paper distribution problem where the
supplier delivers printing paper shipment to customers has been reconsidered. The
necessary data to execute the model is given below:
504 R. Gupta, B. Singh and D. Pandey

Table 1: transit distance (km) between distribution center and customers

Destinations Distribution C1 C2 C3 C4 C5 C6 C7 C8
(Customers) Centre
Distribution 0 10 15 8 5 10 15 20 10
Centre
C1 10 0 5 5 10 15 20 15 15
C2 15 5 0 10 15 20 5 10 5
C3 8 5 10 0 20 5 8 5 20
C4 5 10 15 20 0 8 10 20 15
C5 10 15 20 5 8 0 5 15 20
C6 15 20 5 8 10 5 0 20 15
C7 20 15 10 5 20 15 20 0 8
C8 10 15 5 20 15 20 15 8 0

Table 2: due service time schedules

Destinations C1 C2 C3 C4
(Customers)
Service time e1 u1 l1 e2 u2 l2 e3 u3 l3 e4 u4 l4
Required time 8:00 9:00 10:00 9:00 10:00 11:00 10:00 11:00 12:00 11:00 12:00 13:00
Destinations C5 C6 C7 C8
(Customers)
Service time e5 u5 l5 e6 u6 l6 e7 u7 l7 e8 u8 l8
Required time 12:00 13:00 14:00 13:00 14:00 15:00 14:00 15:00 16:00 15:00 16:00 17:00

The paper supplier confronts a vehicle routing problem, which consists of a fleet of 4
vehicles and 8 customers located in the Northern Taiwan. The objective is to design a
dispatching strategy for distribution of papers to 8 customers by 4 vehicles in the least
time. It is assumed that the load in the vehicle doesn’t exceed its capacity. The
service time taken to serve each customer has been generated randomly to make the
system robust.

4. Experimental Results
Computational experiments were conducted to test the effectiveness of the proposed
algorithm. The basic setting of weights for the evaluation function and the parameters
(pop_size, max_gen) for genetic algorithms is given in Table 4.1.
Fuzzy vehicle routing problem 505

Based on the above settings, the population size was varied from 50 to 150. The
average results over 10 runs for each case are reported in Table 4.2. The results show
that when the population size is larger than 80, the total fitness is better. We get the
best fitness corresponding to population size of 120, in which we achieve the best
value for all the attributes with little compromise on slightly more waiting time.
Based on the above settings, the weights of objective functions (8) were varied as
shown in Table 4.3 to investigate how they impact on the final decision. From the
results we can see that emphasis on the objective of distance minimization will lead
to a much better solution (with total fitness of 0.2).
The chromosome and the best route corresponding to each setting have been
displayed in Table 4.4. The second column represents the chromosomes as a
sequence of customers generated randomly. The third column gives the number of
vehicle used to serve the customers. The last column shows the path which each
vehicle will follow starting from depot, satisfying the time windows of the customers
and coming back to depot..

Table 4.1: Basic Setting of Function parameters and GA parameters

ρ1 ρ2 ρ3 ρ4 pop_size max_gen
0.5 0.3 0.1 0.1 100 500

Table 4.2: Results with varied pop_size

Pop_size Satisfaction Waiting Distance Fleet Total

grade time traveled size fitness
60 0.4835 4.540 116 2 0.579
80 0.4383 2.426 126 2 0.579
100 0.3425 2.714 98 2 0.548
120 0.6001 4.914 86 2 0.467
140 0.4524 2.2415 99 2 0.5196

Table 4.3: Results with varied weights

Sett- ρ1 ρ2 ρ3 ρ4 SG WT DT FS TF
ings
1 0.85 0.05 0.05 0.05 0.168 5.768 99 3 0.563
2 0.05 0.85 0.05 0.05 0.481 1.212 133 2 0.500
3 0.05 0.05 0.85 0.05 0.297 1.756 113 2 0.200
4 0.05 0.05 0.05 0.85 0.606 2.493 106 2 0.397
5 0.50 0.30 0.10 0.10 0.476 1.993 124 2 0.580
506 R. Gupta, B. Singh and D. Pandey

• SG – Satisfaction grade FS - Fleet size (# of Vehicles)

• WT - Waiting time (in minutes) TF - Total fitness
• DT – Distance traveled (in kms)

Table 4.4: Results with chromosomes and corresponding routes

Settings Chromosome Fleet Vehicle Routes (0 indicates depot)

[customers] size (k)
1 [2 3 1 7 4 5 6 8] 2 k=1 {0 2 3 6 7 0}
k=2 {0 1 4 5 1 0}
2 [2 1 3 6 7 4 8 5] 2 k=1 {0 2 3 6 7 8 0}
k=2 {0 1 4 5 0}
3 [2 3 4 6 7 1 5 8] 2 k=1 {0 2 3 4 6 7 8 0}
k=2 {0 1 5 0}
4 [1 5 2 3 6 7 4 8] 2 k=1 {0 1 5 6 7 0}
k=2 {0 2 3 4 8 0}
5 [1 5 2 6 3 8 4 7] 2 k=1 {0 1 5 6 8 0}
k=2 {0 2 3 4 7 0}

5. Conclusions
This algorithm provides many compromised strategies to dispatcher which helps him
taking a decision in any given situation. Looking at the customer’s preference he can
very well plan the priorities of the goals. Reducing the fleet size really helps the
dispatcher in saving company’s transportation cost. The solutions given above are the
best solutions under given circumstances from dispatcher’s point of view. In case any
customer demands an early service or doesn’t want to wait at all, an extra vehicle can
be arranged for that customer which will cost more to the dispatcher but customer’s
satisfaction grade will be high that will help the dispatcher in future business.

References
[1] A. V. Breedam, An analysis of the effect of local improvement operators in
genetic algorithms and simulated annealing for the vehicle routing problem, RUCA
Working Paper 96/14, University of Antwerp, Belgium, (1996).
Fuzzy vehicle routing problem 507

[2] G. Clark and J. V. Wright, Scheduling of vehicles from a central depot to a

number of delivery points, Operations Research, 12(1964), 568-581.

[3] G. B. Dantzig and J. H. Ramser, The truck dispatching problem, Management

Science, 6 (1959), 80.

[4] J. Lin, A GA-based Multi-objective Decision Making for optimal Vehicle

Transportation, Journal of information science and engineering, 24 (2008), 237-260.

[5] J.Y. Potvin, and S. Bengio, The Vehicle Routing Problem with Time Windows
Part II: genetic search, INFORMS Journal on Computing, 8 (1996), 165-172.

[6] M.Gen and R. Cheng, Genetic Algorithms and Engineering Design, John Wiley
and Sons, 1996.

[7] R. Gupta, B. Singh and D. Pandey, Genetic Algorithm Based Vehicle Routing
Problems: The State of the Art, Vikram Mathematical Journal, Ujjain, India, 26
(2006), 89-110.

[8] R. Gupta, B. Singh and D. Pandey, A Genetic Approach for Fuzzy Vehicle
Routing Problems with Time Windows, communicated to Indian Academy of
Mathematics, Indore, India, (2009).

[9] S.R. Thangiah, Vehicle Routing with Time Windows using Genetic Algorithms,
Technical report SRU-CpSc-TR-93-23, Slippery Rock University, Slippery Rock,
PA, 1993.

Received: August, 2009