Hindawi Publishing Corporation

Mathematical Problems in Engineering

Volume 2013, Article ID 392789, 10 pages
http://dx.doi.org/10.1155/2013/392789

Research Article
A Hybrid Bat Algorithm with Path Relinking for
Capacitated Vehicle Routing Problem

Yongquan Zhou,1,2 Jian Xie,1 and Hongqing Zheng1

1
College of Information Science and Engineering, Guangxi University for Nationalities, Nanning, Guangxi 530006, China
2
Guangxi Key Laboratory of Hybrid Computation and IC Design Analysis, Guangxi University for Nationalities,
Nanning, Guangxi 530006, China

Correspondence should be addressed to Yongquan Zhou; [email protected]

Received 8 May 2013; Accepted 24 July 2013

Academic Editor: Praveen Agarwal

Copyright © 2013 Yongquan Zhou et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The capacitated vehicle routing problem (CVRP) is an NP-hard problem with wide engineering and theoretical background. In
this paper, a hybrid bat algorithm with path relinking (HBA-PR) is proposed to solve CVRP. The HBA-PR is constructed based on
the framework of continuous bat algorithm; the greedy randomized adaptive search procedure (GRASP) and path relinking are
effectively integrated into bat algorithm. Moreover, in order to further improve the performance, the random subsequences and
single-point local search are operated with certain loudness (probability). In order to verify the validity of the method in this paper,
and it’s efficiency and with other existing methods, several classical CVRP instances from three classes of CVRP benchmarks are
selected to tested. Experimental results and comparisons show that the HBA-PR is effective for CVRP.

1. Introduction an abundance of methods for the CVRP is a population-

based algorithm and the parameter setting of the algorithm,
The vehicle routing problem (VRP) is a classical combinato- is pretty important; however, the parameter setting of many
rial optimization problem that was proposed in the late 1950s, metaheuristics is not considered in the literature.
and it is still a research hotspot in the field of Operations Bat algorithm (BA) is a fairly new metaheuristics pro-
Research. The capacitated vehicle routing problem (CVRP) posed by Yang [11] in 2010, which inspired by the intelligent
is introduced by Dantzig and Ramser in 1959 [1], which echolocation behavior of microbats when they forage. As
designs a set of customer demands that have to be served we know, many new metaheuristics have been widely used
with a fleet of vehicles from a depot or central node, each and successfully applied to solve the CVRP notwithstanding
vehicle has the uniform capacity, and each customer has a BA has not yet been applied to solve the CVRP, and BA
certain demand that must be satisfied at minimal cost. These has been applied to solve other problems with great success.
costs usually represent distances, traveling times, number of For example, Gandom et al. focus on solving constrained
vehicles employed, or a combination of these factors. optimization tasks [12]. Yang and Gandom apply bat algo-
It is known that the CVRP is an NP-hard problem [2]. rithm to solve many global engineering optimizations [13].
Various approaches have been presented to solve the CVRP A classification mode is proposed by Mishra et al. using
during the last decades, such as linear programming [3], bat algorithm to update the weights of a functional link
several metaheuristics [4–6], and many hybrid heuristics artificial neural network (FLANN) classifier [14]. Meanwhile,
with variable neighborhood search or constructive heuristic some researchers have improved bat algorithm and applied
methods [7–10]. The overview of methods presented in [6] it to various optimization problems. Xie et al. proposed
shows that at least 29 different methods for the CVRP exist; all a DLBA bat algorithm based on differential operator and
achieve more or less comparable performance. Although sev- Lévy flights trajectory to solve function optimization and
eral methods can produce good solutions, the computational nonlinear equations [15]. Wang et al. proposed a new bat
time is long when the scale of instances is large. Meanwhile, algorithm with mutation (BAM) to solve the uninhabited
2 Mathematical Problems in Engineering

combat air vehicle (UCAV) path planning problem [16]. In 2.2. Basic Bat Algorithm. The basic bat algorithm (BA) is a
this paper, we propose a hybrid bat algorithm (HBA-PR) to metaheuristic first introduced by Yang in 2010. In simulations
solve capacitated vehicle routing problem. of BA, under several under idealized rules, the updated rules
The rest of this paper is organized as follows. The problem of bat’s positions 𝑥𝑖 and velocities V𝑖 in a 𝐷-dimensional
description of CVRP and original bat algorithm are described search space are defined. The new solutions 𝑥𝑖𝑡 and velocities
in Section 2. The hybrid bat algorithm (HBA-PR) for CVRP V𝑖𝑡 at generation 𝑡 are given by
is described detailedly in Section 3. The experimental results
of the HBA-PR and comparisons with other previous algo- fr𝑖 = frmin + (frmax − frmin ) 𝛽,
rithms are shown in Section 4. In the last section, we conclude
this paper and point out some future work in Section 5. V𝑖𝑡 = V𝑖𝑡−1 + (𝑥𝑖𝑡 − 𝑥∗ ) fr𝑖 ,

2. Problem Descriptions and Bat Algorithm 𝑥𝑖𝑡 = 𝑥𝑖𝑡−1 + V𝑖𝑡 ,

(3)
2.1. Capacitated Vehicle Routing Problem. The CVRP is con-
sidered to be the classical version of the VRP, which designs where 𝛽 ∈ [0, 1] is a random vector drawn from a uniform
a set of customer demands that have to be served with a distribution and fr𝑖 denotes frequency of each bat. Generally,
fleet of vehicles from a depot or central node, each vehicle the frequency fr𝑖 ∈ [frmin , frmax ]. Here 𝑥∗ is the current global
has the uniform capacity, and each customer has a certain best location (solution) which is located after comparing all
demand. The objective makes the expended cost as minimal the solutions among all the 𝑛 bats.
as possible. Let 𝐺 = (𝑉, 𝐸) be a complete graph with a set After the position updating of bat, a random number is
of vertices 𝑉 = {0, 1, . . . , 𝑘}, where the vertex {0} represents generated; if the random number is greater than the pulse
the depot and the remaining ones represent the customers. emission rate 𝑟𝑖 , a new position will be generated around the
Each edge 𝑒𝑖𝑗 = {𝑖, 𝑗} ∈ 𝐸 has a nonnegative cost 𝑐𝑖𝑗 , and current best solutions; it can be represented by
each customer 𝑖 ∈ 𝑉󸀠 = 𝑉 \ {0} has a demand 𝑑𝑖 . Let 𝑆 =
(1, 2, . . . , 𝑚) be the set of homogeneous vehicles with capacity 𝑥 = 𝑥∗ + 𝜀 × Ld𝑡 , (4)
𝑄. The CVRP consists in constructing a set of up to 𝑘 routes
in such a way that (1) every route starts and ends at the depot; where 𝜀 ∈ [−1, 1] is a random number, while Ld𝑡 = ⟨Ld𝑡𝑖 ⟩ is
(2) all demands are accomplished; (3) the vehicle’s capacity is the average loudness of all the bats at current generation 𝑡.
not exceeded; (4) a customer is visited by only a single vehicle; Furthermore, the loudness Ld𝑖 and the pulse emission
(5) the sum of costs is minimized. The mathematical formulas rate 𝑟𝑖 will be updated, and a solution will be accepted if a
are defined as follows [10]: random number is less than loudness Ld𝑖 and 𝑓(𝑥𝑖 ) < 𝑓(𝑥∗ ).
𝑘 𝑘 𝑚
Ld𝑖 and 𝑟𝑖 are updated by
min Z = ∑ ∑ ∑𝑐𝑖𝑗 𝑒𝑖𝑗𝑠 , (1)
𝑖=0 𝑗=0 𝑠=0 Ld𝑡+1
𝑖 = 𝛼 × Ld𝑡𝑖 , 𝑟𝑖𝑡+1 = 𝑟𝑖0 × [1 − exp (−𝛾 × 𝑡)] , (5)
𝑘 where 𝛼, 𝛾 are constants and 𝑓(⋅) is fitness function. The
s.t. ∑𝑑𝑖 𝑦𝑖𝑠 ≤ 𝑄, 𝑠 = 1, 2, . . . , 𝑚, algorithm repeat until the termination criterion is reached.
𝑖=0
𝑚
∑𝑦𝑖𝑠 = 1, 𝑖 = 1, 2, . . . , 𝑘, 3. Hybrid Bat Algorithm with Path
𝑠=1 Relinking for CVRP
𝑘
(2)
∑𝑒𝑖𝑗𝑠 = 𝑦𝑖𝑠 , 𝑗 = 1, 2, . . . , 𝑘; 𝑠 = 1, 2, . . . , 𝑚, 3.1. Solution Representation in HBA-PR. Since standard BA is
𝑖=0 a continuous optimization algorithm, the standard continu-
𝑘 ous encoding scheme of BA cannot be used to solve CVRP
∑𝑒𝑖𝑗𝑠 = 𝑦𝑖𝑠 , 𝑖 = 1, 2, . . . , 𝑘; 𝑠 = 1, 2, . . . , 𝑚, directly. In order to apply BA to solve CVRP, the first step is
𝑗=0 to devise a suitable representation for the candidate solutions
in designing a hybrid bat algorithm for a particular problem.
where 𝑠 denotes the number of vehicles; 𝑒𝑖𝑗𝑠 = 1 if vehicle 𝑠 is Each individual is a sequence with integer number which is
from 𝑖 to 𝑗, otherwise 𝑒𝑖𝑗𝑠 = 0; the 𝑦𝑖𝑠 = 1 if vehicle 𝑠 is loading the order of visiting these customers; the 0 represents the
(active), otherwise 𝑦𝑖𝑠 = 0. depot. For example, if we have the following three routes:

0 3 5 8 0 0 4 9 0 0 2 6 1 7 0
R number 1: Depot Depot R number 2: Depot Depot R number 3: Depot Depot (6)

the coded individual with integer is 0 → 3 → 5 → 8 → is represented as 3 → 5 → 8 → 4 → 9 → 2 → 6 →

4 → 9 → 2 → 6 → 1 → 7 → 0, and the bat individual 1 → 7.
Mathematical Problems in Engineering 3

3.2. Hybrid Bat Algorithm. Aimed at the capacitated vehicle frequency fr is determined according to the pulse emission
routing problem, based on the idea of bat algorithm, a rate 𝑟, which is represented by
hybrid bat algorithm is proposed, which integrates greedy
randomized adaptive search procedure (GRASP) heuristic 1 − max (0.2, min (0.8, 𝑟)) , rand > 𝑟,
fr = { (8)
and bat algorithm, and the path relinking as an intensification max (0.2, min (0.8, 𝑟)) , rand ≤ 𝑟,
strategy to explore local trajectories connecting elite solutions
obtained by the proposed algorithm. The hybrid bat algo- where rand is a random number and 0.2 and 0.8 are expe-
rithm with path relinking is named as HBA-PR. rience parameters reference [18]. The frequency fr decreases
GRASP [17, 18] is a heuristic already applied to many gradually at firstly and then increases gradually, while the
optimization problems successfully [19–21]. GRASP consists generation 𝑡 increases. Figure 1 is the changing curve of rate
of a two-phase iterative process: construction phase and local 𝑟, Figure 2 is an example of frequency fr, and Algorithm 1
search phase. In the first phase, a greedy randomized solution shows the pseudocode of greedy randomized construction
is built. Due to this solution is not guaranteed to be locally with frequency fr.
optimal; a local search is performed in the second phase. The The local search phase uses the 2-Opt heuristic for
final result is simply the best solution found over all iterations. exchanging. Algorithm 2 shows the local search procedure.
The construction phase can be described as a process The input parameter is an initial solution 𝑆 obtained by the
which stepwise adds one element at a time to a partial Greedy Randomized Construction procedure. The current
(incomplete) solution. According to a greedy function, all route need is divided into 𝑚 subroute according to the load,
elements are sorted, and the Restricted Candidate List (RCL) set the start, and end of suroute said to 0; for example, 𝑆 =
is constructed based on the order, and then from the list {1, 2, 3, 4, 5, 6, 7}, subroute is 0 → 1 → 2 → 3 → 4 → 0,
randomly select a element. In the second phase, a local and 0 → 5 → 6 → 7 → 0.
search is initialized from these points, this iterative process
is repeated until a termination criterion is met, and the best 3.3. Hybrid Bat Algorithm with Path Relinking. Path relinking
solution found over all iterations is taken as the result. was originally proposed by Glover [23]; Laguna and Martı́
RCL is created using a parameter 𝛼 to restrict the size [24] were the first to use path relinking within a GRASP
of this list. Candidate 𝑒 ∈ 𝐶 is sorted according to their strategy. Path relinking generates new solutions by exploring
greedy function value 𝑓(𝑒). In a cardinality-based RCL, the trajectories connecting an initial solution 𝑥𝑠 to an elite
latter is made up by the 𝑘 top-ranked elements. In a value- guiding solution 𝑥𝑡 . The path relinking procedure consists
based construction, the RCL consists of the elements in the in selecting moves that introduce attributes contained in
set {𝑒 ∈ 𝐶 : 𝑓∗ ≤ 𝑓(𝑒) ≤ 𝑓∗ + 𝛼 × (𝑓∗ − 𝑓∗ )}, where 𝑓∗ = the guiding solution 𝑥𝑡 to the initial solution 𝑥𝑠 until the
min(𝑓(𝑒) : 𝑒 ∈ 𝐶), 𝑓∗ = max{𝑓(𝑒) : 𝑒 ∈ 𝐶}, and 𝛼 ∈ [0, 1]. initial solution is completely transformed in the guiding
Since the best value for 𝛼 is often difficult to determine, a solution 𝑥𝑡 . Path relinking may also be viewed as a con-
random value is often assigned. The values for 𝛼 adopted in strained local search strategy applied to the initial solution
the constructive heuristics are set using reactive strategies, 𝑥𝑠 . Furthermore, there are several alternatives that have been
which usually leads to better performance than using fixed considered, which involve tradeoffs between computation
values [22]. time and solution quality. These alternatives include periodi-
In original bat algorithm, the bat individual randomly cal relinking, forward relinking, backward relinking, back and
selects a certain range of frequency, its velocity is updated forward relinking, mixed relinking, randomized relinking, and
according to their selected frequency, and at last a new truncated relinking [18].
position is generated using its velocity and its own position. One important issue in implementing a path relinking
The idea is that the position of bat individual is updated technique is the strategy to construct the elite set (ES). We
by adjusting its frequency of sonic pulse. In this paper, adopted a fixed size elite set, and a solution 𝑥 is inserted into
the position updating of bat adopted GRASP to generate a the ES as follows.
new position, the frequency is used for restricting the size A solution 𝑥 is always inserted into ES if it is not full.
of CRL, frequency equivalent to parameter 𝛼 in GRASP, Otherwise, the generated solution 𝑥 is inserted in ES only if its
and the frequency is variable value. In basic bat algorithm, cost is better than the worst cost solution found in ES, and the
the loudness Ld and the pulse emission rate 𝑟 are updated worst cost solution is replaced by the solution 𝑥. Algorithm 3
according as the iterations proceed. As the loudness usually shows the pseudocode for the algorithm to construct and
decreases, while the rate of pulse emission increases, it maintain the elite set ES.
indicates that the bats approximate their prey (optimum Algorithm 4 shows the pseudocode of path relinking
solution). The pulse emission rate 𝑟 is updated by procedure; an instance is 𝑥𝑠 = {1, 2, 3, 4, 5, 6, 7}, 𝑥𝑡 = {1, 3, 4,
2, 5, 6, 7}, according to process of Algorithm 4, the first
different element is position 2 in 𝑥𝑠 ; after executing replace
−1
−5 operation, the 𝑥1 = {1, 3, 4, 2, 5, 6, 7}; the second different
𝑟 (𝑡) = (1 + exp ( ) ), (7) element is position 3 in 𝑥𝑠 ; after executing replace operation,
𝑡max × (𝑡 − 𝑡max /2)
the 𝑥2 = {1, 2, 4, 3, 5, 6, 7}; the third different element is
position 4 in 𝑥𝑠 ; after executing replace operation, the 𝑥3 =
where 𝑡 denotes the 𝑡th generation and 𝑡max is the maximal {1, 4, 3, 2, 5, 6, 7}; if the 𝑥2 is better than the 𝑥𝑡 , and the only,
generation. The rate 𝑟 is similar to Sigmoid function, and the then return 𝑥2 .
4 Mathematical Problems in Engineering

𝑆 ← 𝑅 rand Chosen Vertex(V); // R is selected at random a vertex as initial solution

𝐶 ← 𝑉 \ V; // The candidate set 𝐶 is initialized
While 𝐶 ≠ ⌀ do
𝐼𝐶 ← 𝐸V𝑎𝑙𝑢𝑎𝑡𝑒 Incremental Costs(𝑆); // the incremental costs are evaluated
𝑖𝑐min ← min ({𝑖𝑐 ∈ 𝐼𝐶});
𝑖𝑐max ← max ({𝑖𝑐 ∈ 𝐼𝐶});
𝑅𝐶𝐿 ← {𝑒 ∈ 𝐶 | 𝑖𝑐 (𝑒) ≤ 𝑖𝑐min + 𝑓𝑟 × (𝑖𝑐max − 𝑖𝑐min )}; // 𝑅𝐶𝐿 is created
𝑠 ← 𝑆𝑒𝑙𝑒𝑐𝑡 Element (RCL); // a vertex 𝑠 is randomly selected from 𝑅𝐶𝐿
𝑆 ← 𝑂𝑏𝑡𝑎𝑖𝑛 Min Solution(𝑆, 𝑠); // the partial tour is updated by inserting the
vertex 𝑠
𝐶 ← 𝐶 \ {𝑠}; // the candidate set 𝐶 is updated
end
return 𝑆

Algorithm 1: Greedy Randomized Construction (𝑓𝑟).

1
𝑆󸀠 ← 𝑆; 0.9
{𝑠𝑟} ← 𝐶𝑜𝑛𝑠𝑡𝑟𝑢𝑐𝑡 Sub-route(𝑆󸀠 );
for each sub-route 𝑠𝑟 do 0.8
𝑠𝑟󸀠 ← 2-opt (𝑠𝑟); // carry out the 2-opt operation 0.7
end
𝑆󸀠 ← 𝐶𝑜𝑛𝑠𝑡𝑟𝑢𝑐𝑡 Indiviaual ({𝑠𝑟󸀠 }); 0.6
Frequency fr

if 𝑓(𝑆󸀠 ) < 𝑓(𝑆) then 0.5

𝑆 ← 𝑆󸀠
end 0.4
return 𝑆 0.3
0.2
Algorithm 2: Local Search Phase (𝑆).
0.1
0
1 0 50 100 150 200
0.9 Iteration number

0.8 Figure 2: Changing curve of frequency fr.

0.7
Pulse emission rate r

0.6
if cardinality(𝐸𝑆) ≠ 𝑝𝑠 then // 𝑝𝑠 is the population size
0.5 𝐸𝑆 ← 𝐸𝑆 ∪ {𝑥}
0.4 else
𝐸𝑆𝑤 ← 𝑡ℎ𝑒 worst solution in 𝐸𝑆
0.3 if 𝑓(𝑥) < 𝑓(𝐸𝑆𝑤 ) then
0.2 𝐸𝑆 ← 𝐸𝑆 \ {𝐸𝑆𝑤 }
𝐸𝑆 ← 𝐸𝑆 ∪ {𝑥}
0.1 end
0 end
0 50 100 150 200 return 𝐸𝑆
Iteration number

Figure 1: Changing curve of rate 𝑟. Algorithm 3: Elite Set (𝑥).

3.4. Subsequence and Single-Point Local Search. In this paper, where constant 0.1 is used for avoiding the denominator is
the loudness Ld𝑖 of bat individual 𝑖 is relative to its own fitness zero, and fit𝑖 is the fitness of individual 𝑖, fit∗ and fit∗ are
fit𝑖 , the better fitness, and the lower loudness. The loudness the minimum and maximum fitness in current population,
can be described by respectively. In HBA-PR, the loudness reflects the quality of
individual. In this algorithm, there are two kinds of local
(fit𝑖 − fit∗ + 0.1) search are embedded into HBA-PR to further improve the
Ld𝑖 = , (9)
(fit∗ − fit∗ + 0.1) performance, random subsequence local search that and
Mathematical Problems in Engineering 5

𝑥𝑠 ← 𝑥; // initial solution
𝑥𝑡 ← 𝐸𝑆∗ ; // guiding solution
𝑓(𝑥𝑡 ) ← 𝑓(𝐸𝑆∗ );
Δ ← 𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒(𝑥𝑠 , 𝑥𝑡 ); // find out the difference position between 𝑥𝑠 𝑥𝑡
for 𝑖 = 1: cardinality(Δ) do
𝑗 ← 𝐹𝑖𝑛𝑑 Position(𝑥𝑠 , 𝑥𝑡 , Δ 𝑖 ); // find the position of the element of 𝑥𝑡 in Δ 𝑖 which in 𝑥𝑠
𝑥𝑖 ← 𝑅𝑒𝑝𝑙𝑎𝑐𝑒(𝑥𝑠 , Δ 𝑖 , 𝑗); // replace the 𝑗 position of 𝑥𝑠 with the element of 𝑥𝑠 in Δ 𝑖
𝑥𝑖 ← 𝑅𝑒𝑝𝑙𝑎𝑐𝑒(𝑥𝑠 , 𝑥𝑡 , Δ 𝑖 ); //replace the Δ 𝑖 position of 𝑥𝑠 with the corresponding element of 𝑥𝑡
if 𝑓(𝑥𝑖 ) < 𝑓(𝑥𝑡 ) then
𝑥∗ ← 𝑥𝑖 ;
𝑓(𝑥𝑡 ) ← 𝑓(𝑥𝑖 );
end
end
return 𝑥∗

Algorithm 4: Path Relinking (𝐸𝑆∗ , 𝑥).

single-point local search. The random subsequence local repeated until the termination criterion is met; Algorithm 7
search includes random subsequence inverse and random shows the pseudocode of HBA-PR for CVRP.
subsequence insert, and the single-point local search includes
single-point insert and single-point swap.
4. Numerical Simulation Results
For random subsequence insert, firstly, randomly selected
an origin of subsequence, and then randomly selected a and Comparisons
length of subsequence which is less than length of individual To test the performance of the proposed HBA-PR which is
𝑆. Secondly, after determining the subsequence 𝑆1, randomly extensively investigated by a large number of experimental
selected an insert point in remainder subsequence 𝑆2, 𝑆 =
studies computational simulations are carried out with some
𝑆1 ∪ 𝑆2; the 𝑆1 is inserted into 𝑆2 location in insert point.
well-studied problems taken from the web http://www.bran-
An example is shown in Figure 2. For random subsequence
chandcut.org/, a reference site which contains detailed infor-
inverse, firstly, randomly selected a subsequence with random
length, and then the inverse operation is performed. An mation regarding a large number of benchmark instances.
example is shown in Figure 3. In this paper, 12 instances from three classes of benchmarks
For single-point swap, choose two different positions are selected. The first class is Augerat et al. Set A instances,
from a permutation randomly and swap them. For single- the second class is Augerat et al. Set P instances, and the
point insert, choose two different positions from a permuta- third class is Augerat et al. Set E instances. So far, these
tion randomly, and the element in first position is insert into problems have been widely used as benchmarks to certify the
the back of second element. Similarly, two instances are performance of algorithms by many researchers.
shown in Figures 4 and 5. All computational experiments are conducted with MAT-
In local search part, the random subsequence local search LAB 2012a, and in our simulation, numerical experiments are
is performed before random single-point local search. The run on a PC with AMD Athlon (tm) II X4 640 Processor
random subsequence insert and random subsequence inverse 3.0 GHz and 2.0 GB memory. In the experiment, the termi-
are preformed according to loudness Ld. In other words, if a nation criterion is set as maximum generation 𝑡max = 200.
random number is greater than the loudness Ld𝑖 , the random Each instance independently run15 times for comparison.
subsequence insert is performed; otherwise, the random sub-
sequence inverse is performed. Similarly, the random single- 4.1. Parameter Analysis. In the subsection, parameters of
point insert and random single-point swap are performed HBA-PR are determined by experiments, and the impact
with the loudness Ld. If a random number is greater than of each parameter is analyzed. In particular, the HBA-PR
the loudness Ld𝑖 , the insert operation is performed (Figure 6);
has few parameters; we only need to test population size
otherwise, the swap operation is performed. Note that where
(𝑝𝑠) in HBA-PR. A small 𝑝𝑠 which may lead to insufficient
local search is operated on the current optimal individual ES∗
population information is provided, and the diversity can-
in elite set ES, local search is performed for each individual.
Algorithms 5 and 6 show the pseudocode of subsequence and not be guaranteed. On the other side, a large one which
single-point local search. indicates diversity is sufficient, but the computing time will
increase and the precision of optimal solution may have
3.5. HBA-PR Framework for CVRP. We propose in this work, lesser improvement. In order to evaluate the sensitivity of
to incorporate greedy randomized adaptive search procedure, parameters 𝑝𝑠, three benchmarks selected from different
path-relinking strategies, subsequence, and single-point local benchmark set are chosen to run 10 times. These benchmarks
search to the bat algorithm by defining distinct ways to solve are A n33 k5, E n23 k3 and P n19 k2, and the statistical result
capacitated vehicle routing problem. This iterative process is and convergence curves are shown in Figures 7, 8, 9, and 10.
6 Mathematical Problems in Engineering

Before 2 6 1 7 3 5 8 4
Subsequence
Insert point
0 0
Subsequence
Depot Depot
After 2 5 8 6 1 7 3 4

Figure 3: Random subsequence insert operation.

Before 2 6 1 7 3 5 8 4

Subsequence
0 0
Subsequence
Depot
Depot
After 2 5 3 7 1 6 8 4

Figure 4: Random subsequence inverse operation.

Before 2 6 1 7 3 5 8 4

Second element Random swap First element

0 First element Second element 0
position is (7, 2)
Depot Depot
After 2 8 1 7 3 5 6 4

Figure 5: Single-point swap operation.

2 6 1 7 3 5 8 4
Before
Random insert
Second element position is (7, 2) First element
0 Second element 0
First element
Depot
Depot
After 2 6 8 1 7 3 5 4

Figure 6: Single-point insert operation.

for 𝑖 = 1: ps do
if rand > 𝐿𝑑𝑖 then
𝑥 ← 𝑆𝑢𝑏-sequence Insert(𝐸𝑆∗ ); // perform random sub-sequence insert operation
else
𝑥 ← 𝑆𝑢𝑏-sequence Inverse(𝐸𝑆∗ ); // perform random sub-sequence inverse operation
end
end
return 𝑥

Algorithm 5: Sub-sequence Local Search (𝐸𝑆∗ ).

for 𝑖 = 1: ps do
if rand > 𝐿𝑑𝑖 then
𝑥 ← 𝑆𝑖𝑛𝑔𝑙𝑒-point Insert(𝐸𝑆∗ ); // perform random single-point insert operation
else
𝑥 ← 𝑆𝑖𝑛𝑔𝑙𝑒-point Swap(𝐸𝑆∗ ); // perform random ingle point swap operation
end
end
return 𝑥

Algorithm 6: Single-point Local Search (𝐸𝑆∗ ).

Mathematical Problems in Engineering 7

(1) Initialize the 𝑝𝑠, bat population and other parameters;

(2) Evaluate fitness for each individual;
(3) 𝐸𝑆 ← 𝐸𝑙𝑖𝑡𝑒 Set(𝑥);
(4) while 𝑡 ≤ 𝑡max do
(5) Compute pulse emission rate by (7);
(6) Determine frequency 𝑓𝑟 by (8);
(7) for 𝑖 = 1: ps do
(8) 𝑥 ← 𝐺𝑟𝑒𝑒𝑑𝑦 Randomized Construction(𝑓𝑟);
(9) 𝑥 ← 𝐿𝑜𝑐𝑎𝑙 Search Phase (𝑥);
(10) end
(11) Evaluate fitness for each individual 𝑥;
(12) 𝐸𝑆 ← 𝐸𝑙𝑖𝑡𝑒 Set(𝑥);
(13) 𝐸𝑆∗ ← 𝑆𝑒𝑙𝑒𝑐𝑡 Best Elite(𝐸𝑆); // select the best individual in elite set 𝐸𝑆
(14) for 𝑖 = 1: 𝑝𝑠 do
(15) 𝑥 ← 𝑃𝑎𝑡ℎ Relinking(𝐸𝑆∗ , 𝑥);
(16) end
(17) Evaluate fitness for each individual 𝑥;
(18) 𝐸𝑆 ← 𝐸𝑙𝑖𝑡𝑒 Set(𝑥);
(19) 𝐿𝑑𝑖 ← 𝐶𝑜𝑚𝑝𝑢𝑡𝑒 Loudness(𝑓(𝑥)); //Compute loudness of each individual by (9);
(20) 𝐸𝑆∗ ← 𝑆𝑒𝑙𝑒𝑐𝑡 Best Elite(𝐸𝑆);
(21) 𝑥 ← 𝑆𝑢𝑏-sequence Local Search (𝐸𝑆∗ ) //carry out random sub-sequence local
search
(22) Evaluate fitness for each individual 𝑥;
(23) 𝐸𝑆 ← 𝐸𝑙𝑖𝑡𝑒 Set(𝑥);
(24) 𝐸𝑆∗ ← 𝑆𝑒𝑙𝑒𝑐𝑡 Best Elite(𝐸𝑆);
(25) 𝑥 ← 𝑆𝑖𝑛𝑔𝑙𝑒-point Local Search (𝐸𝑆∗ ) //carry out random single-point local search
(26) Evaluate fitness for each individual 𝑥;
(27) 𝐸𝑆 ← 𝐸𝑙𝑖𝑡𝑒 Set(𝑥);
(28) 𝑡 = 𝑡 + 1
(29) end
(30) Output result and plot

Algorithm 7: HBA-PR.

Table 1: Comparison of results for Augerat et al. Set A, E, and P instances.

Instance Capacity Tightness I BKS R BSK CWS SR-GCWS CS-GRASP HBA-PR

A-n33-k5 100 0.89 661 662.76 712.05 662.11 662.1452 662.1101
A-n33-k6 100 0.9 742 742.83 776.26 742.69 743.5785 742.6933
A-n37-k5 100 0.81 669 672.59 707.26 672.47 672.4652 672.4652
A-n39-k6 100 0.88 831 833.20 863.08 833.20 — 835.2518
E-n23-k3 4500 0.75 569 — — — 569.7461 568.5625
E-n22-k4 6000 0.94 375 375.28 388.77 375.28 — 375.2798
E-n33-k4 8000 0.92 839 838.72 843.1 837.67 — 837.9253
E-n51-k5 160 0.97 521 524.94 584.64 524.61 — 524.6111
P-n19-k2 160 0.97 212 212.66 237.90 212.66 212.6569 212.6569
P-n20-k2 160 0.97 216 217.42 234.00 217.42 217.4156 217.4156
P-n22-k2 160 0.96 216 217.85 239.50 217.85 217.8522 217.8522
P-n51-k10 80 0.97 741 742.48 790.97 741.50 — 743.2648

The ordinate normalized fitness (log) in Figures 8–10 test cases, the parameter 𝑝𝑠 can be determined. For A n33 k5
is logarithm of normalized fitness; the aim is to show the and P n19 k2, the performance of HBA-PR is better when
convergence curves clearly. The normalization formula of 𝑝𝑠 = 30. For E n23 k3, that 𝑝𝑠 is equal to 50 is best, but the
fitness is (fit − fit∗ )/(fit∗ − fit∗ ), where fit∗ is the best-known performance is good as well, while 𝑝𝑠 = 30. From Figures 8–
solution, and fit∗ is the initial fitness. 10, the information provided by population is sufficient when
Figure 7 represents the relative error of test case A n33 k5, 𝑝𝑠 is greater than 20; however, the convergence rate is better
E n23 k3, and P n19 k2 after 10 times independent running, when 𝑝𝑠 = 30 among the three instances. Considering the
which shows the sensitivity of parameter 𝑝𝑠. From the three tradeoff between the stability of algorithm and the rate of
8 Mathematical Problems in Engineering

690 100
Fitness

680
670

ps = 10 ps = 20 ps = 30 ps = 40 ps = 50

Normalized fitness (log)

10−1
A-n33-k5
Fitness

576 10−2
572
568
ps = 10 ps = 20 ps = 30 ps = 40 ps = 50

E-n23-k3 10−3

220 20 40 60 80 100 120 140 160 180 200

Fitness

216 Iteration number

212 ps = 10 ps = 40
ps = 10 ps = 20 ps = 30 ps = 40 ps = 50 ps = 20 ps = 50
ps = 30
P-n19-k2
Figure 9: Convergence curve for E n23 k3.
Figure 7: Box-and-whisker diagram of selected three instances.

100
100
Normalized fitness (log)
Normalized fitness (log)

10−1

10−1
10−2

10−3
20 40 60 80 100 120 140 160 180 200
20 40 60 80 100 120 140 160 180 200
Iteration number
Iteration number
ps = 10 ps = 40
ps = 10 ps = 40
ps = 20 ps = 50
ps = 20 ps = 50
ps = 30
ps = 30
Figure 10: Convergence curve for P n19 k2.
Figure 8: Convergence curve for A n33 k5.

convergence, the parameter 𝑝𝑠 takes a compromise values, capacity; tightness (demand/capacity); I BKS is integral best-
𝑝𝑠 = 30. known solution (BKS) or “optimal” value according to the
web http://www.branchandcut.org/; R BSK is verifies real
4.2. Comparisons of Simulation Results. In order to show the costs for the best-known solution according to [9]; CWS is the
effective of HBA-PR, we carry out a simulation to compare costs associated with the solution given by the parallel version
HBA-PR with other state-of-art algorithms, that is, a parallel of the CWS heuristic; SR-GCWS is the best solution obtained
version of the classical Clarke and Wright Savings (CWS) by SR-GCWS method; HBA-PR is our best solution, where
heuristic and SR-GCWS (Simulation in Routing via the “—” represents no records in the literature.
Generalized Clarke and Wright Savings heuristic) proposed From the simulationresults obtained by testing HBA-PR,
by Juan et al. [9], CS-GRASP proposed by Zheng et al. [10]. it demonstrates that using the proposed HBA-PR to solve the
Results of these simulations are summarized in Table 1, which CVRP is effective, and the performance of HBA-PR is promi-
contain the following information of each instance: vehicle nent. From Table 1, all instances achieved a good quality
Mathematical Problems in Engineering 9

100 70
28 27 32 29 37
19 36 9 12 44 8 4 36
90 26 9
10 60 21
24 23
80 25 49
27 31 25 2
70 3229 50 28 3 30
24 7 33
33 4 21 15 22
60 18 15 1 47 12 17
3735 20 40 26 51
50 30 161 39
2 19 48 13 10 35
22 13 3 30 6
40 50 31
17 14 5
18 38
30 11 11
8 23 20
42 16
20 14 45 40
5 20
10 43 46 34
10 7 41
6
34
0 0
0 20 40 60 80 100 0 10 20 30 40 50 60 70
Figure 11: Optimal routes of A n37 k5. Figure 13: Optimal routes of P n51 k10.

70 32 1800
27 29 37
44 8 4 36
9
60 21 1600
24 23
49
25 2
50 28 3 30
7 33 1400
15 12 17 22
40 1 47
26
Fitness

51 1200
39
19 48 13 10 35
30 6
50 31
14 5 1000
18 38
11
20
42 16
45 40 800
20
10 43 46 34
41 600
0
0 10 20 30 40 50 60 70 0 50 100 150 200
Iteration number
Figure 12: Optimal routes of E n51 k5.
A-n37-k5
E-n51-k5
P-n51-k10
solution, and solutions of 6 instances are better than the
Figure 14: Convergence curve of three instances.
best-known solution (“optimal” value). HBA-PR has matched
10 of the 12 best-known solutions except for P n51 k10 and
A n39 k6, and the average deviation from the real costs
best-known solution is 0.057%. HBA-PR outperforms CWS In general, the proposed HBA-PR can produce good
for the 12 instances. Comparing with CS-GRASP and HBA- solutions when compared with existing heuristics for solving
PR, HBA-PR outperforms CS-GRASP in terms of A n33 k5, the CVRP, and the convergence rate of HBA-PR is faster.
A n33 k6, and E n23 k3; the other instances have same These results seem to indicate that the hybrid bat algorithm
results. The SR-GCWS and HBA-PR have similar results, and with path relinking is an alternative to solve the capacitated
the gap is very small. vehicle routing problem.
Furthermore, Figure 11 shows the best solution found so
far by using our methodology for the A n37 k5.vrp file, where 5. Conclusions
the depot (using 1 instead of 0) is at the center. Analogously,
Figures 12 and 13 show the best solution found by HBA-PR for The capacitated vehicle routing problem is important in the
the E-n51-k5.vrp file and P n51 k10.vrp file. The convergence fields of Operations Research, which is an NP-hard problem.
curve for some test problems has been shown in Figure 14. Bat algorithm is a continuous metaheuristics; it cannot be
From Figure 14, for instance, A n37 k5, it converges to an used to solve the CVRP directly. In this paper, a hybrid
optimal solution after 43 generation, and it expends about 170 bat algorithm with path relinking (HBA-PR) for solving the
generations, for instance, E-n51-k5 and P n51 k10 when algo- CVRP has been presented. This methodology, which does not
rithm converges to an optimal solution, which demonstrate require any particular fine tuning of parameters or config-
that the HBA-PR has a faster convergence rate. uration process, combines the classical greedy randomized
10 Mathematical Problems in Engineering

adaptive search procedure (GRASP) with bat algorithm; the [10] H. Q. Zheng, Y. Q. Zhou, and Q. F. Luo, “A hybrid cuckoo
path relinking as an intensification strategy to explore local search algorithm-GRASP for vehicle routing problem,” Journal
trajectories connecting elite solutions obtained by proposed of Convergence Information Technology, vol. 8, no. 3, 2013.
algorithm; subsequence and single-point local search are [11] X.-S. Yang, “A new metaheuristic Bat-inspired algorithm.
effectively integrated into HBA-PR. Results show that our Nature inspired cooperative strategies for optimization
methodology is able to provide fine-quality solutions which (NICSO),” Studies in Computational Intelligence, vol. 284, pp.
can compete with the ones provided by some exact and 65–74, 2010.
heuristic approaches. Moreover, because of its simplicity and [12] A. H. Gandom, X. S. Yang, A. H. Alavi, and S. Talatahari, “Bat
flexibility, we think that this methodology can easily be algorithm for constrained optimization tasks,” Neural Comput-
adapted to other variants of the vehicle routing problem ing and Applications, vol. 22, no. 6, pp. 1239–1255, 2013.
and even to other combinatorial problems, for example, the [13] X. S. Yang and A. H. Gandom, “Bat algorithm: a novel approach
vehicle routing problem with time window and traveling for global engineering optimization,” Engineering Computa-
tions, vol. 29, pp. 464–483, 2012.
salesman problem, which is our further work.
[14] S. Mishra, K. Shaw, and D. Mishra, “A new meta-heuristic bat
inspired classification approach for microarray data,” Procedia
Acknowledgments Technology, vol. 4, pp. 802–806, 2012.
[15] J. Xie, Y. Q. Zhou, and H. Chen, “A novel bat algorithm based on
This work is supported by National Science Foundation of
differential operator and Lévy-flights trajectory,” Computational
China under Grant no. 61165015, Key Project of Guangxi Intelligence and Neuroscience, vol. 2013, Article ID 453812, 13
Science Foundation under Grant no. 2012GXNSFDA053028, pages, 2013.
Key Project of Guangxi High School Science Foundation
[16] G. Wang, L. Guo, H. Duan, L. Liu, and H. Wang, “A bat algo-
under Grant no. 20121ZD008, the Funded by Open Research rithm with mutation for UCAV path planning,” The Scientific
Fund Program of Key Lab of Intelligent Perception and World Journal, vol. 2012, Article ID 418946, 15 pages, 2012.
Image Understanding of Ministry of Education of China [17] T. A. Feo and M. G. C. Resende, “Greedy randomized adaptive
under Grant no. IPIU01201100, and the Innovation Project of search procedures,” Journal of Global Optimization, vol. 6, no. 2,
Guangxi Graduate Education under Grant no. YCSZ2012063. pp. 109–133, 1995.
[18] M. G. C. Resendel and C. C. Ribeiro, “GRASP with path-
References relinking: recent advances and applications,” in Metaheuristics:
Progress as Real Problem Solvers, pp. 29–63, Springer, 2005.
[1] G. B. Dantzig and J. H. Ramser, “The truck dispatching prob- [19] P. Festa and M. G. C. Resende, “An annotated bibliography
lem,” Management Science, vol. 6, pp. 80–91, 1959. of GRASP. I. Algorithms,” International Transactions in Oper-
[2] M. Haimovich, A. H. G. Rinnooy Kan, and L. Stougie, “Analysis ational Research, vol. 16, no. 1, pp. 1–24, 2009.
of heuristics for vehicle routing problems,” in Vehicle Routing: [20] P. Festa and M. G. C. Resende, “An annotated bibliography of
Methods and Studies, B. L. Golden and A. A. Assad, Eds., vol. 16, GRASP. II. Applications,” International Transactions in Opera-
pp. 47–61, North-Holland, Amsterdam, The Netherlands, 1988. tional Research, vol. 16, no. 2, pp. 131–172, 2009.
[3] P. Toth and A. Tramontani, “An integer linear programming [21] M. Mestria, L. S. Ochi, and S. de Lima Martins, “GRASP with
local search for capacitated vehicle routing problems,” in The path relinking for the symmetric euclidean clustered traveling
Vehicle Routing Problem: Latest Advances and New Challenges, salesman problem,” Computers and Operations Research, 2012.
pp. 275–295, Springer, New York, NY, USA, 2008. [22] M. G. C. Resende, C. C. Ribeiro, F. Glover, and R. Mart, “Scatter
[4] H. Nazif and L. S. Lee, “Optimised crossover genetic algorithm search and path-relinking: fundamentals, advances, and appli-
for capacitated vehicle routing problem,” Applied Mathematical cations,” in Handbook of Metaheuristics, M. Gendreau and J. Y.
Modelling, vol. 36, no. 5, pp. 2110–2117, 2012. Potvin, Eds., pp. 87–107, Springer, Boston, Mass, USA, 2010.
[5] T. J. Ai and V. Kachitvichyanukul, “Particle swarm optimization [23] F. Glover, “Tabu search and adaptive memory programming-
and two solution representations for solving the capacitated advances, applications and challenges,” in Interfaces in Com-
vehicle routing problem,” Computers and Industrial Engineering, puter Science and Operations Research, R. S. Barr, R. V. Helgason,
vol. 56, no. 1, pp. 380–387, 2009. and J. L. Kennington, Eds., pp. 1–75, Kluwer, 1996.
[24] M. Laguna and R. Martı́, “GRASP and path relinking for 2-layer
[6] W. Y. Szeto, Y. Wu, and S. C. Ho, “An artificial bee colony algo-
straight line crossing minimization,” Journal on Computing, vol.
rithm for the capacitated vehicle routing problem,” European
11, no. 1, pp. 44–52, 1999.
Journal of Operational Research, vol. 215, no. 1, pp. 126–135, 2011.
[7] P. Chen, H.-K. Huang, and X.-Y. Dong, “Iterated variable neigh-
borhood descent algorithm for the capacitated vehicle routing
problem,” Expert Systems with Applications, vol. 37, no. 2, pp.
1620–1627, 2010.
[8] A. Yurtkuran and E. Emel, “A new hybrid electromagnetism-
like algorithm for capacitated vehicle routing problems,” Expert
Systems with Applications, vol. 37, no. 4, pp. 3427–3433, 2010.
[9] A. A. Juan, J. Faulin, R. Ruiz, B. Barrios, and S. Caballé, “The
SR-GCWS hybrid algorithm for solving the capacitated vehicle
routing problem,” Applied Soft Computing Journal, vol. 10, no. 1,
pp. 215–224, 2010.
Copyright of Mathematical Problems in Engineering is the property of Hindawi Publishing
Corporation and its content may not be copied or emailed to multiple sites or posted to a
listserv without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.