SlideShare a Scribd company logo
1 of 6
Download to read offline
Jun Li, Senior Member,
IEEE
School of Automation
Southeast University
Nanjing, China
j.li@seu.edu.cn
Qirui Sun
School of Automation
Southeast University
Nanjing, China
seusunqr@163.com
MengChu Zhou, Fellow IEEE
Dept. Electrical and Computer
Engineering
New Jersey Institute of Technology
New Jersey, USA
zhou@njit.edu
Xianzhong Dai
School of Automation
Southeast University
Nanjing, China
xzdai@seu.edu.cn
Abstract—This work formulates for the first time a multiple
traveling salesman problem (MTSP) with ordinary and exclusive
cities, denoted by MTSP* for short. In the original MTSP, a city
can be visited by any traveling salesman and is thus renamed as
an ordinary one in MTSP*. A new class of cities is introduced in
MTSP*, called exclusive ones. They are divided into groups, each
of which can be exclusively visited by a specified or pre-
determined salesman. To solve MTSP*, a genetic algorithm is
presented. It encodes cities and salesman into two single
chromosomes. Accordingly, three modes of crossover and
mutation operators are designed, i.e., simple city crossover and
mutation (CCM), simple salesman crossover and mutation, and
mixed city-salesman crossover and mutation. All the operations
of crossover and mutation follow the proper relationship between
cities and salesman. With the help of an MTSP* example, the
performance of the proposed algorithm with three modes of
crossover and mutation operators is compared and analyzed. The
simulation results show that the algorithm can solve MTSP* with
rapid convergence with CCM being the best mode of the
operators.
Keywords-Multiple Traveling Salesman Problem; Genetic
Algorithm; optimization
I. INTRODUCTION
The traveling salesman problem (TSP) is a basic routing
problem [1][2]. Its objective is to find the shortest route for a
traveling salesman who, starting from the depot city, has to
visit every city on a given list exactly once and then return to
the depot city. It is NP-hard. The multiple traveling salesman
problem (MTSP) as the generalization of TSP involves
scheduling m (>1) salesmen to visit a set of n (>m) cities such
that each city is visited exactly once and the total travelled
route is the shortest [3] [4]. Similar to TSP, it is also NP-hard.
All cities in MTSP are identical for any salesman in term of
accessibility. Namely, each city allows any salesman to visit.
However, under some circumstances, a portion of cities allow
any salesman to visit while the rest do not but are partitioned
into several groups, each of which allows a specified salesman
to visit. In other words, the salesmen reach an agreement to
carve the cities into several exclusive “spheres of influence” or
called exclusive city groups and a common “sphere of
influence” or called ordinary city groups. The former allows
only a specific/pre-determined salesman to visit exclusively,
while the latter can be visited by all salesmen. The ordinary
group can be viewed as the intersection of all salesmen’s
accessible zones. The previous study of a dual-bridge water-jet
cutting system [5], as shown in Fig. 1, is a typical example of
this problem. Each bridge system has its exclusive processing
area with outlines to be cut (exclusive cities) while there is a
common processing area with outlines (ordinary cities)
between the two adjacent bridge systems that can be visited by
either bridge.
Figure 1. Dual-bridge water-jet cutting system
We call this problem as a multiple traveling salesman
problem with ordinary and exclusive cities (MTSP*).
MTSP* is NP-hard like MTSP. To solve such a problem
of realistic size, a heuristic approach is a good option. The
genetic algorithm (GA) presented by Holland in the early
1970s is a highly parallel, random and self-adaptive search
algorithm based on the mechanism of the natural selection and
genetics inspired by Darwin’s Theory. This algorithm is proven
to be effective and has been widely used to solve TSP [7][8]
and MTSP [9][10]. This paper proposes a GA-based approach
to solve MTSP*.
The remainder of this paper is organized as follows. The
next section defines the single depot MTSP* with an example.
Section 3 designs the GA and the simulation is conducted in
Section 4. The paper is concluded in Section 5.
A New Multiple Traveling Salesman Problem
and its Genetic Algorithm-based Solution
2013 IEEE International Conference on Systems, Man, and Cybernetics
978-1-4799-0652-9/13 $31.00 © 2013 IEEE
DOI
627
2013 IEEE International Conference on Systems, Man, and Cybernetics
978-1-4799-0652-9/13 $31.00 © 2013 IEEE
DOI 10.1109/SMC.2013.112
627
II. PROBLEM STATEMENT
MTSP* with m salesmen and n cities can be formally
defined over a complete graph ( , )
G V E , where the vertex
set 1 2
{ , , , }
n
V v v v
" corresponding to cities is partitioned into
m+1 sets 0 1
, , , m
V V V
" , and each edge ( , ) ,
i j
v v E i j
 z , is
associated with a weight ij
c C
 that represents a visit cost
(distance) from cities i to j . Its objective is to determine m
sequences of Hamiltonian cycles or circuits on G with the least
total cost such that any vertex of each exclusive set is visited
exactly once by a specified salesman and any vertex of the
public set is visited by any salesman exactly once.
The existence of exclusive and ordinary city groups,
makes MTSP* different from TSP and MTSP. A vertex from
an exclusive group in MTSP* can be next to another in the
same or ordinary group. Similarly, a vertex in an ordinary
group can be next to another in the same or an exclusive one.
MTSP* is not a simple composition of TSP and MTSP.
Figure 2. Example of MTSP*
An example MTSP* is shown Fig. 2. It has 4 salesmen
and 51 cities where city 0 is the depot. It is modified from the
classic example-eil51 by preserving its original city coordinate
information. 1
V represents the exclusive city s to be
exclusively visited by salesman 1, which contains cities 1-7;
2
V represents the exclusive city set 2 containing cities 8-14 to
be visited by salesman 2; 3
V represents the exclusive one by
salesman 3, which contains cities 15-22; 4
V the exclusive one
by salesman 4, which includes cities 23-30; and 0
V represents
the ordinary city set for all salesmen, which contains cities 31-
50. The objective of this problem is to determine a minimum
route that all salesmen start form the depot and finally return
to it. Meanwhile, each exclusive city must be visited exactly
once by a specified salesman and each ordinary city is visited
exactly once by any salesman.
Fig. 2 shows a feasible but may not be the best route.
Routes 1-4 for salesmen 1-4 are˖
1) 0ė35ė39ė4ė1ė2ė7ė5ė3ė49ė6ė45ė41ė
31ė0;
2) 040473810128141193413
48420;
3) 03236162017221521181943
33370;
4) 0ė24ė44ė30ė26ė25ė27ė23ė29ė28ė46ė
50ė0.
Ordinary city 44 can be next to exclusive city 24 in route 4
and exclusive city 6 next to ordinary city 49 in route 1.
III. GA DESIGN
A genetic algorithm (GA) is used to search solutions based
on the evolutionary principle. Since it was firstly introduced
by Holland [6], it has been successfully applied to a variety of
NP-hard combinatorial optimization problems, such as TSP
and MTSP. To apply GA to solve MTSP*, the focus is to
establish an effective coding and decoding scheme and design
suitable selection, crossover and mutation operators to ensure
better population evolution.
A. Chromosome coding for MTSP*
Chromosome representation is a crucial basic work when
applying a GA. It directly determines the performance of the
algorithm. The GAs designed to solve MTSP of m traveling
salesman n cities mainly exploit three encoding schemes. The
first is the one-chromosome scheme. It uses a single
chromosome of length n+m-1. In it, n cities are represented by
a permutation of integers from 1 to n. This permutation is
partitioned into m sub-tours by the insertion of m-1 negative
integers from -1 to -(m-1) that represent the transition from
one salesman to the next. The second is the two-part-
chromosome technique proposed by Arthur and Cliff [9]. The
first part of a chromosome is a permutation of n cities while
the second part is of length m and represents the number of
cities assigned to each of m salesmen. The values assigned to
the second part are constrained to be m positive integers that
must sum to the number of cities to be visited (n). The third
one is the two-chromosome technique. It requires two
chromosomes, each being of length n, to represent a solution.
The first one provides a permutation of n cities while the
second one assigns a salesman to each of the cities in the
corresponding position of the first one.
For MTSP*, it is d to use the first two schemes mentioned
above. The third one with two types of chromosomes can be
adapted to MTSP* via some modification. Particularly, the
city chromosome consists of a permutation of integers from 1
to n while a genetic value of salesman chromosome is the
number of a salesman that corresponds to an exclusive or
ordinary city in the same position of the city chromosome. The
exclusive city is assigned to the specified salesman, and the
ordinary city is randomly assigned to the salesman.
A coding example of MTSP* with three salesmen and ten
cities is shown Fig. 3, where six cities are exclusive ones and
three salesmen are assigned three exclusive ones. The rest
cities are ordinary. Namely, exclusive cities of salesmen 1-3
are 1-2, and 3-4, and 5-6. Respectively, ordinary cities are
cities 7-10. According to two chromosomes in Fig. 3, cities 2,
7 and 1 (in that order) are visited by salesman 1. Similarly,
cities 10, 4 and 3 (in that order) are visited by salesman 2, and
cities 9, 5 and 6 (in that order) are visited by salesman 3.
628
628
Figure 3. Example of MTSP* coding
B. Selection operator
Roulette Wheel method and the elitist strategy [11] are
adopted as the selection operation in this work. The former
simply chooses a chromosome in a statistical fashion based
solely upon its fitness value. The elitist strategy copies the
individual with the best fitness at the present generation to the
next one. It can prevent the optimized individuals from being
eliminated after a selection, crossover or mutation operation. It
is critical to ensure the convergence of a GA. A GA containing
an elitist strategy is proven to be globally convergent.
C. Crossover operator
A crossover operator exchanges parts of the genes from
two parent individuals to form two new individuals. It is one
of important features that distinguish a genetic algorithm from
other ones. In MTSP a crossover operator may be one of
partially matched crossover (PMX), ordered crossover (OX),
cycle crossover (CX), two-point crossover, etc.
We design three modes of crossover operators
corresponding to the adopted chromosome coding style, i.e.,
city crossover (CC), salesman crossover (SC), and city-
salesman crossover (CSC).
1) City crossover
In this paper, we ameliorate PMX as a city crossover
operator. It requires randomly selecting two crossing points to
determine a matching section. The corresponding matching
sections in two parents are swapped, thereby resulting in two
new descendants. Then two new individuals are checked if the
exclusive cities are assigned to the specified salesman. If not,
the particular genes in a salesman individual should be
corrected.

Figure 4. Example of CC
The crossover of two chromosomes is shown in Fig. 4. In
Step 1, given two parents, we randomly select a section of a
city individual, then swap its genes with those of another
individual and produce two individual of descendants as
shown in Step 2. The mapping relationship of the selected
sections in two city individuals is 8—3, 9—8, 5—2, 4—7, 7—
1, and 1—10. In Step 3, exchange the redundant genes
according to the selected section, then find that exclusive cities
5, 3, 7, 1, and 6 in the left chromosome and cities 2, 5, and 4 in
the right one are assigned to the wrong salesmen. Then, the
exclusive cities are reassigned to the correct salesmen and two
reasonable generations are produced as shown in Step 4.
2) Salesman crossover
To avoid a number of duplicate genes appearing in a
salesman chromosome, this work adopts traditional two-point
crossover. It also requires randomly determining a matching
section. Then the corresponding sections of two chromosomes
are swapped to generate two descendants. At end, the
matching relationship between the exclusive cities should be
checked and a wrong salesman should be corrected to the
specified one.
An SC process is shown in Fig. 5. In Step 1, there are two
parents. In particular, the randomly selected matching sections
are marked in gray color in the salesman chromosome. After
swapping of the two marked sections, two descendants are
produced in Step 2. However, it is obvious that exclusive cities
5 and 4 of the left chromosome and exclusive cities 3, 2, and 1
of the right chromosome are assigned to the wrong salesmen.
In Step 3, it is corrected by reassigning exclusive cities to the
specified salesmen and obtained two correct descendants.

Figure 5. Example of SC
3) City-salesman crossover
For CSC, a city chromosome applies PMX, while a
salesman chromosome adopts the two-point crossover scheme.

Figure 6. Example of CSC
An example of CSC is shown in Fig. 6. In Step 1, we
select for each kind of parents a couple of random crossing
sections and swap each couple of sections, respectively. It
results in two new city generations and two new salesman
629
629
ones as shown in Step 2. In Step 3, exchange the redundant
genes of the city chromosome, and find that exclusive cities 5
and 3 of the left city chromosome and 2 of the right city one
are assigned to a wrong salesman. After the correction of
salesman genes, the result is shown in Step 4.
D. Mutation operator
A mutation operator plays an import role in improving
local search ability and maintaining variability of the
population. It also prevents the premature termination in GA.
This work adopts swapping mutation. It requires random
selection of two crossing points, and then swaps the selected
points. Finally, check whether the exclusive cities of a new
descendant match with their corresponding salesmen.
Corresponding to the above three crossover schemes, we
design three mutation operators, i.e., city mutation (CM),
salesman mutation (SM), and city-salesman mutation (CSM).
1) City mutation
In CM, only a city-chromosome applies swapping
mutation. For example, first, two selected swapping gene
points are cities 8 and 7, as shown in Fig. 7. After swapping
them, the exclusive cities are proven to be in accord with
salesmen and the mutation is over.

Figure 7. Example of CM
2) Salesman mutation
In the operation of SM swapping mutation only applied for
a salesman chromosome. An example is given in Fig. 8, where
the selected swapping points are genes 2 and 1 marked in gray
color in the salesman-chromosome. A correct descendant is
generated by swapping them.

Figure 8. Example of SM
3) City-salesman mutation
Swapping mutation takes place in both city and salesman
chromosomes in CSM. An example is shown in Fig. 9, where
the swapping genes marked in gray are swapped pairwise,
which results in new feasible generations.
Considering that crossover and mutation should not apply
into different types of chromosomes, this work selects three
compositions of reasonable crossover and mutation operators
from nine ones. By selecting CC as crossover and CM as
mutation operator, we have CC  CM (called CCM), SC 
SM (called SCM), and CSC  CSM (called CSCM).

Figure 9. Example of CSM
E. Fitness function
A fitness function is used in a GA to judge the chance that
an individual (a route) can be selected into the next generation.
It is a limiting factor to the efficiency of evolution. For a GA,
many selection strategies based on the proportion of fitness
require a non-negative fitness and the larger fitness the better
individual. Hence, for a problem with a minimum solution as
its optimization objective, it needs to turn it to a maximum one.
MTSP* takes the minimum length of all salesmen as its
optimization objective. Hence this work takes the reciprocal of
the length as its fitness.
Taking f(x) as the length of the solution, the fitness
function is given as:
1
( )=
1+ ( )
F x
f x
IV. SIMULATION AND RESULT
This paper takes the revised eil51 as an example given in
Section 2 to verify the correctness and performance of our GA
with the three pairs of crossover and mutation operators.
A. Convergence of GA
Convergence is an important indicator to show the
performance of a GA. If a GA is convergent, it indicates its
stability and evolution towards a correct direction. The quality
of the solution is better and better as the evolution goes on. To
examine a GA’s convergence, we conduct the following
experiment, where the generation count is 2000, the crossover
probability 0.6, mutation probability 0.1, the size of population
is 50, and CCM is selected.
The result as shown in Fig. 10 indicates that the
convergence of our GA is good without considerable
fluctuation. The length of total routes is optimized from
1198.23 km of the initial population to 558.511 km of the
2000-th generation. It implies that the effect of optimization is
notable. The evolution process can be divided into three main
stages. The first one is from the initial generation to the 560-th
one, where the GA convergence is the fastest, and the total
length of routes descends from 1198.23 km to 666.21 km. The
second one from the 561-th to 1560-th generations shows a
630
630
slow change of the total length form 666.21 km to 558.51 km.
In the last one from the 1561-th to 2000-th generations, the
result tends to be stable. Hence, our GA has a good
convergence. For the other two pairs of crossover and
mutation operators, i.e., SCM and CSCM, we can obtain the
similar results.
Figure 10. Result of the example of MTSP*
B. Comparison
To compare the performance of our GA with different
crossover and mutation operators, three groups of experiments
are designed where three pairs of crossover and mutation
operators are adopted in turn. The other parameters are set, i.e.,
the size of population 30, the crossover probability 0.6, the
mutation probability 0.1, and the generation count 2000. Each
experiment is carried out for ten times by using a DELL
Inspiron620s computer with Windows 7 and Inter Core i3
CPU at 3.30GHZ. The data are from the revised eil51 example
in Section 2. The experimental result is shown in TABLE 
and Fig. 11.
TABLE I. Result of the three group experiments Unit˖Km
Group CCM SCM CSCM
1 558.25 575.26 568.45
2 556.85 576.45 572.12
3 561.47 580.56 573.89
4 565.25 576.65 567.45
5 558.36 570.25 570.35
6 557.68 590.68 566.25
7 562.45 586.45 576.87
8 568.56 573.89 565.54
9 556.35 584.57 575.58
10 560.58 571.64 566.4
Average 560.58 578.64 570.29
Figure 11. Result of different operators
The best, the worst and the average solutions are observed
and shown in TABLE .
TABLE II. Data analysis of each operator
Crossover and
mutation
operator
best solution
(km)
worst
solution
(km)
Average
solution
(km)
CCM 556.35 568.56 560.58
SCM 570.25 590.68 578.64
CSCM 565.54 576.87 570.29
It shows that the performance of CCM is the best among
three pairs of crossover and mutation operators while that of
SCM is the worst. By observing the results we find that by
SCM, crossover and mutation merely take place in the
salesman chromosomes. It results in a small solution space and
weakens genic recombination due to large duplicate salesman
individuals. Using CCM only the city chromosomes are
crossed and mutated and there is no duplicate gene. Thus the
extent of genic recombination is stronger and the solution
space is larger than that of SCM. Namely, the population
diversity of CCM is better than that of SCM, as shown in the
experiment data. Both a city chromosome and a salesman
chromosome are crossed and mutated by CSCM. The
dynamics of genic recombination is stronger than SCM. Hence,
the performance of CSCM is better than that of SCM.
However, crossover and mutation of both city chromosomes
and salesman chromosomes may destroy the better solution
due to too strong dynamics of genic recombination. Thus,
CSCM is worse than CCM as observed. As for time
consumption, three groups have small differences from each
other. Therefore, we should select CCM in a GA for MTSP*
.
V. CONCLUSIONS
This paper presents a new multiple traveling salesman
problems with exclusive and ordinary cities. To solve it, we
design a genetic algorithm. It revises the two-chromosome
scheme and design three pairs of crossover and mutation
operators, i.e., simple city crossover and mutation, simple
salesman crossover and mutation, and mixed city-salesman
631
631
crossover and mutation. To show the performance of our
algorithm with different operators, we design three groups of
experiments. The results indicate that the proposed GA is
suitable to solve MTSP* and CCM deploying city crossover
and city mutation has the best performance among three
compositions of crossover and mutation operators.
VI. ACKNOWLEDGEMENT
This work was supported in part by the China National
Natural Science Foundation under Grants 61004035 and
61175113.
REFERENCES
[1] E. Lawler, J. Lenstra, A. Rinnooy, D. Shinmoys. The Traveling
Salesman Problem. John Whiley  Sons, vol. 25, no. 4, pp.251-258,
2006
[2] G. Gutin, A. Punnen. The Traveling Salesman Problem and Its
Variations. Dordrecht: Kluwer, vol. 35, no. 2, pp.625-629, 2002
[3] Hou MS, Liu, DB. A novel method for solving the multiple traveling
salesmen problem with multiple depots. Chinese science bulletin,
vol.57, no.2, pp.1886-1892, 2012
[4] T. Bektas. The multiple traveling salesman problem: An overview of
formulations and solution procedures. Omega, vol. 34, no. 3, pp.209-
219, 2006
[5] Jun Li, Qirui Sun, Xianzhong Dai. A Coordination and Optimization
Method for Multi-Bridge Water jet Cutting processes. 42nd
International Conference on Computers and Industrial Engineering,
Cape Down, South Africa, 2012
[6] J. Holland. Adaptation in Natural and Artificial Systems. the University
of Michigan Press, Ann Arbour, USA, 1975
[7] J. Potvin. Genetic algorithm for the traveling salesman problem.
Annals of Operations Research, vol.63, no. 4, pp.330-370, 1999
[8] M. Sallabi, E. Haddad. An Improved Genetic Algorithm to Solve the
Traveling Salesman Problem. World Academy of Science, Engineering
and Technology, vol. 52, no.3, pp.471-474, 2009
[9] E. Carter, T. Ragsdale. A new approach to solving the multiple
traveling salesperson problem using genetic algorithms. European
Journal of Operation Research, vol. 175, no.2, pp.246-257, 2006
[10] Wei Zhou, Yuanzong Li. An Improved Genetic Algorithm for Multiple
Traveling Salesman Problem. 2010 2nd International Asia Conference
on Informatics in Control, Automation and Robotics, Wuhan Hubei,
China, pp.493-495, 2010.
[11] A. De Jong. An analysis of the behavior of a class of genetic adaptive
systems. Michigan: University of Michigan, 1975
632
632

More Related Content

Similar to A new multiple traveling salesman problem

Solving capacity problems as asymmetric
Solving capacity problems as asymmetricSolving capacity problems as asymmetric
Solving capacity problems as asymmetricijaia
 
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...cscpconf
 
A study and implementation of the transit route network design problem for a ...
A study and implementation of the transit route network design problem for a ...A study and implementation of the transit route network design problem for a ...
A study and implementation of the transit route network design problem for a ...csandit
 
A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...
A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...
A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...April Smith
 
Travelling salesman problem using genetic algorithms
Travelling salesman problem using genetic algorithms Travelling salesman problem using genetic algorithms
Travelling salesman problem using genetic algorithms Shivank Shah
 
Optimal Vacation Itinerary Modeling
Optimal Vacation Itinerary ModelingOptimal Vacation Itinerary Modeling
Optimal Vacation Itinerary ModelingGerard Trimberger
 
Algorithms And Optimization Techniques For Solving TSP
Algorithms And Optimization Techniques For Solving TSPAlgorithms And Optimization Techniques For Solving TSP
Algorithms And Optimization Techniques For Solving TSPCarrie Romero
 
Hybrid iterated local search algorithm for optimization route of airplane tr...
Hybrid iterated local search algorithm for optimization route of  airplane tr...Hybrid iterated local search algorithm for optimization route of  airplane tr...
Hybrid iterated local search algorithm for optimization route of airplane tr...IJECEIAES
 
The Traveling Salesman problem ppt.pptx
The Traveling Salesman problem ppt.pptxThe Traveling Salesman problem ppt.pptx
The Traveling Salesman problem ppt.pptxHalimFerchichi
 
tAn improved genetic algorithm with a
tAn improved genetic algorithm with atAn improved genetic algorithm with a
tAn improved genetic algorithm with aijcseit
 
AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...
AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...
AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...ijcseit
 
Using Genetic Algorithm for Shortest Path Selection with Real Time Traffic Flow
Using Genetic Algorithm for Shortest Path Selection with Real Time Traffic FlowUsing Genetic Algorithm for Shortest Path Selection with Real Time Traffic Flow
Using Genetic Algorithm for Shortest Path Selection with Real Time Traffic FlowIJCSIS Research Publications
 
TRUSTS Mobile App Demo Poster (AAMAS 2013)
TRUSTS Mobile App Demo Poster (AAMAS 2013)TRUSTS Mobile App Demo Poster (AAMAS 2013)
TRUSTS Mobile App Demo Poster (AAMAS 2013)Samantha Luber
 
Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...
Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...
Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...csandit
 
An Improved AC-BM Algorithm for Monitoring Watch List
An Improved AC-BM Algorithm for Monitoring Watch ListAn Improved AC-BM Algorithm for Monitoring Watch List
An Improved AC-BM Algorithm for Monitoring Watch ListNooria Sukmaningtyas
 

Similar to A new multiple traveling salesman problem (20)

Solving capacity problems as asymmetric
Solving capacity problems as asymmetricSolving capacity problems as asymmetric
Solving capacity problems as asymmetric
 
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
 
A study and implementation of the transit route network design problem for a ...
A study and implementation of the transit route network design problem for a ...A study and implementation of the transit route network design problem for a ...
A study and implementation of the transit route network design problem for a ...
 
A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...
A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...
A New Approach To Solving The Multiple Traveling Salesperson Problem Using Ge...
 
K046036367
K046036367K046036367
K046036367
 
Travelling salesman problem using genetic algorithms
Travelling salesman problem using genetic algorithms Travelling salesman problem using genetic algorithms
Travelling salesman problem using genetic algorithms
 
Optimal Vacation Itinerary Modeling
Optimal Vacation Itinerary ModelingOptimal Vacation Itinerary Modeling
Optimal Vacation Itinerary Modeling
 
Traveling Eco-Salesman
Traveling Eco-SalesmanTraveling Eco-Salesman
Traveling Eco-Salesman
 
Algorithms And Optimization Techniques For Solving TSP
Algorithms And Optimization Techniques For Solving TSPAlgorithms And Optimization Techniques For Solving TSP
Algorithms And Optimization Techniques For Solving TSP
 
Hybrid iterated local search algorithm for optimization route of airplane tr...
Hybrid iterated local search algorithm for optimization route of  airplane tr...Hybrid iterated local search algorithm for optimization route of  airplane tr...
Hybrid iterated local search algorithm for optimization route of airplane tr...
 
The Traveling Salesman problem ppt.pptx
The Traveling Salesman problem ppt.pptxThe Traveling Salesman problem ppt.pptx
The Traveling Salesman problem ppt.pptx
 
50120140503015
5012014050301550120140503015
50120140503015
 
tAn improved genetic algorithm with a
tAn improved genetic algorithm with atAn improved genetic algorithm with a
tAn improved genetic algorithm with a
 
Optimal combination of operators in Genetic Algorithmsfor VRP problems
Optimal combination of operators in Genetic Algorithmsfor VRP problemsOptimal combination of operators in Genetic Algorithmsfor VRP problems
Optimal combination of operators in Genetic Algorithmsfor VRP problems
 
AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...
AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...
AN IMPROVED GENETIC ALGORITHM WITH A LOCAL OPTIMIZATION STRATEGY AND AN EXTRA...
 
The Short-term Swap Rate Models in China
The Short-term Swap Rate Models in ChinaThe Short-term Swap Rate Models in China
The Short-term Swap Rate Models in China
 
Using Genetic Algorithm for Shortest Path Selection with Real Time Traffic Flow
Using Genetic Algorithm for Shortest Path Selection with Real Time Traffic FlowUsing Genetic Algorithm for Shortest Path Selection with Real Time Traffic Flow
Using Genetic Algorithm for Shortest Path Selection with Real Time Traffic Flow
 
TRUSTS Mobile App Demo Poster (AAMAS 2013)
TRUSTS Mobile App Demo Poster (AAMAS 2013)TRUSTS Mobile App Demo Poster (AAMAS 2013)
TRUSTS Mobile App Demo Poster (AAMAS 2013)
 
Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...
Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...
Hybrid Ant Colony Optimization for Real-World Delivery Problems Based on Real...
 
An Improved AC-BM Algorithm for Monitoring Watch List
An Improved AC-BM Algorithm for Monitoring Watch ListAn Improved AC-BM Algorithm for Monitoring Watch List
An Improved AC-BM Algorithm for Monitoring Watch List
 

Recently uploaded

Team Dynamics: A Journey to Excellence
Team Dynamics: A Journey to ExcellenceTeam Dynamics: A Journey to Excellence
Team Dynamics: A Journey to ExcellenceGDSC PJATK
 
Dehradun🌹Vip ℂall Girls ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...
Dehradun🌹Vip ℂall Girls  ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...Dehradun🌹Vip ℂall Girls  ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...
Dehradun🌹Vip ℂall Girls ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...Mumbai Escorts
 
原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样
原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样
原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样rtgdfgss
 
TEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdf
TEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdfTEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdf
TEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdfmarcuskenyatta275
 
Marketing Management 16 Global Edition by Philip Kotler test bank.docx
Marketing Management 16 Global Edition by Philip Kotler test bank.docxMarketing Management 16 Global Edition by Philip Kotler test bank.docx
Marketing Management 16 Global Edition by Philip Kotler test bank.docxssuserf63bd7
 
Mastering Compassion: A Heavenly Perspective on Leadership
Mastering Compassion: A Heavenly Perspective on LeadershipMastering Compassion: A Heavenly Perspective on Leadership
Mastering Compassion: A Heavenly Perspective on LeadershipbluetroyvictorVinay
 
UX in an Agile World - Scrum Gathering
UX in an Agile World -   Scrum GatheringUX in an Agile World -   Scrum Gathering
UX in an Agile World - Scrum GatheringKaizenko
 
Leading People - Harvard Manage Mentor Certificate
Leading People - Harvard Manage Mentor CertificateLeading People - Harvard Manage Mentor Certificate
Leading People - Harvard Manage Mentor CertificateVICTOR MAESTRE RAMIREZ
 
Management 13th Edition by Richard L. Daft test bank.docx
Management 13th Edition by Richard L. Daft test bank.docxManagement 13th Edition by Richard L. Daft test bank.docx
Management 13th Edition by Richard L. Daft test bank.docxssuserf63bd7
 
W.H.Bender Quote 63 You Must Plan T.O.P Take-Out Packaging
W.H.Bender Quote 63 You Must Plan T.O.P Take-Out PackagingW.H.Bender Quote 63 You Must Plan T.O.P Take-Out Packaging
W.H.Bender Quote 63 You Must Plan T.O.P Take-Out PackagingWilliam (Bill) H. Bender, FCSI
 
DrupalCamp Atlanta 2022 - Effective Project Management
DrupalCamp Atlanta 2022 - Effective Project ManagementDrupalCamp Atlanta 2022 - Effective Project Management
DrupalCamp Atlanta 2022 - Effective Project ManagementNorah Medlin
 

Recently uploaded (11)

Team Dynamics: A Journey to Excellence
Team Dynamics: A Journey to ExcellenceTeam Dynamics: A Journey to Excellence
Team Dynamics: A Journey to Excellence
 
Dehradun🌹Vip ℂall Girls ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...
Dehradun🌹Vip ℂall Girls  ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...Dehradun🌹Vip ℂall Girls  ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...
Dehradun🌹Vip ℂall Girls ❤Heer 931579xxxx💟 Full Trusted ℂALL GIRLS IN Dehradu...
 
原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样
原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样
原版定制(UBC毕业证书)加拿大英属哥伦比亚大学毕业证原件一模一样
 
TEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdf
TEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdfTEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdf
TEST BANK for Operations Management, 14th Edition by William J. Stevenson,.pdf
 
Marketing Management 16 Global Edition by Philip Kotler test bank.docx
Marketing Management 16 Global Edition by Philip Kotler test bank.docxMarketing Management 16 Global Edition by Philip Kotler test bank.docx
Marketing Management 16 Global Edition by Philip Kotler test bank.docx
 
Mastering Compassion: A Heavenly Perspective on Leadership
Mastering Compassion: A Heavenly Perspective on LeadershipMastering Compassion: A Heavenly Perspective on Leadership
Mastering Compassion: A Heavenly Perspective on Leadership
 
UX in an Agile World - Scrum Gathering
UX in an Agile World -   Scrum GatheringUX in an Agile World -   Scrum Gathering
UX in an Agile World - Scrum Gathering
 
Leading People - Harvard Manage Mentor Certificate
Leading People - Harvard Manage Mentor CertificateLeading People - Harvard Manage Mentor Certificate
Leading People - Harvard Manage Mentor Certificate
 
Management 13th Edition by Richard L. Daft test bank.docx
Management 13th Edition by Richard L. Daft test bank.docxManagement 13th Edition by Richard L. Daft test bank.docx
Management 13th Edition by Richard L. Daft test bank.docx
 
W.H.Bender Quote 63 You Must Plan T.O.P Take-Out Packaging
W.H.Bender Quote 63 You Must Plan T.O.P Take-Out PackagingW.H.Bender Quote 63 You Must Plan T.O.P Take-Out Packaging
W.H.Bender Quote 63 You Must Plan T.O.P Take-Out Packaging
 
DrupalCamp Atlanta 2022 - Effective Project Management
DrupalCamp Atlanta 2022 - Effective Project ManagementDrupalCamp Atlanta 2022 - Effective Project Management
DrupalCamp Atlanta 2022 - Effective Project Management
 

A new multiple traveling salesman problem

  • 1. Jun Li, Senior Member, IEEE School of Automation Southeast University Nanjing, China j.li@seu.edu.cn Qirui Sun School of Automation Southeast University Nanjing, China seusunqr@163.com MengChu Zhou, Fellow IEEE Dept. Electrical and Computer Engineering New Jersey Institute of Technology New Jersey, USA zhou@njit.edu Xianzhong Dai School of Automation Southeast University Nanjing, China xzdai@seu.edu.cn Abstract—This work formulates for the first time a multiple traveling salesman problem (MTSP) with ordinary and exclusive cities, denoted by MTSP* for short. In the original MTSP, a city can be visited by any traveling salesman and is thus renamed as an ordinary one in MTSP*. A new class of cities is introduced in MTSP*, called exclusive ones. They are divided into groups, each of which can be exclusively visited by a specified or pre- determined salesman. To solve MTSP*, a genetic algorithm is presented. It encodes cities and salesman into two single chromosomes. Accordingly, three modes of crossover and mutation operators are designed, i.e., simple city crossover and mutation (CCM), simple salesman crossover and mutation, and mixed city-salesman crossover and mutation. All the operations of crossover and mutation follow the proper relationship between cities and salesman. With the help of an MTSP* example, the performance of the proposed algorithm with three modes of crossover and mutation operators is compared and analyzed. The simulation results show that the algorithm can solve MTSP* with rapid convergence with CCM being the best mode of the operators. Keywords-Multiple Traveling Salesman Problem; Genetic Algorithm; optimization I. INTRODUCTION The traveling salesman problem (TSP) is a basic routing problem [1][2]. Its objective is to find the shortest route for a traveling salesman who, starting from the depot city, has to visit every city on a given list exactly once and then return to the depot city. It is NP-hard. The multiple traveling salesman problem (MTSP) as the generalization of TSP involves scheduling m (>1) salesmen to visit a set of n (>m) cities such that each city is visited exactly once and the total travelled route is the shortest [3] [4]. Similar to TSP, it is also NP-hard. All cities in MTSP are identical for any salesman in term of accessibility. Namely, each city allows any salesman to visit. However, under some circumstances, a portion of cities allow any salesman to visit while the rest do not but are partitioned into several groups, each of which allows a specified salesman to visit. In other words, the salesmen reach an agreement to carve the cities into several exclusive “spheres of influence” or called exclusive city groups and a common “sphere of influence” or called ordinary city groups. The former allows only a specific/pre-determined salesman to visit exclusively, while the latter can be visited by all salesmen. The ordinary group can be viewed as the intersection of all salesmen’s accessible zones. The previous study of a dual-bridge water-jet cutting system [5], as shown in Fig. 1, is a typical example of this problem. Each bridge system has its exclusive processing area with outlines to be cut (exclusive cities) while there is a common processing area with outlines (ordinary cities) between the two adjacent bridge systems that can be visited by either bridge. Figure 1. Dual-bridge water-jet cutting system We call this problem as a multiple traveling salesman problem with ordinary and exclusive cities (MTSP*). MTSP* is NP-hard like MTSP. To solve such a problem of realistic size, a heuristic approach is a good option. The genetic algorithm (GA) presented by Holland in the early 1970s is a highly parallel, random and self-adaptive search algorithm based on the mechanism of the natural selection and genetics inspired by Darwin’s Theory. This algorithm is proven to be effective and has been widely used to solve TSP [7][8] and MTSP [9][10]. This paper proposes a GA-based approach to solve MTSP*. The remainder of this paper is organized as follows. The next section defines the single depot MTSP* with an example. Section 3 designs the GA and the simulation is conducted in Section 4. The paper is concluded in Section 5. A New Multiple Traveling Salesman Problem and its Genetic Algorithm-based Solution 2013 IEEE International Conference on Systems, Man, and Cybernetics 978-1-4799-0652-9/13 $31.00 © 2013 IEEE DOI 627 2013 IEEE International Conference on Systems, Man, and Cybernetics 978-1-4799-0652-9/13 $31.00 © 2013 IEEE DOI 10.1109/SMC.2013.112 627
  • 2. II. PROBLEM STATEMENT MTSP* with m salesmen and n cities can be formally defined over a complete graph ( , ) G V E , where the vertex set 1 2 { , , , } n V v v v " corresponding to cities is partitioned into m+1 sets 0 1 , , , m V V V " , and each edge ( , ) , i j v v E i j  z , is associated with a weight ij c C  that represents a visit cost (distance) from cities i to j . Its objective is to determine m sequences of Hamiltonian cycles or circuits on G with the least total cost such that any vertex of each exclusive set is visited exactly once by a specified salesman and any vertex of the public set is visited by any salesman exactly once. The existence of exclusive and ordinary city groups, makes MTSP* different from TSP and MTSP. A vertex from an exclusive group in MTSP* can be next to another in the same or ordinary group. Similarly, a vertex in an ordinary group can be next to another in the same or an exclusive one. MTSP* is not a simple composition of TSP and MTSP. Figure 2. Example of MTSP* An example MTSP* is shown Fig. 2. It has 4 salesmen and 51 cities where city 0 is the depot. It is modified from the classic example-eil51 by preserving its original city coordinate information. 1 V represents the exclusive city s to be exclusively visited by salesman 1, which contains cities 1-7; 2 V represents the exclusive city set 2 containing cities 8-14 to be visited by salesman 2; 3 V represents the exclusive one by salesman 3, which contains cities 15-22; 4 V the exclusive one by salesman 4, which includes cities 23-30; and 0 V represents the ordinary city set for all salesmen, which contains cities 31- 50. The objective of this problem is to determine a minimum route that all salesmen start form the depot and finally return to it. Meanwhile, each exclusive city must be visited exactly once by a specified salesman and each ordinary city is visited exactly once by any salesman. Fig. 2 shows a feasible but may not be the best route. Routes 1-4 for salesmen 1-4 are˖ 1) 0ė35ė39ė4ė1ė2ė7ė5ė3ė49ė6ė45ė41ė 31ė0; 2) 040473810128141193413 48420; 3) 03236162017221521181943 33370; 4) 0ė24ė44ė30ė26ė25ė27ė23ė29ė28ė46ė 50ė0. Ordinary city 44 can be next to exclusive city 24 in route 4 and exclusive city 6 next to ordinary city 49 in route 1. III. GA DESIGN A genetic algorithm (GA) is used to search solutions based on the evolutionary principle. Since it was firstly introduced by Holland [6], it has been successfully applied to a variety of NP-hard combinatorial optimization problems, such as TSP and MTSP. To apply GA to solve MTSP*, the focus is to establish an effective coding and decoding scheme and design suitable selection, crossover and mutation operators to ensure better population evolution. A. Chromosome coding for MTSP* Chromosome representation is a crucial basic work when applying a GA. It directly determines the performance of the algorithm. The GAs designed to solve MTSP of m traveling salesman n cities mainly exploit three encoding schemes. The first is the one-chromosome scheme. It uses a single chromosome of length n+m-1. In it, n cities are represented by a permutation of integers from 1 to n. This permutation is partitioned into m sub-tours by the insertion of m-1 negative integers from -1 to -(m-1) that represent the transition from one salesman to the next. The second is the two-part- chromosome technique proposed by Arthur and Cliff [9]. The first part of a chromosome is a permutation of n cities while the second part is of length m and represents the number of cities assigned to each of m salesmen. The values assigned to the second part are constrained to be m positive integers that must sum to the number of cities to be visited (n). The third one is the two-chromosome technique. It requires two chromosomes, each being of length n, to represent a solution. The first one provides a permutation of n cities while the second one assigns a salesman to each of the cities in the corresponding position of the first one. For MTSP*, it is d to use the first two schemes mentioned above. The third one with two types of chromosomes can be adapted to MTSP* via some modification. Particularly, the city chromosome consists of a permutation of integers from 1 to n while a genetic value of salesman chromosome is the number of a salesman that corresponds to an exclusive or ordinary city in the same position of the city chromosome. The exclusive city is assigned to the specified salesman, and the ordinary city is randomly assigned to the salesman. A coding example of MTSP* with three salesmen and ten cities is shown Fig. 3, where six cities are exclusive ones and three salesmen are assigned three exclusive ones. The rest cities are ordinary. Namely, exclusive cities of salesmen 1-3 are 1-2, and 3-4, and 5-6. Respectively, ordinary cities are cities 7-10. According to two chromosomes in Fig. 3, cities 2, 7 and 1 (in that order) are visited by salesman 1. Similarly, cities 10, 4 and 3 (in that order) are visited by salesman 2, and cities 9, 5 and 6 (in that order) are visited by salesman 3. 628 628
  • 3. Figure 3. Example of MTSP* coding B. Selection operator Roulette Wheel method and the elitist strategy [11] are adopted as the selection operation in this work. The former simply chooses a chromosome in a statistical fashion based solely upon its fitness value. The elitist strategy copies the individual with the best fitness at the present generation to the next one. It can prevent the optimized individuals from being eliminated after a selection, crossover or mutation operation. It is critical to ensure the convergence of a GA. A GA containing an elitist strategy is proven to be globally convergent. C. Crossover operator A crossover operator exchanges parts of the genes from two parent individuals to form two new individuals. It is one of important features that distinguish a genetic algorithm from other ones. In MTSP a crossover operator may be one of partially matched crossover (PMX), ordered crossover (OX), cycle crossover (CX), two-point crossover, etc. We design three modes of crossover operators corresponding to the adopted chromosome coding style, i.e., city crossover (CC), salesman crossover (SC), and city- salesman crossover (CSC). 1) City crossover In this paper, we ameliorate PMX as a city crossover operator. It requires randomly selecting two crossing points to determine a matching section. The corresponding matching sections in two parents are swapped, thereby resulting in two new descendants. Then two new individuals are checked if the exclusive cities are assigned to the specified salesman. If not, the particular genes in a salesman individual should be corrected. Figure 4. Example of CC The crossover of two chromosomes is shown in Fig. 4. In Step 1, given two parents, we randomly select a section of a city individual, then swap its genes with those of another individual and produce two individual of descendants as shown in Step 2. The mapping relationship of the selected sections in two city individuals is 8—3, 9—8, 5—2, 4—7, 7— 1, and 1—10. In Step 3, exchange the redundant genes according to the selected section, then find that exclusive cities 5, 3, 7, 1, and 6 in the left chromosome and cities 2, 5, and 4 in the right one are assigned to the wrong salesmen. Then, the exclusive cities are reassigned to the correct salesmen and two reasonable generations are produced as shown in Step 4. 2) Salesman crossover To avoid a number of duplicate genes appearing in a salesman chromosome, this work adopts traditional two-point crossover. It also requires randomly determining a matching section. Then the corresponding sections of two chromosomes are swapped to generate two descendants. At end, the matching relationship between the exclusive cities should be checked and a wrong salesman should be corrected to the specified one. An SC process is shown in Fig. 5. In Step 1, there are two parents. In particular, the randomly selected matching sections are marked in gray color in the salesman chromosome. After swapping of the two marked sections, two descendants are produced in Step 2. However, it is obvious that exclusive cities 5 and 4 of the left chromosome and exclusive cities 3, 2, and 1 of the right chromosome are assigned to the wrong salesmen. In Step 3, it is corrected by reassigning exclusive cities to the specified salesmen and obtained two correct descendants. Figure 5. Example of SC 3) City-salesman crossover For CSC, a city chromosome applies PMX, while a salesman chromosome adopts the two-point crossover scheme. Figure 6. Example of CSC An example of CSC is shown in Fig. 6. In Step 1, we select for each kind of parents a couple of random crossing sections and swap each couple of sections, respectively. It results in two new city generations and two new salesman 629 629
  • 4. ones as shown in Step 2. In Step 3, exchange the redundant genes of the city chromosome, and find that exclusive cities 5 and 3 of the left city chromosome and 2 of the right city one are assigned to a wrong salesman. After the correction of salesman genes, the result is shown in Step 4. D. Mutation operator A mutation operator plays an import role in improving local search ability and maintaining variability of the population. It also prevents the premature termination in GA. This work adopts swapping mutation. It requires random selection of two crossing points, and then swaps the selected points. Finally, check whether the exclusive cities of a new descendant match with their corresponding salesmen. Corresponding to the above three crossover schemes, we design three mutation operators, i.e., city mutation (CM), salesman mutation (SM), and city-salesman mutation (CSM). 1) City mutation In CM, only a city-chromosome applies swapping mutation. For example, first, two selected swapping gene points are cities 8 and 7, as shown in Fig. 7. After swapping them, the exclusive cities are proven to be in accord with salesmen and the mutation is over. Figure 7. Example of CM 2) Salesman mutation In the operation of SM swapping mutation only applied for a salesman chromosome. An example is given in Fig. 8, where the selected swapping points are genes 2 and 1 marked in gray color in the salesman-chromosome. A correct descendant is generated by swapping them. Figure 8. Example of SM 3) City-salesman mutation Swapping mutation takes place in both city and salesman chromosomes in CSM. An example is shown in Fig. 9, where the swapping genes marked in gray are swapped pairwise, which results in new feasible generations. Considering that crossover and mutation should not apply into different types of chromosomes, this work selects three compositions of reasonable crossover and mutation operators from nine ones. By selecting CC as crossover and CM as mutation operator, we have CC CM (called CCM), SC SM (called SCM), and CSC CSM (called CSCM). Figure 9. Example of CSM E. Fitness function A fitness function is used in a GA to judge the chance that an individual (a route) can be selected into the next generation. It is a limiting factor to the efficiency of evolution. For a GA, many selection strategies based on the proportion of fitness require a non-negative fitness and the larger fitness the better individual. Hence, for a problem with a minimum solution as its optimization objective, it needs to turn it to a maximum one. MTSP* takes the minimum length of all salesmen as its optimization objective. Hence this work takes the reciprocal of the length as its fitness. Taking f(x) as the length of the solution, the fitness function is given as: 1 ( )= 1+ ( ) F x f x IV. SIMULATION AND RESULT This paper takes the revised eil51 as an example given in Section 2 to verify the correctness and performance of our GA with the three pairs of crossover and mutation operators. A. Convergence of GA Convergence is an important indicator to show the performance of a GA. If a GA is convergent, it indicates its stability and evolution towards a correct direction. The quality of the solution is better and better as the evolution goes on. To examine a GA’s convergence, we conduct the following experiment, where the generation count is 2000, the crossover probability 0.6, mutation probability 0.1, the size of population is 50, and CCM is selected. The result as shown in Fig. 10 indicates that the convergence of our GA is good without considerable fluctuation. The length of total routes is optimized from 1198.23 km of the initial population to 558.511 km of the 2000-th generation. It implies that the effect of optimization is notable. The evolution process can be divided into three main stages. The first one is from the initial generation to the 560-th one, where the GA convergence is the fastest, and the total length of routes descends from 1198.23 km to 666.21 km. The second one from the 561-th to 1560-th generations shows a 630 630
  • 5. slow change of the total length form 666.21 km to 558.51 km. In the last one from the 1561-th to 2000-th generations, the result tends to be stable. Hence, our GA has a good convergence. For the other two pairs of crossover and mutation operators, i.e., SCM and CSCM, we can obtain the similar results. Figure 10. Result of the example of MTSP* B. Comparison To compare the performance of our GA with different crossover and mutation operators, three groups of experiments are designed where three pairs of crossover and mutation operators are adopted in turn. The other parameters are set, i.e., the size of population 30, the crossover probability 0.6, the mutation probability 0.1, and the generation count 2000. Each experiment is carried out for ten times by using a DELL Inspiron620s computer with Windows 7 and Inter Core i3 CPU at 3.30GHZ. The data are from the revised eil51 example in Section 2. The experimental result is shown in TABLE and Fig. 11. TABLE I. Result of the three group experiments Unit˖Km Group CCM SCM CSCM 1 558.25 575.26 568.45 2 556.85 576.45 572.12 3 561.47 580.56 573.89 4 565.25 576.65 567.45 5 558.36 570.25 570.35 6 557.68 590.68 566.25 7 562.45 586.45 576.87 8 568.56 573.89 565.54 9 556.35 584.57 575.58 10 560.58 571.64 566.4 Average 560.58 578.64 570.29 Figure 11. Result of different operators The best, the worst and the average solutions are observed and shown in TABLE . TABLE II. Data analysis of each operator Crossover and mutation operator best solution (km) worst solution (km) Average solution (km) CCM 556.35 568.56 560.58 SCM 570.25 590.68 578.64 CSCM 565.54 576.87 570.29 It shows that the performance of CCM is the best among three pairs of crossover and mutation operators while that of SCM is the worst. By observing the results we find that by SCM, crossover and mutation merely take place in the salesman chromosomes. It results in a small solution space and weakens genic recombination due to large duplicate salesman individuals. Using CCM only the city chromosomes are crossed and mutated and there is no duplicate gene. Thus the extent of genic recombination is stronger and the solution space is larger than that of SCM. Namely, the population diversity of CCM is better than that of SCM, as shown in the experiment data. Both a city chromosome and a salesman chromosome are crossed and mutated by CSCM. The dynamics of genic recombination is stronger than SCM. Hence, the performance of CSCM is better than that of SCM. However, crossover and mutation of both city chromosomes and salesman chromosomes may destroy the better solution due to too strong dynamics of genic recombination. Thus, CSCM is worse than CCM as observed. As for time consumption, three groups have small differences from each other. Therefore, we should select CCM in a GA for MTSP* . V. CONCLUSIONS This paper presents a new multiple traveling salesman problems with exclusive and ordinary cities. To solve it, we design a genetic algorithm. It revises the two-chromosome scheme and design three pairs of crossover and mutation operators, i.e., simple city crossover and mutation, simple salesman crossover and mutation, and mixed city-salesman 631 631
  • 6. crossover and mutation. To show the performance of our algorithm with different operators, we design three groups of experiments. The results indicate that the proposed GA is suitable to solve MTSP* and CCM deploying city crossover and city mutation has the best performance among three compositions of crossover and mutation operators. VI. ACKNOWLEDGEMENT This work was supported in part by the China National Natural Science Foundation under Grants 61004035 and 61175113. REFERENCES [1] E. Lawler, J. Lenstra, A. Rinnooy, D. Shinmoys. The Traveling Salesman Problem. John Whiley Sons, vol. 25, no. 4, pp.251-258, 2006 [2] G. Gutin, A. Punnen. The Traveling Salesman Problem and Its Variations. Dordrecht: Kluwer, vol. 35, no. 2, pp.625-629, 2002 [3] Hou MS, Liu, DB. A novel method for solving the multiple traveling salesmen problem with multiple depots. Chinese science bulletin, vol.57, no.2, pp.1886-1892, 2012 [4] T. Bektas. The multiple traveling salesman problem: An overview of formulations and solution procedures. Omega, vol. 34, no. 3, pp.209- 219, 2006 [5] Jun Li, Qirui Sun, Xianzhong Dai. A Coordination and Optimization Method for Multi-Bridge Water jet Cutting processes. 42nd International Conference on Computers and Industrial Engineering, Cape Down, South Africa, 2012 [6] J. Holland. Adaptation in Natural and Artificial Systems. the University of Michigan Press, Ann Arbour, USA, 1975 [7] J. Potvin. Genetic algorithm for the traveling salesman problem. Annals of Operations Research, vol.63, no. 4, pp.330-370, 1999 [8] M. Sallabi, E. Haddad. An Improved Genetic Algorithm to Solve the Traveling Salesman Problem. World Academy of Science, Engineering and Technology, vol. 52, no.3, pp.471-474, 2009 [9] E. Carter, T. Ragsdale. A new approach to solving the multiple traveling salesperson problem using genetic algorithms. European Journal of Operation Research, vol. 175, no.2, pp.246-257, 2006 [10] Wei Zhou, Yuanzong Li. An Improved Genetic Algorithm for Multiple Traveling Salesman Problem. 2010 2nd International Asia Conference on Informatics in Control, Automation and Robotics, Wuhan Hubei, China, pp.493-495, 2010. [11] A. De Jong. An analysis of the behavior of a class of genetic adaptive systems. Michigan: University of Michigan, 1975 632 632