Professional Documents
Culture Documents
ARTICLE IN PRESS
Journal of Computational Science xxx (2013) xxxxxx
a r t i c l e
i n f o
Article history:
Received 31 December 2012
Received in revised form 16 May 2013
Accepted 26 May 2013
Available online xxx
Keywords:
Genetic Algorithm
Population seeding technique
Traveling Salesman Problem
Order distance vector
Combinatorial problem
a b s t r a c t
Genetic Algorithm (GA) is a popular heuristic method for dealing complex problems with very large search
space. Among various phases of GA, the initial phase of population seeding plays an important role in
deciding the span of GA to achieve the best t w.r.t. the time. In other words, the quality of individual
solutions generated in the initial population phase plays a critical role in determining the quality of nal
optimal solution. The traditional GA with random population seeding technique is quite simple and of
course efcient to some extent; however, the population may contain poor quality individuals which
take long time to converge with optimal solution. On the other hand, the hybrid population seeding
techniques which have the benet of good quality individuals and fast convergence lacks in terms of
randomness, individual diversity and ability to converge with global optimal solution. This motivates to
design a population seeding technique with multifaceted features of randomness, individual diversity
and good quality. In this paper, an efcient Ordered Distance Vector (ODV) based population seeding
technique has been proposed for permutation-coded GA using an elitist service transfer approach. One
of the famous combinatorial hard problems of Traveling Salesman Problem (TSP) is being chosen as the
testbed and the experiments are performed on different sized benchmark TSP instances obtained from
standard TSPLIB [54]. The experimental results advocate that the proposed technique outperforms the
existing popular initialization methods in terms of convergence rate, error rate and convergence time.
2013 Elsevier B.V. All rights reserved.
1. Introduction
Genetic Algorithm (GA) is a well-known method for global optimization of complex problem very large search space based on
the survival of the ttest concept of natural evolution [55]. The
signicant features of the GA, which makes it perform competently could be dened as follows: GA operates on a population
of feasible solutions rather than on a single solution; the variety of genetic operators helps to explore unrevealed solutions in
the large search space effectively; possibility to construct problem the specic genetic operators which can offer better solution
search; population diversity helps to avoid the drawback of getting
trapped in local optima and premature convergence. These exible
1877-7503/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jocs.2013.05.009
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
2
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
an optimal solution, number of generations required to evolve optimal solution increases, the search possibility for an optimal solution
decreases and more importantly the convergence rate or quality of
the optimal solution obtained is reduced. Thus, the requirement for
a modied population seeding technique in GA is clear and in fact,
several research works were proposed to support the dispute [61].
Lawrence and Amini [61] discussed about different GA conguration issues and claims that seeding the initial population with
heuristics can improve the efciency of the GA greatly. Togan and
Daloglu [40] believes that performance and convergence ability of
GA are critically inuenced by the population seeding method and
proposed two new self-adaptive member grouping strategies and a
new strategy for population seeding. The large search space is collected into different groups and the list of cross section values are
assigned as initial values to set the initial population automatically.
This method is dedicatedly proposed for application in the area of
structural engineering and tested in truss structures and transmission towers. Nearest Neighbor (NN) tour construction heuristic is
one of the familiar alternatives for random population seeding in
GA, particularly for TSP [3,16,4246]. In NN technique, individuals
in the population seeding are constructed with the city nearest to
the current city and such good individuals can rene the subsequent search in the next generations [3]. Though NN works ne,
it suffers with some critical factors: several cities are not included
in the individuals created initially and have to be inserted at high
costs in the end; neglecting several cities at the population seeding
stage leads to severe errors in optimal solution construction and
the diversity among the individuals created is very minimum.
Yingzi et al. [32] proposed a Greedy GA (GGA), in which the population seeding is performed using Gene Bank (GB). The GB is built
by assembling the permutation of n cities based on their distance.
In GGA method, the population of individuals is generated from the
GB such that the individuals are of above-average tness and short
dening length. In GGA, with the increase in number of cities leads
to augmented problem complexity and performance degradation,
and large collection of GB individuals enlarges the cost of computation at each generation. The improved performance of GGA is
justied using TSP with maximum of hundred cities and its performance deteriorates with large number of cities. In [38], Fuyan
et al. proposed K-means algorithm, based on the work reported in
[46], which is considered to generate much infeasible solutions, to
obtain the initial population in which N number of individuals are
partitioned and assigned to one of K clusters. As a result, using
K-means algorithm, generating infeasible solutions in the population seeding stage is avoided. Performance evaluation is performed
with a maximum of 10 cities with only time based analysis and
be decient to validate the proposed technique for large number of
cities and convergence capability. Yugay et al. [36] proposes a modied GA with sorted initial population method based on theory of
better parents produce better offsprings. In this approach, a large
initial pool of population is generated and ranked in accordance
to their tness values and at last, a certain number of individuals with bad tness are omitted. This approach also suffers with
the issues discussed with NN tour construction heuristic technique
such as premature convergence, reduced search space exploration
and minimum population diversity. Hence, the traditional GA does
not provide effective performance when applied to some of the
combinatorial problems like TSP [33], so each stage in the traditional GA has been modied in order to achieve a better output and
thus resulted in hybrid GAs [2832,34,49,50].
Recently, many researchers proposed modied versions of GA,
particularly for solving TSP using random population seeding techniques [3,16,3032,34,3941,48,6466]. Though several modied
population seeding techniques for GA have been proposed, many
researchers still continue to work with a random population seeding technique because of the complexity nature of algorithm, which
is difcult to understand and implement, problem specic modications are required to apply and problem such as premature
convergence, ineffective search space exploration and less population diversity. This implies that the researchers are interested
in random population seeding technique to accomplish a better
search space exploration and nding best optimal solution at the
cost of high convergence time. In [64], Xing et al. proposed a hybrid
approach combining an improved Genetic Algorithm and optimization strategies using random population seeding technique. An
efcient hybrid mutation Genetic Algorithm has been proposed
using a random population seeding technique in [34]. Chang and
Ramakrishna [39] proposed a GA for shortest path routing problem,
in which author reveals the reason for preferring random population seeding method than heuristic initialization. Although the
mean tness of the individuals generated using heuristic initialization are high so that it may help the GA to obtain the better
solutions faster, but it ends up in exploring a small part of the search
space and never nd the global optimal solutions because of the
lack of diversity in the initial population generated [47]. In some of
the works, the authors used hybrid population seeding technique
which combines random and any of the modied population seeding technique [33,61]. In [62], Qu and Sun proposed a synergetic
approach to GA by adding some new randomly generated individuals into the population after each generation in order to prevent
premature convergence and to obtain nal optimum.
To summarize, the modied population seeding techniques has
the advantages of good quality or generating potential sequence
individuals at an early stage and the ability to nd near optimal
solutions at few generations; however they lacks in randomness,
diversity of individuals generated, ability to explore more search
space and nding the global solution. On the other hand, the random population seeding technique has the advantages of individual
diversity, can explore search space efciently and nding optimal
solution; however they have the disadvantages of individuals with
worst potential sequence and requires longer search time to converge an optimal solution. The controversies between these two
categories of population seeding techniques motivate to propose
an efcient population seeding technique with characteristics of
randomness, individual diversity and potential sequence. Thus in
this paper, an efcient Ordered Distance Vector (ODV) based population seeding technique with three different varieties has been
proposed for permutation-coded GA. The popular combinatorial
optimization problem of Traveling Salesman Problem (TSP) is being
chosen as the testbed to validate and claim the efcacy of the proposed population seeding technique. Experiments were performed
over the different sized benchmark TSP instances obtained form
the TSPLIB [54]. The organization of the paper is as follows: Section 2 offers sufcient background information over GA and TSP to
improve the understanding of this paper. Section 3 describes the
proposed technique and its variants along with the corresponding
algorithms. The different phases of the experiments are summarized in Section 4. This section also reports experimental results
with corresponding analyses. And nally, Section 5 presents the
conclusive remarks of the work reported in this paper.
2. Background information
As stated earlier, this section offers a brief introduction on
Genetic Algorithm (GA) and Traveling Salesman Problem (TSP) to
improve the understandability of this paper.
2.1. Genetic Algorithm
Genetic Algorithms (GAs), a subclass of evolutionary algorithms,
is a stochastic optimization technique based on the principles
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
ARTICLE IN PRESS
If
Gen>Limi
Stop
If Bestindiv is
Fittest
Individual
Select Individual(s)
Perform Reproduction
Perform Mutation
Perform Crossover
n
d(ci , c(i+1) ),
1 (n + 1)
(1)
i=1
(2)
and
1...n! = {ci , c(i+1) , c(i+2) , . . . , cn }
(3)
TSP has to search for the optimal (best) solution which exists
in the permutation set and can be given as,
n
i=1
n
(4)
i=1
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
ARTICLE IN PRESS
Thus, be the required best solution for the considered TSP with
n number of cities.
By considering the importance and extended applications of
TSP in various elds, several meta-heuristic search techniques have
been proposed to solve it, such as Tabu Search [9,10], Genetic Algorithm [1117], Ant Colony Optimization [1822], Particle swarm
Optimization [23,29], Neural Networks [24,25], Simulated Annealing [26], Multi-Agent System [27,35,41,63] and Hybrid-Heuristics
[2830]. Major applications of TSP are Vehicle routing [4], Drilling
of printed circuit boards [5], Overhauling gas turbine engines [6],
X-ray crystallography [7], computer wiring [4], sequencing and
scheduling [47,52] and the order-picking problem in warehouses
[8]. In this paper, TSP is being used as a testbed for demonstrating
the effectiveness of the proposed technique. On the other hand, it
could also be considered that the proposed technique is a better
solution for solving TSP.
(5)
such that,
d(ci , cj ) d(ci , cj+1 ) d(ci , cj+2 ) . . . d(ci , cn1 )
(6)
C1(j)
ODVc2 C2(j)
ODM =
ODVc3 = C3(j)
... ...
ODVcn
PODM
1 (ci )
1 (c(i+1) )
1 (c(i+2) ) . . .
1 (cn )
=
3 (ci ) 3 (c(i+1) ) 3 (c(i+2) ) . . . 3 (cn )
...
...
...
...
o (ci )
o (c(i+1) )
o (c(i+2) )
(10)
o (cn )
3. Proposed system
Cn(j)
C1(j+1)
C1(j+2)
C1(n1)
C2(j+1)
C2(j+2)
C2(n1)
C3(j+1)
C3(j+2)
...
...
Cn (j + 1)
Cn(j+2)
C3(n1)
...
(7)
Cn(n1)
(8)
R(d(c2(n1) , c2 )) = n 1
(9)
Potential sequence: This factor helps to keep the good quality order
of cities in each individual.
Individual diversity: This factor refers to the difference in the permutation of cities in each individual generated. It helps to avoid
the drawback of getting trapped in local optima and premature
convergence.
Potential sequence: For GA, the capability to nd the optimal
solution is critically affected by the tness (potential sequence) of
individuals generated in the population initialization [9]. This factor
can be applied in ways as follows:
Equi-begin (Eb): It is claimed that forcing individual to always
begin the permutation of city at a specic city cx will allow the GA
to take the benet of good building block or high tness sequence
of cities in generating new individuals [7,10].
Vari-begin (Vb): This method is used in many existing techniques
in which the beginning of the permutation of the city is not necessarily the same city. Creating the individuals beginning with
different cities provide the possibility of creating new individuals, the size of the initial population can be more, repetition of
individual sequence can be avoided and mainly it helps to avoid
the algorithm to get trapped in premature convergence.
Individual diversity: Individual diversity refers to the variation
in the permutation of cities in each individual in the population.
The initial population of the individual must offer a wide diversity because preserving diversity in the population, particularly
at the initial stages of GA, represents a condition for avoiding the
search from getting stuck in local optima and also from premature
convergence.
Best adjacent (ba) number: A detailed study on existing work
[5658] helps to derive an assumption that, in an optimal solution,
any city ci is connected to city cj such that cj is one of the ci s nearest
ba number of cities. The value of ba (an integer) can be expressed
in range and it highly depends on the size of the population n. The
range of ba value can be assigned as follows:
If 1 n 10, then 2 ba 3.
If 11 n 100, then 2 ba 4.
If 101 n 1000, then 2 ba 5.
If n 1001, then 2 ba 6.
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
technique as shown in Fig. 2 and the comparison among those variants are given in Table 1. These variants are explained as follows:
Max(tot(PODM )) = ba(n1)
(11)
The above said ranges are not exact and are derived as the
experimental outcomes of the literature study. It shows that if the
number of cities n is between 101 n 1000, then in the optimal
solution, every city would be connected to any of its ba 2 ba 5
cities which are nearest to it.
In ODV population seeding technique, the value of ba number plays a vital role in providing individual diversity. The adjacent
cities in each individual of the initial population depends on the
ba value. An integer value generated between the ranges of ba
decides which cities are kept as adjacent in each individual. There
are two different ways of initializing population based on diversity
in ba number generation,
where tot(PODM ) refers to the total number individuals in the population PODM and ba refers to the Best Adjacent number assigned
and n refers to the total number of cities.
The PODM generated using this type can be represented as:
where the rst city remains same for each individual (i.e.) 1 (1)
2 (1) 3 (1), . . . , o (1).
This type of population should be handled with suitable
crossover operators because it may be suffered by pre-convergence
problem due to lack of much diversity.
PODM
1 (1)
1 (c2 )
1 (c3 ) . . .
1 (cn )
=
3 (1) 3 (c2 ) 3 (c3 ) . . . 3 (cn )
...
...
...
...
o (1)
o (c2 )
o (c3 )
(12)
o (cn )
(13)
where tot(PODM ) refers to the total number individual in the population PODM and ba refers to the Best Adjacent number assigned
and n refers to the total number of cities.
Table 1
Comparison between the variants of ODV population seeding technique.
Population seeding techniques
Characteristics
Initial city of individuals
Individual diversity
Disadvantages
Type 1 (EV)
Type 2 (VE)
Type 3 (VV)
Same
Random bax value
Huge diversity among
individuals
Different
(n 1) o
ba(n1)
High potential sequence of
cities
Limited size of population
Different
Same bax value
Little diversity among
individuals
Equal
n ba
n ba
Limited size of population
Different
Random bax value
Huge diversity among
individuals
Different
no
ban
High potential sequence of
cities
No cyclic repetition problem
No pre-convergence problem
Very large size of population
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
ARTICLE IN PRESS
Fig. 3. List of variables used in the algorithm for ODV based population seeding technique.
PODM
1 (1)
(1)
2
.
.
.
ba (1)
(2)
1
=
.
..
(2)
ba
1 (n)
.
..
ba (n)
1 (c2 )
1 (c3 )
2 (c2 )
2 (c3 )
..
.
..
.
ba (c2 )
ba (c3 )
2 (c2 )
3 (c3 )
..
.
..
.
ba (c2 )
ba (c3 )
2 (c2 )
2 (c3 )
..
.
..
.
ba (c2 )
ba (c3 )
1 (cn )
..
ba (cn )
3 (cn )
..
ba (cn )
2 (cn )
..
.
2 (cn )
Max(tot(PODM )) = ban
(15)
where tot(PODM ) refers to the total number individual in the population PODM and ba refers to the Best Adjacent number assigned
and n refers to the total number of cities.
The PODM generated using this type can be represented as,
(14)
ba (cn )
PODM
(c ) (c ) (c ) . . . (c )
1 1
1 2
1 3
1 n
2 (c1 ) 2 (c2 ) 2 (c3 ) . . . 2 (cn )
= 3 (c1 ) 3 (c2 ) 3 (c3 ) . . . 3 (cn )
...
o (c1 )
...
o (c2 )
...
o (c3 )
...
o (cn )
This type of population seeding technique is the most recommended for effective search for best solution. The main difculty
concerned with this technique is to determine the limit of population size o, to support small population size techniques, because
good initial population individuals which may occur outside the
limit are left unrevealed.
As the last, the fourth possible type population seeding technique, Equi-begin with Equal diversity (EE), which is not considered
as an effective one since it can produce the initial population with
only n individuals. It is very hard to nd the optimal solution with
less initial population containing un-potential sequence of cities,
which leads to pre-mature convergence and large number of generations.
3.4. Algorithm for ODV population seeding technique
The algorithm for ODV population seeding technique consists
of two stages. First an Order Distance Matrix (ODM) will be created from the Distance Matrix (DM) and followed by generating
the initial population based on the ODM. The second stage of the
technique can be performed in three ways; EV, VE and VV depends
on the type of population seeding technique to be chosen. List of
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
Fig. 4. Algorithm for ODV based population seeding technique and its variants.
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
ARTICLE IN PRESS
Table 2
Experimental results of random population initialization method.
S. no. Instance Optimal
tness
1
2
3
4
5
6
7
Eil51
Pr76
KroA100
Pr144
Gil262
Fl417
Pr1002
426
108,159
21,282
58,537
2378
11,861
259,045
Time taken
(s)
0.51
0.61
1.25
1.35
3.51
7.12
15.04
803.56
195,678.73
39,789.3
111,987.32
4469.94
22,646.99
511,527
1176.34
247,890.85
58,475.04
154,899.24
6402.06
33,757.7
755,689.63
Average
tness
931.38
21,7356.19
44,782.09
124,689.52
5173.77
27,848.63
627,364.63
Error rate
Convergence
rate
Best indiv.
(%)
Worst
indiv. (%)
Best indiv.
(%)
Worst
indiv. (%)
88.63
80.92
86.96
91.31
87.97
90.94
97.47
176.14
129.19
174.76
164.62
169.22
184.61
191.72
11.37
19.08
13.04
8.69
12.03
9.06
2.53
76.14
29.19
74.76
64.62
69.22
84.61
91.72
Convergence
diversity (%)
Average
convergence
(%)
87.51
48.27
87.80
73.31
81.25
93.67
94.25
18.63
0.96
10.42
13.01
17.57
34.79
42.18
Table 3
Experimental results of Nearest Neighbor population initialization method.
S. no. Instance Optimal
tness
1
2
3
4
5
6
7
Eil51
Pr76
KroA100
Pr144
Gil262
Fl417
Pr1002
426
108,159
21,282
58,537
2378
11,861
259,045
Time taken
(s)
1.18
1.26
1.96
2.01
5.62
12.46
20.12
Best tness
495.77
121,567.43
24,421.8
62,952.97
2622.75
14,123.99
312,135.76
Worst
tness
688.01
186,540.75
32,689.65
82,577.56
3532.06
18,554.38
409,843.75
Average
tness
572.65
169,746.4
29,765.76
72,458.43
2734.75
17,754.67
379,744.72
Error rate
Convergence
rate
Best indiv.
(%)
Worst
indiv. (%)
Best indiv.
(%)
Worst
indiv. (%)
16.38
12.40
14.75
7.54
10.29
19.08
20.49
61.50
72.47
53.60
41.07
48.53
56.43
58.21
83.62
87.60
85.25
92.46
89.71
80.92
79.51
38.50
27.53
46.40
58.93
51.47
43.57
41.79
Convergence
diversity (%)
Average
convergence
(%)
45.13
60.07
38.85
33.53
38.24
37.35
37.72
65.58
43.06
60.14
76.22
85.00
50.31
53.41
Convergence
diversity (%)
Average
convergence
(%)
48.25
29.30
51.28
36.45
37.31
39.81
35.29
60.58
52.59
60.66
72.96
68.03
45.32
47.70
Table 4
Experimental results of Gene Bank population Initialization method.
S. no. Instance Optimal
tness
1
2
3
4
5
6
7
Eil51
Pr76
KroA100
Pr144
Gil262
Fl417
Pr1002
426
108,159
21,282
58,537
2378
11,861
259,045
Time taken
(s)
1.14
1.03
1.96
2.18
4.56
9.96
16.62
Best tness
505.65
130,237.83
25,059.4
69,189.73
2937.24
15,190.99
340,468.3
Worst
tness
711.21
161,930.85
35,973.04
90,528.27
3824.45
19,913.1
431,877.73
Average
tness
593.91
159,436.19
29,654.83
74,363.72
3138.27
18,346.11
394,532.29
Error rate
Convergence
rate
Best indiv.
(%)
Worst
indiv. (%)
Best indiv.
(%)
Worst
indiv. (%)
18.70
20.41
17.75
18.20
23.52
28.08
31.43
66.95
49.72
69.03
54.65
60.83
67.89
66.72
81.30
79.59
82.25
81.80
76.48
71.92
68.57
33.05
50.28
30.97
45.35
39.17
32.11
33.28
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
S. no.
Instance
Optimal
tness
ba value
Time taken
(s)
Best tness
Worst
tness
Average
tness
Error rate
Best indiv.
(%)
Convergence
rate
Worst
indiv. (%)
Best indiv.
(%)
Worst
indiv. (%)
Convergence
diversity (%)
Average
convergence
(%)
Eil51
426
2
3
4
5
1.04
1.13
1.13
1.16
480.23
468.72
459.91
486.02
648.01
675.38
660.55
674.68
552.38
555.20
565.52
586.51
12.73
10.03
7.96
14.09
52.11
58.54
55.06
58.38
87.27
89.97
92.04
85.91
47.89
41.46
44.94
41.62
39.38
48.51
47.10
44.29
70.33
69.67
67.25
62.32
Pr76
108,159
2
3
4
5
0.89
1.04
0.96
0.98
122,815.73
129,988.13
122,222.99
117,737.35
161,930.85
180,861.62
181,133.91
185,199.42
150,166.19
152,180.65
156,608.47
160,039.16
13.55
20.18
13.00
8.86
49.72
67.22
67.47
71.23
86.45
79.82
87.00
91.14
50.28
32.78
32.53
28.77
36.16
47.04
54.47
62.37
61.16
59.30
55.21
52.03
KroA100
21,282
2
3
4
5
1.83
2.36
1.96
2.57
22,579.30
23,835.01
24,705.75
27,226.07
31,671.04
33,867.96
36,461.39
36,658.67
26,797.09
28,967.86
30,623.82
31,916.62
6.10
12.00
16.09
27.93
48.82
59.14
71.33
72.25
93.90
88.00
83.91
72.07
51.18
40.86
28.67
27.75
42.72
47.14
55.24
44.32
74.09
63.89
56.10
50.03
Pr144
58,537
2
3
4
5
1.97
1.89
1.96
2.47
62,952.97
65,118.87
67,942.88
68,776.48
77,253.26
84,841.73
93,087.71
104,878.52
68,031.87
71,417.70
78,660.40
81,943.07
7.54
11.24
16.07
17.49
31.97
44.94
59.02
79.17
92.46
88.76
83.93
82.51
68.03
55.06
40.98
20.83
24.43
33.69
42.96
61.67
83.78
78.00
65.62
60.01
Gil262
2378
2
3
4
5
4.61
4.56
4.94
4.50
2695.27
2775.68
3039.58
3094.87
3419.06
3608.30
3797.79
3860.18
3138.27
3293.49
3395.44
3504.37
13.34
16.72
27.82
30.15
43.78
51.74
59.71
62.33
86.66
83.28
72.18
69.85
56.22
48.26
40.29
37.67
30.44
35.01
31.88
32.18
68.03
61.50
57.21
52.63
Fl417
11,861
2
3
4
5
9.64
9.83
9.98
9.85
13,320.99
13,742.98
14,738.10
15,034.46
18,086.68
18,992.64
19,650.10
20,188.81
16,149.93
16,856.82
17,295.57
17,858.83
12.31
15.87
24.26
26.76
52.49
60.13
65.67
70.21
87.69
84.13
75.74
73.24
47.51
39.87
34.33
29.79
40.18
44.26
41.41
43.46
63.84
57.88
54.18
49.43
Pr1002
259,045
2
3
4
5
15.95
20.43
24.53
19.97
311,056.00
318,419.76
344,430.94
360,853.18
362,563.17
381,566.91
392,676.91
406,318.12
335,898.99
338,586.44
367,596.27
380,040.78
20.08
22.92
32.96
39.30
39.96
47.30
51.59
56.85
79.92
77.08
67.04
60.70
60.04
52.70
48.41
43.15
19.88
24.38
18.62
17.55
70.33
69.29
58.10
53.29
ARTICLE IN PRESS
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
Table 5
Experimental results of EV population seeding method.
G Model
ARTICLE IN PRESS
10
individual representation of tour of cities in the TSP into Path, Ordinal, Position Listing adjacency and adjacency list representation
schemes. Among these schemes, the Path representation of an individual is a most widely used scheme for the TSP [12,35], which is
considered for the experiments reported in this paper.
4.1.2. Performance factors
There are four performance factors used to investigate the signicance of the proposed technique and they are summarized as
follows:
(i) Convergence rate: Convergence rate of a solution can be dened
as the percentage of tness attained by the solution w.r.t. the
known optimal solution for the problem. It can be given as,
Convergence rate(%)=1
FitnessOptimal tness
100 (17)
Optimal tness
100
(18)
(19)
(iv) Convergence diversity: The convergence diversity of the population is the difference between the convergence rate of the
best and worst solutions found in the population. This factor
shows the diversity among the individuals in the population,
which is an important factor used to overcome the pre-mature
convergence problem and it can be represented as,
Convergence diversity(%) = CRBest Indiv CRWorst Indiv
(20)
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
S. no.
Instance
Optimal
tness
ba value
Time taken
(s)
Best tness
Worst
tness
Average
tness
Error rate
Convergence
rate
Best indiv.
(%)
Worst
indiv. (%)
Best indiv.
(%)
Worst
indiv. (%)
Convergence
diversity (%)
Average
convergence
(%)
Eil51
426
2
3
4
5
0.85
0.68
0.79
0.64
514.61
510.67
510.67
510.67
635.02
635.02
638.16
688.21
571.92
568.51
570.44
583.60
20.80
19.88
19.88
19.88
49.07
49.07
49.80
61.55
79.20
80.12
80.12
80.12
50.93
50.93
50.20
38.45
28.26
29.19
29.93
41.68
65.75
66.55
66.09
63.00
Pr76
108,159
2
3
4
5
0.58
0.71
0.66
0.67
142,396.36
142,396.36
142,396.36
142,396.36
159,821.61
170,453.18
187,050.56
187,050.56
152,344.17
155,229.69
161,093.74
161,892.75
31.65
31.65
31.65
31.65
47.77
57.60
72.94
72.94
68.35
68.35
68.35
68.35
52.23
42.40
27.06
27.06
16.11
25.94
41.29
41.29
59.15
56.48
51.06
50.32
KroA100
21,282
2
3
4
5
1.50
1.36
1.54
1.59
25,125.22
25,125.22
25,125.22
25,125.22
28,541.55
31,983.78
36,397.00
37,333.38
26,544.47
27,817.70
30,195.35
31,326.19
18.06
18.06
18.06
18.06
34.11
50.29
71.02
75.42
81.94
81.94
81.94
81.94
65.89
49.71
28.98
24.58
16.05
32.23
52.96
57.36
75.27
69.29
58.12
52.80
Pr144
58,537
2
3
4
5
1.31
1.50
1.48
1.39
61,496.32
61,496.32
61,496.32
61,496.32
80,851.07
80,851.07
84,518.93
97,687.98
68,600.74
71,180.47
72,375.21
72,375.21
5.06
5.06
5.06
5.06
38.12
38.12
44.39
66.88
94.94
94.94
94.94
94.94
61.88
61.88
55.61
33.12
33.06
33.06
39.33
61.83
82.81
78.40
76.36
76.36
Gil262
2378
2
3
4
5
3.87
3.85
4.34
4.51
3002.62
3002.62
3002.62
3002.62
3593.52
3685.80
3973.25
4120.66
3199.15
3343.51
3407.15
3487.54
26.27
26.27
26.27
26.27
51.12
55.00
67.08
73.28
73.73
73.73
73.73
73.73
48.88
45.00
32.92
26.72
24.85
28.73
40.82
47.02
65.47
59.40
56.72
53.34
Fl417
11,861
2
3
4
5
8.94
7.96
8.23
8.11
15,104.49
15,104.49
15,104.49
15,104.49
19,043.95
19,043.95
19,043.95
19,607.30
16,820.81
17,326.59
17,628.11
17,492.75
27.35
27.35
27.35
27.35
60.56
60.56
60.56
65.31
72.65
72.65
72.65
72.65
39.44
39.44
39.44
34.69
33.21
33.21
33.21
37.96
58.18
53.92
51.38
52.52
Pr1002
259,045
2
3
4
5
17.40
17.10
17.11
18.90
315,597.59
315,597.59
315,597.59
315,597.59
351,116.55
377,213.31
410,111.38
436,384.90
329,395.18
345,410.40
360,368.97
374,376.88
21.83
21.83
21.83
21.83
35.54
45.62
58.32
68.46
78.17
78.17
78.17
78.17
64.46
54.38
41.68
31.54
13.71
23.79
36.49
46.63
72.84
66.66
60.89
55.48
ARTICLE IN PRESS
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
Table 6
Experimental results of VE population seeding method.
G Model
JOCS-208; No. of Pages 21
12
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
of the city x. Experimental results of NN population seeding technique are shown in Table 3. From Table 3, the following observation
can be made:
Observation 3: For all the problem instances, NN technique
produces the individual with least error rate, consequently the best
individual, when compared to the other existing techniques.
4.2.1.3. Gene Bank (GB) technique. In this technique, the gene bank
is built by assembling the permutation of N cities based on their
distance. The population of individuals is generated from the gene
bank such that the individuals are of above-average tness and
short dening length. Table 4 shows the experimental results of GB
population seeding technique and from that the following observation can be made:
Observation 4: It can be observed that the quality of best individuals generated decreases with increase in the size of problem
instance. It is because of the constant gene size used for the generation of initial population.
4.2.2. Experiments on proposed techniques
4.2.2.1. Type 1: EV technique. The experimental results of EV
method of ODV based population seeding technique is presented in
Table 5. In Table 5, the best results are emphasized in bold. The following observations can be made based on the statistical analyses
shown in Table 5.
Observation 5: For all the problem instances and sizes, the EV
population seeding technique tends to give the best t individual
with the ba value of 2.
Observation 6: The error rate of the best individuals generated using the EV population seeding technique increases with an
increase in the number of cities in the instance.
Observation 7: The Convergence rate of the best individuals created using the EV population seeding technique decreases with an
increase in the number of cities in the instance. The maximum and
minimum values obtained are 93.90% for instance KroA100 with
ba value of 2 and 80.92% for instance Pr1002 with ba value 2.
This is because with the same population size of 100, the initial
convergence rate of the technique decreases with an increase in the
number of cities in the instance. However, at least 80.92% of convergence has been achieved at the population initialization stage
of the GA for solving TSP problem.
Observation 8: The convergence diversity of the individuals
remains constant for all the instances except Pr1002 for which
it decreases comparatively. The most suitable practical diversity
range can be 3545%. The maximum and minimum values obtained
are 62.37% for instance of Pr76 with ba value of 2 and 17.55% for
Pr1002 with ba value of 5.
Observation 9: The average convergence of EV technique
remains between the range of 6050%, which shows that the
population contains a good composition of good and bad quality
individuals leads for exploring more search space.
The ranges of variations in convergence rate and error rate of the
population generated by EV technique are illustrated in Figs. 5 and 6
respectively.
4.2.2.2. Type 2: VE technique. The experimental results of the VE
method of ODV based population seeding technique is recorded in
Table 6 and the best results are emphasized in bold. The following
observations are made based on the results presented in Table 6.
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
S. no.
Instance
Optimal
tness
ba value
Time taken
(s)
Best tness
Worst
tness
Average
tness
Error rate
Best indiv.
(%)
Convergence
rate
Worst
indiv. (%)
Best indiv.
(%)
Worst
indiv. (%)
Convergence
diversity (%)
Average
convergence
(%)
Eil51
426
2
3
4
5
1.16
0.96
1.12
1.30
481.93
466.72
464.15
488.30
650.60
681.36
677.47
694.68
562.88
556.12
572.71
589.75
13.13
9.56
8.96
14.63
52.72
59.94
59.03
63.07
86.87
90.44
91.04
85.37
47.28
40.06
40.97
36.93
39.59
50.38
50.08
48.45
67.87
69.46
65.56
61.56
Pr76
108,159
2
3
4
5
0.96
1.10
1.24
0.85
132,815.73
116,449.76
121,277.86
133,252.43
164,233.18
172,650.39
184,779.43
186,572.67
151,647.85
153,976.41
157,896.94
162,671.50
22.80
7.67
12.13
23.20
51.84
59.63
70.84
72.50
77.20
92.33
87.87
76.80
48.16
40.37
29.16
27.50
29.05
51.96
58.71
49.30
59.79
57.64
54.01
49.60
KroA100
21,282
2
3
4
5
2.09
2.29
2.33
2.03
23,081.90
24,763.78
25,720.73
27,389.15
32,911.35
34,533.24
37,224.15
37,511.53
27,110.27
29,062.30
30,894.57
31,862.64
8.46
16.36
20.86
28.70
54.64
62.27
74.91
76.26
91.54
83.64
79.14
71.30
45.36
37.73
25.09
23.74
46.19
45.90
54.05
47.56
72.61
63.44
54.83
50.28
Pr144
58,537
2
3
4
5
1.81
2.23
1.83
1.91
63,213.76
65,249.00
67,995.22
68,069.09
78,883.82
82,901.59
92,248.32
107,500.38
68,178.38
72,129.39
79,161.49
78,761.06
7.99
11.47
16.16
16.28
34.76
41.62
57.59
83.65
92.01
88.53
83.84
83.72
65.24
58.38
42.41
16.35
26.77
30.16
41.43
67.36
83.53
76.78
64.77
65.45
Gil262
2378
2
3
4
5
5.04
5.29
4.85
5.14
2639.70
2728.50
3054.15
3145.82
3443.90
3731.19
3772.09
3986.36
3125.53
3294.28
3407.21
3523.60
11.01
14.74
28.43
32.29
44.82
56.90
58.62
67.63
88.99
85.26
71.57
67.71
55.18
43.10
41.38
32.37
33.82
42.17
30.19
35.35
68.56
61.47
56.72
51.82
Fl417
11,861
2
3
4
5
9.90
10.97
10.40
11.03
12,994.83
14,309.41
14,806.25
15,081.88
19,207.45
19,140.47
19,705.09
20,633.11
16,214.79
16,761.76
17,464.90
17,881.99
9.56
20.64
24.83
27.16
61.94
61.37
66.13
73.96
90.44
79.36
75.17
72.84
38.06
38.63
33.87
26.04
52.38
40.73
41.30
46.80
63.29
58.68
52.75
49.24
Pr1002
259,045
2
3
4
5
21.54
21.84
22.07
18.02
306,292.97
317,395.30
345,069.13
360,241.58
369,643.84
375,577.45
390,673.04
404,670.15
336,460.31
352,951.65
367,582.75
380,772.15
18.24
22.53
33.21
39.07
42.69
44.99
50.81
56.22
81.76
77.47
66.79
60.93
57.31
55.01
49.19
43.78
24.46
22.46
17.60
17.15
70.12
63.75
58.10
53.01
ARTICLE IN PRESS
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
Table 7
Experimental results of VV population seeding method.
G Model
ARTICLE IN PRESS
14
Table 8
Performance order for average computation time.
Random Generation
GeneBank
VE
Instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Random VE VV EV GB NN
Random VE VV EV GB NN
Random VE EV VV NN GB
VE Random VV EV NN GB
Random VE EV VV GB NN
Random VE EV VV GB NN
Random EV GB VE VV NN
Random Generation
GeneBank
VE
90
Average Convergence ( %)
S. no.
Nearest Neighbour
EV
VV
70
50
30
10
-10
-30
Nearest Neighbour
EV
VV
-50
25
Eil51
Pr76
KroA100
Pr144
Gil262
Fl417
Pr1002
TSP Instance
20
Time (sec)
10
0
Eil51
Pr76
KroA100
Pr144
Gil262
Fl417
Pr1002
TSP Instance
Fig. 11. Computation time of different population seeding techniques.
Random Generation
GeneBank
VE
100
Nearest Neighbour
EV
VV
Convergence Diversity ( %)
90
80
70
60
50
40
30
20
10
0
Eil51
Pr76
KroA100
Pr144
Gil262
Fl417
Pr1002
TSP Instance
Fig. 12. Convergence diversity of different population seeding techniques.
Table 9
Performance order for convergence diversity.
S. no.
Instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Random VV EV GB NN VE
Random EV NN VV VE GB
Random VE EV VV GB NN
Random VV VE EV GB NN
Random VE VV EV NN GB
Random VV EV GB VE NN
Random VV EV GB NN VE
Table 10
Performance order for average convergence.
S. no.
Instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
EV VV VE NN GB Random
VV EV VE GB NN Random
VE EV VV GB NN Random
EV VV VE NN GB Random
NN VV EV VE GB Random
VV EV VE NN GB Random
VE VV EV NN GB Random
Table 11
GA conguration parameters.
S. no.
Parameter
Value/technique
1
2
3
4
5
6
7
8
Population size
Generation limit
Crossover method
Crossover probability
Mutation method
Mutation probability
Elitism
Termination condition
100
200
Ordered crossover
0.6 [3,35]
Swap Mutation [35]
0.02 [3,32]
True (3 individuals)
Generation Limit
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
ARTICLE IN PRESS
Optimal
Nearest Neighbour
EV
VV
Random Generation
GeneBank
VE
850
Random Generation
GeneBank
VE
200000
700
650
600
550
500
180000
Fittness Value
750
Fittness Value
Fittness Value
800
160000
140000
120000
450
400
100000
3
Random Generation
GeneBank
VE
Optimal
Nearest Neighbour
EV
VV
Random Generation
GeneBank
VE
Optimal
Nearest Neighbour
EV
VV
4500
22000
90000
80000
4000
3500
70000
3000
60000
2500
50000
Fittness Value
24000
Fittness Value
5000
110000
100000
Random Generation
GeneBank
VE
20000
18000
16000
14000
12000
10000
2000
4
c. KroA100
120000
Random Generation
GeneBank
VE
b. Pr76
Optimal
Nearest Neighbour
EV
VV
Optimal
Nearest Neighbour
EV
VV
a. Eil51
Fittness Value
42000
40000
38000
36000
34000
32000
30000
28000
26000
24000
22000
20000
15
d. Pr144
e. Gil262
f. Fl417
Optimal
Nearest Neighbour
EV
VV
Random Generation
GeneBank
VE
Fittness Value
550000
500000
450000
400000
350000
300000
250000
2
g. Pr1002
Fig. 14. Fitness graph for TSP instances using different population seeding techniques.
Random Populaton
GeneBank
VE
125
Nearest Neighbour
EV
VV
105
Time (sec)
85
65
45
25
5
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
TSP Instance
Fig. 15. Computation time of different methods analyzed.
Average
4.2.3.1. Computation time. The total time taken for EV, VE and VV
population seeding methods is completely depends on the number of cities in the instance and number of ba value generated
for creating each individual. As a result, VV technique takes more
computation time than EV method and it is further decreased in VE
method. The total number of times ba value generated for EV, VE
and VV methods is (n 1), n ba and n o respectively, where n is
the number of cities of corresponding instance, ba is the best adjacent number and o is the population size. The comparison in terms
of computation time of proposed and existing population seeding
techniques with different TSP instances is shown in Fig. 11. The
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
16
Table 12
Experimental results of random population seeding method.
S. no.
TSPLIB instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Optimal length
426
108,159
21,282
58,537
2378
11,861
259,045
Average tness
499.94
133,483.31
25,443.33
74,833.01
2840.79
14,505.18
335,942.32
Average time
7.74
11.43
15.97
22.13
39.79
71.03
167.41
16.45
20.10
18.60
26.25
18.02
22.14
27.66
83.55
79.90
81.40
73.75
81.98
77.86
72.34
Average time
10.93
16.14
22.56
31.25
56.18
100.29
236.38
9.24
11.50
9.65
9.85
13.64
11.60
11.73
90.76
88.50
90.35
90.15
86.36
88.40
88.27
Table 13
Experimental results of NN population seeding method.
S. no.
TSPLIB instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Optimal length
426
108,159
21,282
58,537
2378
11,861
259,045
Average tness
467.96
121,067.36
23,708.35
64,367.21
2800.92
14,000.61
294,051.26
Table 14
Experimental results of GB population seeding method.
S. no.
TSPLIB Instance
Optimal Length
Average tness
Average Time
Error Rate
Conv. Rate
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
426
108,159
21,282
58,537
2378
11,861
259,045
507.44
126,919.34
24,786.43
67,806.50
2869.72
13,988.40
315,000.77
10.02
14.79
20.67
28.63
51.48
91.89
216.59
14.42%
16.54%
14.80%
14.99%
18.57%
16.64%
16.76%
85.58%
83.46%
85.20%
85.01%
81.43%
83.36%
83.24%
Table 15
Experimental results of EV population seeding method.
S. no.
TSPLIB instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Optimal length
426
108,159
21,282
58,537
2378
11,861
259,045
Average tness
458.19
117,443.94
23,054.95
63,541.92
2678.76
13,654.68
291,997.80
Average time
9.17
13.53
18.91
26.20
47.11
84.10
198.22
5.12
6.08
5.67
6.78
10.38
10.63
11.93
94.88
93.92
94.33
93.22
89.62
89.37
88.07
Average time
7.43
10.96
15.32
21.22
38.16
68.12
160.55
18.78
20.80
17.74
19.50
22.72
20.89
21.00
81.22
79.20
82.26
80.50
77.28
79.11
79.00
Average time
10.86
16.02
22.39
31.02
55.78
99.57
234.68
1.02
3.22
4.13
3.81
5.80
7.11
8.18
98.98
96.78
95.87
96.19
94.20
92.89
91.82
Table 16
Experimental results of VE population seeding method.
S. no.
TSPLIB instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Optimal length
426
108,159
21,282
58,537
2378
11,861
259,045
Average tness
518.83
136,677.89
25,527.27
70,895.65
2979.44
15,316.37
324,520.74
Table 17
Experimental results of VV population seeding method.
S. no.
TSPLIB instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Optimal length
426
108,159
21,282
58,537
2378
11,861
259,045
Average tness
431.73
112,775.72
22,771.79
61,076.87
2555.70
12,906.91
284,340.72
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
ARTICLE IN PRESS
40
Nearest Neighbour
EV
VV
35
30
25
20
15
17
Table 20
Performance order for convergence rate.
S. no.
Instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
VV EV NN VE GB Random
VV EV NN GB Random VE
VV EV NN GB Random VE
VV EV NN GB VE Random
VV EV NN Random GB VE
VV EV NN GB VE Random
VV NN EV GB VE Random
10
5
0
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Average
TSP Instance
Fig. 16. Error rate of different methods analyzed.
Random Populaton
GeneBank
VE
Nearest Neighbour
EV
VV
100
95
90
85
80
75
70
65
60
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
Average
TSP Instance
Fig. 17. Convergence rate of different methods analyzed.
Table 18
Performance order for computation time.
S. no.
Instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
VE Random EV GB VV NN
Random VE EV GB NN VV
VE Random EV GB NN VV
VE Random EV GB VV NN
Random VE EV GB VV NN
Random VE EV GB VV NN
Random VE EV GB NN VV
Instance
1
2
3
4
5
6
7
Eil51
Pr76
Kroa100
Pr144
Gil262
Fl417
Pr1002
VV EV NN VE GB Random
VV EV NN GB Random VE
VV EV NN GB Random VE
VV EV NN GB VE Random
VV EV NN Random GB VE
VV EV NN GB VE Random
VV NN EV GB VE Random
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
18
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
Fig. 18. Convergence graph for TSP instances using different population seeding techniques.
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
19
4.4. Discussion
Appendix A. Supplementary data
In summary, the investigations in the Phase I shows that the
VV population seeding method offers better convergence diversity
and initial convergence rate than the EV which is better than NN,
GB, VE and random techniques. It is also noted that, the computation time of VV method is relatively more than the random, EV
and VE methods. The Phase II investigation proves that the VV and
EV methods of population seeding outperforms the random, GB and
NN, which are claimed to be the better population seeding methods.
Thus, the required characteristics of population seeding like randomness, individual diversity and potential sequence are assured
in the proposed ODV based VV and EV population seeding methods. Though, the VV method works best, it is necessary to point
that it could not achieve the complete 100% of convergence for any
of the test instances after certain point of generations. The tness
value of best solution generated by the VV method at the population seeding stage for instance Eil51 is 91.04%, for which the tness
of the best solution after 200 generations of crossover is 98.98%.
This implies that the OX operator does not completely exploit the
potential sequence of individuals generated at the initial stage.
Therefore, genetic operators like edge-recombination and partition
crossover [59,60], which perform well with local optima solutions,
can be used to make use of potential solutions generated using ODV
population seeding technique. And also, designing an ODV based
crossover operator that exclusively exploit the advantages of individuals generated using ODV methods may enable to attain the
100% of convergence, particularly for large sized TSP instances.
5. Conclusion
In this paper, an efcient Ordered Distance Vector (ODV) based
population seeding technique has been proposed for GA to enhance
its overall performance. The proposed technique consists of three
different methods namely EV, VE and VV, which generate a population of individuals with characteristics such as randomness,
diversity and potential sequence. The Traveling Salesman Problem
(TSP) has been chosen as the testbed and the experiments are performed on different sized TSP benchmark datasets obtained from
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
20
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
[39] W.A. Chang, R.S. Ramakrishna, A genetic algorithm for shortest path routing
problem and the sizing of populations, IEEE Transactions on Evolutionary Computation 6 (December (6)) (2002) 566579, ISSN:1089-778X.
[40] V. Togan, A.T. Daloglu, An improved genetic algorithm with initial
population strategy and self-adaptive member grouping, Journal of
Computers and Structures 86 (June (1112)) (2008) 12041218, ISSN:
0045-7949.
[41] S. Venkatesan, P. Dhavachelvan, C. Chellapan, Performance analysis of mobile
agent failure recovery in e-service applications, in: International Journal of
Computer Standards and Interfaces, Vol. 32, Issue 12, Elsevier, ISSN:09205489, 2005, pp. 3843.
[42] C.F. Tsai, C.W. Tsai, Yang T, A modied multiple-searching method to genetic
algorithms for solving traveling salesman problem, IEEE International Conference on Systems, Man, and Cybernetics 3 (2002) 69.
[43] G. Reinelt, The traveling salesman: computational solutions for TSP
applications, Lecture Notes in Computer Science (1994) 840, Springer
Verlag.
[44] J.L. Bentley, Fast algorithms for geometric traveling salesman problems, ORSA
J Computing 4 (4) (1992) 387411.
[45] D.S. Johnson, L.A. McGeoch, The Traveling Salesman Problem: A Case Study
in Local Optimization. Local Search in Combinatorial Optimization, Wiley and
Sons, New York, 1996.
[46] Y. Lu, S. Lu, F. Fotouhi, Y. Deng, S. Brown, Incremental genetic K-means algorithm and its application, Gene Expression Data Analysis. Submitted to BMC
Bioinformatics (2004).
[47] X. Hue, Genetic Algorithms for Optimization: Background and Applications,
Edinburgh Parallel Computing Centre, Univ. Edinburgh, Edinburgh, Scotland,
Ver 1.0, 1997 February.
[48] J. Lu, B. Feng, B. Li, Finding the optimal gene order for Genetic Algorithm, in:
Fifth World Congress on Intelligent Control and Automation, Vol. 3, WCICA
2004, June, ISBN:0-7803-8273-0, 2004, pp. 20732076.
[49] Z. Jianxin, C.T. Beijing, Solving TSP with novel local search heuristic genetic algorithms, in: Fourth International Conference on Natural Computation, ICNC08,
Vol. 1, Jinan, ISBN:978-0-7695-3304-9, 2008, pp. 670674.
[50] L.-Y. Wang, Z. Jie, H. Li, An improved genetic algorithm For TSP, in: International
Conference on Machine Learning and Cybernetics, Vol. 2, Hong Kong, August,
ISBN:978-1-4244-0973-980, 2007, pp. 925928.
[51] P. Larranaga, C. Kuijpers, R. Murga, I. Inza, Dizdarevic S, Genetic algorithms for
the traveling salesman problem: a review of representations and operators,
Articial Intelligence Review 13 (1999) 129170.
[52] Goldberg DE, Genetic Algorithm in Search, Optimization and Machine
Learning, Machine Learning, Addison-Wesley, New York, 1989, pp.
188.
[53] S.S. Ray, S. Bandyopadhyay, S.K. Pal, New Operators of Genetic Algorithms for
Traveling Salesman Problem, ICPR, Cambridge, UK, 2004, pp. 497500, ICPR-04
2.
[54] http://comopt.i.uni-heidelberg.de/software/TSPLIB95/ (accessed 21.09.12).
[55] J.H. Holland, Adaptation in Natural and Articial Systems, The University of
Michigan Press, Ann Arbor, MI, 1975.
[56] S.S. Ray, S. Bandyopadhyay, S.K. Pal, Genetic operators for combinatorial optimization in TSP and microarray gene ordering, Applied Intelligence 26 (3)
(2007) 183195.
[57] H. Maaranen, K. Miettinen, M.M. Makela, Quasi-random Initial Population for
Genetic Algorithms, Computers and Mathematics with Applications 47 (12)
(2004) 18851895.
[58] S. Lin, B.W. Kernighan, An effective Heuristic Algorithm for the Traveling Salesman Problem, Operations Research 21 (2) (1973) 498516.
[59] D. Whitley, D. Hains, A. Howe, Tunneling between optima: partition
crossover for the traveling salesman problem, in: Proceedings of the 11th
ACM Annual Conference on Genetic and Evolutionary Computation, 2009,
pp. 915922.
[60] D. Whitley, D. Hains, A.E. Howe, A hybrid genetic algorithm for the traveling salesman problem using generalized partition crossover, PPSN 1 (2010)
566575.
[61] J.S. Lawrence, M.M. Amini, Performance characteristics of alternative genetic
algorithmic approaches to the traveling salesman problem using path representation: an empirical study, European Journal of Operational Research (April
(108)) (1998) 551570, Elsevier, ISSN: 0377-2217/98.
[62] L. Qu, R. Sun, A synergetic approach to genetic algorithms for solving traveling
salesman problem, Information Sciences (July (117)) (1999) 267283, Elsevier,
ISSN:0020-0255.
[63] P. Dhavachelvan, G.V. Uma, Reliability enhancement in software testing: an
agent-based approach for complex systems, in: 7th ICIT 2004, Springer Verlag
Lecture Notes in Computer Science (LNCS), Vol. 3356, ISSN:0302-9743, 2004,
pp. 282291.
[64] L.-N. Xing, Y.-W. Chen, K.-W. Yang, F. Hou, X.-S. Shen, H.-P. Cai, A
hybrid approach combining an improved genetic algorithm and optimization
strategies for the asymmetric traveling salesman problem, in: Engineering
Applications of Articial Intelligence, Elsevier, Issue 21, April, ISSN:0952-1976,
2008, pp. 13701380.
[65] P. Victer Paul, N. Saravanan, S.K.V. Jayakumar, P. Dhavachelvan, R. Baskaran, QoS
enhancements for global replication management in peer to peer networks,
Future Generation Computer Systems 28 (3) (2012) 573582.
[66] P. Dhavachelvan, G.V. Uma, Multi-agent based framework for intra-class testing
of object-oriented software, International Journal on Applied Soft Computing
5 (2) (2005) 205222, Elsevier.
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009
G Model
JOCS-208; No. of Pages 21
ARTICLE IN PRESS
P. Victer Paul et al. / Journal of Computational Science xxx (2013) xxxxxx
Dr. R. Baskaran is currently associate professor, Department of Computer Science and Engineering, Anna
University. He completed his B.Tech. in electrical and
electronics engineering (2000) and M.Tech. in computer
science and engineering (2001). He obtained his Ph.D. in
computer science and engineering in 2007 from Anna University, Chennai. He has around a decade of experience
in the education and research. His research areas include
evolutionary computing, service computing and multimedia databases. He has published more than 60 research
articles in International & National Journals, Conferences
and Books. He is the member of various National and
International bodies like The Institution of Electronics and
Telecommunication Engineers, Computer Society of India (CSI), International Network for Engineering Education and Research, International Association of Engineers
and International Congress for Global Science and Technology.
21
Dr. P. Dhavachelvan is working as the professor, Department of Computer Science, Pondicherry University, India.
He completed his B.Tech. in electrical and electronics engineering in Madras University (1997), India. He obtained
his M.Tech. in computer science and engineering (2000)
and Ph.D. in computer science and engineering (2007)
from Anna University, Chennai. He is having around 15
years of experience as an Academician, Researcher and
Administrator. Presently he is heading the Department of
Computer Science, Pondicherry Central University, India.
His research areas include Software Engineering, Web
Service Computing and Evolutionary Algorithms. As the
Main and Coauthor, he has more than 100 publications in
his credit. The publication list includes National and International Journals, National
and International Conferences, Books and Book Chapters.
Dr. K.Vivekanandan is working as the senior professor
in the Department of Computer Science and Engineering, Pondicherry Engineering College, India. He obtained
his B.E. degree in electronics and communication engineering from Coimbatore Institute of Technology in 1986
and M.Tech. in computer science and engineering from
Indian Institute of Technology, Bombay (India) in 1991. He
received his Ph.D. in computer science and engineering
from Pondicherry University, India in 2005. He is having around 21 years of experience in teaching, research
and administration. He worked in Indian Institute of Science, Bangalore and Central Research Laboratory, Bharat
Electronics Limited, Bangalore. Since 1992 he has been
working in Pondicherry Engineering College. His research interest includes Software Engineering, Web Service Computing and Information Security. In his credit,
he is having around 50 National and International publications in various forums
like Journals, Conferences and Books.
Dr. R. Subramanian is the senior professor in the
Department of Computer Science, Pondicherry Central
University, India. He completed his B.Sc. in mathematics in the Madurai Kamaraj University, India in 1982. He
received his M.Sc. and Ph.D. in mathematics from Indian
Institute of Technology, Delhi, India in 1984 and 1989
respectively. He is having around 23 years of experience
in teaching and research. As a part of administration, he
had been the HOD of the Department of Computer Science
and currently he is the Dean of the School of Engineering
and Technology, Pondicherry Central University, India. His
specialization includes parallel and distributed systems,
robotics and evolutionary algorithms. He has published
more than 50 National & International Journal & Conference publications, Books
and Book Chapters.
Please cite this article in press as: P. Victer Paul, et al., A new population seeding technique for permutation-coded Genetic Algorithm:
Service transfer approach, J. Comput. Sci. (2013), http://dx.doi.org/10.1016/j.jocs.2013.05.009