Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Meta-Heuristic Hybrid Algorithmic Approach

for Solving Combinatorial Optimization


Problem(TSP)

Usman Ashraf1 , Jing Liang1 , Aleena Akhtar1 , Kunjie Yu1 , Yi Hu1 , Caitong
Yue1 , Abdul Mannan Masood2 , and Muhammad Kashif3
1
School of Electrical Engineering, Zhengzhou University, Zhengzhou 450001, China
2
School of Computer Science and Engineering, Jiangsu University of Science and
Technology, Jiangsu, China
3
School of Computer Science and Technology, Beijing Institute of Technology,
Beijing, China
usmanjadoon@ymail.com

Abstract. Solving and optimizing combinatorial problems require high


computational power because of their exponential growth and require-
ment of high processing power. During this study, a hybrid algorithm
(Genetic Ant Colony Optimization Algorithm) is proposed in comparison
with standard algorithm (Ant Colony Optimization Algorithm). Further
the parameters for Ant Colony Optimization Algorithm are instinctively
tuned to different levels of all heuristics to obtain suboptimal level, then
multiple crossovers and mutation operators are used alongside those se-
lected parameters while generating results with hybrid algorithm. The
main emphasis of the proposed algorithm is the selection and tuning of
parameters, which is extremely influential in this case. The algorithm has
been tested on six benchmarks of TSPLIB. The results were compared
with standard ACO algorithm, the hybrid algorithm outperformed the
standard ACO algorithm.

Keywords: Ant Colony Algorithm (ACO) · Genetic Algorithm (GA)


· Traveling Salesman Problem (TSP) · Heuristics · Meta-Heuristics ·
Optimization.

1 Introduction

Finding optimal solution is the main focus of combinatorial optimization from a


finite set of objects. These optimization problems have always been a prime focus
in every field of Engineering. With the increasing importance of Optimization
problems the main focus is to utilize less resources and make it more efficient
[1]. Traveling Salesman Problem, TSP for short, can be best defined as “Given
a number of specific cities along with the cost from Point A to all other points,
finding the efficient and cheapest way back to initial point after visiting all cities
once and only once” [2]. One complete cycle through all the points once and
2 Usman Ashraf et al.

only once by a traveling salesman is called a tour.

Traveling Salesman Problem (TSP) is known Non Polynomial(NP) hard


problem which is excessively studied for the research work in field of computer
science and especially Artificial Intelligence [3]. The general description of the
symmetric TSP can be seen in Fig. 1. There are 5 cities which are called nodes
and the cost from one node to every other node is also labeled [2][4].

Fig. 1. General TSP Representation.

TSP is generally interpreted by the complete edge weighted graph,

G = (V, E) (1)

Where V is absolute number set, for number of cities or nodes while E ⊆ V × V


which represents the set of edges. Each are (i, j) ∈ E assign value dij to the
cities from city i to city j where i and j belongs to V . The set of TSP data can
either be symmetric or asymmetric. The difference between the symmetric and
asymmetric is that the distance between two nodes remains constant from i to j
and j to i presented dij = dji while in asymmetric the i to j and j to i is never
the same which can be interpreted as dij 6= dji . Optimal solution for TSP is the
permutation (π) of the cities indices from {1...n} such that f (π) is the possible
minimum which can be represented mathematically by[5][6][7];
n−1
X
f (π) = d π (i) π (i + 1) + d π (n) π (1) (2)
i−1

In recent decades, researchers have focused on different computational intelli-


gence algorithms to solve the traveling salesman problem, such as neural network,
simulated annealing method, genetic algorithm GA, particle swarm optimization
and so on[8][9]. An Italian scholar M.Dorigo first proposed it, taking full advan-
tage by introducing similarity between both, travelling salesman problem TSP
and ant colony search process[10]. Recently researchers are using ant colony
Meta-Heuristic Hybrid Algorithm for TSP 3

algorithm to solve non-deterministic polynomial (NP) problems, e.g. Travel-


ing Salesman Problem[11], wireless sensor networks[12], Capacitated Location
Routing Problem, software project scheduling problem[13], Course Timetabling
Problem[14], time series prediction and dynamic topology optimization[15], etc.
In a research for improved ACO for solving TSP, local optimal search strat-
egy and change of parameters were introduced to avoid premature stagnation
phenomenon of standard ACO[16]. In another study researchers compared three
algorithms to solve TSP Ant colony algorithm, Evolution based and Annealing
algorithm and the results concluded that the nature inspired algorithm ACO
has outperformed the Evolution based and Annealing algorithm[17].

Multiple literatures have discussed the use of GA for solving TSP, and many
literatures generally focused on designing the representation of solution, and
initializing an initial population [18][19], designing crossover, mutation and se-
lection operator and designing a self-adaptive method to use multiple mutation
or crossover operators[20-22]. Research in which a new method called greedy
permuting method (GPM) was introduced to initialize an initial population and
then the efficiency of proposed method was tested on some TSP benchmark
problems, with promising results[23].

2 Methodology
Generally, the parameter tuning for GA is problem oriented and set as per
requirements but in case of TSP it has been tuned as per the work requirement.
The selection of parameter which should be tuned are selection, cross over and
mutation probability which in general varies from 0.1 to 1. GA specifically uses
fixed length of chromosome strings for coding the solution of the problem. There
should be a proper balance between exploitation and exploration. GA always
fails because of the not being balanced properly between crossover and mutation
probability. Proper allocation of values will help the algorithm not to be trapped
in local solution.

2.1 Ant Colony Optimization Algorithm


For the purpose of finding the best parameters for ACO TSPLIB’s instance
Berlin52 was taken for instance and parameters were tuned to see the effects on
the results. The factors which effect the results are discussed as follows. At this
point basic parameters used are as shown in Table 1.

Table 1. Basic parameters

α β ρ % M axIt nAnT
1 2 0.9 10 1000 30
4 Usman Ashraf et al.

a. Pheromone Rate First, the other parameters will be kept constant, only
the value of α will be changed. Result can be seen as shown below in Table 2.

Table 2. Pheromone rates

α β ρ % M axIt nAnT Best Solution


1 2 0.9 10 1000 30 7548.9927
3 2 0.9 10 1000 30 7758.7176
5 2 0.9 10 1000 30 8971.4955
7 2 0.9 10 1000 30 9090.2158
9 2 0.9 10 1000 30 8804.8284

b. Heuristic Rate Now keeping other parameters constant while changing the
value of β, and observe the changes in results as shown Table 3.

Table 3. Effects of Heuristic rates

α β ρ % M axIt nAnT Best Solution


1 1 0.9 10 1000 30 8083.8025
1 2 0.9 10 1000 30 7548.9927
1 3 0.9 10 1000 30 7663.5851
1 5 0.9 10 1000 30 7721.2979
1 7 0.9 10 1000 30 7681.4537
1 9 0.9 10 1000 30 7681.4537

c. Pheromone Evaporation Rate Now keeping other parameters constant


while changing the value of ρ, and observe the change in results as shown in
Table 4.

Table 4. Pheromone rates

α β ρ % M axIt nAnT Best Solution


1 2 0.1 10 1000 30 7677.6608
1 2 0.3 10 1000 30 7662.8103
1 2 0.5 10 1000 30 7601.0492
1 2 0.7 10 1000 30 7629.4472
1 2 0.9 10 1000 30 7548.9927
Meta-Heuristic Hybrid Algorithm for TSP 5

2.2 Genetic Ant Colony Optimization Algorithm

The hybrid algorithm solving TSP can be expressed as follows:

Algorithm 1 Genetic Hybrid Ant Colony Optimization Algorithm


1) c ←0 (c is iteration number).
2) Generate 100 tours through Ant Colony Optimization algorithm,
a. nc ←0 (nc is iteration number).
b. Choose the next city j according to
 α β
τij (t).ηi j
f j ∈ allowedk
P
pkij (t)
β
= τiαs (t).ηi s
s ∈ allowedk
0 otherwise

c. Update trail (pheromone) values.


τij (t + n) = ρτij (t) + ∆τij
Pm k
∆τij = k=1 ∆τij
 Q
if k − th ant uses edge(i, j) in its tour
∆τikj = Lk
0 otherwise

d. nc ← nc+1
e. If the iteration number nc reaches the maximum iteration number, then
go to Step a. Otherwise, go to Step 3.
3) Choose the better 30 tours based on the cost Lk (length of tour done by ant
k) from these 100 tours, and pheromone laid on edge of these 30 better tours.
4) Crossover and Mutate the tours and calculate the new evaluated tours,
a. mc ←0 (mc is iteration number).
b. Choose the next city j
c. mc ← mc+1
d. If the iteration number mc reaches the maximum iteration number, then
go to Step a. Otherwise, go to Step 5.
5) Compute Lk (k = 1, 2, .., m) (Lk is the length of tour done by ant k). Save the
current best tour.
6) c ← n+1
7) If the iteration number c reaches the maximum iteration number, then go to Step
8. Otherwise, go to Step 2.
8) Print the current best tour.

In simple ACO Algorithms, in start the value of pheromone matrix is equal.


Multiple iterations process is needed by Ants to find the best tour. Trail can be
found on each visited edge, after an ant completes a tour. No matter the tour is
better or worse, even for the worse tour ants lay the trail on each edge, disturb-
ing the ants which are following them which results in the slower convergence
speed of ACO algorithm. At first the length of the tour is calculated and then it
is compared with the given value. If the calculated value is lesser than the given
6 Usman Ashraf et al.

value then the trail value is updated, else it remains the same. This results in
improvement of ants laying on the better tours, and also affect the ants which
are following. A large number of tours can be generated (e.g. 100), out of which
some better tours (e.g. 30) are then selected. These selected best tours undergo
mutation and crossover operations under Genetic Algorithm, and after these op-
erations are performed then we can calculate the cost of newly mutated tours.

A. Genetic Ant Colony Optimization Parameters Results from basic Ant


Colony Optimization Algorithm shows that the following parameters would be
proved to get the better results. Parameters are shown in Table 7.

B. Genetic Algorithm Parameters In this methodology we only used two


fundamental Genetic Algorithm parameters, namely mutation and crossover.
Several Mutation methods are used to improve the discussed hybrid algorithm.

Mutation
1. Mutation A
Strategy: Swap a randomly selected city with next visiting city.
Detail: Choose one city j1 randomly and then swap it with the next visiting city
in the tour C0 , this will generate a new tour calling C1 .
2.Mutation B
Strategy: swap a randomly selected city with the city after next visiting city.
Detail: Choose one city j1 randomly and then swap it with the city after next
visiting city in the tour C0 , this will generate a new tour calling C1 .
3. Mutation C
Strategy: Swap a randomly selected city and the city adjacent to it with the
next visiting city.
Detail: Choose random city j1 , then swap j1 and next adjacent city with after
coming next city the tour C0 , this will generate a new tour calling C1 .
4. Mutation D
Strategy: Swap randomly chosen 2 adjacent cities with 2 other random adjacent
cities.
Detail: Choose two adjacent cities j1 and j2 randomly and then swap them ad-
jacently at a randomly selected place in the tour C0 , this will generate a new
tour calling C1 .
5. Mutation E
Strategy: Put randomly chosen 2 adjacent cities at a random place.
Detail: Choose two adjacent cities j1 and j2 randomly and then put them adja-
cently at a randomly selected place in the tour C0 , this will generate a new tour
calling C1 .
6. Mutation F
Strategy: 2 different randomly selected cities put adjacently at a place adjacently.
Meta-Heuristic Hybrid Algorithm for TSP 7

Detail: Choose two random cities j1 and j2 , then put them adjacently at a ran-
domly selected place in the tour C0 , this will generate a new tour calling C1 .
7. Mutation G
Strategy: 2 different randomly selected cities put adjacently at 2 different ran-
dom locations.
Detail: Choose two random cities j1 and j2 , then swap them at randomly selected
places separately in the tour C0 , this will generate a new tour calling C1 .

Crossover
Similarly, several Crossover methods were used to improve the currently dis-
cussed hybrid algorithm. Let’s suppose we have two parent tours given by old1
and old2. A substring from old2 for crossover purpose will be selected and it will
be called as donor.
1. Crossover A
Strategy: Replace random length substring at the same place in the old1.
Detail: We will swap a substring of random length generated from old2. This
substring will be then replaced at the exact starting location of the old1.
2. Crossover B
Strategy: Replace random length substring at the beginning at the old1.
Detail: A substring is selected randomly of random length. Then we insert the
substring into the beginning of old1.
3. Crossover C
Strategy: Replace random length substring at the end at the old1.
Detail: A substring is selected randomly of random length. Then we insert the
substring into the end of old1.
4. Crossover D
Strategy: Replace random length substring at the random place at the old1.
Detail: A substring is selected randomly of random length. Then we insert the
substring at a random place of old1.
5. Crossover E
Strategy: Replace random length substring starting in first half and ending in
second half of old2, placed at a random location.
Detail: substring will be selected from random place of first half of the string
and ending in second half of the string. The length of this substring will random.
Then we insert the substring into random location of old1.
6. Crossover F
Strategy: Replace substring of 1/4th length of old2 starting in first half of old2,
placed at a random location.
Detail: this substring will be selected from random place of first half of the string.
The length of this substring will be 1/4th of the length of full string. Then we
insert the substring into random location of old1.
7. Crossover G
Strategy: Replace substring of 1/4th length of old2 starting in first 1/3rd part of
old2, placed at a random location.
Detail: This substring will be selected from random place of first 1/3rd part of
8 Usman Ashraf et al.

the string. The length of this substring will be 1/4th of the length of full string.
Then we insert the substring into random location of old1.
8. Crossover H
Strategy: Replace substring of 1/4th length of old2 starting in middle half part
of old2, placed at a random location.
Detail: this substring will be selected from random place of first middle half of
the string. The length of this substring will 1/4th of the length of full string.
Then we insert the substring into random location of old1.
9. Crossover I
Strategy: Replace substring of 4 adjacent cities from old2 starting in middle half
part of old2, placed at a random location.
Detail: this substring will be selected from random place of first middle half of
the string. The length of this substring will be of 4 cities. Then we insert the
substring into random location of old1.
10. Crossover J
Strategy: Replace substring of 4 adjacent cities from old2, placed at a random
location.
Detail: this substring will be selected from random place of first middle half of
the string. The length of this substring will be of 4 cities. Then we insert the
substring into random location of old1.

3 Results
Ant Colony Optimization and Genetic Ant Colony Optimization, which are dis-
cussed above have been run for 10 times with the best parameters known as per
previous testing. After applying the best parameters, the algorithms have been
run for 10 times on the six Benchmarks of TSPLIB varying from 16 cities to 120
cities. The Benchmarks used as the test bed are Ulysses16, Ulysses22, bayg29,
att48, berlin52 and g120m. All results are rounded up to two decimal points.
The results obtained for the benchmark of TSP are shown below. The Results
are compared with basic ACO Algorithm, and it can be clearly noticed that the
hybrid Algorithm outperformed the basic ACO algorithm. Environment used to
run the above said strategies was suitable enough to obtain good results and has
the following characteristics.

• Processor: INTEL(R) core(TM) i7-6500U CPU@2.50GHZ


• RAM: 16 Gb
• Graphic Card: AMD Radeon R5 M335
• Operating System: Microsoft Windows 10
• Simulation Software: MATLAB R2015a
All the benchmarks of the TSPLIB varying from 16 cities to 120 cities ran
ten times to get the better results. Each of the TSPLIB has been tested with
all the combinations of proposed mutations and crossovers that produces 70 dif-
ferent variations of Hybrid Algorithm. To check the difference between different
Meta-Heuristic Hybrid Algorithm for TSP 9

Table 5. Minimum result comparison

Algorithms Ulysses16 Ulysses22 bayg29 att48 berlin52 g120m


Basic ACO algorithm 53.75 56.61 9054.3 35945 8180.36 612.45
Crossover operator A +
53.65 56.52 8988.2 35676 7835.3 604.61
Mutation operator A
Crossover operator B +
51.65 56.76 8497.09 21089.2 8024.58 1245.26
Mutation operator A
Crossover operator B +
55.21 65.59 9625.45 35298.62 5970.83 912.95
Mutation operator D
Crossover operator B +
54.25 57.76 9871.96 13908.4 8561.05 893.93
Mutation operator F
Crossover operator E +
56.92 59.5 9258.99 27373.26 7893.41 852.49
Mutation operator D
Crossover operator J +
53.22 65.57 6357.52 37003.63 8540.21 1437.02
Mutation operator F
Minimum Values 51.65 56.52 6357.52 13908.4 5970.83 604.61

variations of Hybrid Algorithm, 5 minimum values have been selected from each
TSPLIB. Further to refine these results only highly productive variations of Hy-
brid Algorithm were selected that have been shown in Table 5 and then plotted
in Figure 2. Where it can be easily deduced that as the number of cities or the
distance between the cities increases hybrid algorithm produces better results
than basic ACO.

Further to check the stability of the Hybrid Algorithm, statistical analysis


has been performed on the data, and the values of mean and standard deviation
are presented in Table 6.

Table 6. Mean(µ) and standard deviation(σ)

Algorithms Ulysses16 Ulysses22 bayg29 att48 berlin52 g120m


Crossover Operator A+ 59.12 61.96 9328.06 37144.06 8076.23 698.14
Mutation Operator A ±4.41E+0 ±4.33E+0 ±6.52E+2 ±2.63E+3 ±5.65E+2 ±4.88E+1
Crossover Operator B+ 72.09 69.67 10968.28 25714.56 9418.77 1428.16
Mutation Operator A ±5.04E+0 ±4.87E+0 ±7.68E+2 ±1.80E+2 ±6.59E+2 ±9.99E+2
Crossover Operator B+ 78.89 91.11 12397.15 52689.81 8698.34 1174.95
Mutation Operator D ±5.52E+0 ±6.37E+0 ±8.67E+2 ±3.69E+3 ±6.08E+2 ±8.22E+1
Crossover Operator B+ 77.26 79.03 11944.33 20122.17 10440.93 1146.93
Mutation Operator F ±5.40E+0 ±5.53E+0 ±8.36E+3 ±1.41E+3 ±7.30E+2 ±8.02E+1
Crossover Operator E+ 83.99 82.77 13042.37 37047.24 9916.16 1248.59
Mutation Operator D ±5.88E+0 ±5.79E+0 ±9.13E+2 ±2.59E+3 ±6.94E+2 ±8.74E+1
Crossover Operator J+ 79.44 84.09 8556.23 48696.73 10385.86 1639.78
Mutation Operator F ±5.56E+0 ±5.89E+0 ±5.99E+2 ±3.41E+3 ±7.27E+2 ±1.15E+2
10 Usman Ashraf et al.

Fig. 2. Results Comparison

The ANT Colony Optimization Algorithm has been run for the above men-
tioned Benchmarks for 10 times and then those results are used as benchmarks
to compare with other results generated from hybrid Genetic Ant Colony Algo-
rithm. Best solution has been taken from the 10 test runs. The parameter set
for the algorithms are as follows in Table 7.

Table 7. Genetic Ant colony optimization parameters

Paramter Value
Pheromone Exponential weight (α) 1
Heuristic Exponential weight (β) 2
Evaporation Rate (ρ) 0.9
% 10
Number of Ants 30
Selected Best tours from ACO 30
Number of tours Generated (M axIt) 1000
Number of Mutations 1000

4 Conclusion
This research proposed a hybrid algorithm to see the change in results with re-
spect to standard algorithm (Ant Colony Optimization Algorithm) and Hybrid
algorithm (Genetic Ant Colony Optimization Algorithm). Further the perime-
ters for Ant Colony Optimization Algorithm and values are instinctively tuned
to different levels of all heuristics to obtain suboptimal level. And then multi-
ple crossovers and mutations are used alongside those selected parameters while
Meta-Heuristic Hybrid Algorithm for TSP 11

generating results with hybrid algorithm. From the results in section 3 the fol-
lowing information can be concluded for the selection of crossover and mutation
as shown in Table 8. The results show that if the mutation, crossover and their

Table 8. Selection of Crossover and Mutation

Crossover Mutation
Ulysses16 B A
Ulysses22 A A
bayg29 J F
att48 B F
berlin52 B D
g120m A A

combination in the proposed hybrid Genetic Ant Colony Algorithm is chosen


accordingly, this hybrid algorithm can generate better results. There are other
hybrid algorithms that have been presented with the passage of time which have
performed efficiently for many cases. But during this study it is very clear that
selection and tuning of parameters is extremely influential for the purpose of
getting good results. There is always a room for improvement, so does these al-
gorithms can be improved individually and as a hybrid. We can always improve
this hybrid of two algorithms with the help of introducing another algorithm.

5 Acknowledgement
This work is supported by the National Natural Science Foundation of China
(61922072, 61876169, 61673404, 61976237).

References
1. K. Ilavarasi and K. S. Joseph, Variants of travelling salesman problem: A survey
”International Conference on Information Communication and Embedded Systems
(ICICES2014),” 2014.
2. David Applegate and William Cook, ”The Traveling Salesman Problem: A Compu-
tational Study,” in Princeton Series in Applied Mathematics, Princeton, NJ, USA,
Princeton University Press, 2007, pp. Chapter 1–5,12–17.
3. Malik Muneeb Abid and Iqbal Anjum Muhammad, ”Heuristic Approaches to Solve
Traveling Salesman,” TELKOMNIKA Indonesian Journal of Electrical Engineering,
vol. 15, no. 2, p. 390 396, August 2015.
4. Cook and William, ””History of the TSP.” The Traveling Salesman Problem.,” Geor-
gia Tech, October 2009. [Online]. Available: http://www.tsp.gatech.edu/index.html.
5. Amanur Rahman Saiyed, ”The Traveling Salesman problem,” Indiana State Uni-
versity USA, April 11, 2012.
6. N. M. Sureja and B. V. Chawda, ”Random Travelling Salesman Problem using SA,”
International Journal of Emerging Technology and Advanced Engineering, vol. 2,
no. 3, 2012.
12 Usman Ashraf et al.

7. S. Wang and A. Zhao, ”An Improved Hybrid Genetic Algorithm for Traveling Sales-
man problem,” IEEE Proc. Int. Conf. Computational Intelligence and Software En-
gineering, 2009.
8. Y. Nagata, S. Kobayashi, ”A powerful genetic algorithm using edge assembly
crossover for the traveling salesman problem”, Informs Journal on Computing, vol.
25, no. 2, pp. 346-363, 2013.
9. P. Tian, Z. H. Yang, ”An improved simulated annealing algorithm with genetic
characteristics and the traveling salesman problem”, Journal of Information and
Optimization Sciences, vol. 14, no. 3, pp. 241-255, 2013.
10. M. Birattari, P. Pellegrini, M. Dorigo, ”On the invariance of ant colony optimiza-
tion”, IEEE Transactions on Evolutionary Computation, Vol.11, No.6, 732-742,
2008.
11. E. Lizárraga, O. Castillo, J. Soria, ”A method to solve the traveling salesman
problem using ant colony optimization variants with ant set partitioning”, Studies
in Computational Intelligence, vol. 451, no. 1, pp. 237-246, 2013.
12. J. H. Ho, H. C. Shih, B. Y. Liao, S. C. Chu, ”A ladder diffusion algorithm using
ant colony optimization for wireless sensor networks”, Information Sciences, Vol.192,
No.6, 204-212, 2012.
13. J. Xiao, X. T. Ao, Y. Tang, ”Solving software project scheduling problems with
ant colony optimization”, Computers and Operations Research, vol. 40, no. 1, pp.
33-46, 2013.
14. C. Nothegger, A. Mayer, A. Chwatal, G. Raidl, ”Solving the post enrolment course
timetabling problem by ant colony optimization”, Annals of Operations Research,
Vol.194, No.1, 325-339, 2012.
15. K. S. Yoo, S. Y. Han, ”A modified ant colony optimization algorithm for dynamic
topology optimization”, Computers and Structures, vol. 123, no. 4, pp. 68-78, 2013.
16. Xue Yang, Jie-sheng Wang “Application of improved ant colony optimization al-
gorithm on traveling salesman problem” in IEEE 2017 International Conference on
Robotics and Automation Sciences (ICRAS)
17. Kashif Muhammad, Shang Gao, “Comparitive analysis of meta-hueristic algo-
rithms for solving optimization problems”, Advances in Intelligent Systems Re-
search, volume 163, Proceedings of the 2018 8th International on Management,
Education and Information (MEICI 2018).
18. Y. Deng, Y. Liu, D. Zhou, ”An Improved Genetic Algorithm with Initial Population
Strategy for Symmetric TSP”, Mathematical Problems in Engineering, vol. 2015,
no. 3, pp. 1-6, 2015.
19. Y. Luo, B. Lu, F. Liu, ”Neighbor field method for population initialization of TSP”,
Journal of Chongqing University, vol. 32, no. 11, pp. 1311-1315, 2009.
20. Ahmad B. Hassanat, E. Alkafaween, ”On Enhancing Genetic Algorithms Using
New Crossovers”, International Journal of Computer Applications in Technology,
vol. 55, no. 3, pp. 202-212, 2017.
21. Noraini Mohd Razali, John Geraghty, ”Genetic algorithm performance with dif-
ferent selection strategies in solving TSP”, Proceedings of the World Congress on
Engineering 2011 (WCE 2011), vol. 2, pp. 1134-1139, 2011.
22. Martin C. Serpell, James E. Smith, ”Self-adaptation of mutation operator and
probability for permutation representations in genetic algorithms”, Evolutionary
Computation, vol. 18, no. 3, pp. 491-514, 2010.
23. Junjun Liu, Wenzheng Li “Greedy Permuting Method for Genetic Algorithm on
Travelling Salesman Problem” in IEEE 8th International Conference on Electronics
Information and Emergency Communication (ICEIEC) 2018

You might also like