Uploaded by NixieWar

6-Jurnal-Submit

advertisement
Hybrid of Ant Colony Optimization (ACO) and
Tabu Search (TS) Algorithm to Solve the
Traveling Salesman Problem (TSP)
1st Rico Wijaya Dewantoro,
Computer Science Department,
Faculty of Computer Science and
Information Technology, Universitas
Sumatera Utara,
Jl. Dr. Mansur No.9, Padang Bulan,
Medan,
[email protected]
2nd Poltak Sihombing,
Computer Science Department,
Faculty of Computer Science and
Information Technology, Universitas
Sumatera Utara,
Jl. Dr. Mansur No.9, Padang Bulan,
Medan,
[email protected]
Abstract—In this research, we propose the combination of
Ant Colony Optimization Algorithm and Tabu Search Algorithm
as local search to solve Traveling Salesman Problem. This is a
hybrid method of ACO to find the shortest path value and get a
better running time. One of the popular case that can be adopted
is TSP. We still develop the prototype of the software by using the
combination of ACO and TS in order to find the best routes and
better running time. In this research, we compare the hybrid of
ACO-TS and ACO. The result, we got the best routes and a
better running time from hybrid of ACO-TS compared to the
ACO itself. It means that combination of ACO-TS is better than
ACO itself. Therefore, to get a better running time in solution of
TSP and more optimal length of route, we recommended the
ACO-TS algorithm than ACO itself.
3rd Sutarman,
Computer Science Department,
Faculty of Computer Science and
Information Technology, Universitas
Sumatera Utara,
Jl. Dr. Mansur No.9, Padang Bulan,
Medan
[email protected]
pheromones on the way while traveling from the nest to food,
in order to communicate with one to another to find the shortest
pathway. The ants have to make a decision that they should go
left or right, and the choice that is made is an erratic decision.
The accumulation of pheromones is faster on the shorter
pathway. The difference of pheromones value in each path
from time to time makes the ants choose the shortest path based
on the number of pheromones on the shortest path [2], [3].
Keywords— TSP, ACO, Tabu Search, Optimization, ACOTS
I. INTRODUCTION
In the case of the ACO algorithm, theoretical analysis of the
run time behavior of ACO has begun and has always been a
discussion as a future development so that the ACO algorithm
becomes more efficient. In TSP problems, ACO is one
algorithm that is often used. ACO is always a topic of
discussion from a theoretical perspective because of its
behavior[1]. ACO is inspired by the intelligence of ants in the
search of the shortest path to food sources. ACO applies an
optimization problem solving method based on the principle of
communication between ant colonies. Basically, all ants will
leave traces of special substances known as pheromones. This
pheromone will then be a guideline for other ants in searching.
If the shorter the path visited, the less evaporation occurs and
the higher the pheromones in the path or track. Ants will tend
to move along a track that has a high trace of pheromones.
Fig.1 shows, the ant’s colony find the shortest path. In Fig.1
A, the ant colony has to decide that they should to turn left or
right randomly because they don’t know which is the best
choice. In Fig. 1B, 1C and 1D show, the ants put the secrete
XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE
Fig. 1. The ants’ colony find the shortest path
In the early 90s the ACO algorithm had increased attention
by a number of researchers. ACO has solved various of
optimization problems such as: traveling salesman problem
(TSP), vehicle routing problem (VRP), job-shop scheduling
problem (JSP), quadratic assignment problem (QAP), WeaponTarget Assignment problems (WTA) etc [3]. ACO can be also
used to provide solutions for difficult optimization problems.
Although ACO is able to provide solutions for optimization
problems, ACO has problems with stagnation and premature
convergence and ACO convergence speeds that are always
slow. These problems will become clearer when the size of the
problem increases. Therefore, several improvements to the
ACO algorithm have been introduced for years. Various
adaptations are carried out, such as: dynamic control of
solution construction and emergence of local search [4]–[6],
new pheromone updating strategies [7], max-min ant system
[8], a strategy is to partition artificial ants into two
groups: scout ants and common ants[2], using candidate lists
strategies [9], [10], dynamic ant colony system with three level
updates [11], [12] and using the path selection controlled
by information entropy [13] studied to improve the quality of
the final solution and continue to lead to improved algorithms.
All of these studies that have contributed to optimizing the
ACO Algorithm have only reached a certain limit, but have a
few obvious effects on increasing the speed of convergence and
obtaining global optimal solutions. To improve the
performance of the ACO algorithm, we used Tabu Search
Algorithm as a local search. The results of the combination of
ACO-TS Algorithm show that this combination can improve
the performance of ACO algorithm.
A simple description of the TSP understanding can be
shown in Fig. 2.
A
YA
E
C
B
G
YB
D
F
II. THE METHOD
XB
This paper proposes the combination of the ACO algorithm
and Tabu Search algorithm as a local search to solve TSP
problems.
A.
Traveling Salesman Problem (TSP)
One of the well-known and widely studied problems in
discrete optimization or combination is the Traveling
Salesman Problem (TSP) [14]. The TSP can be applied to
various activities such as routing. The main problem of TSP is
that a salesman must visit a number of cities that are known to
be one distance from each other and return to their hometown.
A TSP can be defined as follows: there is a set of cities
{C1, C2, C3, ... Cn} and d {Ci, Cj} is the distance between
cities i and j. The purpose is to find the order π from the
following formula to get the most minimal value.
(1)
XA
Fig. 2. The position of the cities to be passed
Each city on Fig. 2 has coordinates (x, y), so that the distance
of each city can be calculated by formula (2).
(2)
After the distances that connecting each city are known, then
find the shortest route from the path to be passed to return to
the initial city is shown in Fig. 3.
A
E
Start
The results of the formula are referred to as the length of
the journey of the salesman visits the cities in the order π
where after visiting all cities, the salesman returned to his
hometown.
C
G
There are two kinds of problem in TSP, i.e.: Symmetric
Salesman Problem commonly called Traveling Salesman
Problem and Asymmetric Salesman Problem. On the
Symmetric Salesman Problem, d {Ci, Cj} = d {Cj, Ci}, if
Asymmetric Salesman Problems then d {Ci, Cj} ≠ d {Cj, Ci}
for 1≤i and j≥n.
TSP is one of the optimization problems which is an NPcomplete problem (nondeterministic polynomial complete
problem), which is a problem whereby increasing variables
linearly, the time to solve the problem increases exponentially.
This is also called combinatorial exponential.
TSP can be simply defined as the process of finding the
most efficient and shortest paths from several cities presented,
passing through each of these cities and returning to the initial
city. Every city can only be visited once. The problem facing
the TSP is how to plan a minimum total distance. To solve this
problem, it is not easy to do because there is a search room
from a set of permutations in a number of cities, then the TSP
is then known as the Non Polynomial problem.
B
D
F
Fig. 3. Path in the TSP
B.
Ant Colony Optimization (ACO)
The ACO algorithm was presented by Marco Dorigo. ACO
algorithm inspired by the intelligence of ants in the search of
the shortest path to food sources. ACO applies an optimization
problem solving method based on the principle of
communication between ant colonies. Basically, all ants will
leave traces of special substances known as pheromones. This
pheromone will then be a guideline for other ants in searching.
If the shorter the path visited, the less evaporation occurs and
the higher the pheromones in the path or track. The ants will
tend to move along a track that has a high trace of pheromones.
Steps in completing the optimal path search using the ACO
algorithm, i.e.:
1. Input the TSP route data set
2. Initialize initial parameters, namely α, β, ρ, τ0 = τij, Q, NC
max.
3. Fill in the first node in the tabu list.
4. Preparation of visited routes for each ant in each of the
points passed.
5. Visit other nodes that are not in the tabu list based on
random proportional rule. The node with the greatest
probability is the node to be visited. Then add it to the tabu
list.
6. Calculate the number of pheromone in the tabu list.
7. Perform renewal of pheromones in each path and empty
the tabs of each ant list.
8. Repeat the fifth process until the sixth process using
renewed pheromones as the initial pheromones of each
segment or stop the process if the NC max has been
fulfilled.
9. Then specify a tabu list with the best value with the highest
probability value.
C.
Tabu Search
The Tabu Search algorithm is one of the algorithm that is
within the scope of the heuristic method. This algorithm uses
short-term memory to keep the search process from being
trapped at the local optimum value [15]. We then use the tabu
list to store a set of solutions that have just been evaluated.
During the optimization process of each iteration, the solution
to be evaluated will be matched to the contents of the tabu list
in order to see if the solution is in the tabu list or not. If the
solution is already in the tabu list, then the solution will not be
evaluated anymore in the next iteration. And if there are no
more solutions that will become members of the tabu list, then
the best value just obtained that this is the real solution. The
two components that are very important in the Tabu Search
algorithm are the intensification strategies and diversification
strategies. Tabu Search can be applied directly to verbal and
symbolic statements of various kinds of decision-making
problems without the need to change them into a form of
mathematical formula. However, it is also useful to use
mathematical notation to describe a larger scope of problems as
a basis for explaining some things in Tabu Search. Tabu Search
characterizes part of the problem with the optimized goal of the
function f (x) with x € X, where f (x) can be either linear or
non-linear and summary X containing the decision value x.
Basically the searches on the tabu search algorithm are the
same as ordinary local or neighborhood search, starts the
iteration process from one point (solution) to another until the
criteria or conditions stop being reached. Each x € X associated
with neighborhood N (x) € X and each solution x € N (x) is
obtained from x from the operation called move.
Fig. 4. Flowchart of ACO
Fig. 5. Flowchart of Tabu Search
III. IMPLEMENTATION
A. ACO-TS Algorithm
The objective of this research is to optimize a running time
in the ACO Algorithm by using the Tabu Search algorithm as
its local search.
Fig. 6. Flowchart of ACO-TS
B. Data set
The data used for this study is a Hassler Whitney
TSPLIB95 A280 data set [16]. The data source can be
displayed in table 3.1. The data source has 280 points (nodes),
namely from C1 to C 280. Each point (node) consists of X
(ordinate) and Y (abscissa), for example city 1 (C1) at
coordinates X = 288 and Y = 149.
Table 1. Data set of Hassler Whitney TSPLIB95 A280
V. CONCLUSION AND FUTURE WORKS
Table 2. Data set of Hassler Whitney TSPLIB95 A280
IV. RESULT AND DISCUSSION
The results obtained are the best routes based on the closest
distance and the best running time as in the Fig.6. and Fig. 7.
This research is a part of our ongoing research. So far, we
have still develop the prototype of implementation ACO-TS
Algorithm in order to know which is the best route in another
case. The temporarily result, we got a better running time using
ACO-TS compared to the ACO itself. It means that the
combination of ACO-TS is better than ACO itself. Therefore,
to get a better running time in solution and a more optimal
route length, we recommended the ACO-TS algorithm than
ACO itself.
ACKNOWLEDGMENT
The authors would like to thank The Rector of Universitas
Sumatera Utara (USU), the Chair of the Research Center of
USU, and the Dean of Fac. of Computer Science and
Information Technology, for The Research Award. This
Research is Supported by “DRPM Kemenristekdikti of
Republic Indonesia Year 2019”, under Contract No:
227/UN5.2.3.1/PPM/KP-DRPM/2019.
REFERENCES
Fig. 6. ACO Algorithm
[1]
Pan Junjie and Wang Dingwei, “An Ant Colony
Optimization Algorithm for Multiple Travelling
Salesman Problem,” vol. 2, no. 8, pp. 210–213, 2006.
[2]
R. Gan, Q. Guo, H. Chang, and Y. Yi, “Improved ant
colony optimization algorithm for the traveling
salesman problems,” J. Syst. Eng. Electron., vol. 21,
no. 2, pp. 329–333, 2014.
[3]
L. Li, S. Ju, and Y. Zhang, “Improved ant colony
optimization for the traveling salesman problem,” in
Proceedings - International Conf. on Intelligent
Computation Technology and Automation, ICICTA
2008, 2008.
[4]
H. M. Rais, Z. A. Othman, and A. R. Hamdan,
“Improved Dynamic Ant Colony System (DACS) on
symmetric Traveling Salesman Problem (TSP),” in
2007 International Conf. on Intelligent and Advanced
Systems, ICIAS 2007, 2007.
[5]
J. Han and Y. Tian, “An Improved Ant Colony
Optimization Algorithm Based on Dynamic Control of
Solution Construction and Mergence of Local Search
Solutions,” in Proceedings - 4th International Conf.
on Natural Computation, ICNC 2008, 2008, vol. 7, pp.
490–495.
[6]
T. N. Bui and M. Colpan, “Solving geometric TSP
with ants,” p. 271, 2005.
[7]
C. M. Pintea and D. Dumitrescu, “Improving Ant
Systems using a local updating rule,” in Proceedings Seventh International Symposium on Symbolic and
Numeric Algorithms for Scientific Computing,
SYNASC 2005, 2005.
In Fig. 6. Using ACO Algorithm, we got optimal distance:
188.53 and running time: 0.165 second.
Fig. 7. ACO-TS Algorithm
In Fig. 7. Using the hybrid of ACO-TS, we got optimal
distance: 186.08 and running time: 0.146 second.
[8]
T. Stützle and H. Hoos, “{{\cal MAX}-{\cal MIN}}
{A}nt {S}ystem and Local Search for the {T}raveling
{S}alesman {P}roblem,” in Proceedings of the fourth
International
Conference
on
Evolutionary
Computation, 1997.
[9]
C. Wang, D.-W. Cui, Y.-K. Zhang, and Z.-R. Wang,
“A Novel Ant Colony System Based on Delauney
Triangulation and Self-adaptive Mutation for TSP,”
2006.
[10]
Z. A. Othman, H. M. Rais, and A. R. Hamdan,
“Strategies DACS3 increasing its performances,” Eur.
J. Sci. Res., vol. 27, no. 4, pp. 488–499, 2009.
[11]
Z. A. Othman, H. Rais, and A. R. Hamdan, “DACS3 :
Embedding Individual Ant Behavior in Ant Colony
System,” no. 4, pp. 25–30, 2009.
[12]
H. Md. Rais, Z. A. Othman, and A. R. Hamdan,
“Improvement DACS3 searching performance using
local search,” 2009 2nd Conf. Data Min. Optim. DMO
2009, no. October, pp. 160–166, 2009.
[13]
X. M. Wei, “Improved ant colony algorithm based on
information entropy,” in Proceedings - 2010
International Conference on Computational and
Information Sciences, ICCIS 2010, 2010.
[14]
M. Dorigo, “Luca Maria Gambardella Ant Colony
System: A Cooperative Learning Approach to the
Traveling Salesman …,” Belgium TR/IRIDIA/1996-,
vol. 1, no. 1, p. 53, 5AD.
[15]
M. M.-U.- Rashid Khan and M. Asadujjaman, “A
Tabu Search Approximation for finding the Shortest
distance using Traveling Salesman Problem,” IOSR J.
Math., vol. 12, no. 05, pp. 80–84, 2016.
[16]
Sitanggang. Delima, “Population Initialization in
Genetic Algorithms Using Simple Hill Climbing (Shc)
For Traveling Salesman Problem (Tsp),” J. TIMES,
vol. 4, no. 2, pp. 40–44, 2016.
Download