Vol. 3 No.1. / June 2012
14
Business Systems Research
Performance analysis of the partial use of a local
optimization operator on the genetic algorithm for
the Travelling Salesman Problem
Milan Djordjevic, Marko Grgurovič and Andrej Brodnik
Department of Information Science and Technology, University of Primorska, Koper, Slovenia
Background: The Travelling Salesman Problem is an NP-hard problem in combinatorial optimization
with a number of practical implications. There are many heuristic algorithms and exact methods for
solving the problem. Objectives: In this paper we study the inuence of hybridization of a genetic
algorithm with a local optimizer on solving instances of the Travelling Salesman Problem. Methods/
Approach: Our algorithm uses hybridization that occurs at various percentages of generations of a
genetic algorithm. Moreover, we have also studied at which generations to apply the hybridization
and hence applied it at random generations, at the initial generations, and at the last ones. Results:
We tested our algorithm on instances with sizes ranging from 76 to 439 cities. On the one hand, the
less frequent application of hybridization decreased the average running time of the algorithm from
14.62 sec to 2.78 sec at 100% and 10% hybridization respectively, while on the other hand, the quality
of the solution on average deteriorated only from 0.21% till 1.40% worse than the optimal solution.
Conclusions: In the paper we have shown that even a small hybridization substantially improves the
quality of the result. Moreover, the hybridization in fact does not deteriorate the running time too
much. Finally, our experiments show that the best results are obtained when hybridization occurs in
the last generations of the genetic algorithm.
Keywords: genetic algorithm, travelling salesman problem, hybridization, optimization, grafted
genetic algorithm
JEL classication: C61
Paper type: Research article
Recieved: 28, April, 2012
Revised: 30, May, 2012
Accepted: 29, June, 2012
Citation: Djorjevic, M., Grgurović, M., Brodnik, A. (2012). “Performance analysis of the partial use of a
local optimization operator on the genetic algorithm for the Travelling Salesman Problem”, Business
Systems Research, Vol. 3, No. 1, pp. 14-22.
DOI: 10.2478/v10305-012-0002-4
Introduction
Genetic Algorithms (GA) use mechanisms inspired by biological evolution (Holland, 1975; Haupt and Haupt,
2004). They are applied on a nite set of individuals called population. Each individual in a population
represents one of feasible solutions in a search space. Mapping between genetic codes and the search
space is called encoding and can be binary or over some alphabet of higher cardinality. Good choice
of encoding is a prime condition for successful application of a genetic algorithm. Each individual in the
population is assigned a value called tness. Fitness represents a relative indicator of quality of an individual
compared to other individuals in the population. Selection operator chooses individuals from the current
population and takes the ones that are transferred to the next generation. Thereby, individuals with better
tness are more likely to survive in the next generation. The recombination operator combines parts of
genetic code of the individuals (parents) into codes of new individuals (offspring).
The components (operators) of the genetic algorithm software system are: Genotype, Fitness function,
Recombinator, Selector, Mater, Replacer, Terminator, and in our system (Local) Optimizer which is a new
extended component.
Abstract
Unauthenticated
Download Date | 1/8/18 3:28 PM