A Clearing Procedure as a Niching Method for Genetic Algorithms

The clearing procedure is a niching method inspired by the principle stated by J.H. Holland in 1975: the sharing of limited resources within subpopulations of individuals characterized by some similarities. But, instead of evenly sharing the available resources among the individuals of a subpopulation, the clearing procedure supplies these resources only to the best individuals of each subpopulation. The clearing is naturally adapted to elitist strategies. This can significantly improve the performance of genetic algorithms applied to multimodal optimization. Moreover the clearing procedure allows the GA to efficiently reduce the genetic drift when used with an appropriate selection operator. Some experimental results are presented for a massively multimodal deceptive function optimization. I. GENETIC ALGORITHMS AND MULTIMODAL OPTIMIZATION A simple genetic algorithm [1] (SGA) is suitable for searching the optimum of unimodal functions in a bounded search space. However, both experiments and analysis show that the SGA cannot find the multiple global maxima of a multimodal function [1][2]. This limitation can be overcome by a mechanism that creates and maintains several subpopulations within the search space in such a way that each highest maximum of the multimodal function can attract one of them. These mechanisms are referred to as “niching methods” [2]. The clearing procedure described in this paper derives from the niching principle stated by J.H. Holland in 1975 [3]. A niche is characterized by a limited amount of renewal resources available for individuals which presents similarities. Each individual in a niche can consume a fraction of the available resources: the greater the subpopulation size of the niche, the smaller the fraction. This leads towards a steady state in which the subpopulation sizes are proportional to the amount of the corresponding available resources. D.E. Goldberg and J. Richardson [4] presented an implementation of this concept known as the “sharing method”. This paper first presents the clearing procedure and subsequently an elitist variant. This technique is then compared with the sharing method from the subpopulation size point of view. The behavior of the clearing procedure facing the genetic drift with the operators “Roulette Wheel Selection” (RWS) or “Stochastic Universal Selection” [5] (SUS), is also discussed. Finally, the test of the clearing procedure on a massively multimodal deceptive function is presented. II. CLEARING THE SEARCH SPACE