An explicit spatial model for niching in genetic algorithms

Abstract

A niching technique is an important component of the genetic algorithm when attempting to solve problems that have multiple optimal solutions. Traditional niching techniques use an explicit concept of similarity to perform the actual niche formation. Often, the definition of this similarity function is difficult or requires a priori knowledge of the problem domain. This paper investigates the use of an explicit spatial structure to perform niching. This technique differs from other niching techniques in that it does not require a definition of similarity between individuals in order to form niches. Early results indicate that using this technique can allow a GA to maintain multiple peaks in some multi-modal functions.