Genetic Algorithms in Multimodal Search Space

Abstract

Genetic algorithms (GAs) (Holland, 1975; Goldberg, 1989) try to find the solution for a problem using an initial group of individuals?the population?where each one represents a potential solution. Actually they are successfully applied in very different and actual fields (Yang, Shan, & Bui, 2008; Yu, Davis, Baydar, & Roy, 2008); nevertheless, GAs have some restrictions on a search space with more than a global solution or a unique global solution, together with multiple local optima. A classical GA faced with such a situation tends to focus the search on the surroundings of the global solution; however, it would be interesting to know a higher number of possible solutions for several reasons: precise information about the search space, easy implementation of the local solutions compared with the global one, simple interpretation of certain solutions compared with others, and so forth. To achieve that knowledge, an iterative process will be executed until reaching the desired goals. Such process will start with the grouping of the individuals into species that will independently search a solution in their environments; following, the crossover operation will involve individuals from different species in order not to leave unexplored any search space area. The process will be repeated according to the goals achieved.