An Adaptive GA in Partitioned Search Space

Abstract

Evolutionary algorithms are population based meta-heuristics inspired from natural survival of fittest phenomena. Despite their reasonable performance, these algorithms suffer from some weaknesses including the need for finding the values of their parameters that affect their performance. A new algorithm is proposed that divide the search space into equal sized partitions. Each partition is assigned with two parameters that determine the intensification and diversification rates. The partitions will be intensified or diversified adaptively with regards to the corresponding parameters. Traditional crossover and mutation operators are replaced with two new parameter-free operators. The experiments conducted on a wide range of multi-modal and epistatic problems showed the superiority of the proposed method in comparison to other algorithms in literature.