A self-configuring metaheuristic for control of multi-strategy evolutionary search

Abstract

There exists a great variety of evolutionary algorithms (EAs) that represent different search strategies for many classes of optimization problems. Real-world problems may combine several optimization features that are not known beforehand, thus there is no information about what EA to choose and which EA settings to apply. This study presents a novel metaheuristic for designing a multi-strategy genetic algorithm (GA) based on a hybrid of the island model, cooperative and competitive coevolution schemes. The approach controls interactions of GAs and leads to the self-configuring solving of problems with a priori unknown structure. Two examples of implementations of the approach for multi-objective and non-stationary optimization are discussed. The results of numerical experiments for benchmark problems from CEC competitions are presented. The proposed approach has demonstrated efficiency comparable with other well-studied techniques. And it does not require the participation of the human-expert, because it operates in an automated, selfconfiguring way.