Multi-context cooperative coevolution in particle swarm optimization

Abstract

A novel multi-context cooperatively coevolving particle swarm optimization (MCC-PSO) algorithm is proposed for the large-scale global optimization (LSGO) problems. As most optimization algorithms lose to find the global optimum on LSGO due to the curse of dimensionality, the famous cooperative co-evolution (CC) framework is proposed to overcome such weakness. In the basic CC framework, a single context vector is utilized for cooperatively but greedily coevolving different subcomponents, which sometimes loses its effectiveness. In this study, a novel multi-context cooperative coevolution framework and its application in PSO is proposed, in which more than one context vectors are employed to provide robust and effective co-evolution, as well as a new PSO updating rule based on the subpopulation in subcomponent (SPSC) structure and Gaussian distribution. On a comprehensive set of benchmarks (up to 1000 dimensionalities), the performance of MCC-PSO can rival several state-of-the-art evolutionary algorithms. Experimental results indicate that the novel multi-context CC framework is effective to improve the performance of PSO on LSGO and can be generally extended in other evolutionary algorithms.