Enhancing PSO for Dealing with Large Data Dimensionality by Cooperative Coevolutionary with Dynamic Species-Structure Strategy

摘要

It is widely recognized that the performance of PSO in dealing with multi-objective optimization problem is deteriorated with an increasing of problem dimensionality. The notions of cooperative coevolutionary (CC) allow PSO to decompose the large-scale problem into the multiple subcomponents. However, the combination of CC and PSO tends to perform worse if interrelation among variables is exhibited. This paper proposes the dynamic species-structure strategy for improving search ability of CC and incorporates it with PSO algorithm. The resulting algorithm, denoted as "DCCPSO", is tested with the standard test problems, widely known as "DTLZ", with 3-6 optimized objectives. In the large-scale problem setting, the experimented results reveal that our proposed decomposition strategy helps enhancing performance of PSO and overcoming problems pertaining to interrelation among variables issues.