Development of Forked Divergence Particle Swarm Optimization with Exploration and Exploitation

Fukuhara S and Arakawa M

Published on: 2025-07-15

Abstract

There are more and more issues, requirements, and complexities that engineering design problems must take into account in industrial applications. Therefore, the number of behavior and design variables tends to increase, making optimization problems harder. In such large-scale optimization problems (LSOPs), conventional PSO algorithms often suffer from premature convergence and search stagnation due to the high dimensionality. To address these challenges, we propose Forked Divergence Particle Swarm Optimization with Exploration and Exploitation (FDPSO-EE). The proposed method introduces a dual-population structure in which the swarm is divided into two subgroups: one for exploration and another for exploitation. These groups operate in parallel and interact to maintain search diversity and precision. Furthermore, a dimensionality reduction mechanism is employed, allowing each particle to focus on a reduced subspace during optimization. Extensive experiments on 128-dimensional benchmark functions demonstrate that FDPSO-EE significantly outperforms existing algorithms such as SPSO, FDPSO, RIPSO, BLX-α, and SPLX. The proposed method achieves lower mean objective values and standard deviations, particularly on multimodal functions, showing strong global search ability and convergence accuracy. These results highlight the effectiveness of FDPSO-EE in solving complex LSOPs with both robustness and efficiency.