Dimension-Selected Orthogonal Particle Swarm Optimization for High-Dimensional Problems

초록

Particle swarm optimization (PSO) is a widely used population-based optimization algorithm, but it often suffers from premature convergence when solving complex optimization problems. Orthogonal experimental design (OED) has shown potential in improving the exploration capability of PSO; however, its application to high-dimensional optimization requires large orthogonal arrays, resulting in excessive fitness evaluations and reduced efficiency. To address this issue, this paper proposes a dimension-selected orthogonal particle swarm optimization (DSOPSO) algorithm. The proposed method employs a ReliefFbased dimension importance estimation strategy and a kneepoint-based dimension selection mechanism to identify informative dimensions for orthogonal search adaptively. By performing OED only on the selected dimensions, DSOPSO reduces the number of dimensions involved in OED while preserving the search effectiveness of orthogonal recombination. Experimental results on the CEC2017 benchmark suite demonstrate that the proposed algorithm achieves competitive optimization performance and reduces the number of dimensions involved in OED on highdimensional optimization problems.

키워드