Comprehensive Learning Particle Swarm Optimization Algorithm With Local Search for Multimodal Functions

TLDR

The study proposes a comprehensive learning particle swarm optimizer enhanced with local search and an adaptive start strategy based on a quasi‑entropy index to improve optimization performance. The algorithm combines CLPSO with a quasi‑entropy‑guided local search, is evaluated on multimodal benchmark functions, and its robustness is shown through parameter sensitivity analysis. Experimental results demonstrate that the proposed method achieves faster convergence and higher accuracy than CLPSO and other state‑of‑the‑art particle swarm variants.

Abstract

A comprehensive learning particle swarm optimizer (CLPSO) embedded with local search (LS) is proposed to pursue higher optimization performance by taking the advantages of CLPSO's strong global search capability and LS's fast convergence ability. This paper proposes an adaptive LS starting strategy by utilizing our proposed quasi-entropy index to address its key issue, i.e., when to start LS. The changes of the index as the optimization proceeds are analyzed in theory and via numerical tests. The proposed algorithm is tested on multimodal benchmark functions. Parameter sensitivity analysis is performed to demonstrate its robustness. The comparison results reveal overall higher convergence rate and accuracy than those of CLPSO, state-of-the-art particle swarm optimization variants.