A ranking-system-based switching particle swarm optimizer with dynamic learning strategies

In this paper, a novel ranking-system-based switching particle swarm optimizer (RSPSO) is proposed. In particular, according to a ranking system, the swarm is divided into elite and normal group, then each particle has been assigned a fitness-based (for normal group member) or a distance-based neighborhood (for elite group member). It is remarkable that neighborhood of a particle is time-varying so that communication among swarm during whole searching process is greatly enhanced. In addition, searching process is divided into four stages by the switching framework, where learning strategies and parameter settings are changed in an adaptive way. Moreover, a newly proposed dimensional learning strategy has been hybridized in RSPSO so as to preserve useful information in the swarm and differential evolution algorithm is employed for a further exploration and also diversifying the swarm. Proposed RSPSO is comprehensively evaluated on series of benchmarks including uni-modal, multi-modal and rotated multi modal functions. Furthermore, ablation study and sensitivity analysis are performed, where influences of the ranking system, time-varying neighborhood as well as other key parameters have been discussed in detail. Experimental results demonstrate the superiority of proposed algorithm which outperforms other five PSO variants on several indicators regarding to both solution accuracy and convergence performance on most benchmarks in a statistic sense. (c) 2022 Elsevier B.V. All rights reserved.

Automated summary

In this paper, a novel ranking-system-based switching particle swarm optimizer (RSPSO) is proposed. The algorithm enhances communication among swarm and improves optimization performance through strategies like switching stages and dynamic neighborhood. Experimental results demonstrate that the proposed algorithm performs excellently on various benchmark tests.