A Grouping Particle Swarm Optimizer with Personal-Best-Position Guidance for Large Scale Optimization

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Swarm Intelligence

Background:

• Canonical Particle Swarm Optimization (PSO) often suffers from premature convergence and local optima due to poor population diversity maintenance.
• Existing PSO variants struggle with effectively preserving diversity in complex optimization landscapes.

Purpose of the Study:

• To propose a novel PSO variant, Grouping PSO with Personal-Best-Position Guidance (GPSO-PG), designed to enhance population diversity.
• To address the limitations of canonical PSO in maintaining diversity and avoiding local optima for large-scale optimization problems.

Main Methods:

• Implemented a Grouping PSO with Personal-Best-Position Guidance (GPSO-PG) algorithm.
• Utilized a uniform random allocation strategy for assigning particles into groups, where 'losers' learn from 'winners'.
• Employed personal historical best positions (Pbest) for social learning, replacing the global best particle to increase exemplar diversity.

Main Results:

• GPSO-PG demonstrated competitive performance in maintaining population diversity.
• The algorithm achieved satisfactory results on large-scale optimization problems (LSOPs) from CEC 2008 and CEC 2010 benchmarks.
• Comparative analysis showed improved diversity preservation compared to several peer algorithms.

Conclusions:

• GPSO-PG effectively enhances population diversity by preserving exemplar diversity.
• The proposed method mitigates premature convergence and local optima issues in PSO.
• GPSO-PG offers a promising approach for tackling large-scale optimization challenges.