New Selection Schemes for Particle Swarm Optimization

Abstract

In Evolutionary Algorithms (EA), the selection scheme is a pivotal component, where it relies on the fitness value of individuals to apply the Darwinian principle of survival of the fittest. In Particle Swarm Optimization (PSO) there is only one place employed the idea of selection scheme in global best operator in which the components of best solution have been selected in the process of deriving the search and used them in generation the upcoming solutions. However, this selection process might be affecting the diversity aspect of PSO since the search infer into the best solution rather than the whole search. In this paper, new selection schemes which replace the global best selection schemes are investigated, comprising fitness-proportional, tournament, linear rank and exponential rank. The proposed selection schemes are individually altered and incorporated in the process of PSO and each adoption is realized as a new PSO variation. The performance of the proposed PSO variations is evaluated. The experimental results using benchmark functions show that the selection schemes directly affect the performance of PSO algorithm. Finally, a parameter sensitivity analysis of the new PSO variations is analyzed.