Capitalizing Diversity for Efficiency Enhancement in Multi-Population Swarm Algorithms

Abstract

Swarm algorithms are an efficient means to optimize various real-life problems. Their efficiency is influenced by diversity, which helps it to escape any local optima. There are multiple ways to increase diversity like mutation, repulsion, multi-population, replacement, etc. Of these, multi-population and replacement based techniques work without changing the internal functioning of any algorithm. In this paper we study three different replacement based techniques for multi-population swarm algorithms. The paper also proposes techniques (variant/new) to improve diversity in a multi-population scenario. All the techniques:-studied and proposed are subsequently tested with bat algorithm over 30 benchmark functions. The most efficient amongst them is further tested over six other swarm algorithms. The comparative results indicate that the identified technique is significantly better in terms of efficiency and diversity for most algorithms. Hence it can be concluded that it is an efficient means to improve the diversity and efficiency of a multi-population swarm algorithm.