Adapting Particle Swarm Optimisation for Fitness Landscapes with Neutrality

Abstract

The concept of 'neutral networks' in fitness landscapes - where amongst any rugged terrain there may also be connected regions or pathways over which solution fitness does not change - has been shown to be of significance to the use of evolutionary algorithms. To our knowledge this important aspect of the fitness landscape has not previously been examined in the context of particle swarm optimisation (PSO). The standard PSO algorithm is here shown to be inadequate for optimisation tasks where such neutrality exists; we investigate modifications of a standard PSO and compare their performances on various novel fitness landscapes that contain neutrality. One simple modification to the standard PSO algorithm is shown to enable significantly improved functionality upon neutral landscapes, with no compromise to operation upon non-neutral terrains