Diversity preservation using excited particle swarm optimisation
Shannon S. Pace, Clinton J. Woodward · 2011
The particle swarm optimisation (PSO) algorithm suffers from the possibility of premature convergence. This problem has historically been addressed ab intra---manipulating velocity and swarm topology---yet the judicious addition of external mechanisms has been shown to adjust search behaviour to yield significantly improved results across many problems. This paper introduces an addition to the canonical particle swarm algorithm, designed to preserve the diversity typically lost by attraction to suboptimal positions. The proposed excited PSO method stimulates exploration upon the discovery of a candidate solution by manipulating the position to which particles are attracted. It is shown to maintain a suitable degree of diversity for the duration of an experiment, as well as an ability for self-scaling. Comparisons to the canonical PSO algorithm demonstrate improved solutions in both unimodal and multimodal spaces.