Using Mutual Information to Build Dynamic Neighbourhoods for Particle Swarm Optimisation
Ángel Arturo Rojas-García, Arturo Hernández-Aguirre · 2016
A proposal to build dynamic neighbourhoods in PSO based on mutual information is presented in this paper. The relationship among the paths the particles follow in the search space along the iterations is measured with mutual information and particles are linked to produce a graph (a tree) with maximum mutual information. Another graph investigated is ring topology. The performance of the approach is tested using a set of thirteen benchmark functions available in the specialised literature. A comparison with canonical PSO (a static fully connected neighbourhood) and two state of the art dynamic neighbourhood PSOs are reported. The results show a fast convergence (a lesser number of function evaluations) in unimodal functions; this convergence speed is more remarkable in high dimensional problems.