Particle swarms and population diversity

Abstract

The optimisation of dynamic optima can be a difficult problem for evolutionary algorithms due to diversity loss. However, another population based search technique, particle swarm optimisation, is well suited to this problem. If some or all of the particles are ‘charged’, an extended swarm can be maintained, and dynamic optimisation is possible with a simple algorithm. Charged particle swarms are based on an electrostatic analogy—inter-particle repulsions enable charged particles to swarm around a nucleus of neutral particles. This paper proposes a diversity measure and examines its time development for charged and neutral swarms. These results facilitate predictions for optima tracking given knowledge of the amount of dynamism. A number of experiments test these predictions and demonstrate the efficacy of charged particle swarms in a simple dynamic environment.