Frequency Distribution of Candidate Solutions in Angle Modulated Particle Swarms

This paper investigates the frequency distribution of candidate solutions in the search space when angle modulation is applied to particle swarm optimisation (PSO). It is shown that angle modulated particle swarm optimisers (AMPSO) have non-uniform solution frequency distributions. A new technique is introduced to ensure that the frequency distribution of candidate solutions is uniform. The new technique is compared with AMPSO and three AMPSO variants, as well as binary PSO (BPSO) on a number of problem cases. It is shown that AMPSO algorithms obtain lower average fitness values on binary minimisation problems whose optimal solutions contain repetition. However, when optimal solutions do not contain repetition, AMPSO and its variants are at a clear disadvantage.