Cognitive Bare Bones Particle Swarm Optimisation with Jumps
The ‘bare bones' (BB) formulation of particle swarm optimisation (PSO) was originally advanced as a model of PSO dynamics. The idea was to model the forces between particles with sampling from a probability distribution in the hope of understanding swarm behaviour with a conceptually simpler particle update rule. ‘Bare bones with jumps' (BBJ) proposes three significant extensions to the BB algorithm: (i) two social neighbourhoods, (ii) a tuneable parameter that can advantageously bring the swarm to the ‘edge of collapse' and (iii) a component-by-component probabilistic jump to anywhere in the search space. The purpose of this paper is to investigate the role of jumping within a specific BBJ algorithm, cognitive BBJ (cBBJ). After confirming the effectiveness of cBBJ, this paper finds that: jumping in one component only is optimal over the 30 dimensional benchmarks of this study; that a small per particle jump probability of 1/30 works well for these benchmarks; jumps are chiefly beneficial during the early stages of optimisation and finally this work supplies evidence that jumping provides escape from regions surrounding sub-optimal minima.