Particle Swarm Optimization with Moving Particles on Scale-Free Networks

Abstract

PSO is a nature-inspired optimization algorithm widely applied in many fields. In this paper, we present a variant named MP-PSO, in which some particles are allowed to move on a scale-free network and change the interaction pattern during the search course. In contrast to traditional PSOs with fixed interaction sources, MP-PSO shows better flexibility and diversity, where the structure of the particle swarm could change adaptively and balance exploration and exploitation to a large extent. Experiments on benchmark functions show that MP-PSO outperforms other PSO variants on solution quality and success rate, especially for multimodal functions. We further investigate effects of the moving strategy from a microscopic view, finding that the cooperation mechanism of particles located on hub and non-hub nodes plays a crucial role during the optimization process. In particular, owing to the movement of particles on non-hub nodes, the exploration can be guaranteed to some extent even in the final stage, which may be benefit for optimization. We demonstrate the applicability of MP-PSO by using it to solve an important optimization problem, arrival sequencing and scheduling, in the field of air traffic control.