Network-Structured Particle Swarm Optimizer That Considers Neighborhood Distances and Behaviors

抄録

This study proposes a network-structured particle swarm optimizer (NS-PSO), which considers neighborhood distances. All particles of the NS-PSO are connected to adjacent particles in the neighborhood of topological space, and NS-PSO utilizes the connections between them not only to share local best position but also to increase swarm diversification. Each NS-PSO particle is updated depending on the positions of the local best and current best particles. In NS-PSO, the neighborhood distance in the topological space from each particle to the current best position is also considered. This effect promotes the diversification of solutions and avoids the solutions from becoming trapped at local optima. Simulation results and comparisons with conventional particle swarm optimization show that the proposed NS-PSO can effectively enhance the searching efficiency by measuring in terms of accuracy, robustness and parameterdependence. Furthermore, we consider various network topologies, grid, hexagonal, cylinder and toroidal. We investigate their behaviors and evaluate the kind of topology that would be the most appropriate for each benchmark.