IMPROVED PARTICLE SWARM OPTIMIZATION AND NEIGHBORHOOD FIELD OPTIMIZATION BY INTRODUCING THE RE-SAMPLING STEP OF PARTICLE FILTER

摘要

A technique of introducing the re-sampling step of particle filter is proposed to improve the particle swarm optimization (PSO) algorithm, a typical global search algorithm. The re-sampling step can decrease particles with low weights and duplicate particles with high weights, given that we define a type of suitable weights for the particles. To prevent the identity of particles, the re-sampling step is followed by the existing method of particle variation. Through this technique, the local search capability is enhanced greatly in the later searching stage of PSO algorithm. More interesting, this technique can also be employed to improve another algorithm of which the philosophy is "learning from neighbors", i.e., the neighborhood field optimization (NFO) algorithm. The improved algorithms (PSO-resample and NFO-resample) are compared with other metaheuristic algorithms through extensive simulations. The experiments show that the improved algorithms are superior in terms of convergence rate, search accuracy and robustness. Our results also suggest that the proposed technique can be general in the sense that it can probably improve other particle-based intelligent algorithms.