A novel multimodal-problem-oriented particle swarm optimization algorithm

In this paper, we present a novel particle swarm optimization (PSO) variant named scatter learning PSO algorithm (SLPSOA) for solving multimodal problems. SLPSOA takes full account of the distribution information of exemplars while following the basic framework of PSO. It constructs an exemplar pool (EP) which is composed of a certain number of relatively high-quality solutions scattering in the solution space, and allows each particle to select a solution from EP as the exemplar using the roulette wheel rule, with the aim of leading the particles to promising solution regions. In addition, SLPSOA employs Solis and Wets? algorithm as a local searcher to enhance its fine search ability in the newfound solution regions. SLPSOA was tested on 16 benchmark functions, and compared with five existing typical PSO algorithms. Computational results demonstrate that it can manage to prevent premature convergence and produce competitive solutions.