A composite particle swarm algorithm for global optimization of multimodal functions

Abstract

During the last decade, many variants of the original particle swarm optimization (PSO) algorithm have been proposed for global numerical optimization, but they usually face many challenges such as low solution quality and slow convergence speed on multimodal function optimization. A composite particle swarm optimization (CPSO) for solving these difficulties is presented, in which a novel learning strategy plus an assisted search mechanism framework is used. Instead of simple learning strategy of the original PSO, the proposed CPSO combines one particle’s historical best information and the global best information into one learning exemplar to guide the particle movement. The proposed learning strategy can reserve the original search information and lead to faster convergence speed. The proposed assisted search mechanism is designed to look for the global optimum. Search direction of particles can be greatly changed by this mechanism so that the algorithm has a large chance to escape from local optima. In order to make the assisted search mechanism more efficient and the algorithm more reliable, the executive probability of the assisted search mechanism is adjusted by the feedback of the improvement degree of optimal value after each iteration. According to the result of numerical experiments on multimodal benchmark functions such as Schwefel, Rastrigin, Ackley and Griewank both with and without coordinate rotation, the proposed CPSO offers faster convergence speed, higher quality solution and stronger robustness than other variants of PSO.