An adaptive dynamic multi-swarm particle swarm optimization with stagnation detection and spatial exclusion for solving continuous optimization problems

Particle swarm optimization (PSO) is a simple and efficient optimization method that has been used in many fields. However, PSO has some flaws including premature convergence and poor population diversity. To solve these problems of PSO, an adaptive dynamic multi-swarm particle swarm optimization with stagnation detection and spatial exclusion (ADPSO) is proposed. The newly proposed ADPSO is based on a dynamic multi-swarm PSO framework that cooperates with stagnation detection mechanism (SDM) and spatial exclusion strategy (SES). Firstly, the whole population is divided into multiple equal sub-swarms, which can be regrouped during evolution. The best particle of each sub-swarm, lbest, is used to evaluate the evolutionary state of each sub-swarm. If the lbest cannot improve its solution continuously without reaching the regroup period, the SDM will be triggered. In this case, a vitality particle is generated to help the stagnant sub-swarm to search for promising areas again. To keep the population diversity, the vitality particle is constructed according to the excellent historical information of all particles in the whole population. Secondly, to enhance the diversity of the population, the SES is proposed to avoid premature adsorption of all sub-swarms. Finally, the effect of ADPSO is evaluated using CEC2013, CEC2017 and four engineering optimization problems. The results shows that the proposed ADPSO is suitable for solving most optimization problems, and ADPSO has better results compared to the state-of-the-art PSO variants (i.e., MSCPSO, HCDMPS) and other popular evolutionary algorithms.