Multi-population random mean particle swarm optimization algorithm

Particle swarm optimization (PSO), as an efficient and simple optimization algorithm, is widely used in numerical and engineering optimization problems. Because of the single learning mechanism, PSO is prone to premature convergence and local optimality. In order to improve the optimization ability of PSO algorithm, a multi-population random mean particle swarm optimization algorithm (MRMPSO) is proposed. In MRMPSO, the particles are divided into ordinary, elite and leader populations according to the fitness value of the particles, and the subgroup size is adjusted according to the time variation of the iterative process. Different populations implement different learning mechanisms, and the random selection of each ordinary particle is better than its own particle learning. The elite subgroup adopts random mean value strategy to make each elite particle fly to the center of a group of random particles. The leading particles adopt self-learning and random dimension learning strategies. In addition, a reverse learning exit strategy and random trailing particle elimination strategy are designed. The performance of seven advanced PSO variants was compared with the proposed MRMPSO using the benchmark function in CEC2013 and CEC2017. Experiments show that MRMPSO has good global exploration ability and can effectively avoid local optimization.