A PSO Algorithm with Multi-Grouping and Two-Layer Structure

Abstract

A PSO (particle swarm optimization) algorithm with multiple groups and two-layer structure(MTPSO) is proposed. First, the whole particle swarm is divided into several groups, each one is optimized. For enhance efficiency in the optimization process, each particle adopts a mixed algorithm of negative gradient. Secondly, according to the function value of each particle solution, the particle group is divided into two structures: the elite layer and the ordinary layer. In the algorithm initialization phase, the particles with the best value in each group are selected into the elite layer. The particles in the elite layer are relatively fixed and regularly selected (similar to the parliamentary mechanism in human society), that is, each elite layer as a whole. After several iterations of optimization calculations in the same way as the normal layer, the new elite layer is formed again according to the previou method. The optimal solution calculated by the elite layer is added as a global optimal solution to the position update formula of each particle. Similarly, the optimal solution for the common layer is also added to the update formula for each particle in the elite layer. It reflects the upper and lower particle’s mutual penetration and mutual influence of. Simulation experiments show the proposed algorithm is able to avoid premature and powerful global optimization ability and fast convergence, PSO’s computing efficiency.