An improved constriction factor PSO to prevent premature convergence

摘要

In order to prevent premature convergence of constriction factor particle swarm optimization (cfPSO), an improved version of cfPSO is presented in this paper. Firstly, a new standard of premature convergence was set up to determine the algorithm relapsing into local extremum. Then an improved re-initialization method was design to insure the uniform distribution of swarm in search space. In this method, the search space was divided into S, the size of swarm, sub-spaces. Each particle was reinitialized in respectively sub-space. At the same time, all particles were compelled to clear its memory. Simulation results show that the performance was improved significantly.