A Master-Slave Particle Swarm Optimization Algorithm for Solving Constrained Optimization Problems

Penalty function based PSO converts constrained optimization problems into non-constrained optimization problems, but slow convergence and premature convergence easily happen because of inappropriate penalty coefficients. Modified PSO by tracking best feasible particle can not facilitate particles exploring unknown feasible region from known infeasible region, so the global exploration ability is greatly limited. Therefore, finding better unknown feasible solution by flying through infeasible region is critical to the performance of PSO. This paper proposes master-slave particle swarm optimization (MSPSO), a novel approach for solving constrained optimization problems, in which particles in master swarm fly toward better feasible particles, particles in slave swarm fly toward better infeasible particles, and particles in two swarms help each other flying by sharing information of better feasible and infeasible particles. The proposed algorithm was tested on 11 benchmark constrained optimization problems. The test results show that MSPSO can significantly improve the globe exploration ability and effectively avoid being trapped into local optimum. By comparison with other evolutionary algorithms, MSPSO performs better for constrained optimization problems