Non-uniform Variance Fuzzy Guided Particle Swarm Algorithm

The particle swarm optimization (PSO) attracts many researchers´ interests. However, its quick convergence often implies a rapid loss of diversity, which inevitably results in prematurity. Borrowed the idea of non-uniform mutation from evolutionary algorithm (EA) a non-uniform variance fuzzy guided PSO (fGuiPSO) is presented in this paper. The new particle updating strategy is based on the concept of fuzzy global best to deal with the premature convergence and diversity maintenance within the swarm. The variance of the global best determines its perturbing neighbor size and the non-uniform mutation operation has the merits of searching the space uniformly initially and very locally at later stage. Therefor the non-uniform idea is applied to adjust the variance of the global best particle. This non-uniform variance fuzzy guided particle updating scheme is also theoretically analyzed. Eight benchmarks are used to validate fGuiPSO and experiments indicate that fGuiPSO performs much better than PSO both in quality of solutions and robustness. The experiments confirm us that the non-uniform fuzzy particle updating strategy is a useful attempt for stochastic heuristic algorithms.