Particle swarm optimizers with grow-and-reduce structure
Eiji Miyagawa, Toshimichi Saito · 2008
This paper presents an improved version of PSO having grow-and-reduce structure. When a particle is trapped into a local optimum, a new particle is born at a position away from the trap and is connected to some/all of existing particles. If a particle can not escape from the trap, the particle is deleted in order to suppress excessive swarm grows. We have adopted three basic population topology: complete graph, ring and tree. Performing basic numerical experiments, the algorithm performance is investigated. The results suggest that the “grow-and-reduce” is very effective for escape from a trap and the tree topology has effective flexibility to realize the optimization.