A Global Robust Particle Swarm Optimization by Improving the Learning Strategy

In consideration of stagnation phenomenon in the later phase of the particle swarm optimization (PSO) caused by diversity scarcity of particles, a new learning strategy for improving the global and local exploration capability of particle swarm optimization is proposed in the paper. The new learning strategy is inspired by the mass migration behaviors of animal swarms that each individual has the ability of keeping its inertia movement and learning from another randomly selected individual that is nearer to the destination. Therefore, in the modified PSO, both of the inertia weight and the learning rate coefficients in the velocity update formula are replaced by random sequences multiplied with positive constants, and each particle learns from a randomly selected particle which has better performance in stead of learning from the previous best positions of itself and the population. Comparison results with the basic PSO on the examination of some well-known benchmark functions show the perfective and robustness of the modified PSO.