On the properties of search directions in the Particle Swarm Optimization system
K. Masuda, Kenzo Kurihara · 2008
In this paper, the properties of Particle Swarm Optimization (PSO) with respect to the velocity of particles are studied empirically. This study aims at clarifying why PSO works so well for global optimization. The velocity of PSO particles is compared with other local or global search direction obtained analytically, by introducing a measure of the similarity between the velocity of particles and such an analytic direction. Numerical experiments provide some noteworthy properties of the PSO system in the concentration to the global optimum and escapement from local optima.