Cautious particle swarm
Hiroshi Someya · 2008
Behavior and search performance of the proposed cautious particle swarm, in which the particles are cautious about imitating the best successful one, are investigated with a question: “Is it really the best strategy for a group that every individual in which obediently imitates the top one?” Their neutral or disobedient attitudes to the global best are expressed by replacing the conventional uniform distribution for random numbers with either of the two probability distributions: slided uniform distribution and asymmetrical normal distribution. Empirical analyses on the behavior of such a single particle in one-dimensional search space have aroused an expectation that appropriate cautiousness balance may accomplish both exploration-oriented search for avoiding local minima and exploitation-oriented search for finding the global optimum near the global best. For discussion on the optimization performance of the cautious particle swarm, experiments in typical test functions were performed. The experimental results have presented an acceptable parameter range of obedience for appropriate cautiousness.