A Bare Bones Particle Swarm Optimization Algorithm with Dynamic Local Search

摘要

Swarm intelligence algorithms are wildly used in different areas. The bare bones particle swarm optimization (BBPSO) is one of them. In the BBPSO, the next position of a particle is chosen from the Gaussian distribution. However, all particles learning from the only global best particle may cause the premature convergence and rapid diversity-losing. Thus, a BBPSO with dynamic local search (DLS-BBPSO) is proposed to solve these problems. Also, because the blind setting of local group may cause the time complexity an unpredictable increase, a dynamic strategy is used in the process of local group creation to avoid this situation. Moreover, to confirm the searching ability of the proposed algorithm, a set of well-known benchmark functions are used in the experiments. Both unimodal and multimodal functions are considered to enhance the persuasion of the test. Meanwhile, the BBPSO and several other evolutionary algorithms are used as the control group. At last, the results of the experiment confirm the searching ability of the proposed algorithm in the test functions.