Particle swarm optimization algorithm with reverse-learning and local-learning behavior

To resolve conflict between convergence and diversity in particle swarm optimization (PSO) algorithm, an improved PSO algorithm which called reverse-learning and local-learning PSO (RLPSO) algorithm is introduced. In RLPSO, a reverse-learning behavior is adopted by some particles and local-learning behavior is adopted by elite particles. In RLPSO, some inferior particles of initial population and each particle's historical worst position are reserved. Furthermore, the hamming distance among the inferior particles is no less than a rejection distance that predefined. While population has being trapped into a local optimum, the inferior particles and a particle's historical worst position can attract the particle to leap out of the local optimums in a high speed. This action is called reverse-learning behavior which can preservation population diversity and improve RLPSO's exploration ability. Furthermore, in each generation, the difference between the best particle and the second-best particle is adopted to guide the best one to carry out a local search process called local learning behavior by which exploitation ability of population can be improved. In the local learning behavior that can parallel execute with population's evolution, the local scale factor is dynamic adjusted during the evolution. The results achieved by RLPSO were compared with some modified PSO algorithm, which indicated that RLPSO has better global searching ability and higher convergence speed especially in high dimension functions. ©, 2015, Science Press. All right reserved.