Re-diversified Particle Swarm Optimization
The tendency to converge prematurely is a main limitation which affects the performacne of evolutionary computation algorithm, including particle swarm optimization (PSO). To overcome the limitation, we propose an extended PSO algorithm, called re-diversified particle swarm optimization (RDPSO). When population diversity is small, i.e., particless velocity approches zero and the algorithm stagnates, a restart approach called diversification mechanism begins to work, which disperses particles and lets them leave bad positions. Based on the diversity calculated by the particles current positions, the algorithm decides when to start the diversification mechanism and when to return the usual PSO. We testify the performance of the proposed algorithm on a 10 benchmark functions and provide comparisons with 4 classical PSO variants. The numerical experiment results show that the RDPSO has superior performace in global optimization, especially for those complex multimodal functions whose solution is difficult to be found by the other tested algorithm.
Re-diversified particle swarm optimization population diversity benchmark function swarm intelligence local optima
Jie Qi Shunan Pang
College of Information Science and Techenology, Donghua University, Shanghai, China
国际会议
无锡
英文
30-39
2010-09-17（万方平台首次上网日期，不代表论文的发表时间）