Ensemble learning is a methodology that combines multiple individual learners, and can get better results. In this paper, ensemble learning was applied to improve the performance of Particle Swarm Optimization (PSO) algorithm in three aspects: on particle level or dimension level; in the whole iteration process or periodical iteration process; on single population or multiple subpopulations. The experimental results of three typical function optimization problems show that PSO algorithm based on ensemble learning can improve performance compared with the standard PSO and some improved PSO algorithms.