For the complex multi-peaks function with high dimension, the improved particle swarm optimization algorithm with reverse forecast and repulsion(RFRPSO) is proposed on the basis of analyzing the problem of premature. Firstly, this method improves the speed renewal way by introducing the reverse forecast factor in order to decreas the probability of premature convergence. Furthermore, the repulsion factor is introduced to make the swarm even distribution in search space, which can avoid local optimum. Finally, experimental results show that the performance of RFRPSO algorithm is superior to the traditional PSO algorithm.