A diversity guided PSO combined with BP for feedforward neural networks

摘要

In this paper, a diversity guided particle swarm optimization (DGPSO-BP) guided by diversity and fitness value is firstly proposed to address two problems: premature convergence in the standard PSO and longer searching time brought by the optimization of the PSO. Further, the DGPSO-BP is combined with back-propagation (BP) for feed forward neural networks to avoid the problem of being trapped into local minima in the BP and combines PSO's strong local search ability and BP's good local search ability meanwhile. Compared with the traditional learning algorithms, the improved learning algorithm has much better convergence performance. Finally, the experimental results are given to verify the efficiency and effectiveness of the proposed algorithm.