Alternative combination of improved particle swarm and back propagation neural network

As the back propagation neural network (the BP neural network) can easily be trapped in the local optimal solutions and have slow convergence and the particle swarm optimization (PSO) is weak on the precision of the convergence, this paper proposes a new method to improve the performance with the combination of the two algorithms. This paper applies both of them in a new alternating optimization of neural networks. Besides, taking into account the drawbacks of the single PSO algorithm, the PSO improved by the simulated annealing algorithm (SA) is also applied. The improved algorithm can be used when building a fuzzy link and making a rough prediction just the same as what traditional neural networks do. But it is far more efficient and the error is smaller. To confirm its superiority, this paper uses three specific datasets to test its performance, with a comparison followed. The results prove that this new model is desirable.