Abstract:
Aiming at the random of the determination of initial weight and the number of neurons of the hidden layers for BP neural network, this paper utilizes hybrid particle swarm optimization to optimize the initial weight and structure for the neural network.It is used to get a better search space by hybrid particle swarm optimization firstly, and then the network is trained and learned in the solution space by BP.Thereby, it searches out the optimal network structure and weight.This model is used to train and test by the use of Iris pattern classification, Wine pattern classification and generalized XOR problem, compared with other algorithm such as genetic algorithm, it can get higher recognition rate.The result shows that the way of this paper is feasible.