“…In order to better solve the problems of determine the initial weights and threshold, can adopt the weights and threshold based on genetic algorithm, using the global search ability of genetic algorithm to determine the initial weights and threshold.This paper adopts three layer BP network to determine the initial solution space, setting training number and training error of the network. Set the input population size, crossover probability ( c P ), mutation probability ( m P ), the network layers, each layer neural metadata, and use GA to optimize the weights of neural network repeatedly, until the average value is no longer meaningful increase so far, at this time the decoded parameter combination has sufficiently close to the optimum combination of parameters, and then BP algorithm reoptimization connection weights and threshold of the network in the small solution space , search out the optimal solution [4]. Because GA is based on the population, not to search base on a single point , can also obtain a plurality of extreme value from different points, so it is not easy to fall into local optimum, which can effectively solve the existing problem in BP neural network, and effectively improve the generalization performance of neural network.…”