为克服 BP算法中存在的网络学习收敛速度慢 ,以及容易陷入局部极小的问题 ,本文在神经网络训练过程中 ,加入一个局部极小判别式 ,以确定网络是否陷入局部极小点 ,若陷入局部极小点 ,则利用遗传优化算法进行权值的修正。以机械设备故障诊断为例 ,应用此算法对其进行了故障诊断研究 ,从而证明了该算法的有效性。 In order to overcome the problem of slow learning convergence and easy falling into local minima in BP algorithm, this paper adds a local minima discriminant in neural network training to determine whether the network is in a local minimum or not, Local minima, the use of genetic optimization algorithm for weight correction. Taking the mechanical equipment fault diagnosis as an example, this algorithm is applied to the fault diagnosis research, which proves the effectiveness of the algorithm.