为提高BP神经网络训练的预测能力,采用有助于提高BP神经网络逼近精度的Metropolis准则来克服BP神经网络训练学习过程中容易陷入局部极小值的问题;考虑到两类误分的代价不同,利用两个惩罚系数C1和C2,对两类误分给予不同程度的惩罚;采用轮换法的策略来避免因样本不均衡分布带来的负面影响.研究结果表明:所构建的神经网络模型效果令人满意,是行之有效的做法. In order to improve the prediction ability of BP neural network training, the Metropolis criterion that can help to improve the approximation accuracy of BP neural network is adopted to overcome the problem that BP neural network is apt to fall into local minimum during training and learning. Considering the difference of two types of misclassification , Using two penalty coefficients C1 and C2 to punish the two kinds of misclassification to different degrees.The strategy of rotation method is used to avoid the negative influence caused by the unbalanced distribution of samples.The results show that the effect of the constructed neural network model Satisfactory, is an effective approach.