通过静态质量损失法腐蚀试验获取BP神经网络的样本数据,利用MATLAB的工具箱函数建立了拓扑结构为5×8×10×1的BP神经网络,并对网络模型的预测精度进行探讨和研究。结果表明,在样本集和训练条件下,5×8×10×1 BP网络对稀土合金铸铁碱腐蚀深度的预测精度较高,较好的解决主要合金成分、腐蚀时间、介质温度与静态腐蚀深度之间的非线性关系,可用于稀土合金铸铁在烧碱液中的静态腐蚀性能的预测分析和研究。 The BP neural network sample data was obtained by static mass loss corrosion test. The BP neural network with topological structure of 5 × 8 × 10 × 1 was established by MATLAB toolbox function, and the prediction accuracy of the network model was discussed. The results show that the prediction accuracy of 5 × 8 × 10 × 1 BP network on the alkali corrosion depth of rare earth alloy cast iron is high under the sample set and training conditions, which can better solve the main alloy composition, corrosion time, medium temperature and static corrosion depth The nonlinear relationship between the rare earth alloy cast iron can be used in the static corrosion of caustic soda in the prediction of corrosion analysis and research.