针对设施与装备选材时对金属材料大气腐蚀预测的需求,应用神经网络理论,在改进算法的基础上,建立了一种基于贝叶斯正则化算法的BP神经网络预测模型,探讨了小样本条件下的网络泛化问题,分析了算法对泛化精度的提高能力,并通过Matlab实现了仿真验证.通过对试验数据的测试,证实了改进算法和模型的有效性. Aiming at the demand of atmosphere corrosion prediction of metal materials in the selection of facilities and equipment, a BP neural network prediction model based on Bayesian regularization algorithm is established based on the improved algorithm and neural network theory. The small sample conditions The generalization of the network is analyzed, and the ability of the algorithm to improve the generalization accuracy is analyzed, and the simulation verification is realized by Matlab.Through the test of the test data, the effectiveness of the improved algorithm and the model is verified.