乳制品是人们日常生活中一种重要的营养食品，为了提高对乳制品质量安全风险预测的准确性，保障乳制品质量安全，本文基于检测产品和检验数据的随机性、模糊性以及信息不完全性，将所得不同地区的乳制品检测数据通过改进的softmax公式进行等级划分，并按自然日进行分箱处理，通过风险权重等比例映射法得到风险等级，充分利用了乳制品灰色数据，对检验合格数据中的潜在风险进行挖掘。采用小波分解（Wavelet Decomposition，WD）和长短期记忆神经网络（Long Short-Term Memory，LSTM）模型结合的方式，对不同地区的乳制品检测数据进行风险预测。结果表明，该组合模型的平均准确率达97.54%，标准偏差为0.03，与经验模态分解（Empirical Mode Decomposition，EMD）-LSTM模型和有选择性重构且间隔为2的WD-LSTM模型相比准确率更高，稳定性更好，可实现对乳制品质量风险的预测和防控，能为乳制品的风险监管提供有利参考和技术支撑。

Dairy products are important nutritious food items in people’s daily life. To improve the analysis of dairy product quality and safety risk prediction, the randomness, fuzziness, and incomplete information of detection products and inspection data obtained during dairy product evaluation in different regions were divided into grades using an improved softmax formula and processed in boxes according to the natural day. The risk grades were determined from the risk weight and other proportional mapping methods, and the dairy gray data were used to determine the potential risks in the qualified inspection data. Wavelet decomposition combined with the long short-term memory (LSTM) model was used to predict the risk of dairy product detection data in different regions. The results showed that the combined model exhibited an average accuracy of 97.54% and a standard deviation of 0.03, indicating higher accuracy and better stability compared to those of the empirical mode decomposition-LSTM model and wavelet decomposition-LSTM model with selective reconstruction and an interval of 2. Thus, risks associated with dairy product can be predicted and prevented. These results provide a reference and technical support for supervising risks associated with dairy products.

国家重点研发计划项目（2018YFC1603602）