基于BP神经网络的UHT纯牛奶包装货架期预测

Shelf Life Prediction of UHT Milk Packaging Based on BP Neural Network

为探究初始蛋白质与脂肪含量、贮藏温度对UHT纯牛奶包装货架期的影响,以三种UHT纯牛奶为研究对象,试验测定23、30和37℃贮藏过程中样品褐变指数、蛋白水解度指标。将数据集整合,根据其在预测集上的表现确定具体的输入参数,开展基于BP神经网络的UHT纯牛奶包装货架期预测。结果表明,BP神经网络模型对UHT牛奶褐变指数、蛋白水解度指标的拟合度为0.9412、0.9527,相较于传统多元线性回归模型的0.8799和0.9211,经优化隐含层神经元数的BP神经网络模型对UHT纯牛奶贮藏期间的特征指标变化预测精度更高,为不同配方UHT纯牛奶货架期的快速准确预测提供技术支持。

To investigate the effects of initial protein,fat content,and storage temperature on the shelf life of UHT pure milk packaging,three types of UHT pure milk were used as research objects to experimentally measure sample browning index and protein hydrolysis index during storage at 23,30,and 37℃.Integrate the dataset and determine specific input parameters based on its performance on the prediction set,and carry out UHT pure milk packaging shelf life prediction based on BP neural network.The results showed that the fitting degrees of the BP neural network model for the browning index and protein hydrolysis index of UHT milk were 0.9412 and 0.9527,respectively,and compared with traditional multiple linear regression model’s number of 0.8799 and 0.9211,the BP neural network model with optimized hidden layer neuron numbers had higher prediction accuracy for the changes in characteristic indicators during the storage period of UHT pure milk,providing technical support for rapid and accurate prediction of the shelf life of UHT pure milk with different formulas.