基于氢谱核磁共振的岭回归算法快速预测煎炸油氧化指标

Rapid Prediction of Oxidation Index of Frying Oil Based on Ridge Regression Algorithm of ^(1)H NMR Spectroscopy

通过氢谱核磁共振技术检测并分析大豆油煎炸时的氧化产物,经主成分分析后确定Z,E-共轭结构、E,E-共轭结构、直链饱和醛、(E,E)-2,4-二烯醛、(E)-2-烯醛5种特征基团作为特征矩阵,采用岭回归算法构建了煎炸油同时检测过氧化值与极性组分的快速预测模型,结果表明,交叉验证下正则化参数α为0.077,过氧化值与极性组分模型的决定系数(R2)分别为0.968与0.957,均方根残差(RMSE)分别为0.208与5.249,经重现性检验,预测值与国标值相对标准偏差小于5%,说明预测法重现性和精度良好。同时,采用两种方法检测时没有显著性差异,说明基于氢谱核磁共振和机器学习算法联用,构建氧化指标的预测模型,可为煎炸油品质监测指标提供一种新思路。

Oxidation products of soybean oil during frying are detected and analyzed by ^(1)H nuclear magnetic resonance(^(1)H NMR) spectroscopy. After principal component analysis, five characteristic groups(Z,E-conjugated structure, E,E-conjugated structure, linear saturated aldehyde,(E,E)-2,4-dialdehyde,(E)-2-enaldehyde) are determined as the characteristic matrix. A fast prediction model for simultaneous detection of peroxide value and polar components in frying oil is established by ridge regression algorithm. The results show that under cross validation, the regularization parameter α is 0.077, the determination coefficients(R^(2)) of peroxide value and polar component model are 0.968 and 0.957 respectively, and the root mean square residuals(RMSE) are 0.208 and 5.249 respectively. Through the reproducibility test, the relative standard deviation between the predicted value and the national standard value is less than 5%, indicating that the prediction method has good reproducibility and precision. At the same time, there is no significant difference when using the two methods, which shows that the prediction model of oxidation index based on the combination of ^(1)H NMR and machine learning algorithm can provide a new idea for the quality monitoring index of frying oil.