食用油油酸的近红外特征谱区优选

Research on Selecting Characteristic NIR Regions of Oleic Acid in Edible Oil

为提高食用油油酸的近红外定量分析模型的预测性能,采用4种波长变量优选方法:移动窗口偏最小二乘算法(MWPLS)、间隔偏最小二乘法(i PLS)、向后间隔偏最小二乘法(Bi PLS)、组合间隔偏最小二乘算法(Si PLS),优选食用油油酸近红外光谱特征区间,建立57份食用油样本的油酸定量分析模型。试验结果表明,相较于全谱建模,4种变量优选方法都能在有效地减少建模所用的变量数的同时提高模型性能,其中采用Si PLS优选变量所建的油酸定量模型的预测性能最优,决定系数R2为0.995 0,交叉校验均方根误差(RMSECV)为1.037 2,预测均方根误差(RMSEP)为0.924 6。

To improve the prediction performance of NIR model to oleic acid in edible oil, 4 kinds of wave-length variable selection methods, moving window partial least square algorithm （MWPLS）, interval partial least squares （iPLS）, backward interval partial least squares （BiPLS） Synergy interval Partial Least Squares algorithm （ SiPLS）, have been employed respectively to select characteristic NIR intervals of oleic acid in edible oil. The oleic acid quantitative models of 57 samples were built based on characteristic intervals chosen by the 4 methods above. The experiment results showed that compared with the model built by full spectrum, the 4 methods were effective in reducing the variable numbers and improve the models＇performance. Among them, the best model was established by SiPLS. Determination coefficient （R^2） , root mean square error of cross validation （RMSECV） and root mean square error of prediction （RMSEP） were 0. 995 0, 1. 037 2 and 0. 924 6.