为了提高白酒酒醅主要成分的快速检测技术水平, 利用近红外光谱法对白酒酒醅中的水份、 淀粉、 酸度和酒精度进行定量分析, 通过PLS建立酒醅成分定量分析模型, 同时采用CARS、 间隔偏最小二乘变量选择法(iPLS)和无信息变量消除法(UVE), 实现光谱特征波段选择, 并利用多元散射校正(MSC)、 导数以及标准正态变量变换(SNV)等预处理方法进行模型优化。 用决定系数(R2)、 预测标准偏差(RMSEP)以及最佳主因子数对模型进行评价, 确定最佳建模方法。 结果表明: 波段筛选能够对模型起到优化作用, 并提高模型运算速度, 其中CARS优化效果最为明显, 酒醅水份、 淀粉、 酸度和酒精度的R2分别达到0.885, 0.915, 0.951, 0.954; RMSEP分别为0.630, 0.519, 0.228, 0.234, 有效地提高了模型的稳定性与准确度。 经过优化, 所建模型预测效果好, 能够满足白酒酒醅主要成分的快速检测要求, 在实际应用中有一定的参考价值。

In order to improve the technical level of the rapid detection of liquor fermented grains, in this paper, use near infrared spectroscopy technology to quantitative analysis moisture, starch, acidity and alcohol of liquor fermented grains. Using CARS, iPLS and no information variable elimination method (UVE), realize the characteristics of spectral band selection. And use the multiple scattering correction (MSC), derivative and standard normal variable transformation (SNV) pretreatment method to optimize the models. Establish models of quantitative analysis of fermented grains by PLS, and in order to select the best modeling method, using R2, RMSEP and optimal number of main factors to evaluate models. The results showed that the band selection is vital to optimize the model and CARS is the best optimization of the most significant effect. The calculation results showed that R2 of moisture, starch, acidity and alcohol were 0.885, 0.915, 0.951, 0.885 respectively and RMSEP of moisture, starch, acidity and alcohol were 0.630, 0.519, 0.228, 0.234 respectively. After optimization, the model prediction effect is good, the models can satisfy the requirement of the rapid detection of liquor fermented grains, which has certain reference value in the practical.