中红外光谱技术在浓香型白酒原酒关键指标分析中的应用

基于中红外光谱分析技术对浓香型白酒原酒中总酸及己酸乙酯的关键指标含量进行检测分析，采用标准正态变换（SNV）方法对数据进行预处理，结合组合间隔偏最小二乘波段选择法（SiPLS）及遗传算法（GA）对各指标的特征吸收波长进行筛选，最终采用偏最小二乘法（PLS）建立分析模型，通过决定系数R2、预测标准偏差（RMSEP）、性能和标准差之比（RPD）以及实际生产的独立测试样品对所建模型效果进行评价，探索中红外光谱分析技术结合波段筛选用于白酒原酒关键指标的定量分析效果。结果表明：采用中红外全光谱397~4 000 cm-1波段所建模型效果并不理想，总酸及己酸乙酯模型的RMSEP值分别达到0.156及0.981，R2仅有0.666及0.453，RPD值为1.731及1.213，代表模型整体的相关拟合效果并不理想，预测误差较大，无法应用于实际生产。而采用GA在SiPLS波段优化基础上进一步进行特征变量筛选后，所建模型效果有了非常显著的提升，总酸及己酸乙酯两项指标的GA-SiPLS模型均体现出了较高的预测精度，R2分别提升到了0.993及0.997，RMSEP值分别降低到0.023及0.077，RPD值提升至11.739及15.455，变量数也分别从935个降低到55及40个，在保留关键信息变量，体现原酒中总酸及己酸乙酯指标特征吸收的同时，有效减轻了模型的复杂程度，同时提高了模型运算速度及预测效果，充分体现了波段筛选对于中红外光谱分析技术应用于白酒原酒关键指标定量分析的重要性，也说明了中红外光谱技术结合波段筛选在白酒品质控制及原酒成分分析中的巨大潜力。考虑到白酒成分复杂，大部分关键品质指标含量相对较低，而中红外吸收区域的基频吸收强度具有比倍频、合频吸收强数十倍的特点，中红外光谱可能比其余光谱技术更适用于白酒液体样品的快速分析，为白酒酿造过程的品质控制提供技术借鉴，同时为酒类品质的快速分析方法开拓提供新的思路。

Application of Mid-Infrared Spectroscopy in the Analysis of Key Indexes of Strong Flavour Chinese Spirits Base Liquor

Quantitative analysis of total acid and ethyl caproate content of key indicators in strong flavour Chinese spirits base liquor（SACSL）based on mid-infrared spectroscopy technology, data preprocessing using standard normal transformation (SNV) method, combined with combined interval partial least squares band selection method (SiPLS) and genetic algorithm (GA) to screen the characteristic absorption wavelength of each index, and finally use partial least squares (PLS) to establish an analysis model, through the determination coefficient R2, prediction standard deviation (RMSEP), performance and standard deviation ratio (RPD) and the actual production of independent test samples to evaluate the effect of the built model, and explore the effect of mid-infrared spectroscopy analysis technology combined with band screening for quantitative analysis of key indicators of SACSL. The results show that the effect of the model built using the mid-infrared full spectrum 397~4 000 cm-1 band is not ideal. The RMSEP values of the total acid and ethyl caproate models reach 0.156 and 0.981, respectively, the R2 is only 0.666 and 0.453, and the RPD value is 1.731 and 1.213, the overall fitting effect of the representative model is not ideal, the prediction error is large, and it cannot be applied to actual production. After the further screening of characteristic variables based on the optimization of the SiPLS band using GA, the effect of the built model has been significantly improved. The GA-BiPLS model with two indicators of total acid and ethyl caproate showed higher prediction accuracy, R2 is increased to 0.993 and 0.997, RMSEP value is reduced to 0.023 and 0.077, RPD value is increased to 11.739 and 15.455, and the number of variables is also reduced from 935 to 55 and 40, respectively, while retaining key information variables to reflect the base wine. While the index characteristics of total acid and ethyl caproate are absorbed, it effectively reduces the complexity of the model, and at the same time improves the calculation speed and prediction effect of the model, which fully reflects the application of band screening to the application of mid-infrared spectrum analysis technology to the quantification of key indicators of SACSL. The importance of analysis also illustrates the great potential of mid-infrared spectroscopy technology combined with chemometric methods in the quality analysis of liquor. Considering the complex composition of liquor, most of the key quality indicators are relatively low, and the fundamental frequency absorption intensity of the mid-infrared absorption region is dozens of times stronger than the frequency doubled and combined frequency absorption. Mid-infrared spectroscopy may be more suitable than other spectroscopy techniques. The rapid analysis of liquor liquid samplesprovide technical reference for the quality control of the liquor brewing process, and provide new ideas for the rapid analysis of liquor quality.