Preliminary exploration and application research on the model of gathering distillate according to the quality based on Fourier transform near infrared spectroscopy

In order to establish a fast, efficient and accurate technology based on gathering distillate according to the quality, the volatile flavor components in different grades of original liquor (initial distillate, middle distillate, last distillate) were analyzed by gas chromatography-mass spectrometry (GC-MS), the principal components were selected by principal component analysis (PCA). The spectrum was obtained by Fourier transform near-infrared spectroscopy (FT-NIR), and the spectrum pretreatment and wavelength screening were performed, the regression prediction model was established based on the principal components, and the model of gathering distillate according to the quality was constructed by random forest (RF). Results showed that 17 principal components were selected by PCA. Multiplicative scatter correction (MSC), competitive adaptive reweighting algorithms sampling (CARS) and support vector regression (SVR) were better methods to construct the regression prediction model, with coefficient of determination R² and root mean square error (RMSE) mean values of 0.8951 and 0.03, respectively. The model of combining with RF was the optimal, and the accuracy rate, precision and recall rates of the model were 99.10%, 99.62% and 99.78%.