Study On Classification Identification And Pattern Recognition Of Base Liquors Based On Trace Components

As a traditional national industry,Chinese liquor has a long history and profound cultural heritage.As the mother liquor for Chinese liquor blending,the status and function of base liquor cannot be ignored.However,there are a series of problems such as adulteration,forgery,difficulty in tracing the origin of the liquor age and so on.Therefore,how to use the high sensitivity,high resolution,and rapid detection of base liquor to achieve the identification,classification,and pattern recognition of different grades,ages,and flavor types base liquor is the focus of current research.And so in this work,GC and GC/Q-TOFMS were employed to quantitatively detect the trace components in base liquor.The principal component analysis(PCA),hierarchical cluster analysis(HCA),linear discriminant analysis(LDA)and neural network analysis(ANN)were used to identify and classify the base liquor to build the model.The main work of this study is as follows:(1)Thirty three trace components,including ethyl acetate,acetaldehyde and?-phenylethanol in liquor with different grades and flavor type,were quantitatively determined by gas chromatography(GC)internal standard method.And the precision and recovery of the analytical method were evaluated.The data were analyzed by ANOVA,PCA,HCA and LDAs.The analysis results showed that ten trace components of acetaldehyde,acetal,methanol,ethyl butyrate,secondary butanol,n-propanol,isobutanol,n-butanol,ethyl acetate and ethyl lactate had significant effects on different grade base liquor.Ten trace components of acetaldehyde,ethyl acetate,acetal,methanol,ethyl butyrate,secondary butanol,isobutanol,isoamyl alcohol,ethyl acetate and ethyl lactate had significant effects on different flavor base liquor.The base liquor with different grades and flavors can be successfully differentiated using PCA and HCA without any misclassification.And the LDA results suggested that the successful classification for the unknown samples without anymistakes.(2)The GC/Q-TOFMS internal standard method was used to quantitatively detect79 trace components of(eg.ethyl formate,acetal,formic acid)in base liquor with different grades and age.The accuracy and repeatability of as-presented methods were evaluated by the precision and recovery test.Forty three kinds of trace components,including ethyl formate,propionic acid,n-propanol and 37 trace components(eg.ethyl formate,butyric acid and 4-methylphenol),respectively displaying the ages and grades of base liquor were screened out by single factor variance analysis test.The ages of base liquor can be correctly differentiated using the PCA,HCA and LDA towards above mentioned 43 trace components.And accuracy rate of 96.7%for unknown samples can be obtained by LDA.(3)The comprehensive characteristic index G corresponding to different grade base liquor were obtained by PCA of 37 kinds of trace components(displaying significant influence on different grade base liquor).Linear regression,quadratic regression,cubic regression,logarithmic regression and index regression were carried out using the comprehensive index G as independent variable and sensory evaluation score U as dependent variable.To acquire the regression equation U=f g.And a grade evaluation model of basic liquor with different quality was established,U=0.00013g~3-0.0182g~2+0.993g+72.19.The model was verified and the coincidence rate reached 95%.(4)After the data of 43 trace components(displaying significant influence on ages of base liquor)processed with Z standardization,the wine age recognition model was established by neural network.And the model was evaluated by confusion matrix and F1-score score.The results show that the accuracy of neural network is only 20%at the beginning of training.After the iteration of more than 40,the training accuracy reaches100%.The lowest probability of the modeland the classification accuracy are74.6%and100%,respectively.The forecast accuracy rate for all categories is 100%.