Identification Of Storage Time And Quality Of Baijiu Based On Profiling Of Volatiles

Baijiu is unique traditional distilled liquor with specific taste and value.The volatiles which account for about 2%determines the style and quality of Baijiu.Storage process is also an important way for quality improvement of Baijiu,and the value of Baijiu is closely related to the storage time and quality.At present,the discrimination of Baijiu quality mostly depends on a few physical and chemical indicators,and relay on sensory evaluation.the identification of Baijiu storage time is still in the initial stage.It is urgent to establish a easeand accurate method to discriminate storage time and quality of Baijiu.In this study,the storage time-indicating compounds were identified,and a regression and classification model is used to predict the aging time and quality temporarily.A method for rapid screening time-indicating compounds is established for time identification,and a regression model is used to predict the aging time.The influencing factors of Baijiu quality are complex,and by whole dataset a classification model was established to realize the the quality grade of Baijiu.The main contents and conclusions of this study are as follows:(1)Using HS-SPME/GC-MS technology to separate and identify the volatile components of 110 Lu-flavor Baijiu(70 raw Baijiu and 40 commercial Baijiu)with different storage time and 113 different quality sauce-flavor Baijiu.A total of 98 volatile compounds are detected from raw Baijiu,including 6 alcohols,9 aldehydes,7 acids,and 68 esters,3ketones and 5 others.A total of 174 volatile compounds are detected from different quality sauce-flavor Baijiu,including 19 alcohols,20 aldehydes,9 acids,and 80 esters,16 ketones and 30 others.(2)Analyzing the volatile compounds data of different alcohol content Baijiu during the aging process.Three methods including PLS-DA,Spearman Correlation and Random Forest are used to screen time-indicating compounds,and the ethyl oleate is the key compound.The key compounds of ethyl laurate,ethyl myristate,ethyl pentadecanoate and ethyl palmitate were selected from 52%vol raw Baijiu after two years of storage.The number of carbon atoms of these long-chain fatty acid ethyl esters are positively correlated with the storage time of detected Baijiu samples.The above conclusions are verified and confirmed using commercial Baijiu with a storage time of four years,and the peak area and detection rate of long-chain fatty acid ethyl esters in the samples are higher with longer storage time.The detection rate of ethyl oleate is 0 in 0-1year Baijiu and 90%in 3-4 year Baijiu,ethyl palmitate increases from 60%to 100%,ethyl pentadecanoate increases from 40%to 90%.In this study,for the first time long chain fatty acid ethyl esters were identified as time-indicative compounds in Baijiu,which provided crucial information for the prediction of the storage time of Baijiu.(3)The XGBoost algorithm is used to establish regression model to discriminate the storage time of Baijiu temporarily.Using the extreme random forest variable importance evaluation,sklearn feature selection module of F＿Region and mutual＿info＿regression,the important feature variables of the model were screened and the effective modeling variables were determined.According to the model evaluation results,the accuracy reached 95.83%.The storage time of Baijiu is predicted by test data and output in the form of R2,which is 0.987.This result demonstrated the reliability of the prediction model and further confirmed that the time-indicating compounds screened above can be applied to optimize the discriminant model of Baijiu age.(4)According to the sensory evaluation,the collected 113 sauce-flavor Baijiu of different quality are divided into three grades(excellent,grade A,grade B),which was applied as the grouping label for the quality discrimination model.Based on the sample collected and data obtained in this study,the grade of sauce-flavor Baijiu is positively correlated with the detected volatiles.The PLS-DA model is established to screen grade difference compounds,the R~2X,R~2Y and Q~2 of PLS-DA are 0.312,0.721,0.6,respectively,which shows that the discriminate model can effectively screen compounds with varied grade.According to the rule that parameters with VIP value greater than 1,49 key compounds were generally considered as important parameters,causing grade difference.The highest VIP value was methane,diethoxy-(1.829),followed by hexanoic acid,propyl ester(1.780)and 2-Furancarboxaldehyde,5-methyl-(1.588).(5)The SVM algorithm is used to establish a classification model to discriminate the quality of sauce-flavor Baijiu.Comparing the modeling performance of the original volatile components data,the pairwise ratio of the volatile components data,and the combination of the above two data set.The combined data set get the highest accuracy of 0.7.Comparing different classification methods,including linear kernel,polynomial kernel and RBF,the highest score was obtained by polynomial kernel,which is 0.603.The Grid Search CV module is further applied to optimize the model,using test set data to predict quality of Baijiu,and the result of AUC is 0.844.