The Establishment Of Chinese Base Liquor Analysis Model Based On Near Infrared Spectroscopy

In recent years,near infrared spectrum analysis technology has become one of the most rapid high-tech analysis technology in chemical field,with high efficiency,nondestructive,and at the same time analysis of multicomponent,high reproducibility,has been widely used in food,petrochemical,pharmaceutical analysis,and other fields.This topic selection of superior grade A,grade B,grade 1 A and 1 B grade four different grades of base liquor has the Dukang liquor as the research object,by gas chromatography was developed for the determination of 16 kinds of flavoring substances in liquor base physical and chemical values,and the content and quantity for each material composition than the relationship of law to make analysis and discussion,as the liquor provides theory basis for the classification;Dimension reduction using principal component analysis(PCA),linear discriminant method,markov distance discriminant partial least squares method,BP artificial neural network modeling discriminant methods such as qualitative analysis,qualitative analysis was carried out on the near infrared spectrum,realize the rapid and sizing liquor base liquor;Combined with partial least squares method,using different pretreatment methods,established the near-infrared quantitative model of five kinds of flavor substances,and sweet taste of liquor base liquor composition of rapid detection methods are discussed in this paper.(1)The use of gas chromatography technology for 16 kinds of flavor components in liquor base ethyl caproate and ethyl lactate,ethyl butyrate,ethyl acetate,ethyl heptanoic acid,acetic acid,caproic acid,valeric acid,butyric acid and isoamyl alcohol,isoamyl alcohol,n-butyl alcohol,normal propyl alcohol,aldehyde,acetal,furfural content were tested,esters,alcohols,acids,and aldehyde four types of material at different levels of liquor base content change rule is discussed and analyzed.Results:ethyl caproate base liquor in four levels,the ratio of the average content of ethyl lactate and ethyl acetate were 1.2:1:0.7,0.8:1:0.6,0.6:1:0.5 and 0.4:1:0.4(the content of ethyl lactate average value as the reference standard).The base wine isoamyl alcohol and amyl alcohol in the four level of the ratio is about 1.1 ~ 1.2:1,and caproic acid pentanoic acid is 1.4 ~ 1.75:1,acetaldehyde and acetal ratio is about 1.9 ~ 2.0:1,the amyl alcohol,acetic acid,acetaldehyde content did not change significantly.(2)The use of near infrared spectral technology acquisition of the base has thedukang liquor sample spectrum data,the original spectrum of different spectral preprocessing methods,application of principal component analysis combined with the linear discriminant method,markov distance discriminant partial least squares method,BP artificial neural network established the model identification of the liquor base,including LD-MD sample classification accuracy is above 98.47%;Using cross validation method to establish the DPLS model calibration and validation sets the correlation coefficients of above 0.987,average sample identification accuracy is above 99.15%;Using PCA to extract the best principal component as the BP artificial neural network input variables,prediction accuracy of 100%.Comparing the classification results of the model,the above three methods of discriminant analysis can be more accurate to classification of liquor base,MSC-PCA-BPANN classification effect is best,as a better fast classification method of base liquor;As well as the content of flavor components in the liquor base change rule as near infrared classification theory on the basis of the analysis.(3)With liquor base of heptanoic acid ethyl ester,respectively is amyl alcohol,butyric acid,pentanoic acid and furfural content on the basis of the data,analyze the near infrared spectrum data,using different pretreatment methods,combined with partial least squares,respectively,set up five kinds of typical flavor correction model and predictive model.Calibrating the determination coefficient(R2)were 0.981,0.963,0.93,0.963 and 0.963,the root mean square error of cross-validation(RMSECV)of2.99 mg/100 ml,0.101 mg/100 ml,0.296 mg/100 ml,0.849 mg/100 ml and 1.95 mg /100 ml;Validation set decision coefficient(R2)were 0.977,0.928,0.95,0.928 and 0.95,predict root mean square error(RMSEP)were 3.18 mg/100 ml,0.49 mg/100 ml,0.366mg/100 ml,1.165 mg/100 ml and 2.58 mg/100 ml.Show that the model of quality index has reached the ideal expectations,to establish a quantitative model to predict the accuracy,stability and good performance.