Model Optimization For Quality Of Chinese Rice Wine Based On Near Infrared Spectroscopy

In this paper, Shaoxing rice wine is the main study objective. UsingFT-NIR with a combination of chemometrics analysis, the major quality model, suchas alcohol model, total acid model and total sugar model, of rice wine was researched.Meanwhile utilizing local model, total acid model and total sugar model wereoptimized. Consider the influence of interaction among principal ingredients, such asalcohol, total acid and total sugar, on quality model, Finally discuss the decision toestablish the material composition of the wine style model and its main qualityclassification model is to optimize,The following ideas and conclusions in thisarticle:(1) Firstly, using a leveraged residual value and the student’s t-test with acombination of the principle of recycling one by one, abnormal samples werediscriminated and eliminated. Then calibration sets of rice wine major quality modelwere chosen and compared by Random, KS and SPXY. Secondly, by interval partialleast squares wavelength range was selected and the major quality model wasestablished, in which alcohol model’s accuracy was the best and total acid model andtotal sugar model’s accuracy, were not good enough, should be improved. Finally,based on principal component analysis (PCA) and PCA-Ward, the local model oftotal acid and total sugar was established. With a randomly selected10samples fromlocal model, total sugar’s average relative deviation was reduced from7.28%to4.62%and3.59%, respectively and total acid’s average relative deviation decreasedfrom6.41%to3.44%and2.55%, respectively. It has improved the predictive abilityeffectively.(2) Owing to the complexity of spectra and ingredients of rice wine, it issignificant to consider the influence of interaction among principal ingredients onquality model. It concluded that increasing the role of component concentration willlead to action component model accuracy decreases, the impact and role of substance concentration, alcohol is most obvious, followed by total sugar, total acid weakest;through wavelength selection, to avoid the strong interaction region can weaken theimpact on the quality of the interaction between the main components of the model,so that the model optimization.(3) PCA and PLS-DA discriminant analysis models were established, in whichPLS-DA discriminant model was more robust and its recognition rate was100%, itcan discriminate the difference of three types of rice wine successfully. Through theclassification of rice wine style, the influence of main ingredient of decision style onrice wine quality model can be reduced effectively, which also makes the modeloptimization.