Modelling And Optimisation Of Laboratory Simulated Baijiu Solid-state Bilateral Fermentation

Baijiu is the most representative ancient distilled liquor in China,also known as China’s"national liquor",and occupies a unique position in Chinese traditional culture.The traditional Chinese liquor craftsmanship has a history of thousands of years,and it has always mainly depended on personal operation skills and experience.In recent years,liquor production has also begun to be combined with some advances in modern biotechnology,automation and electronic information technology to improve traditional process methods.At present,the development of liquor processing technology is developing in the direction of full mechanization,automation and intelligence,and at the same time,it is necessary to ensure the quality and flavor of liquor.In this paper,starting from the realization of the identification of the state of the fermented grains entering the pond and the control of the fermentation process of the liquor,the near-infrared spectroscopy machine learning model of the fermented grains entering the pond and the kinetic model of the solid-state bilateral alcohol fermentation of the liquor are sought to be established.In the aspect of near-infrared spectral model,the preprocessing and feature screening strategies of near-infrared spectral data are discussed.In terms of kinetic model,the influence of the introduction of enzyme kinetic model on model fitting results was discussed,and the differences in parameter results,fitting degree and time-consuming of two different algorithms were compared.The main research conclusions are as follows:(1)In the accumulation stage,the quality of the original wine obtained by distillation according to the proportion of different upper-layer fermented glutinous rice grains is significantly different.The overall wine yield of each sample is positively correlated with the upper-layer fermented grains ratio.The ensemble(subspace discrimination)classification accuracy was the highest among the near-infrared spectroscopy machine learning classification models of wine grains obtained by using 24 different model presets,reaching 93.3%.The above conclusions can indicate that this study successfully established a near-infrared spectroscopy machine learning model for the liquor grains before entering the pond,which can realize a more accurate discrimination of the liquor grains before entering the pond.(2)In the modeling results of 2520 groups of data preprocessed by different preprocessing strategies,no baseline correction is used,MSC algorithm is used for scattering correction,and then S-G smoothing(windows:11 pt,order:4)is used to eliminate noise,and finally The data obtained by the preprocessing strategy without scaling processing obtained the highest model classification accuracy of 98.4%.According to the results,it can also be shown that not every preprocessing step must be performed,and redundant preprocessing operations may not necessarily affect the model.Model predictions are favorable.In the results of modeling the original spectral data using two different feature screening methods,PCA and LASSO,when LASSO selected 4 wavelength data for classification training,the classification accuracy of the model reached 93.3%,and it was found that this method The models that can achieve the highest classification accuracy after feature selection data modeling are preset as Naive Bayes(Gaussian)and SVM(fine Gaussian),which are different from those before feature screening.The above conclusions can indicate that this study successfully proposed an effective near-infrared spectrum preprocessing strategy and feature selection method,which improved the accuracy of model classification and reduced the complexity of data and analysis efficiency.(3)The change process of yeast growth,glucose conversion,product ethanol formation and substrate starch hydrolysis in the laboratory simulated liquor solid-state bilateral alcohol fermentation process was analyzed,and it was concluded that the fermentation period could be controlled within 100 h,and the ethanol concentration was found.It has an inhibitory effect on yeast growth and ethanol production.Based on the experimental results,a kinetic model of liquor solid-state bilateral alcohol fermentation was established,and the objective function of the subsequent optimization process was constructed.The above conclusions can play a certain reference role in the construction and optimization of other fermentation kinetic models.(4)Before the introduction of the enzyme kinetic model,the comprehensive goodness of fit R~2 of the model was 0.828,but after the introduction of the enzyme kinetic model,the comprehensive goodness of fit R~2 of the model increased to 0.969,and the goodness of fit of the starch()model R~2 Reaching 0.985 09,in the premise of no enzyme concentration data,it is proved that the parameter identification results in the enzyme kinetic model are reasonable,and the introduction of the enzyme kinetic model has greatly improved the overall fitting effect.In the UGO fitting results,except that the estimation of the fitting results of the glucose conversion model(R~2=0.992 65)is slightly smaller than that of GA(R~2=0.992 77),the fitting results of the other three models are all better than GA.The fitting result is better,and the algorithm operation efficiency of GA is higher than that of UGO.The parameters obtained by the two algorithms in the ethanol kinetic model and the glucose conversion model have little difference,but there are certain differences in the starch hydrolysis model,yeast growth model and enzyme kinetic model.From a practical point of view,the parameters obtained by GA are obviously different.It has practical significance,and the parameters obtained by UGO are more reasonable,and the introduction of the enzyme kinetic model is also helpful to the rationality of the parameters.The goodness-of-fit R~2 of the four fitting curves of starch hydrolysis,glucose conversion,yeast growth and ethanol production obtained by UGO reached 0.986 52,0.992 65,0.910 52 and 0.996 45,respectively.Except for the yeast growth model(0.910 52),other The goodness of fit R~2 of the models is far greater than 0.95,indicating that the model is successful in a statistical sense,and has a certain reference value for the optimization and control of the future solid-state bilateral fermentation process of liquor,and has a certain significance for the automation development of liquor production process.