Detection Method Via Near-infrared Spectroscopy For The Component Contents During Ethanol And Monosodium Glutamate Fermentation Process

The demand for ethanol is increasing year by year due to the advantages of low carbon dioxide emissions,low dust emissions,biodegradability and renewables.The bio-fermentation process is one of the important ways of ethanol production.MSG is one of the essential condiments in human life,and real-time monitoring and control of its fermentation process is very representative and practical.In this paper,based on the research of ethanol fermentation process and monosodium glutamate fermentation process,the on-line monitoring of its fermentation process based on near infrared spectroscopy technology is adopted to realize the rapid and non-destructive online detection of the compound concentration in the biological fermentation process.The experimental platform and near infrared spectrum prediction model designed and built can overcome the shortcomings of traditional offline detection.First,the principle and main process of near infrared spectroscopy detection technology are analyzed.For the abnormal points of near infrared spectroscopy caused by interference and noise during the fermentation process of ethanol and monosodium glutamate,spectral data preprocessing methods are given to achieve smooth processing of the spectrum.Correction.Secondly,for the prediction model,three data modeling methods are given.Among them,the classical PLS algorithm model can solve the collinearity problem of spectral data.In addition,the least squares-support vector machine(LSSVR)model based on three different kernel functions can use the principle of structural minimization to avoid overfitting problems.Finally,a spectral modeling method based on the random forest algorithm is given.The parameter setting of the random forest tree and the number of features is realized by the out-of-bag error rate(OOB),and the wavelength is optimized based on the Gini coefficient(G)reduction to optimize the random forest model.In order to compare the prediction effects of the established model,performance evaluation indexes are given.For the ethanol fermentation process,an ethanol fermentation process experiment platform based on FT-NIR spectroscopy technology was built,and six batches of ethanol fermentation experiments using Saccharomyces cerevisiae 4126 as strains were carried out.,The target product ethanol and the OD value reflecting biomass are analyzed by near infrared spectroscopy and the quantitative model is established.The partial least squares algorithm,LSSVR algorithm based on different kernel functions and random forest algorithm are used to model the preprocessed spectrum,and the prediction indexes of root mean square error(RMSEP)and correlation coefficient(R2)are used to evaluate different models.Comparative analysis.The experimental results verify that the three modeling methods have a predictive effect on the ethanol fermentation process,among which the random forest has the best modeling effect on the three target parameters of the ethanol fermentation process.For the actual industrial process of monosodium glutamate fermentation in the Shandong Linghua Group Monosodium Glutamate Factory,we designed and built a follow-up spectrum detection experimental platform in the fermentation tank workshop.Based on the collected 7 batches of near-infrared spectral data of the fermentation broth of different fermentation tanks at different time periods,the spectral data is preprocessed,using partial least squares algorithm,LSSVR algorithm based on different kernel functions and random forest algorithm for modeling The model of the model predicts and analyzes the substrate glucose,the target product,glutamic acid,and the OD value reflecting biomass by near infrared spectroscopy.Finally,the model evaluation indicators RMSEP and R2 are used to evaluate and analyze different models.The results verify that the near-infrared spectroscopy technique based on random forest algorithm has a better model prediction effect.