Research On Process Monitoring Methods Of Solid-state Fermentation Of Bioethanol Based On Molecular Spectral Fusion Technique

Solid-state fermentation(SSF)is one of the key steps in the production of bioethanol.In order to improve the monitoring efficiency and product yield of ethanol SSF process,this work innovatively proposes an idea for monitoring the ethanol SSF process based on molecular spectral fusion technique.The specific study contents are as follows:(1)Research on the monitoring methods of yeast growth status based on molecular spectral fusion technique.First,the Raman spectra and near-infrared(NIR)spectra of the yeast fermentation samples were acquired using Raman and NIR spectrometers,and the spectra were preprocessed by Savitzky-Golay(SG)filtering combined with standard normal variate(SNV)method.Second,the variable combination population analysis(VCPA)was utilized to optimize the characteristic wavelengths of the two preprocessed molecular spectra,and then fused at the feature layer.Finally,support vector machine(SVM)models based on fused spectral features were established for quantitative monitoring of yeast growth status.Experimental results showed that,compared with the optimal VCPA-SVM model based on single-molecule spectral features,the best VCPA-SVM model based on fused spectral features could obtain better prediction performance.The root mean square error of prediction(RMSEP),coefficient of determination(R_P~2)and relative percent deviation(RPD)of the best model were 0.468,0.978 and 6.693,respectively.The overall results demonstrate that the molecular spectral fusion technique can effectively improve the monitoring accuracy of yeast growth status.In addition,the VCPA method has a promising potential in the mining of molecular spectral feature wavelengths,which can effectively reduce the dimension of fusion features.(2)Research on the state recognition methods of ethanol SSF process based on molecular spectral fusion technique.First,the Raman spectra and NIR spectra of the ethanol SSF process samples were acquired using the spectrometers.Second,the SG filtering and SNV were used to preprocess the collected Raman spectra and NIR spectra,respectively,and principal component analysis(PCA)was employed to perform feature dimensionality reduction of the two preprocessed molecular spectra,and then fused on the feature layer.Finally,the k-nearest neighbor(KNN),extreme learning machine(ELM)and SVM recognition models of ethanol SSF process based on fused features were established,respectively.Experimental results showed that,compared with the single-molecule spectral recognition models,the recognition models based on fusion features all showed obvious advantages.Among them,the PCA-SVM model based on fused features had the best recognition performance.When the number of principal components(PCs)was 5,the correct recognition rates of the best PCA-SVM model were both 100%in the training set and prediction set.The overall results sufficiently demonstrate that it is feasible to realize the high precision identification of ethanol SSF process by using the molecular spectral fusion technique and appropriate pattern recognition methods.(3)Research on the detection methods of the process parameters in ethanol SSF based on molecular spectral fusion technique.First,the SG+SNV was used to preprocess the Raman spectra and NIR spectra collected by the experiments.Second,a competitive adaptive reweighted sampling(CARS)method was utilized to optimize the feature wavelengths of the preprocessed Raman spectra and NIR spectra,and then fused at the feature layer.Finally,the SVM quantitative detection models based on the molecular spectral fusion features were established for quantitative determination of the process parameters in ethanol SSF.Experimental results showed that,compared with the best CARS-SVM model based on single-molecule spectral features,the performance of the best CARS-SVM model based on fusion features has been significantly improved.For glucose content detection,the RMSEP,R_P~2 and RPD of the optimal CARS-SVM model were 5.398,0.957 and 4.922,respectively.For ethanol content detection,the RMSEP,R_P~2 and RPD of the optimal CARS-SVM model were 4.394,0.977 and 6.758,respectively.The overall results sufficiently demonstrate that molecular spectral fusion technique combined with appropriate chemometric methods is feasible to achieve high-precision quantitative detection of the process parameters in ethanol SSF.This work carried out the research on the monitoring methods of the ethanol SSF process based on molecular spectral fusion technique.The study results can provide theoretical support and method reference for the monitoring of the ethanol SSF process,and also provide technical basis and experimental reference for the development of portable monitoring equipment for ethanol SSF process.