Research And Application On Process Detection Of Solid-state Fermentation Of Bioethanol Using Near-infrared Spectroscopy (NIRS) Technique

In order to effectively improve the monitoring efficiency of solid-state fermentation(SSF)process,the research and application of the SSF process based on near infrared spectroscopy(near-infrared spectroscopy,NIRS)technique were studied.The specific research contents are as follows:Study ?: The method of NIRS detection in yeast culture was discussed.First,the raw spectral of all samples were preprocessed by using the standard normal variate transformation(SNV),and then the feature variables could be selected by competitive adaptive reweighted sampling(CARS).Finally,the Gaussian mixture regression(GMR)was used for quantitative prediction of yeast growth process.Experimental results showed as follows: 30 feature wavelength variables of NIR data were selected by CRAS algorithms,and the root mean square error of prediction(RMSEP)and correlation coefficient(2R)of GMR are 0.073 and 0.988.The overall results demonstrate that near infrared spectroscopy is feasible to realize the quantitative description of yeast growth process.In addition,before the model is established,the extraction of spectral feature information with CARS can not only help to improve the prediction accuracy of the model,but also reduce the complexity of the modeling.Study ?: quantitative detection method of process parameters in SSF of bioethanol based on NIRS was discussed.First,the reference value of ethanol and glucose content in the sample were obtained by conventional physical and chemical analysis,and then the original spectrum was preprocessed by SNV method.Finally,the deep learning method named stacked denoising auto-encoder neural network(SDAE-NN)and partial least squares(PLS)are effectively fused to achieve spectral feature learning and accurate detection model construction.The experimental results based on SDAE-PLS were achieved as follow: RMSEP,2PR and residual predictive deviation(RPD)are 1.651,0.978 and 6.776 for glucose model,and RMSEP,2PR and RPD are 1.382,0.969 and 5.734 for alcohol model.The research shows that it is feasible to use the near infrared spectroscopy to detect the process parameters of SSF of bioethanol.In addition,the SDAE could dig deeper information when learning spectral data,which is helpful to improve the prediction accuracy of the model.Study ?: The qualitative identification method of process state in SSF of bioethanol based on NIRS was discussed.On the basis of study II,the experiment of abnormal ethanol solid-state fermentation was added,and the two kinds of spectral data were mixed with a deep fusion algorithm named stack extreme learning machine auto-coder(SELM-AE)network to realize qualitative identification of normal and abnormal state of SSF of bioethanol.The experimental results show that the accuracy of the model in the training set and test set were up to 97.222% and 100%,and the model training time is only 2.652 s.The overall results sufficiently demonstrate that the effective fusion of NIRS and SELM-AE can achieve high accuracy identification of process state of ethanol SSF.This study provides the new methods for building on-line monitoring model of NIRS in solid-state fermentation process,aiming at improving the accuracy and timeliness of the monitoring of ethanol solid fermentation process.The research results can provide a technical basis for the research and development of NIRS portable monitoring equipment for ethanol solid-state fermentation.