Soft Sensor Modeling Of Key Process Of Butanol Fermentation Based On Bayesian And Support Vector Machine

With the progress of industrialization,energy has become an indispensable part of social development.However,non-renewable resources such as petroleum are becoming less and less and environmental pollution is large.Therefore,research on renewable new energy has become one of the focuses of modern development.Butanol has the advantages of high octane number,low hygroscopicity,and can be mixed with gasoline at any ratio.It is an outstanding representative of new energy.In order to avoid the problems of environmental pollution,complicated operation and high cost caused by the traditional chemical synthesis method of butanol,a method for synthesizing butanol by microbial fermentation using straw as a substrate is proposed.The straw is hydrolyzed by various enzymes to obtain a saccharification solution containing a variety of sugars,and the saccharification solution is inoculated with Clostridium beijerinckii through the metabolism to synthesize butanol.After researching the process of straw fermentation to synthesize butanol,the key biomass in the fermentation process was determined,and the key influencing factors were screened out from the dynamic analysis of the key biomass to determine the optimal fermentation conditions.In order to improve fermentation efficiency,real-time and accurate measurement of key biomass is required,and soft sensing is introduced to predict key biomass that cannot be directly measured.In the initial stage of microbial fermentation,there are few sample data,and Support Vector Machines(SVM)has a high generalization ability for small samples.Therefore,algorithm optimization is based on support vector machines.The Bayesian method has good scalability and can be integrated with a variety of algorithms,and the Bayesian method can solve the problems of data deviation and data loss.Therefore,the Bayesian method is used to optimize the support vector machine and the resulting Bayesian method is used.Support Vector Machine Optimized by Bayesian(BSVM)and SVM are compared.The results by analyzed show that the BSVM algorithm has better accuracy,but the stability of lower accuracy needs to be improved.Since the actual fermentation data will have faulty errors,the data will be pre-processed using dynamic time warping;since the soft-sensing technology is based on the regression algorithm,the Support Vector Machine Regression(SVR)algorithm is introduced on the basis of SVR Build the soft-sensing model;as the sample data gradually increases as fermentation proceeds,the deep learning concept is introduced using the hierarchical characteristics of the Bayesian method to build a multi-layer support vector machine regression and its Bayesian optimized multi-model layer support vector machine regression:(DB-SVR),the simulation analysis of the DB-SVR soft-sensing model is carried out first.The results show that the model has good stability and fast response speed.Then the DB-SVR model is applied to the fermentation of straw to synthesize butanol the key biomass prediction experiment in the process,after comparison and analysis with the BSVR model,found that the DB-SVR soft sensor model has good stability,fast response speed,higher accuracy and strong robustness.