Research On Soft-sensor For Pichia Pastoris Fermentation Process Based On PCA-ARLSSVM

Inulinase is a hydrolytic enzyme capable of hydrolyzing beta-2,1-D-fructan glycosidic bonds,often used in the bio-fermentation industry to refine inulin,producing bioethanol,butanol,single-cell lipids,fructooligosaccharides,lactic acid and other industrial products.Therefore,inulinase shows great application prospect in medicine,health care,food and bioenergy.The parameters to be measured in the high-density fermentation process of Pichia pastoris include chemical parameters,biological parameters and physical parameters.The concentration of inulinase can only be quantified by offline measurement of enzyme activity,which not only requires a lot of time and energy,but also affects the controlling decision and technology improvement in the fermentation process.To solve this problem,this dissertation proposed a soft-sensor model to predict the concentration of inulinase during the fermentation process.The main contents of this dissertation are as follows:1.In this dissertation,high-density fermentation with Pichia pastoris KM71/pPic9k-Eninu was compeletd after full analysis of the fermentation process of Pichia pastoris to obtain the raw data of key variables.A standard least squares support vector machine(LSSVM)model was esTab.lished to predict inulinase concentration online.2.To solve the problems of high dimension of input data,strong correlation between input data,poor model robustness,long operation time and low prediction accuracy of the standard LSSVM model,this dissertation proposed a hybrid model(PCA-ARLSSVM)based on PCA and adaptive robust least squares support vector machine.Firstly,through the principal component analysis of the input sample data,the dimensionality and the correlation of the input data was reduced.As a result,the operation process of the model was simplyfied and the operation time was reduced.Secondly,based on LSSVM model,the PCA-ARLSSVM can improve the robustness of the model by using degree of membership function.Finally,in order to make the soft-sensor model better adapt to the working conditions and improve the generalization ability of the model,this dissertation takes the total prediction error of the model as the threshold value,so that the model can constantly update the sample data with the changes of the working conditions,so as to update the model parameters and further improve the accuracy of the model.Experimental results show that the performance of the PCA-ARLSSVM model is better than LSSVM model.3.In view of the punishment coefficient and the width coefficient have an importantinfluence on the prediction accuracy and generalization ability of the model,this dissertation proposed a hybrid algorithm of Glowworm Swarm Optimization-Fruit fly Optimization Algorithm(GSO-FOA)searching for the optimal parameters of PCA-ARLSSVM model based on the further study of Glowworm Swarm Optimization(GSO)and Fruit fly Optimization Algorithm(FOA)with their advantages and disadvantages,making the prediction accuracy and generalization ability to achieve the best balance.The experimental results show that GSO-FOA has less optimization time and higher prediction accuracy,compared with GSO.In conclusion,the soft-sensor model based on PCA-ARLSSVM can be effectively applied in the online prediction of inulinase concentration in the fermentation process of Pichia pastoris,providing guidance and reference for the controlling decision and technology improvement in.