Research On Soft Sensor Modeling Method Of Marine Lysozyme Fermentation Process

As a high salt-based hydrolyzed protein,Marine lysoenzymes have better metabolic ability than terrestrial lysoenzymes.It has the characteristics of non-toxic,no side effects and strong antiviral sterilization ability.At present,it has been successfully applied to medical biology,food industry and animal husbandry and other fields.In the fermentation process of marine lysozyme,some key parameters that directly reflect the quality of the fermentation process(such as cell concentration,matrix concentration,and relative enzyme activity)are difficult to be measured online.In actual fermentation,most of these key parameters are measured by manual sampling and offline measurement,such as dry weight method,direct staining method,optical density method and cell counting method,etc.These methods are complex operation,long lag,large measurement error,easy to introduce human pollution,and measurement precision affected by cell death,measurement error and other factors,cannot timely reflect the current state of the fermentation process,difficult to meet the requirements of Marine lysozyme real-time dynamic regulation.According to the above problems,this dissertation proposed the soft measurement of Marine lysozyme fermentation process based on DICA-FNN and the migration of it under different fermentation conditions.The specific research contents are as follows:Firstly,the properties of different enzyme producing strains were analyzed and compared,then Marine Bacillus S-12-86 was selected as the fermentation strain.The changes of each stage of marine lysozyme fermentation process were described.The process flow of marine lysozyme fermentation and the treatment methods of each stage of fermentation process were introduced.The environmental impact parameters of fermentation process were studied and analyzed,and the optimal range of fermentation parameters was determinedSecondly,aiming at the problem that the key parameters of marine lysozyme fermentation process are difficult to be detected online,a soft sensing modeling method based on DICA-FNN is proposed.Based on the basic principles and parameter learning method of FNN,the imperialist competitive algorithm(ICA)is used to solve the problems of slow convergence speed and easy to fall into local optimization during the optimization process of FNN network structure parameters.The dynamic grouping strategy and adaptive offset assimilation are used to solve the ICA global search ability weak and local optimal problems.The model was applied to use in the online prediction of key parameters of marine lysoase fermentation process,the experimental results show that the improved soft measurement model prediction accuracy improved by more than 3%.Finally,aiming at the problem of model invalidation in the fermentation process of marine lysozyme with multiple condition(take different initial parameter as example),an transfer soft sensor model based on MMK-MMD is proposed.The Mahalanobis distance is used to calculate the predicted output distance of FNN normalization layer in source domain and target domain,then the MK-MMD measure criterion is used to project the data samples of the predicted output distance of the source domain and the target domain into RKHS,the unbiased estimation of samples in RKHS is used to reduce the spatial distribution distance of samples,which reduces the probability of edge distribution between the target domain and the source domain,so as to re select and adjust the connection weights of soft sensor model,then the model parameters were transfered.The transfer soft sensor model is applied to the prediction of key parameters of marine lysozyme fermentation process with multiple condition.The experimental results show that the training time of the transfer soft sensor model is reduced by about 35%,and the prediction accuracy is improved by more than 2%.