Research On Soft-Sensor Methods In Erythromycin Fermentation Process Based On LS-SVM

With the development of biotechnology,microbial fermentation engineering is playing more and more important role in the national economy and social life.For the purpose of improving product output,quality and cost reduction,it is a necessity to optimize the fermentation process control.However,due to the complexity of the biochemical process,many important parameters can not be measured in real-time. Therefore,advanced control algorithm and strategy can't be applied efficiently in the fermentation industry.With the fermenting process of erythromycin as the object,this research adopted regress algorithm of Least Squares Support Vector Machine (LS-SVM) to set up the soft sensor models to predict three important parameters in the fermenting process.The detailed work is as follows.Firstly,on the basis of reading literatures and studying the fermentation mechanism of erythromycin,the soft sensor models of LS-SVM are established respectively to predict the parameters in the erythromycin fermenting process,such as mycelium concentration,sugar concentration and production concentration,these models are compared and studied with soft sensor models of Support Vector Machine(SVM).Secondly,analyze the impact of the model parameters to the model accuracy, research the method of grid search with cross-validation and to determine the LS-SVM model parameters.Finally,in order to meet the requirement of variable measurement and optimal control in the fermenting process,design the data acquisition system to collect data of key parameters such as temperature,pH value,dissolved oxygen,and design the erythromycin fermentation process control system structure.