SVM based soft sensor for antibiotic fermentation process

In this paper, we introduce a SVM based soft sensor method for resultant concentration of antibiotic fermentation. Non-linear black-box models for resultant concentration are established used SVM with poly and Gaussian kernels respectively. Some treatments such as the outliers´ elimination and data smoothness are done before modeling. Simulation results with Matlab toolbox show that the SVM with poly and Gaussian kernels both can do the soft sensor work. Compare to BP neural network, the SVM with Gaussian kernel based soft sensor has better performance in generalization ability and training speed. According to the analysis, the method of SVM based soft sensor for resultant concentration is feasible.