Modeling Based on MOGA and the Dynamic ε -SVM for Fermentation Process

To found the suitable models to describe the behavior of biochemistry systems, the dynamic epsiv-SVM method was proposed on the basis of SVM. Each training sample uses different error. The existed methods for selecting the parameters of SVM not only consume time, but also are difficult to find the optimal parameters. The optimal parameters were automatically decided by using multi-object genetic algorithm (MOGA). A new modeling method that combined MOGA with the dynamic epsiv-SVM was presented. The model for penicillin titer pre-estimate was developed by it in Matlab 6.5 with data collected from real plant. The model possesses the strong capability of fitting and generalization. Experiments show that the dynamic epsiv-SVM is superior to the standard SVM modeling method. MOGA is very feasible and efficient too