Modeling of Fermentation Process Based on MOACO and Epsilon-SVM

To establish suitable models to describe the behavior of biochemistry systems, a new modeling method was introduced, combining multiple objective ant colony optimization(MOACO) with the dynamic Epsilon-SVM. The hyper-parameters of Epsilon-SVM were automatically decided by using multiple objective ant colony optimization(MOACO). Each training sample used different error. The model for penicillin production´s prediction was developed using the method with data collected from real plant in Matlab7.0. The model possesses strong capability of fitting and generalization. Experiments also show that the dynamic Epsilon-SVM is superior to the standard SVM modeling method. MOACO is very feasible and efficient too.