Abstract: On the basis of the standard SVM for regression, the dynamic ε-SVM method was proposed to establish precise mathematical models to describe the behavior of biochemistry systems, namely each training sample used different error. At the same time, an improved multi-objective Genetic Algorithm (MOGA) was used to automatically select the dynamic ε-SVM parameters. A new modeling method that combined improved MOGA with dynamic ε-SVM regression was presented. The model for titer pre-estimate was developed in Matlab6.5 with data collected from real plant. The model possessed the strong capability of fitting and generalization. It is shown that the method achieves significant improvement in the generalization performance in comparison with the modeling method based on MOGA and the standard SVM.