The accuracy of SVM in fermentation process is mainly impacted by two factors input variable selection and parameter setting in SVM training procedures. In this paper, a novel method is proposed to solve the problem． The selection problem of SVM parameters and input variables is considered as a compound optimization problem. A new compound optimal objective function based on Akaike information criterion is constructed. In this paper, we propose a new method of soft sensor constructed with generalized support vector machine for microbiological fermentation. Experiment results demonstrate this method is an effective approach for parameter selection and input variable selection and has good performance for soft sensor modeling in microorganism fermentation process.