Model optimization of SVM for a fermentation soft sensor

Support Vector Machine (SVM) is a novel machine learning method of soft sensor modeling in fermentation process, which has the ability to approximate nonlinear process with arbitrary accuracy. Learning results and generalization ability are key performance indicators of a soft sensor model. Parameters settings and input variable selection are crucial for SVM learning results and generalization ability. In this paper, input variable selection and parameter setting are regarded as a combinatorial optimization problem, and a combinatorial optimal objective function is constructed based on the Akaike Information Criterion (AIC). Genetic simulated annealing algorithm (GSAA) is used to search the an optimal model with the function extremum. Simulations show that the proposed soft sensor modeling method based on SVM has good performance in fermentation process.