Soft Sensor Development in Fermentation Processes Using Recursive Gaussian Mixture Regression Based on Model Performance Assessment

Abstract

The soft sensor modeling method based on moving window method (MW) and Gaussian mixture regression (GMR) has been developed in fermentation processes. However, the MW method always results in low computational efficiency because of frequency updating. GMR has the multi-model structure. In order to reduce calibration frequency of recursive GMR modeling methods, a recursive GMR soft sensor based on model performance assessment (MPA) is developed. According to the results of the model performance assessment, the model updating is selectively activated. The developed model was investigated to estimate biomass concentration in an industrial Erythromycin fermentation process. Compared with the GMR model, the prediction accuracy is improved obviously.