Dynamic soft sensor modeling based on multi-criterion method and Gaussian process regression

A dynamic soft sensor modeling method based on multi-criterion method and Gaussian process regression is presented to overcome the problems of low prediction accuracy and poor robustness in static soft sensors. The multi-criterion method, as a theoretical basis of determing model order, takes into account several traditional criterions. The application of the proposed soft sensor to an erythromycin fermentation process is presented. Results show that the proposed dynamic soft sensor has high prediction accuracy and small predicted confidence intervals. © 2015, Southeast University. All right reserved.