Soft sensor modeling using RVM and PCA in fermentation process

With massive data of a fermentation process, a SVM-based soft sensor modeling method suffers from heavy burden calculation and complicated parameter setting. A novel soft sensor using Relevance Vector Machine(RVM) based on Principal component analysis(PCA) algorithm is proposed. Firstly, feature extraction and dimensionality reduction of sample data are achieved by PCA. Secondly, a RVM is used to construct soft sensor models. Compared with SVM, RVM doesn´t need penalty factor parameter, constrains the weight coefficient using hyper parameter and leads to sparser model with better generalization ability. The proposed modeling method is used to construct a novel soft sensor model for an erythromycin fermentation process. Case studies show that the approach has better performance compared to the conventional SVM model.