Semi-Supervised Process Data Regression and Application Based on Latent Factor Analysis Model

This article investigates the problem of modeling regression with labeled and unlabeled data samples commonly found in industrial processes. By incorporating additional information on unlabeled data samples, a new semi-supervised latent factor analysis model is developed. Compared with the purely supervised regression models which can only use the information of labeled dataset, the semi-supervised model can efficiently extract useful information for improvement of the regression performance. Furthermore, the proposed basic semi-supervised model has been extended to a mixture form, which is capable of describing data from more complex processes. For predictive modeling for key/mass variables, two soft sensors are constructed based on the semi-supervised models. Then, three case studies are used to evaluate the performance of the proposed soft sensing methods.