A Framework and Modeling Method of Data-Driven Soft Sensors Based on Semisupervised Gaussian Regression

Abstract

Soft sensors have been widely used in industrial processes to predict uneasily measured important process variables. The core of data-driven soft sensors is to construct a soft sensor model by using recorded process data. This paper analyzes the geometry and characteristics of soft sensor modeling data and explains that soft sensor modeling is essentially semisupervised regression rather than widely used supervised regression. A framework of data-driven soft sensor modeling based on semisupervised regression is introduced so that information on all recorded data, including both labeled data and unlabeled data, is involved in the soft sensor modeling. A soft sensor modeling method based on a semisupervised Gaussian process regression is then proposed and applied to the estimation of total Kjeldahl nitrogen in a wastewater treatment process. Experimental results show that the proposed method is a promising method for soft sensor modeling.