Adaptive soft sensor for online prediction based on enhanced moving window GPR

Abstract

Process nonlinearity and time-varying behavior of industrial systems are the main factors for poor performance of online soft sensors. To ensure high predictive accuracy, adaptive soft sensor is a common practice. In this paper, an adaptive soft sensor based on moving window Gaussian process regression (GPR) is presented. To make the moving window strategy more efficient, a just-in-time learning (JITL) algorithm is used to enhance the performance, which avoids the selection of a window size that original moving window approaches have to select . The effectiveness of the proposed method is demonstrated by an example concerning the H 2 S concentrations of tail gas in the sulfur recovery unit (SRU). Compared with other soft sensor methods, the proposed JITL based moving window GPR has higher accuracy.