Optimal design of soft sensors and bias updating scheme based on rank-constrained optimization

Abstract

Soft sensors have been widely applied in many different industrial fields to predict the values of quality variables, which cannot be measured online. However, it is likely that most of processes are affected greatly by time-varying changes. Thus, the bias updating mechanism is frequently introduced into the maintenance of soft sensors in industrial processed. However, the soft sensors models are developed in a static sense, and it is questionable that their performance is optimal under bias update. To address this issue, we propose an optimal design of soft sensors and bias updating scheme based on rank-constrained optimization. To efficiently solve the optimization problem, an algorithm based on the difference-of-convex programming is proposed. Compared with classical static least squares equipped with bias update, the new approach turns out to more accurate and robust, which is demonstrated by a simulation study.