Development of a Soft Sensor using Subspace Identification

The Kalman filter is a robust knowledge-based model to deal with random measurement noises in industrial processes. In order to reduce the influence of measurement noise, a data-driven method, i.e. expectation-maximization algorithm based subspace identification (EM SID) soft sensor is presented in this paper. Firstly, system matrices are estimated by using subspace identification (SID). Secondly, the expectation-maximization (EM) algorithm is used for calibrating such matrices. Applications to a simulated numerical process and the Tennessee Eastman (TE) process validate the performance of the proposed method.