Development of a soft sensor for processes with multiple operating regimes using adaptive multi-state partial least squares regression

This paper proposes an adaptive multi-state partial least squares (MSPLS) algorithm for multivariate chemical processes over a wide range of operating conditions. In the proposed algorithm, the state variable with the maximum variation is first selected from the defined key variables. The system is then divided into several states according to the rank of this state variable. The deviation is subtracted from the process variables in each state, resulting in a set of unified process variables that are then combined to form the PLS model. Finally, an adaptive scheme is designed to generalize the performance of MSPLS. Applications to a continuous stirred tank reactor and a real industrial process are used to evaluate the proposed algorithm. (C) 2016 Taiwan Institute of Chemical Engineers. Published by Elsevier B.V. All rights reserved.