Filtering Approaches for Inequality Constrained Parameter Estimation

Parameter estimation of a dynamic system is an important task in process systems engineering. The utilization of an augmented system offers the approach of estimating process states and parameters simultaneously. In practice, the parameters often satisfy certain constraints which should be incorporated to improve the estimation performance. This thesis focuses on the inequality constrained parameter estimation problem. We introduce a method of constructing inequality constraints on parameters from routine steady-state operation data. A constraint implementation method with the unscented Kalman filter (UKF) is proposed that yields faster recovery of parameter estimates than the conventional projection method. The appropriate use of projection method with the ensemble Kalman filter (EnKF) is introduced. Also, a constrained estimation method with the EnKF is proposed which results in improved performance compared to the projection method. For the moving horizon estimation (MHE), we propose an alternative approach for constrained parameter estimation, which provides better performance than the directly constrained MHE. The efficacies of the proposed approaches in this thesis are evaluated using several simulated process examples.