CONTROL-RELEVANT DYNAMIC DATA RECONCILIATION AND PARAMETER-ESTIMATION

Chemical process plants typically have a large number of measured variables that contain some degree of error. Accurate knowledge of the process states and parameters can be obtained through the use of data reconciliation techniques. The objective of this paper is to show the importance of dynamic data reconciliation for better closed-loop performance. We propose a successively linearized horizon-based estimation strategy for the estimation of the states and parameters. In addition, we also propose a two-level strategy for the estimation of process inputs and outputs that are corrupted by measurement errors. The estimation of process inputs is decoupled from the estimation of process outputs and parameters in the two-level strategy, resulting in a significant reduction in the computation time. Open-loop and closed-loop simulation studies are performed on a continuous stirred tank reactor to demonstrate the effectiveness of the proposed horizon-based estimation strategy. Satisfactory results are obtained using this strategy for state and parameter estimates at open-loop stable and unstable operating points, and in a region exhibiting limit cycle behavior. We demonstrate the performance of a nonlinear predictive controller using the estimates obtained by the successively linearized horizon approach. The control system performance using this approach is virtually identical to a nonlinear programming-based approach that uses the full nonlinear model.