Nonlinear dynamic data reconciliation: In-depth case study

Data reconciliation is a well-known method in on-line process control engineering aimed at estimating the true values of corrupted measurements under constraints. Early nonlinear dynamic data reconciliation (NDDR) studies considered models that were simple and of low order. In such cases the ability to run the NDDR algorithm in real time for relatively slow processes is not a serious problem, despite the heavy computational burden imposed by NDDR. In this study a much more difficult process was treated and the method presented by Laylabadi and Taylor [1], [2] was explored, refined and extended to increase efficiency (reduce computation time). In addition, a new hybrid NDDR method is proposed and a demonstrative example performed to show the promise of this approach in reducing the computational burden and handling industrial processes for which a realistic dynamic model does not exist. This contribution makes NDDR more feasible for a wider variety of applications.