A framework for integrating diagnostic knowledge with nonlinear optimization for data reconciliation and parameter estimation in dynamic systems

Abstract

Dynamic data reconciliation and parameter estimation are challenging problems for large, nonlinear process systems due to problem size and complexity, and the effects of nonlinearities. Recently, an elegant nonlinear optimization formulation has been proposed in the literature. In this work, we extend the nonlinear reconciliation problem to include the detection of the biased parameters. The central idea in this framework is the recognition that the biased parameter identification problem can be viewed as a diagnostic problem, and methods from fault diagnosis literature may be brought in to improve the performance. Once the biased parameter is identified, then the estimation of the bias is performed using nonlinear optimization methods. Using several case studies, this framework is shown to both, detect and produce acceptable estimates of the biased parameters. Since, the bias detection and estimation are decoupled, this framework is shown to provide faster and more accurate estimates for real-time applications. (C) 2001 . .