Dynamic data reconciliation using process simulation software and model identification tools

Existing data reconciliation strategies for nonlinear and dynamic systems require solution of the process model expressed as a differential and algebraic equation system. Although it is difficult to model a complex process by customized equations, the availability of dynamic simulation software packages, such as HYSYS, makes process modeling and simulation easier. Because of interfacing problems between these software tools and nonlinear programming algorithms and computational inefficiency, dynamic data reconciliation with these software tools has not been feasible. In this dissertation two new approaches developed using model identification tools and HYSYS dynamic simulation software are presented. These methods are based on an analogy to the previously developed Nonlinear Dynamic Data Reconciliation method based on a finite difference approximation and time series analysis for the process model. A simplified local model can be employed to incorporate the simulation software with a nonlinear programming algorithm to obtain an algorithm that is computationally more efficient. Several techniques can be applied to identify the input-output model including regression model, parametric time series models, time series adaptive regression splines and artificial neural networks. Proposed methods are compared with each other and with existing data processing techniques to validate these approaches. Finally, the effect of covariance matrix estimation on data reconciliation is investigated and it was shown by an example that covariance update during data reconciliation provides better estimates.