Dynamic data reconciliation considering model structure uncertainty

Abstract

There are two types of uncertainty in mathematical representation of process models, namely model structure uncertainty and parameter uncertainty. Uncertainty in parameters was considered in most previous approaches for dynamic data reconciliation. In the present study, an efficient strategy is proposed to solve dynamic data reconciliation containing nonlinear variables and model structure uncertainty. A penalty function is introduced to address the model structure uncertainty. The problem is formulated to include the uncertainty of model structure and solved by simultaneous method. Dynamic data reconciliation was performed using nonlinear programming (NLP) with a multistep ordinary differential equation (ODE) solver. The ability of the proposed strategy is compared with that of other solution strategies: the extended Kalman filter (EKF) and a simultaneous method using high order onestep method. It is found that the proposed method shows good performance in the nonlinear region since it accelerates the computational speed without sacrificing the error reduction ability compared to onestep method. Computational load of this approach decreases to a third of the one-step method.