Estimation in Large Mechanistic Models

Abstract

The main objective in this thesis is to model an offshore processing plant, and adapt it to a given dataset by off-line parameter estimation and then perform an on-line state and parameter estimation. The aim is to use the model in model based control to increase production in the process, and it therefore needs to be suited for on-line estimation and have good predictive qualities.The modeling was performed in Dymola based on Modelica models. Selecting appropriate parameters for both the off-line and on-line estimation compose a great deal of the work load in this thesis, as it is an iterative process which require a lot of knowledge about the system. With a suitable set of parameters and measurements, the off-line estimation was be performed by a software tool called ModelFit, based on sequential quadratic programming, while the on-line estimation was be performed by a software toolbox called CENIT based on an extended Kalman filter. Both ModelFit and CENIT are developed by Cybernetica AS. It appears as the modeling and choice of parameters for both the off-line and on-line estimations are reasonable. The off-line and on-line estimation results are satisfactory as there are small deviations between the predicted outputs from the model and the measured outputs from real data. It therefore seems like the model has promising qualifications in terms of being used in model based control. The model is however highly coupled and have many local solutions making it sensitive to initial conditions and non-trivial to estimate.