Parameter identification in large-scale models for oil and gas production

Models used for model-based (long-term) operations as monitoring, control and optimization of oil and gas reservoirs are often first principles models. They are the result of partial differential equations being discretized, leading to nonlinear models that are largescale in terms of number of states and parameters. Estimating a large number of parameters from measurement data leads to problems of identifiability and consequently to inaccurate identification results. This is problematic since the models are used for model predictions and control strategies. In this paper options are given to deal with the lack of identifiability: approximating the model structure while retaining the physical interpretation of the parameters, redefining the model structure using parameters with e.g. a geological meaning, and thirdly adding additional prior information to the identification problem. These options are illustrated with examples taken from oil and gas reservoir engineering