Geological parameterisation of petroleum reservoir models for improved uncertainty quantification
Arnold, Daniel Peter
MetadataShow full item record
As uncertainty can never be removed from reservoir forecasts, the accurate quantification of uncertainty is the only appropriate method to make reservoir predictions. Bayes’ Theorem defines a framework by which the uncertainty in a reservoir can be ascertained by updating prior definitions of uncertainty with the mismatch between our simulation models and the measured production data. In the simplest version of the Bayesian methodology we assume that a realistic representation our field exists as a particular combination of model parameters from a set of uniform prior ranges. All models are believed to be initially equally likely, but are updated to new values of uncertainty based on the misfit between the historical and production data. Furthermore, most effort in reservoir uncertainty quantification and automated history matching has been applied to non-geological model parameters, preferring to leave the geological aspects of the reservoir static. While such an approach is the easiest to apply, the reality is that the majority of the reservoir uncertainty is sourced from the geological aspects of the reservoir, therefore geological parameters should be included in the prior and those priors should be conditioned to include the full amount of geological knowledge so as to remove combinations that are not possible in nature. This thesis develops methods of geological parameterisation to capture geological features and assess the impact of geologically derived non-uniform prior definitions and the choice of modelling method/interpretation on the quantification of uncertainty. A number of case studies are developed, using synthetic models and a real field data set, that show the inclusion of geological prior data reduces the amount of quantified uncertainty and improves the performance of sampling. The framework allows the inclusion of any data type, to reflect the variety of geological information sources. ii Errors in the interpretation of the geology and/or the choice of an appropriate modelling method have an impact on the quantified uncertainty. In the cases developed in this thesis all models were able to produce good history matches, but the differences in the models lead to differences in the amount of quantified uncertainty. The result is that each quantification would lead to different development decisions and that the a combination of several models may be required when a single modelling approach cannot be defined. The overall conclusion to the work is that geological prior data should be used in uncertainty quantification to reduce the uncertainty in forecasts by preventing bias from non-realistic models.