Learning pressure flow dynamic is of primary importance towards reliable predictive modelling for hydrocarbon reservoirs. For instance, in gas condensate systems, productivity impairment due to reservoir pressure drop causes condensation buildup around the wellbore. In this study, we present a data-driven based approach to a gas reservoir high-dimensional pressure field to achieve a low-dimensional representation that enables the visualization of spatiotemporal modes of the pressure field. Our proposed approach uses a Dynamic Mode Decomposition (DMD) approach, applied to pressure field data of a gas reservoir single-phase flow model. We demonstrated that the method is capable of decomposing the reservoir pressure field into spatiotemporal modes, and reconstructing the original pressure field from the decomposed modes with a minimal Root Mean Squared Error of 0.0622. The proposed method also allows for the approximation of the reservoir average pressure drop over time. Comparison of results was made between the proposed approach (DMD) and one of the widely used dimensionality reduction techniques namely the Principal Component Analysis (PCA). Our proposed approach outperformed PCA in both pressure field reconstruction and average pressure drop. Since pressure drop has a significant effect on well productivity, the proposed approach can be used as a fast and accurate tool for prediction and estimation of reserves for natural gas reservoirs.
Key words: Data-driven modelling, Dynamic mode decomposition, Natural gas reservoirs, Reduced order modelling.
|
