Data-Driven Proxy Modeling of Water Front Propagation in Porous Media

Abstract

In the water flooding process, determining the location of the injected water front is as one of the most critical variables, which is the basis of many subsequent predictions. Despite the importance and use of this parameter in a vast range of flooding-related assessments, there are no alternative methods to traditional analytical modeling or time-consuming numerical 3D simulation for its determination. This study introduces a data-driven proxy modeling approach based on two powerful deep learning algorithms for real-time determination of the injected water front location on the grid scale. The developed proxy models have realized the possibility of modeling the location of the flow front by minimally using the data extracted from the numerical simulators and only relying on commonly available field data. The proposed proxy models successfully simulated the breakthrough time in production wells and water arrival time in certain reservoir grids in new blind scenarios.