A Digital Twin for Reservoir Simulation

Abstract

Abstract We have developed a physical room-scale porous media flow rig for operating, measuring, and visualizing reservoir flows in real time – the FluidFlower. The flow rig scale is large enough to achieve true multiphase flow effects (including phase mixture, gravity segregation and geological heterogeneities), while small enough to work on weekly time-scales, and allow for repeatable experiments. Mirroring the FluidFlower, we have constructed a prototype of a digital twin for porous media flow – the PoroTwin. Essentially, we demonstrate that it is possible to achieve real-time transmissions of laboratory data from the FluidFlower to a cloud-based simulation- and machine learning environment, and complete the loop with applying optimal control algorithms to steer the experiment. As part of the proof of concept, we also demonstrate that the machine learning environment can identify, and learn to correct for, incomplete physical descriptions within a reservoir simulator. The PoroTwin thus shows the potential of a fully integrated experimental and automated learning environment.