A learning-based multiscale model for reactive flow in porous media

We study the transport of reactive flow through permeable geological formations, with a focus on advection-dominated transport. As the fluid flows through the permeable medium, it reacts with the medium, changing the morphology, transport, and material properties of the medium; this in turn, affects the flow condition and chemistry. We present a computationally efficient and quantitatively accurate learning-based multiscale framework for reactive transport with volume reactions. We introduce a surrogate of the history-dependent lower-scale behavior that can be used directly at the upper scale and train it using one-time off-line data generated by repeated calculations of the lower-scale problem. The dataset used for training the recurrent neural operator is provided.