Machine learning of dual porosity model closures from discrete fracture simulations

Fine-scale discrete fracture simulations provide a natural means to quantify the matrix-fracture fluxes and specify reference solutions for upscaling approaches such as dual porosity/dual permeability models. Since typically fine-scale are computationally demanding, fractured reservoirs highly heterogeneous, it is desirable parametrize geometry obtain coarse-scale model closures using precomputed results. We show that this can be done case of two-dimensional geometries compressible single-phase flows. Specifically, set parameters linked grid block mapped underlying via convolutional neural network. In particular, if transfer function parametrized with number spatially varying on coarse scale, shape per learned from simulations.