Machine learning modeling of permeability in 3D heterogeneous porous media using a novel stochastic pore-scale simulation approach

Accurate predictions of rock permeability is critical for resource exploration and environmental management. To improve on existing approaches to prediction, this study employed a stochastic pore-scale simulation approach. The petrophysical properties needed the implementation approach are porosity pore size distribution (PSD) samples which can be obtained easily from mercury injection capillary pressure measurements. was tested four carbonate five siliciclastic cores. consider wide range possible connectivity scenarios that associated same PSD porosity, involves generation hundreds 3D microstructures but different connectivity. Permeability calculated by averaging flow simulations through generated microstructures. estimations were closer measured with than deterministic empirical model equations. Machine learning used reduce required number 157 times reproduced estimated mean absolute percentage error 10%.