A Stochastic 3D Pore-Scale Simulation Method to Predict Permeability of Carbonate Rocks from Pore Size Distribution Data

Case ID:

2023-020

BACKGROUND

Accurately predicting the permeability of rocks is critical for developing and managing fossil fuel and water resources. Traditional approaches to permeability prediction involves using deterministic model equations and petrophysical data such as porosity and pore size distribution, gathered from mercury injection capillary pressure measurements (MICP). While these techniques work well with siliciclastic rocks, they typically fail when employed in carbonate rocks. This is due to the complex heterogenous pore microstructures which can vary up to five orders of magnitude, making the traditional deterministic empirical modelling approach inaccurate. With increasing scarcity of fossil fuels and water resources for certain areas, there is a clear need for improvement of permeability estimation tools for carbonate rocks.

SUMMARY OF TECHNOLOGY

Researchers at OSU have developed software for predicting permeability of carbonate rocks through the usage of machine learning algorithms. The invention employs a stochastic pore-scale simulation approach, using porosity and pore size distribution (PSD) of rock samples for input data, easily obtained from MICP. To consider the wide range of possible pore connectivity scenarios, the developed stochastic pore-scale simulation approach involves generating hundreds to thousands of 3D pore microstructures of equivalent PSD and porosity but different stochastic pore connectivity. Permeability is calculated by averaging the permeability distribution obtained from pore-scale flow simulations through the generated 3D pore microstructures, which is more accurate than deterministic model equations also using MICP data. This invention allows for the implementation of a machine learning (ML) approach, reducing the number of pore-scale simulations required. This ML method reduces computation hundreds of times and reproduces permeability estimates from pore-scale flow simulations with a mean absolute percentage error of 10%. Moreover, a pre-trained ML model using this invention can be incorporated with nuclear magnetic resonance PSD information obtained while drilling, providing instantaneous permeability information of penetrated formations.