Abstract
Despite recent advancements in computational methods, it is still challenging to properly model fracture properties, such as relative permeability and hydraulic aperture, at the field scale. The challenge is in determining the most representative fracture properties, concluded from multi-scale data. In this study, we demonstrate how to capture fracture properties at the field scale from core-scale and pore-scale data through multi-scale uncertainty quantification, and assess how pore-scale processes can significantly impact the recovery factor. There are three components within our workflow: 1) performing high-resolution Navier-Stokes (NS) simulation at pore-scale to obtain hydraulic aperture of discrete single fractures, 2) embedding pore-scale parameters into core-scale for predicting field-scale objective, such as recovery factor, and 3) performing Monte Carlo simulations to determine the relationship effect of the pore-scale parameters to the field scale responding. At pore-scale, we start with four parameters that characterize the fractures: mean aperture, relative roughness, tortuosity, and the ratio of minimum to mean apertures. We then construct hydraulic aperture surrogates using an Artificial Neural Network (ANN). At the field scale, we deploy Long Short-Term Memory (LSTM) to capture the recovery factor at field-scale. The final results are the time-varying recovery factor and its sensitivity analysis. Monte Carlo simulation is performed on the final surrogate to produce the recovery factor value for various time-step. The result is beneficial for risk assessment and decision-making during the development of fractured reservoirs. Our method is the first to quantitatively estimate multi-scale parameters’ effect on recovery factors in two-phase flow in fractured media. This method also shows how we accommodate and deal with multi-scale parameters.
