A Robust Machine Learning Model to Predict the Relative Permeability of an Oil Porous Medium at Elevated Temperatures

Abstract Thermal Enhanced Oil Recovery (TEOR) such as SAGD, CSS and other steam injection processes have been employed in heavy oil reservoirs of North-American and Middle-East countries for oil recovery. Elevation of temperature during this process leads to wettability alteration, IFT variation, viscosity reduction, asphaltene and resin precipitation. These variations during TEOR impact relative permeability to each fluid phase in the reservoirs. Therefore, available models like the Corey model and Stone's model for estimating the relative permeability cannot be directly used for reservoir simulation/modelling study of such reservoir where TEOR is implemented. Hence, an attempt has been made to develop a reliable, accurate, and robust data-driven model for two-phase oil/water relative permeability using the XG-Boost machine learning algorithm which accounts for the temperature's effect. For this study, numerous sets of oil and relative permeability data have been sourced, compiled and validated using our proposed model via the supervised XG-Boost approach. For model construction, 1270 oil relative permeability and 1230 water relative permeability data points were obtained from literature covering different rock/fluid and reservoir conditions. This study presents a new data-driven model developed using the XG-Boost algorithm to predict two-phase oil/water relative permeability over a wide range of temperatures in unconsolidated sand and sandstone formations. Moreover, the proposed model gave us better results based on the statistical error analysis.