Machine learning-based prediction of oil–water relative permeability using core flooding and CT-scan data
Taha Zarin, Hossein Ahmadi Eshkaftaki, Amin Sharifi
Abstract
Relative permeability (Kr) is a crucial parameter in reservoir engineering, influencing fluid flow dynamics and recovery efficiency. It quantifies the effective permeability of in-situ fluids, playing a vital role in enhanced oil recovery (EOR) assessments. Traditionally, Kr curves data are derived from core flooding experiments, which are time-intensive, costly, and require specialized laboratory setups. Due to financial and logistical constraints, acquiring extensive Kr data remains a significant challenge. To overcome these limitations, various models and empirical formulas have been developed to estimate Kr, though they introduce uncertainties. This study presents a machine learning (ML)-based approach for predicting oil-water relative permeability (Kro-Krw) curves using a dataset comprising 105 Kr curves generated from diverse reservoir conditions. The dataset includes results from unsteady-state core flooding tests that measure Kr across a range of water saturations. By aggregating 105 Kr curves from different wells, the study compiles over 6,450 data points. A novel ML framework integrating regression trees, neural networks, and support vector machines is introduced to enhance permeability prediction accuracy. The proposed methodology significantly outperforms traditional models, offering a reliable, efficient, and cost-effective alternative for reservoir engineers. Beyond permeability prediction, this research highlights the transformative potential of artificial intelligence (AI) in petrophysical analysis, paving the way for broader ML applications in the oil and gas industry.