Novel Soft Computing Saturation Exponent Modeling in Tight Sand Reservoirs Based on Experimental Data

ABSTRACT The precise evaluation of the saturation exponent is essential for dependable hydrocarbon‑saturation estimates in oil and gas reservoirs. Conventional laboratory techniques, however, are often slow and expensive. This study carried out controlled experiments on a range of tight sandstone cores and then built several advanced machine‐learning (ML) models to predict the saturation exponent directly from nuclear magnetic resonance (NMR) measurements. Porosity, permeability, and NMR T 2lm served as the predictive inputs. Outliers were detected using the leverage method, and a sensitivity assessment clarified how strongly each variable influenced the exponent. Model consistency was confirmed using k ‑fold cross‑validation, and among all evaluated methods, AdaBoost achieved the strongest performance, showing the highest predictive accuracy and the lowest errors relative to the experimental data. The analysis also showed that permeability exerted the greatest influence on the saturation exponent, a conclusion that aligned closely with detailed core analyses and laboratory results.