Water saturation determination is among the most challenging tasks in petrophysical well-logging, which directly impacts the decision-making process in hydrocarbon exploration and production. Low-field nuclear magnetic resonance (LF-NMR) measurements can provide reliable evaluation. However, quantification of oil and water volumes is problematic when their NMR signals are not distinct. To overcome this, we developed two machine learning frameworks for predicting relative water content in oil-sand samples using LF-NMR spin-spin (T2) relaxation and bulk density data to derive a model based on Extreme Gradient Boosting. The first one facilitates feature engineering based on empirical knowledge from the T2 relaxation distribution analysis domain and mutual information feature extraction technique, while the second model considers whole samples' NMR T2-relaxation distribution. The NMR T2 distributions were obtained for 82 Canadian oil-sands samples at ambient and reservoir temperatures (164 data points). The true water content was determined by Dean-Stark extraction. The statistical scores confirm the strong generalization ability of the feature engineering LF-NMR model in predicting relative water content by Dean-Stark-root-mean-square error of 0.67% and mean-absolute error of 0.53% (R2 = 0.90). Results indicate that this approach can be extended for the improved in-situ water saturation evaluation by LF-NMR and bulk density measurements.
© 2022. The Author(s).