Evaluation and development of a predictive model for the analysis of geophysical well log data in reservoir characterization

Abstract

This study addresses the growing demand for efficient and accurate petrophysical interpre-tation in the oil and gas industry through the integration of artificial intelligence into reservoir characterization workflows. Traditional deterministic and empirical methods, while valuable, often prove time-consuming and limited in their ability to handle geological heterogeneity. To vercome these challenges, we developed an intelligent system designed to predict and visualize key reservoir properties directly from well log data. The methodological pipeline encompassed data reprocessing, feature engineering, supervised model training, and interpretability anal-ysis. Ensemble learning algorithms demonstrated strong predictive performance: Gradient Boosting for shale volume (Vsh) achieved R2 = 0.914, MAE = 0.0642; XGBoost for porosity (PHIE) yielded R2 = 0.801, MAE = 0.0248; and Extra Trees for water saturation (Sw) reached R2 = 0.895, MAE = 0.0127. These results confirm the ability of AI models to capture non-linear relationships inherent in petrophysical data and to outperform conventional approaches. The system was implemented as a user-riendly desktop application, enabling geoscientists and engi-neers to obtain rapid, reliable insights for reservoir evaluation. Overall, this work demonstrates the feasibility and impact of AI-powered solutions in advancing reservoir interpretation, sup-porting informed decision-making, and opening perspectives for future real-time and scalable deployment.