Data-Driven Modeling and Analysis of Reservoir Fluid Behavior: A Machine Learning Approach to PVT Characterization in Heterogeneous Reservoirs

Accurate Pressure-Volume-Temperature (PVT) analysis is crucial for understanding reservoir fluid behavior and optimizing hydrocarbon production. This study develops a robust model for PVT analysis to enhance the characterization of reservoir fluids and improve reservoir management. The study employed regression analysis, Decision Tree Regressor, and a comparative Neural Network approach to evaluate relationships between critical parameters such as gas-oil ratio (GOR), reservoir temperature, gas specific gravity, oil gravity, and the oil formation volume factor (OFVF). Findings revealed complex non-linear correlations, with gas specific gravity and bubble point pressure emerging as the most influential predictors. The Decision Tree Regressor achieved high accuracy (R² = 96.44%), while the Neural Network provided comparable performance. An expanded error analysis, numerical example, and discussion on reservoir stabilization are presented. The study underscores the significance of high-quality reservoir fluid sampling and data interpretation, emphasizing representative data to reduce uncertainties in modeling. Comparative tables and enriched references position this work as an original contribution bridging machine learning and petroleum engineering.