Enhancing Fractional Flow Curve Modeling with Advanced Data-driven Techniques: A Comparative Evaluation of Machine Learning Frameworks

Modeling fractional flow curves accurately is essential for optimizing reservoir performance and improving hydrocarbon recovery. This study introduces a robust analytical framework utilizing advanced computational techniques to predict fractional flow behavior. The model leverages Gradient Boosted Decision Trees (GBDT) and integrates key physical parameters such as water saturation, viscosity ratios, and relative permeability. The performance of the proposed framework was evaluated using data from reservoir simulations and experiments. The model demonstrated high predictive accuracy, achieving a Root Mean Square Error (RMSE) of 0.005, a Coefficient of Determination (R<sup>2</sup>) of 0.99, and a Mean Absolute Percentage Error (MAPE) of 1%. Compared to conventional fractional flow models based on Buckley-Leverett theory, which yielded an RMSE of 0.16 and a MAPE of 12.8%, the new approach showed significant improvement. Additionally, it outperformed other computational approaches, including Random Forest (RMSE: 0.02, MAPE: 10.4%) and Artificial Neural Networks (RMSE: 0.016, MAPE: 6.0%), providing both enhanced accuracy and consistency. A sensitivity analysis confirmed the robustness of the model across a range of viscosity ratios, showing strong alignment with physical principles, such as shock front behavior and saturation constraints. The practical utility of this model lies in its ability to accurately predict fractional flow under varying conditions, bridging gaps between analytical methods and data-driven techniques, while remaining computationally efficient. This development enhances the tools available for reservoir engineers, offering new insights for waterflooding strategies, enhanced oil recovery (EOR), and other multi-phase flow applications, with direct relevance to field operations.