Leveraging ML for On-Site Bubble Point Pressure Prediction in Sudanese Crude Oil: A Resilient Approach During Conflict

This study proposes a novel machine learning (ML)-driven framework for on-site Bubble Point Pressure (BBP) prediction tailored to Sudanese crude oil. Given the current disruption of traditional laboratory-based Pressure-Volume-Temperature (PVT) analysis due to the conflict in Sudan, there is an urgent need for reliable, decentralized solutions within the energy sector. Empirical correlations, though widely used in this context, often provide inaccurate predictions for BBP due to the unique properties of Sudanese crude. In this work, we introduce and evaluatemultiple ML models, with a focus on optimizing real-time BBP predictions. Among the evaluatedmodels, the Multi-Layer Perceptron (MLP) and XGBoost demonstrated superior predictive accuracy, with MLP achieving the most significant improvements over traditional empirical methods. This advancement directly addresses the limitations of existing models by offering enhanced precision and adaptability. The contribution of this study lies in the development of a field-deployable, ML-powered solution that enables real-time BBP predictions without the need for centralized laboratory infrastructure. This approach not only ensures continuity of oil analysis & production during the conflict but also provides a robust foundation for post-conflict recovery, supporting long-term operational and economic resilience in the Sudanese oil sector.