Machine learning-based Asphaltene Precipitation prediction: Case study- HMD field -

Abstract

Asphaltene precipitation is a major flow assurance challenge in the Hassi Messaoud (HMD) field, Algeria, where crude oils containing small amount of asphaltenes can cause severe wellbore plugging. This study develops a machine learning-based predictive framework for asphaltene content using SARA fractionation and PVT data collected from several wells. six supervised algorithms were trained and evaluated via Leave-One-Out Cross-Validation (LOOCV). Gradient Boosting achieved the best performance (R² = 0.969, MSE = 0.000584, MAE = 0.01347), outperforming Decision Tree, Random Forest, XGBoost, Linear Regression, and SVR. Feature importance analysis consistently identified the Asphaltene-to-Resin (A/R) ratio as the dominant predictor across all tree-based models, confirming the primacy of colloidal stability in governing asphaltene behavior. The proposed framework offers a reliable, data-driven complement to conventional thermodynamic models for asphaltene management in Algerian oilfields.