Robust hybrid tree-based machine learning-assisted optimization of well parameters to reduce asphaltene precipitation risk in oil fields

Abstract Asphaltene precipitation is a persistent flow-assurance issue in carbonate oil wells, leading to increased intervention frequency and production losses. This study utilizes multi-decade surveillance and operational data from a mature carbonate field (1983–2023) to train and optimize five tree-based models: Decision Tree, Random Forest, Extra Trees, Gradient-Boosting Decision Tree, and CatBoost, employing a Tree-Structured Parzen Estimator. These models, constrained by operational parameters, were integrated to identify optimal settings that can minimize the frequency of asphaltene precipitation and the subsequent cleanups. The novelty of this research lies in the direct integration of interpretable tree ensembles with an optimizer that adheres to operational limitations, transforming historical field behavior into practical, field-ready, well-parameterized guidance. Key risk factors considered include production rate, gas-oil ratio, trajectory type (vertical versus deviated), producing layer, surface position, and completion type of the wells. The results indicate that the most effective strategy for reducing asphaltene risk is to implement moderate production rates and utilize lower-skin configurations (e.g., open-hole (OH); vertical in Asmari), in line with reservoir-engineering principles aimed at reducing near-wellbore drawdown. This study offers realistic, field-applicable guidelines to mitigate or significantly reduce the risk of asphaltene in future wells within this field or other mature carbonate reservoirs.