Evaluation of Reservoir Inter-Well Connectivity Using Machine Learning Techniques: A Case Study of the Algerian Mesdar Oil Field

Abstract

Artificial intelligence (AI) has become an increasingly important tool across multiple industries, with machine learning (ML) in particular offering new opportunities for optimizing oilfield operations. In petroleum reservoir management, traditional approaches to evaluating inter-well connectivity rely heavily on physical and numerical simulations, which can be time-consuming, computationally intensive, and constrained by model assumptions. In contrast, machine learning techniques establish data-driven correlations through iterative training, enabling more accurate and efficient characterization of reservoir connectivity.

    This study applies machine learning to evaluate inter-well connectivity in the Mesdar Field, Algeria. Two ensemble learning algorithms—Extreme Gradient Boosting (XGBoost) and Random Forest—were implemented due to their demonstrated robustness in handling complex, sequential datasets. Model interpretability was incorporated through agnostic interpretation methods, with training performed using a comprehensive reservoir database. Empirical results from field data reveal that both algorithms perform strongly in predicting production rates, achieving accuracy levels exceeding 97%. Moreover, the connectivity patterns inferred through “Permutation Feature Importance” and “Feature Importance” align closely with those obtained using conventional static methods. These findings suggest that ML-based approaches not only enhance predictive reliability but also provide a cost-effective and scalable alternative for estimating inter-well connectivity in petroleum reservoirs. The integration of such techniques has the potential to improve reservoir management strategies, reduce operational risks, and contribute to more efficient hydrocarbon recovery.