Automated Integrated Workflow for Enhanced Reservoir Characterization and History Matching Using Machine Learning and Ember in Complex Reservoir Systems

The increasing complexity of modern reservoirs has emphasized the need for advanced, integrated, and automated workflows that efficiently manage geological and dynamic uncertainties. This study presents a comprehensive methodology that couples static and dynamic reservoir modeling within a single simulation environment, using machine learning and Python scripting to enhance automation, accuracy, and speed. The proposed workflow includes automated static property modeling using the EMBER algorithm and dynamic model calibration through a machine learning-assisted history matching process. Over 20 equilibrium regions, commingled production, and two reservoir fluids were modeled under limited data conditions, using probabilistic scenarios, experimental design, and geostatistical inputs to reduce uncertainty and improve model fidelity. The results demonstrated that the integrated workflow not only accelerated the generation of simulation ensembles but also significantly improved model predictability and calibration quality. SHAP-based sensitivity analysis enabled the identification of high-impact variables, and visual tools such as tornado and worm plots enhanced interpretability. Compared to traditional approaches, this method reduced computational time and human error, resulting in a more reliable basis for field development planning. The novelty of this approach lies in the seamless integration of machine learning, probabilistic modeling, and workflow automation in a unified platform, offering a replicable strategy for efficient and scalable reservoir management in highly uncertain environments.