Machine Learning Assisted History Matching Application in the North Kuwait Jurassic

History Matching (HM) is a critical process in reservoir simulation and management, aiming to calibrate reservoir models to historical production data to predict future performance and guide development decisions. However, traditional HM processes are often time-consuming due to the complexity of reservoir systems and rely on manual trial-and-error approaches. In North Kuwait’s Jurassic Fields, an annual manual HM process across 15 reservoir models has been applied historically which has been time-intensive with limited range of uncertainties. The increasing complexity of reservoirs, characterized by a large number of parameters and complicated parameter-response relationships, necessitates innovative approaches. To address these challenges, this work introduces an integrated framework, a novel approach, combining Assisted History Matching (AHM) and Machine Learning (ML) techniques to achieve efficient and reliable history matching, enhance probabilistic forecasting, and improve well-placement strategies. By leveraging the power of ML, we address the challenges associated with large-scale, complex reservoir models and high uncertainty. Our proposed methodology automates the parameter calibration process, reducing the reliance on human intervention and significantly accelerating the HM workflow. This approach ensures the preservation of geological complexity while considering uncertainties in subsurface parameters. By coupling AHM with ML, we not only expedite the history-matching process, achieving results up to 5–10 times faster depending on model size and complexity, but also deliver ensemble-based history-matching outputs that enable probabilistic forecasts. This facilitates the identification of infill drilling locations based on robust probabilistic assessments, improving decision-making under uncertainty. Two case studies from the North Kuwait Jurassic Field demonstrates the effectiveness of our approach. Machine learning-assisted history matching was applied to a Jurassic carbonate reservoir with complex heterogeneities and high subsurface uncertainties. By employing advanced ML algorithms, the workflow efficiently calibrated the reservoir simulation model to historical production data, providing a better match to observed field performance compared to conventional methods. The resulting probabilistic forecast models captured the full spectrum of uncertainties, enabling more reliable predictions of future reservoir behavior [4]. Furthermore, the outputs of ML-assisted history matching can seamlessly be integrated into an automated machine learning framework [3] for well-placement optimization. This automated process identifies optimal locations for infill wells, considering both geological and operational constraints. The integration of probabilistic forecasts ensures that well-placement decisions are aligned with the full range of possible reservoir scenarios, reducing the risk of suboptimal field development.