Machine Learning-Driven History Matching: A Scalable Approach for North Kuwait Jurassic Fields

Summary The paper addresses the challenges of history matching (HM) in the North Kuwait Jurassic (NKJ) reservoirs, characterized by their complex, dual-porosity carbonate formations and high uncertainty. Traditional HM methods are inefficient and subjective, failing to capture nonlinearities in such reservoirs. To overcome these limitations, the study introduces a Machine Learning-Assisted History Matching (ML-AHM) workflow, integrating algorithms like XGBoost and Bayesian Optimization within ensemble-assisted HM. This approach combines rigorous data preprocessing, feature engineering (e.g., PCA), and ML surrogates to accelerate model calibration and reduce reliance on computationally expensive simulations. The ML-AHM framework was validated across multiple NKJ fields, achieving significant improvements in accuracy (50–85% RMSE reduction) and computational efficiency (5–10× faster). It also enables probabilistic forecasting, enhancing risk assessment and decision-making under uncertainty. Challenges include data quality, computational costs, and model interpretability, though techniques like SHAP are proposed to address these. Overall, the ML-AHM approach offers a scalable, efficient solution for managing complex, multi-field assets.