Integrated AutoML-based framework for optimizing shale gas production: A case study of the Fuling shale gas field

This study introduces a comprehensive and automated framework that leverages data-driven methodologies to address various challenges in shale gas development and production. Specifically, it harnesses the power of Automated Machine Learning (AutoML) to construct an ensemble model to predict the estimated ultimate recovery (EUR) of shale gas wells. To demystify the “black-box” nature of the ensemble model, KernelSHAP, a kernel-based approach to compute Shapley values, is utilized for elucidating the influential factors that affect shale gas production at both global and local scales. Furthermore, a bi-objective optimization algorithm named NSGA-Ⅱ is seamlessly incorporated to optimize hydraulic fracturing designs for production boost and cost control. This innovative framework addresses critical limitations often encountered in applying machine learning (ML) to shale gas production: the challenge of achieving sufficient model accuracy with limited samples, the multidisciplinary expertise required for developing robust ML models, and the need for interpretability in “black-box” models. Validation with field data from the Fuling shale gas field in the Sichuan Basin substantiates the framework's efficacy in enhancing the precision and applicability of data-driven techniques. The test accuracy of the ensemble ML model reached 83 % compared to a maximum of 72 % of single ML models. The contribution of each geological and engineering factor to the overall production was quantitatively evaluated. Fracturing design optimization raised EUR by 7 %–34 % under different production and cost tradeoff scenarios. The results empower domain experts to conduct more precise and objective data-driven analyses and optimizations for shale gas production with minimal expertise in data science.