Coupling EUR prediction with fracturing optimization: An integrated machine learning framework for shale gas development

As conventional hydrocarbon resources continue to deplete, shale gas has become a key driver of the global energy transition due to its production potential and economic viability. However, significant reservoir heterogeneity and the complex evolution of fracture networks introduce uncertainties in single-well production forecasts, making hydraulic fracturing designs heavily dependent on empirical judgment. To address these challenges, this study proposes a data-driven, multi-objective integrated evaluation framework that links feature selection, EUR prediction, and optimal fracturing scheme generation. The framework employs a binary-encoded genetic algorithm (GA) for feature selection, balancing linear and nonlinear dependencies among geological and engineering variables. A heterogeneous ensemble of models, including CatBoost, NGBoost, and TabPFN, is fused using a two-level stacking strategy, significantly improving EUR prediction accuracy. The framework optimizes decision variables, such as fluid and proppant volumes, and uses NSGA-II to solve the bi-objective problem of maximizing EUR while minimizing fluid-proppant consumption, yielding Pareto-optimal designs. Validation on 231 shale gas wells in the Sichuan Basin demonstrates a 10–30% improvement in , a reduction in , a 40–270% increase in EUR, and a 10–20% reduction in fracturing costs for medium-to-low-yield wells. SHAP analysis identifies key factors such as FSL, TIRLD, and HRT as strongly, nonlinearly, and positively correlated with EUR, offering valuable insights for precise production enhancement. The framework shows robustness and transferability, providing essential decision support for shale gas development across diverse geological settings. • Integrated Framework for Shale Gas Development: We propose a novel, data-driven framework that integrates EUR prediction and fracturing optimization for shale gas development, improving decision-making efficiency and cost-effectiveness. • Machine Learning-Enhanced EUR Prediction: We combine feature selection via genetic algorithms with ensemble machine learning models (CatBoost, NGBoost, and TabPFN) to enhance the accuracy of EUR production forecasts, achieving 10-30% improvements over traditional methods. • Multi-Objective Optimization of Fracturing Design: We utilize NSGA-II to optimize fracturing parameters (fluid and proppant volumes) by balancing EUR maximization and cost reduction, demonstrating up to a 270% increase in EUR and a 10-20% reduction in fracturing costs. • Cross-Regional Validation: We validate the framework’s effectiveness across different shale gas datasets, including datasets from the U.S., confirming its robustness and transferability to diverse geological conditions.