A new study published in Engineering offers a systematic overview of artificial intelligence-driven subsurface hydraulic fracturing, outlining its technical framework, field implementations, and future directions amid the global energy transition and the expansion of unconventional oil and gas development. The research presents a three-stage technical evolution pathway centered on data-driven analysis, dynamic optimization, and autonomous decision-making, aiming to replace conventional experience-based operations with data-intensive and intelligent workflows.

The study points out that traditional hydraulic fracturing, constrained by linear percolation theory and simplified homogeneous models, struggles to capture strong reservoir heterogeneity, and static designs often fail to adapt to complex subsurface dynamics. To address these limitations, the research proposes a four-layer intelligent architecture covering data perception and integration, intelligent modeling, dynamic control, and autonomous execution, forming a closed-loop system from real-time sensing to automatic operation adjustment. Key applications include intelligent fracture characterization, real-time process warning, post-fracturing production forecasting, and full-lifecycle regulation.

Researchers developed Dy-Fracture-Net, a deep learning model for three-dimensional fracture propagation prediction, which integrates multi-source heterogeneous data and supports efficient dynamic forecasting of fracture geometries. A dual-model collaborative framework is established to identify fracturing events and predict downhole pressures, supporting real-time anomaly warning and operational adjustment. Meanwhile, Dy-Production-Net is designed to jointly forecast post-fracturing reservoir parameter evolution and production performance, supporting long-term development planning. By combining these models with intelligent optimization algorithms, a closed-loop decision system is constructed to regulate pumping parameters and adapt to changing formation conditions.

The paper also identifies current bottlenecks, including insufficient downhole monitoring data, limited model interpretability of purely data-driven approaches, and incomplete autonomous control loops. Future developments are suggested in three areas: miniaturized multi-modal sensing agents, mechanism-data fusion modeling with improved interpretability, and fully autonomous closed-loop control systems. These efforts are expected to support the digital transformation of the oil and gas industry and improve the efficiency and sustainability of unconventional resource development.

The paper “Artificial Intelligence-Driven Subsurface Hydraulic Fracturing Engineering: Connotation and Practices,” is authored by Bin Yuan, Mingze Zhao, Wei Zhang, Siwei Meng, Aoran Jin, Birol Dindoruk. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.12.024. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.