EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Rapid Prediction of Fracture Surface Morphology and Identification of Key Controlling Parameters in Compressed Air Energy Storage Underground Caverns Using Tabular In-Context Learning

Yuan Xu, Zulkifl Ahmed, Chen Ding, Kaixi Xue, Fahad Alshawmar

Energies · 20.11.2026 · 19 (22) · 5624

DOI: 10.3390/en19225624 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Energies, online erschienen am 20.11.2026. Verfasst von Yuan Xu und 4 weiteren Autor:innen. Die Arbeit nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Basic Scientific Research Project of the Education Department of Liaoning Province, Doctoral Scientific Research Foundation of Liaoning Institute of Science and Technology, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Underground caverns have emerged as a promising solution for large-scale energy storage owing to their substantial storage capacity and economic advantages. During the operation of compressed air energy storage (CAES) systems, fluctuations in internal pressure can induce and propagate fractures in the surrounding rock mass, potentially reducing its load-bearing capacity and compromising cavern stability. To investigate fracture development under complex in situ stress conditions, this study develops an integrated framework combining the discrete element method (DEM) and Tabular In-Context Learning (TabICL). A two-dimensional grain-based discrete element model (DEM2D-GBM) is employed to construct a numerical dataset incorporating eight governing parameters: in situ stress ratio (k), cohesion (c), internal friction angle (ϕ), cavern depth (H), cavern diameter (D), Geological Strength Index (GSI), maximum internal pressure (Pmax), and stiffness ratio (Ei/Em). Fracture surface morphology is characterized using two indicators: fracture initiation angle (α) and fracture surface inclination angle (β). The results demonstrate that the TabICL model achieves testing R2 of 0.98704 for α and 0.984 for β. For fracture initiation angle prediction, the model obtains a testing mean absolute error (MAE) of 0.06100 and a root mean square error (RMSE) of 0.08400. SHapley Additive exPlanations (SHAP) analysis identifies the in situ stress ratio as the most influential parameter, with mean absolute SHAP values of 16.27 for α and 21.41 for β. Consistently, Sobol global sensitivity analysis shows total-order sensitivity indices of 0.731 and 0.748 for α and β, respectively, confirming the dominant contribution of the in situ stress ratio. Maximum internal pressure represents the second most influential parameter in the Sobol analysis, with total-order sensitivity indices of 0.284 for α and 0.298 for β. Additionally, SHAP analysis reveals that cavern diameter and maximum internal pressure contribute substantially to fracture initiation angle, whereas internal friction angle plays a more important role in determining fracture surface inclination angle. These findings demonstrate the potential of the integrated TabICL framework for efficient prediction of fracture morphology and identification of influential geological and operational parameters in CAES underground caverns.

Förderung

  • Basic Scientific Research Project of the Education Department of Liaoning Province
  • Doctoral Scientific Research Foundation of Liaoning Institute of Science and Technology
  • National Natural Science Foundation of China
  • Jiangxi Provincial Research Center for Digital Risk Management of Underground Engineering Open Foundation

Zitieren

APA

Xu, Y., Ahmed, Z., Ding, C., Xue, K., & Alshawmar, F. (2026). Rapid Prediction of Fracture Surface Morphology and Identification of Key Controlling Parameters in Compressed Air Energy Storage Underground Caverns Using Tabular In-Context Learning. Energies, 19(22), 5624. https://doi.org/10.3390/en19225624

Vancouver

Xu Y, Ahmed Z, Ding C, Xue K, Alshawmar F. Rapid Prediction of Fracture Surface Morphology and Identification of Key Controlling Parameters in Compressed Air Energy Storage Underground Caverns Using Tabular In-Context Learning. Energies. 2026;19(22):5624. doi:10.3390/en19225624

BibTeX
@article{Xu2026,
  author = {Xu, Yuan and Ahmed, Zulkifl and Ding, Chen and Xue, Kaixi and Alshawmar, Fahad},
  title = {Rapid Prediction of Fracture Surface Morphology and Identification of Key Controlling Parameters in Compressed Air Energy Storage Underground Caverns Using Tabular In-Context Learning},
  journal = {Energies},
  year = {2026},
  volume = {19},
  number = {22},
  pages = {5624},
  doi = {10.3390/en19225624}
}
Kostenlos starten →