EUVIMEDEUVIMED entdecken
Crossref

Geologically Constrained Machine Learning for Hydraulic Flow Unit Identification in Tight Sandstone Reservoirs: A Case Study from the Shaximiao Formation, Zitong Area

Xiaolong He, Bing Zhang, Chuan Xu, Kai Yang, Yifan He, Zhuo Li, Dongxing Wang, Ken Cheng, Yangsen Gao

SPE Journal · 13.05.2026 · 31 (05) · 3255-3270

DOI: 10.2118/232823-pa ↗

Kurz gefasst

Fachartikel in der Zeitschrift SPE Journal, online erschienen am 06.03.2026. Verfasst von Xiaolong He und 8 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The tight sandstone reservoirs of the first member of the Shaximiao Formation in the Zitong area, northwestern Sichuan Basin, exhibit strong heterogeneity, posing a major challenge to efficient hydrocarbon development. To decipher its genesis and property variations, we integrated core-based analyses from six wells, encompassing detailed mineralogical/textural characterization of more than 120 samples and a comprehensive suite of porosity-permeability measurements yielding 687 data points. The results revealed that reservoir heterogeneity is fundamentally controlled by the interplay of deposition and diagenesis. Sedimentary environments with different hydrodynamic conditions established distinct sandbody compositions. Subsequent diagenetic processes, particularly compaction and variable cementation, drove divergent pore evolution pathways, ultimately leading to the formation of distinct hydraulic flow units (HFUs). Building upon this geological model, we developed a novel machine learning (ML) workflow to achieve continuous well-based prediction of HFUs. The 687 data points, each classified into an HFU based on core-derived geological and petrophysical criteria, served as the training and validation labels. We demonstrated that morphological parameters extracted from the gamma ray (GR) log effectively capture primary sedimentary energy signals. These geologically meaningful attributes were then integrated with conventional log suites to form the input feature set. An extreme gradient boosting (XGBoost) classification model was subsequently trained, achieving a prediction accuracy of 85.4% on the test set and consistently greater than 80% on blind well validation. The principal novelty of this work lies in its successful fusion of geological process understanding with data-driven predictive analytics. This methodology provides a reliable and interpretable workflow that can be adapted to characterize tight sandstone reservoirs with similar geological complexity elsewhere.

Zitieren

APA

He, X., Zhang, B., Xu, C., Yang, K., He, Y., Li, Z., Wang, D., Cheng, K., & Gao, Y. (2026). Geologically Constrained Machine Learning for Hydraulic Flow Unit Identification in Tight Sandstone Reservoirs: A Case Study from the Shaximiao Formation, Zitong Area. SPE Journal, 31(05), 3255-3270. https://doi.org/10.2118/232823-pa

Vancouver

He X, Zhang B, Xu C, Yang K, He Y, Li Z, et al. Geologically Constrained Machine Learning for Hydraulic Flow Unit Identification in Tight Sandstone Reservoirs: A Case Study from the Shaximiao Formation, Zitong Area. SPE Journal. 2026;31(05):3255-3270. doi:10.2118/232823-pa

BibTeX
@article{He2026,
  author = {He, Xiaolong and Zhang, Bing and Xu, Chuan and Yang, Kai and He, Yifan and Li, Zhuo and Wang, Dongxing and Cheng, Ken and Gao, Yangsen},
  title = {Geologically Constrained Machine Learning for Hydraulic Flow Unit Identification in Tight Sandstone Reservoirs: A Case Study from the Shaximiao Formation, Zitong Area},
  journal = {SPE Journal},
  year = {2026},
  volume = {31},
  number = {05},
  pages = {3255--3270},
  doi = {10.2118/232823-pa}
}
Kostenlos starten →