EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Machine-learning prediction of flow partitioning, bottom-hole pressure, and fracture-tip pressure in horizontal water injectors with a single transverse fracture

Ranvijay Singh, Ajay Suri

Energy Geoscience · 2026-12 · 7 (6) · 100602

DOI: 10.1016/j.engeos.2026.100602 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Energy Geoscience, 2026-12. Verfasst von Ranvijay Singh und 1 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 57 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Horizontal water injectors are critical to sustaining reservoir pressure and maximising recovery, yet accurate prediction of injectivity and fracture behaviour remains challenging. Conventional analytical models depend on oversimplified assumptions, while fully coupled numerical simulations incur prohibitive computational cost. This study develops three supervised machine-learning regression models to predict the key parameters governing injectivity and fracture propagation in horizontal water injectors with a single transverse fracture under a constant-pressure boundary scenario: bottom-hole pressure, fracture-tip pressure, and the percentage of injection rate entering the fracture. Pressure targets are expressed as well-to-boundary pressure drop (delpres) and fracture-tip-to-boundary pressure drop (delptip), from which absolute pressures are readily obtained. A comprehensive dataset of 30,000 high-fidelity numerical simulations was generated with a constant-pressure lateral boundary and no-flow top and bottom boundaries, capturing near-wellbore and fracture-tip dynamics across diverse reservoir, fracture, and operational conditions. Key input features include reservoir thickness, porosity, permeability anisotropy (permeabilities in x, y and z directions), rock and water compressibilities, water viscosity, horizontal wellbore radius and length, fracture half-length and width, fracture conductivity, skin factor, and near-fracture damage. Eleven algorithms were rigorously trained and evaluated on held-out test sets - CatBoost, LightGBM, XGBoost, Random Forest, Gradient Boosting, Support Vector Regression, Linear, Ridge, Lasso, Elastic Net, and K-Nearest Neighbours using MAE, RMSE, and R2 metrics. CatBoost achieved best performance across all three targets, attaining MAE of approximately 30 psi and R2 of 99%, substantially outperforming all competing algorithms. LightGBM and XGBoost ranked closely behind, while linear and kernel-based methods showed markedly inferior accuracy, underscoring necessity of ensemble gradient-boosting architectures for this non-linear, high-dimensional problem. Global sensitivity analysis identified permeability anisotropy, fracture half-length, fracture conductivity, and near-fracture damage as the most influential parameters. Although trained on static well-fracture configurations, the surrogates can be embedded within time-stepping workflows that evolve skin and fracture geometry, enabling real-time system response simulation in field-scale injection operations at computational costs orders of magnitude lower than fully coupled numerical simulation. This work is among the first to develop ML surrogates trained on 30,000 high-fidelity simulations of a horizontal well with a single transverse fracture under a constant-pressure boundary condition to simultaneously predict bottom-hole pressure, fracture-tip pressure, and flow partitioning between wellbore and fracture. The combined surrogate set provides practising engineers with a computationally efficient, deployable toolkit for injectivity assessment, fracture-risk evaluation, water-quality specification, and coupling with dynamic skin and fracture-growth models - capabilities unavailable in existing analytical or standalone simulation approaches.

Zitieren

APA

Singh, R., & Suri, A. (2026). Machine-learning prediction of flow partitioning, bottom-hole pressure, and fracture-tip pressure in horizontal water injectors with a single transverse fracture. Energy Geoscience, 7(6), 100602. https://doi.org/10.1016/j.engeos.2026.100602

Vancouver

Singh R, Suri A. Machine-learning prediction of flow partitioning, bottom-hole pressure, and fracture-tip pressure in horizontal water injectors with a single transverse fracture. Energy Geoscience. 2026;7(6):100602. doi:10.1016/j.engeos.2026.100602

BibTeX
@article{Singh2026,
  author = {Singh, Ranvijay and Suri, Ajay},
  title = {Machine-learning prediction of flow partitioning, bottom-hole pressure, and fracture-tip pressure in horizontal water injectors with a single transverse fracture},
  journal = {Energy Geoscience},
  year = {2026},
  volume = {7},
  number = {6},
  pages = {100602},
  doi = {10.1016/j.engeos.2026.100602}
}
Kostenlos starten →