An integrated physics–machine learning framework for time-dependent injectivity decline and fracture growth in horizontal water-injection wells
Results in Engineering · 2026-06 · 30 · 110211
DOI: 10.1016/j.rineng.2026.110211 ↗Kurz gefasst
Fachartikel in der Zeitschrift Results in Engineering, 2026-06. Verfasst von Ranvijay Singh und 1 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
A hybrid physics-machine learning framework is presented to predict time-dependent injectivity decline and fracture half-length evolution in horizontal water injection wells subjected to particulate-laden injection. The framework couples mechanistic formation damage and filtration models with machine learning surrogates trained on over 30000 CMG-IMEX reservoir simulations, enabling rapid, physically consistent evaluation of fracture-initiation risk and injection performance across a wide range of reservoir and operating conditions. The framework integrates physics-based filtration models for near-wellbore and near-fracture damage covering internal pore plugging and external filter-cake buildup with machine learning (ML) surrogates (CatBoost, LightGBM, and XGBoost) trained on 30000 high-fidelity CMG-IMEX simulations. Surrogates predict bottom-hole pressure, fracture-tip pressure, and flow partitioning for static well-fracture states. A sequential operator-split time-marching scheme couples damage evolution with quasi-static fracture growth governed by a minimum horizontal stress admissibility criterion. Simulation scenarios are organised into throughput-based injection systems to ensure physically interpretable sensitivity analysis across constrained, moderate, and high-injectivity reservoir-well configurations. Three representative throughput systems (B1-B3) spanning constrained to high injectivity were analysed under base particulate concentration (20 ppmv). Base injection rates were designed to suppress fracture initiation under clean-water conditions, ensuring all observed fracture growth was attributable solely to particulate-induced pressure buildup. Fracture initiation and progressive half-length growth were observed in all systems, with final fracture half-lengths of 14, 40 and 86 ft for B1, B2 and B3 at 1000 days, respectively. Bottom-hole pressure and fracture-tip pressure rose sharply at early time due to near-wellbore and fracture-face damage before stabilising slightly above the minimum horizontal stress. Well skin increased monotonically, with final values of 51, 76, and 148 across the three systems. Sensitivity analysis identified minimum horizontal stress gradient, reservoir thickness, horizontal permeability, water viscosity, injected solids concentration, and injection duration as the dominant controlling parameters across all throughput systems. The framework reproduces coupled injectivity-fracture behaviour at computational costs orders of magnitude lower than fully implicit numerical simulators, enabling practical uncertainty analysis and fracture-risk assessment. The throughput-system design methodology provides a rigorous and operationally meaningful basis for injection-well optimisation and water-quality specification. This framework is among the first to couple physics-based filtration damage models for horizontal wells and fractures with ML surrogates trained on 30000 CMG-IMEX simulations for rapid, time-resolved prediction of injectivity decline and fracture growth. The throughput-system scenario design methodology enables physically interpretable uncertainty analysis, offering practising engineers a computationally efficient tool for injection-well optimisation, water-quality specification, fracture-risk assessment, and evaluation of remedial treatment timing.
Zitieren
APA
Singh, R., & Suri, A. (2026). An integrated physics–machine learning framework for time-dependent injectivity decline and fracture growth in horizontal water-injection wells. Results in Engineering, 30, 110211. https://doi.org/10.1016/j.rineng.2026.110211
Vancouver
Singh R, Suri A. An integrated physics–machine learning framework for time-dependent injectivity decline and fracture growth in horizontal water-injection wells. Results in Engineering. 2026;30:110211. doi:10.1016/j.rineng.2026.110211
BibTeX
@article{Singh2026,
author = {Singh, Ranvijay and Suri, Ajay},
title = {An integrated physics–machine learning framework for time-dependent injectivity decline and fracture growth in horizontal water-injection wells},
journal = {Results in Engineering},
year = {2026},
volume = {30},
pages = {110211},
doi = {10.1016/j.rineng.2026.110211}
}
EUVIMED