Scholarlink

<p>Reliable heat extraction from enhanced geothermal systems (EGS) requires accurate characterisation of complex hydraulic fracture networks in low-permeability granite. Existing approaches struggle to reconstruct three-dimensional fracture geometries non-destructively, to enforce thermodynamic consistency, and to transfer laboratory observations to reservoir scale. To address these gaps, we develop an entropy-regularised control-volume physics-informed neural network (cvPINN) framework constrained by acoustic emission (AE) and distributed acoustic sensing (DAS) measurements, and we couple it to a novel multi-fidelity transfer-learning module that upscales laboratory-derived fracture patterns to km-scale discrete fracture networks (cvPINN-TL-DFN). A time-dependent Open Stimulation Fracture (OSF) damage variable governs permeability enhancement and stiffness degradation within a fully coupled thermo-hydro-mechanical (THM) formulation, with thermodynamic admissibility enforced by a hinge-loss penalty on the local entropy production. AE events provide micro-crack locations and timing, while DAS captures dynamic strain and flow-induced vibrations along the borehole; the two modalities are fused as physics-informed observational anchors. The framework is calibrated on a single-stage hydraulic stimulation of a 300&#xa0;mm granite block under true triaxial stress, achieving a reconstruction error below 2% and a connectivity match exceeding 92%. The cvPINN-TL-DFN upscaling is validated against EGS case studies in granitic reservoirs, reproducing microseismic cloud geometry, pressure response, and fracture-orientation statistics within quantified 95% uncertainty bounds. Computational cost is reduced by approximately 25% relative to finite-volume baselines. The framework identifies stress-controlled vertical conduits as the dominant pathways for fluid and heat transport, offering a quantitative basis for optimising permeability enhancement and heat recovery in hot dry rock reservoirs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advanced cvPINN-TL-DFN reconstruction and multi-scale transfer learning of complex fracture networks in granite for enhanced geothermal systems applications

  • 期刊论文

  • 发表日期：

    2026年7月27日

查看全文
  • Mohammed Ali Badjadi,
  • Haiyan Zhu,
  • Peng Zhao,
  • Shouwei Zhou,
  • Twine Usito Bakesigakyi,
  • Muhsan Ehsan

摘要

Reliable heat extraction from enhanced geothermal systems (EGS) requires accurate characterisation of complex hydraulic fracture networks in low-permeability granite. Existing approaches struggle to reconstruct three-dimensional fracture geometries non-destructively, to enforce thermodynamic consistency, and to transfer laboratory observations to reservoir scale. To address these gaps, we develop an entropy-regularised control-volume physics-informed neural network (cvPINN) framework constrained by acoustic emission (AE) and distributed acoustic sensing (DAS) measurements, and we couple it to a novel multi-fidelity transfer-learning module that upscales laboratory-derived fracture patterns to km-scale discrete fracture networks (cvPINN-TL-DFN). A time-dependent Open Stimulation Fracture (OSF) damage variable governs permeability enhancement and stiffness degradation within a fully coupled thermo-hydro-mechanical (THM) formulation, with thermodynamic admissibility enforced by a hinge-loss penalty on the local entropy production. AE events provide micro-crack locations and timing, while DAS captures dynamic strain and flow-induced vibrations along the borehole; the two modalities are fused as physics-informed observational anchors. The framework is calibrated on a single-stage hydraulic stimulation of a 300 mm granite block under true triaxial stress, achieving a reconstruction error below 2% and a connectivity match exceeding 92%. The cvPINN-TL-DFN upscaling is validated against EGS case studies in granitic reservoirs, reproducing microseismic cloud geometry, pressure response, and fracture-orientation statistics within quantified 95% uncertainty bounds. Computational cost is reduced by approximately 25% relative to finite-volume baselines. The framework identifies stress-controlled vertical conduits as the dominant pathways for fluid and heat transport, offering a quantitative basis for optimising permeability enhancement and heat recovery in hot dry rock reservoirs.