A segmented surrogate model for THMC responses in the near-wellbore zone of super-long gravity heat pipes

Abstract

The super-long gravity heat pipe, as an emerging technology for deep geothermal exploitation, has attracted considerable attention owing to its "pumpless" heat extraction feature. The heat extraction process involves complex coupled thermal-hydraulic-mechanical-chemical (THMC) Multiphysics processes in the near-wellbore region. Constrained by the extremely large length-to-diameter ratio of the heat pipe and the need for detailed characterization of near-wellbore fractures, conventional simulation methods face inherent trade-offs between accuracy and efficiency. To address this challenge, this study proposes a segmented THMC coupled modeling strategy that captures the dynamic evolution of fracture aperture under coupled thermal, mechanical, and chemical effects. A deep neural network (DNN) surrogate model is further developed based on high-fidelity numerical simulations to accelerate calculation. The proposed method achieves high accuracy with an overall MAE of 2.786 W/m and R² of 0.998 against numerical benchmarks. The DNN surrogate reduces computation time from 847 min to 98 min, representing an 88% reduction. Field validation against the Xiong'an 4200 m gravity heat pipe test yields an average relative error of 6.11%, confirming the model's practical applicability. By synergistically integrating segmented modeling, THMC coupling with dynamic fracture aperture evolution, and DNN-based acceleration, this work provides a high-fidelity and computationally efficient solution for refined numerical simulations in deep geothermal exploitation.