Key points are not available for this paper at this time.
Accurate estimation of borehole thermal resistance ( R b ) is essential for the efficient design of ground heat exchanger systems. In coil-type energy piles, densely packed pipe arrangements induce significant thermal interference, which traditional 1D or 2D models fail to capture adequately. This study aims to develop a reliable machine learning (ML)-based model for predicting R b in coil-type energy piles by incorporating 3D thermal interaction effects. R b values were first estimated using in-situ thermal performance test data reported in previous studies for coil-type energy piles with different coil pitches (200 mm and 500 mm). Then, A 3D numerical heat transfer model was developed and validated using COMSOL Multiphysics, demonstrating strong agreement with experimental measurements with errors less than 5.26 %. Based on the validated model, 875 simulation cases were conducted to construct a parametric database, varying coil-shaped pipe and pile configurations. Three ML algorithms were then applied to develop predictive models for R b . While XGBoost achieved the highest test-phase accuracy (MAPE = 0.41 %), LightGBM exhibited the best generalization performance on unseen data (MAPE = 0.0497 %). Consequently, the LightGBM-based model was identified as the most effective, offering a strong balance between predictive accuracy and robustness. • 3D heat transfer model was developed to capture thermal interference in coil-type energy piles. • Machine learning models were trained using 875 simulation database for borehole thermal resistance. • LightGBM identified as the optimal model considering training accuracy and generalization performance.
Park et al. (Sun,) studied this question.