Volume 59

Geological Stratification Effects on the Prediction Accuracy of Neural Networks for Energy-Pile Heat Exchange Weiyi Zhang, Haibin Yang, Junwei Liu, Hongzhi Cui, Jinyue Yan

Abstract

Energy piles integrate building foundations with ground heat exchangers to harvest shallow geothermal energy, providing low-carbon heating and cooling. Subsurface geological conditions—particularly vertically stratified soil layers with contrasting thermal properties—can influence the heat-exchange performance of piles; yet their impact on the accuracy of neural-network predictions remains underexplored. This study quantifies how geological layering affects predictive performance by employing a hybrid spatio-temporal neural network that captures both spatial and temporal characteristics of a pile’s thermal field. The model maintains high predictive accuracy even in layered soils. These findings indicate that, in the absence of groundwater flow or other perturbations, geological layering alone exerts a limited effect on neural-network-based prediction accuracy of energy-pile performance.

Keywords Geothermal, energy pile, stratified soil, heat-exchange prediction