Prediction of thermal conduction of energy piles using various machine learning models

Abstract

ABSTRACT The global energy crisis has led to a shift towards sustainable energy sources, such as geothermal energy, which can be efficiently used in energy piles, walls, and tunnels. This paper compares six machine learning algorithms for predicting the thermal performance of energy piles. A random 80:20 split of collected data points from the literature was used for training and testing. The input parameters were the number of tubes, inner and outer tube diameters, tube thermal conductivity, pile diameter, pile thermal conductivity, and soil thermal conductivity. The output was energy pile thermal conductance. Among all the ML models used for training and testing, the Gaussian Process Regression (GPR) model outperforms with the highest R2 value of 0.994 and RMSE of 1.886. After sensitivity analysis, the thermal conductivity of the pile was the key factor reducing energy pile thermal resistivity, followed by the number of U-tubes and pipe diameter.