Data-driven prediction model for the heat performance of energy tunnels

To date, it has been challenging to quickly and accurately quantify the heat performance of energy tunnels under unknown conditions. This study innovatively introduces an intelligent prediction model for the thermal performance of energy tunnels to address the above difficulties. Five machine learning (ML) prediction models for energy tunnel heat flux were established based on a database sourced from various regions, various conditions, and various operations. The prediction results were compared with the measured heat flux of the energy tunnel to determine the prediction performance of these ML models, and the sensitivity of the input variables was also analysed. The results indicate that the established database has reliable representation, as the selected variables (features) in this database are independent and relatively random. Furthermore, these ML models can accurately capture the trends of energy tunnel heat flux values under unknown conditions, with the random forest model demonstrating the best prediction performance, generalization ability, and great accuracy among these five ML models. These 14 input variables in the database are categorized into three groups according to the sensitivity analysis: thermal variables, design variables, and other variables (environmental and test variables). These findings provide confidence for the intelligent prediction of energy tunnel heat performance.