Evaluation of machine learning approach for base and subgrade layer temperature prediction at various depths in the presence of insulation layers

To protect pavement in cold regions, insulation layers have been used in pavement construction. The insulation layer blocks the heat exchange, resulting in different pavement temperature distributions at different layers. Pavements in cold regions are subject to frost heave, thaw weakening, and freeze–thaw cycling, and the moisture content of unbound materials is affected by the temperature of the pavement. Therefore, the base and subgrade temperature can affect the pavement performance, and the insulation layer influences the performance of the pavement. This work aims to demonstrate the capability of the machine learning method to predict the pavement temperature of the base and subgrade layers in the presence of insulation layers. Pavement temperature measured from sensors embedded in the pavement at the IRRF test road and environmental factors collected from a weather station were used to train machine learning models. The results obtained from machine learning models indicated that the air temperature and day of the year were the most robust input prediction parameters at each depth. Machine learning models outperformed the existing model and proved that it could be a method to improve pavement temperature prediction with insulation layers, especially when the base layer temperature is below 0°C.