Application of machine learning techniques to predict the temperature distribution in semi-rigid pavement with a cement-treated base

Due to the visco-elastic properties of asphalt concrete (AC), its strength is reduced exponentially along with the increased temperatures. Including cement-treated base (CTB) will improve the whole pavement structure strength in general but potentially lead to warmer asphalt surface and lower AC layer stiffness. This study applied machine learning (ML) models to precisely predict the temperature distribution inside the AC layer laid over the CTB at various depths: 2 cm, 5 cm, 7 cm, 10 cm, and 13 cm from the AC top surface. Thermal sensors were installed at such depths for around one year to collect temperature data, which was then combined with air temperature and solar radiation data from the local environmental monitoring station to develop temperature prediction models. The Ensembles of trees were selected as the best model with RMSE = 0.16 and R-squared = 0.97 from various ML models in the MATLAB Regression Learner App. The developed Ensembles of trees model has provided a higher prediction performance than other BELLS models and can be adapted for AC temperature prediction in tropical regions.