Abstract
This study compares the feasibility of using artificial neural networks (ANN) and multi-linear regression (MLR) for predicting an hourly temperature of the pavement, considering the depth, time, and air temperature as independent variables. Accurate prediction of the pavement temperature is critical for road maintenance, pavement design, and near-surface microclimate environment to overcome the problems caused by the fluctuations in temperature such as rutting in higher temperature and thermal cracking in lower temperature. A dataset containing 7200 measurements of the pavement temperature was used. Thermal instruments were used to measure the asphalt pavement temperature every two hours with different variables during the four seasons of the year in an attempt to model pavement temperature by utilising MLR and ANN. The target used for prediction was modelling the pavement temperature profile. In autumn, the regression square (R-2) value predicted by the ANN model is 0.95, while the R-2 value predicted by the MLR model is 0.92 as the most significant value in autumn and the lowest value for MLR and ANN analysis is 0.83 and 0.85, respectively in the winter. Results show that the temperature predictions made by ANN are more accurate than those made by MLR. Nonetheless, these models are recommended to be used in the same region since both models presented in this study showed high correlation coefficients.