Establishment of icing prediction model of asphalt pavement based on support vector regression algorithm and Bayesian optimization

To improve the icing prediction accuracy of asphalt pavement, a prediction method for asphalt pavement icing based on the support vector regression (SVR) algorithm is proposed in this study. A prediction model was established using the SVR method to predict the icing time and thickness of the pavement on bridge at the low solar radiation (rainy), and analyze the effects of the external natural environment (ambient temperature, wind speed and water depth) on the freezing time and icing thickness. The Bayesian optimization algorithm (BOA) was also used to automatically adjust the parameters of the prediction model, which fully considered the coupling correlation of the influencing factors of the asphalt pavement icing. Finally, the prediction performances of the BOA-SVR models with different kernel functions were compared. The results show that the training-set prediction accuracy of the icing time and thickness reaches 99.2% and 92.9%, respectively, and the testing-set prediction accuracy of the icing time and thickness reach 97.7% and 84.4%, respectively. Therefore, the BOA-SVR model has high prediction accuracy. The water depth has the greatest influence on the icing time and thickness of the asphalt pavement, followed by the ambient temperature and wind speed. Overall, the BOA-SVR model can predict the icing time and thickness of the asphalt pavement more precisely compared to existing methods.