Icing Prediction Model of Asphalt Pavement in Cold Regions Based on an Improved BP Neural Network

To better understand the factors influencing asphalt pavement icing and its impact trends, a research approach combining indoor experiments, field meteorological factor investigations, and statistical analysis was adopted. An indoor simulation test platform for asphalt pavement icing was constructed to replicate icing under various natural conditions. The icing time and thickness were observed to quantitatively analyze the influence of different factors on icing. The findings indicate that particle size distribution has a more significant impact on icing thickness than on icing duration. The primary contributing factors to icing are water accumulation depth, ambient temperature, wind speed, water temperature, and radiation intensity. Using the results of the indoor road icing simulation as the data basis, a backpropagation (BP) neural network algorithm was employed to build a regression prediction model for the icing duration and thickness of asphalt road surfaces. The model comparison results showed that the BP neural network prediction model for icing time and thickness exhibited superior fitting accuracy, reaching 99.1% and 91.2%, respectively. Utilizing field meteorological data, a classification prediction model for asphalt road surface state was established using the support vector machine algorithm and optimized with the Bayesian optimization algorithm. The results indicated that the support vector machine classification algorithm based on Bayesian optimization could effectively classify and predict the state of the asphalt road surface, with an accuracy of 96.2% and an area under the curve of 0.99.