To address traffic safety hazards from asphalt pavement icing in Xinjiang’s cold regions and inefficiencies of conventional deicing and imprecise geothermal deicing systems, this study focused on local asphalt surfaces. Using “outdoor qualitative screening and indoor quantitative verification”, key variables were identified via controlled tests and their coupling effects on the time to complete icing were quantified through an L16(44) orthogonal test (a 4-factor, 4-level design encompassing 16 test groups). A Backpropagation (BP) neural network model (3 inputs, 5 hidden neurons, and a learning rate of 0.7) optimized with 64 datasets was established to predict the time to complete icing of asphalt pavements, achieving a prediction accuracy (PA) of 90.7% for the time to complete icing and a mean error of merely 0.71 min. Dynamic icing risk thresholds (high/medium/low) were established via K-means clustering and statistical tests, enabling data-driven precise activation and on-demand regulation of geothermal deicing systems. This resolves energy waste and deicing delays, offering technical support for efficient geothermal utilization in cold-region transportation infrastructure, and provides a scalable “factor screening + model prediction” framework for asphalt pavement anti-icing practice.
Mo et al. (Wed,) studied this question.