A Study on Overhead Contact Line Icing Prediction Method Integrating a Hybrid Physical-Transformer Deep Learning Model

This paper presents a novel approach that integrates physical modeling with Transformer architecture to predict overhead contact line icing. This hybrid method overcomes the limitations of conventional weather models when dealing with complex weather and terrain conditions. By incorporating physical elements such as temperature, humidity, wind speed, and terrain features to establish an initial physical prediction model, and leveraging the Transformer model’s strengths in handling time-series data to capture the nonlinear dynamics of icing, this approach achieves remarkable results. Experimental evidence demonstrates that compared to traditional forecast models, this method delivers higher prediction accuracy across various weather conditions, particularly under extreme weather and complex terrain. Consequently, this study offers a reliable technical solution for disaster prevention, mitigation, and early-warning systems in the transportation sector, holding significant practical value for engineering applications.