A Study on Catenary Icing Prediction Method Integrating Physical Modeling and Transformer-Based Deep Learning

This paper proposes a novel method for predicting icing on overhead contact lines by integrating physical modeling with Transformer-based deep learning, addressing the limitations of traditional meteorological models in complex weather conditions and terrains. The method combines physical factors such as meteorological data (e.g., temperature, humidity, wind speed) and topographic features to construct a physical model for initial predictions, while leveraging the Transformer model's robust capability in processing time-series data to capture the nonlinear dynamics of the icing process. Experimental results demonstrate that the proposed method significantly outperforms traditional single meteorological models in prediction accuracy across various weather conditions, particularly excelling in extreme weather and complex terrain scenarios. This approach provides reliable technical support for disaster prevention, mitigation, and early warning systems in the transportation sector, offering substantial practical value for engineering applications.