Data-driven modeling and explanatory analysis of ice-galloping in railway catenary systems considering multidimensional structural parameters

As a key power supply facility for railways, the operation state of the catenary directly determines the quality of current collection between the pantograph and the catenary. Exposed to harsh outdoor environments, the catenary is prone to severe icing, which will induce large-amplitude and low-frequency ice galloping under the coupling of structural parameters and environmental factors, seriously endangering the safety of high-speed railways. Traditional nonlinear finite element methods suffer from low computational efficiency, while conventional data-driven methods lack interpretability and cannot characterize strong multi-parameter coupling. To address these shortcomings, this study proposes an interpretable prediction framework for catenary ice galloping. Firstly, a high-fidelity numerical model is established based on the Absolute Nodal Coordinate Formulation (ANCF). Secondly, a Transformer-LSTM hybrid model is developed to accurately capture the nonlinear mapping relationship between structural parameters and galloping responses. Finally, the LIME method is adopted to quantify the key influence mechanisms. The results show that the proposed model achieves an R 2 of 0.94569 in the test, which is significantly superior to CNN and LSTM models. The unequal-span optimization scheme effectively reduces the galloping amplitude, providing a scientific basis for accurate prevention and control.