An LSTM-Transformer-Based Surrogate Model for Predicting Pantograph-Catenary Interaction in Railway Systems

As the primary energy collection mechanism for electric trains, the pantograph-catenary interface is crucial for operational safety and dynamic stability. The quality of their interaction directly influences power supply reliability and train performance. In previous studies, the evaluation of pantograph–catenary interaction has primarily relied on numerical simulation methods, which are often computationally intensive and time-consuming. To overcome this challenge, we introduce a surrogate modeling approach leveraging deep neural networks to efficiently evaluate pantograph-catenary interaction dynamics. The model enables fast and accurate mapping from system design parameters to dynamic response characteristics. The core architecture of the predictive model integrates both LSTM and Transformer networks, effectively capturing both global and local features from the training data. Results demonstrate that the predicted contact forces from the proposed surrogate model are in close agreement with those obtained from numerical simulations, confirming the model’s accuracy. Comparative experiments with several conventional neural network architectures further highlight the superiority of the proposed model design.