A Pantograph-Catenary Arcing Detection Method Based on the Fusion of Current and Electromagnetic Radiation

The pantograph-catenary system is crucial for the power collection of high-speed trains. However, arcs caused by contact loss between the pantograph and catenary could create safety risks for the normal operation of high-speed railways. Accurately detecting whether pantograph-catenary arcing occurs at any moment during train operation is essential for improving maintenance efficiency. Nevertheless, existing methods primarily rely on traditional single-signal detection, which is highly sensitive to interference, leading to significant errors in arc detection. This study proposes a novel pantograph-catenary arc detection method that integrates current and electromagnetic radiation signals. First, we set up a pantograph-catenary arc simulation platform and used low-cost custom sensors to collect arc current and electromagnetic radiation data. Then, time-frequency features of the arc data were extracted by integrating ensemble empirical mode decomposition and the rolling time-window method. The improved Fisher score was applied for feature dimensionality reduction. Comparative analysis of different algorithm combinations demonstrated that the Grey Wolf Optimizer -Support Vector Machine model achieved the best performance. The results indicate that dual- information fusion outperforms single- information detection in terms of arc detection accuracy and F1-score. Moreover, the detection accuracy of the key parameter for percentage of arcing, arc duration, reaches 97.65%. This study combines dual-information fusion, signal processing techniques, and machine learning. It achieves high-precision detection of the pantograph-catenary arcing state at any moment, providing a solution to detection errors in single-information methods.