Fault Detection for Covered Conductors With High-Frequency Voltage Signals: From Local Patterns to Global Features

摘要: The detection and characterization of partial discharge (PD) are crucial for the insulation diagnosis overhead lines with covered conductors. With release a large dataset containing thousands naturally obtained high-frequency voltage signals, data-driven analysis fault-related PD patterns on an unprecedented scale becomes viable. high diversity background noise interferences motivates us to design innovative pulse shape method based clustering techniques, which can dynamically identify set representative PD-related pulses. Capitalizing those pulses as referential patterns, we construct insightful features develop novel machine learning model superior performance early-stage conductor faults. presented outperforms winning in Kaggle competition provides state-of-the-art solution detect real-time disturbances field.