Abstract:

A series arc fault detection method based on an improved lightweight network MobileNet is proposed to adress the unsatisfactory detection performance of series arc faclts with noulinear loads and complex combination loads. According to the asymmetric distribution of magnetic flux, the neutral and phase lines are simultaneously passed through the current transformer to obtain the high-frequency current coupling signals. By simulating a large number of arc fault experiments in the laboratory, various arc fault samples are obtained for 7 types of loads specified in the national standard and 5 types of loads commonly used in household appliances. Data augmentation is used to generate datasets for training and testing. Finally, the arc fault detection network model is mounted on the arc fault protection prototype for online detection testing.The experimental results show that the proposed method can effectively identify series arc faults and is conducive to the productization of arc fault protectors.

Key words: series arc fault, asymmetrical distribution, lightweight network, fault detection