On Synchro-Waveform Data Analytics for High Impedance Fault Identification in Distribution Networks

It is a challenging task to detect High Impedance Faults (HIF) in distribution networks due to its small, random, and nonlinear characteristics. Utilizing the synchro-waveform data, this paper proposes a data-driven method named time and frequency convolutional network (TFCNet) for HIF detection in distribution networks. The proposed method consists of two learning streams. The first stream extracts the temporal features from the original waveform data using convolutional operations, while the second stream extracts the frequency-domain features. The two type features are fused and fed into the classifier, which can distinguish the event category amongst normal condition, HIF, and transient event. The proposed method is verified based on the IEEE 33-bus system and it outperforms several baseline methods.