Physics Regulated Neural Network for High Impedance Faults Detection

High impedance faults (HIFs) in distribution grids may cause wildfires and threaten human lives. Still, more than 10\% HIFs fail to be detected by conventional protection relays. Existing methods require sufficient labeled datasets and heavily rely on measurements of relays at substations. Considering the insufficiency of labeled events, we construct a physics regulated convolutional auto-encoder (PRCAE) to detect HIFs without labeled HIFs for training. Our PRCAE introduces a physical regularization, derived from the elliptical trajectory of voltages-current characteristics, to distinguish HIFs from other abnormal events even in highly noisy situations. Also, we formulate a system-wide detection framework of combining multiple nodes' local detection results and a $\mu$PMU placement algorithm for the partially observed system. The proposed approaches are validated in the IEEE 34-node test feeder simulated through PSCAD/EMTDC. Our PRCAE shows superior detection performance than existing works in various scenarios, and is robust to different observability, noise, and low sampling rates.