Reducing hardware requirements for AI-based high-impedance ground fault detection

The detection of high-impedance ground faults (HIFs) in resonant-grounded grids is challenging due to the reduction of fault currents and high fault impedance. Convolutional neural networks (CNNs) have been successfully applied for detection at 14, 400 Hz sampling rate. The expensive hardware required for high sampling rates poses an obstacle for larger-scale deployment. These requirements arise from the fact that it has not yet been conclusively determined which information content is relevant for HIF detection. This investigation aims to find the optimal sampling rate that balances detection accuracy and hardware requirements. A large set of measurement data is downsampled to lower rates to simulate reduced sampling rate measurements and upsampled back to the sampling rate expected by the CNN. Results of this study indicate that lower sampling rates achieve comparable performance and and can be used to reduce the high hardware requirements of AI-based HIF detection.