Neural Network and Correlation based Earth-Fault Localization utilizing a Digital Twin of a Medium-Voltage Grid

Fast localization of earth faults in medium voltage grids is required in order to avoid subsequent faults and to quickly restore the normal grid operation. We propose a localization approach utilizing a signature database with high-resolution transient voltages. The database is created based on a digital twin of a live medium-voltage grid for which measurements of voltages during actual earth faults of known location are available. The robustness and accuracy of two different realizations of the signature based fault localization are investigated: (1) a comparison approach using a correlation metric; (2) a neural network that has been trained by the signatures provided by the digital twin. The performance of our approach is assessed based on artificially generated earth fault events as well as real field measurements from the electrical grid.