Location of faults based on deep learning with feature selection for meter placement in distribution power grids

A problem of great interest for power distribution companies is ensuring uninterrupted service in extensive power distribution systems. Thus, the monitoring of networks and identification of system faults become essential. This work focuses on identifying a fault's occurrence from a small number of low-cost measurements in a power distribution system. The determination of sensor locations is based on the recent feature selection approach LassoNet, where the measurement locations are ranked. It provides the most informative measures during a fault resulting in a shortening data set. It is used as input to a deep neural network without a significant loss in accuracy. We validate our method on the IEEE 13 and 34 node test feeders for distribution systems to conduct the suggested approach's experimental studies.