A Deep Learning Approach to Earth Fault Classification and Source Localization

A portion of electrical feeders in distribution grids are prone to faults, often resulting in different types of earth faults, power quality disturbances as well as damaged equipment and outages. While in developed countries the amount of such feeders can be relatively low, the quota reaches as high as similar to 20% for many developing countries. Tackling this issue requires (i) understanding the current status of the grid and the faults that occur and (ii) identifying the origin of the fault for preventing similar future faults. This process is however costly and time consuming as it requires many hours of tedious manual work from engineers, operators or field experts. In an effort to tackle this issue, we present in this work a machine learning based framework for automatized fault type classification and faulty feeder identification. We provide an empirical evaluation of our proposed framework on a dataset of recordings from a real grid, showing encouraging results.