Data-Driven Fault Localization of a DC Microgrid with Refined Data Input

This paper proposes an online fault localization method for low voltage DC microgrids. This method is based on Artificial Neural Network (ANN) and only requires real-time measurements of a local power converter to locate a fault. During a DC fault, the current component fed by the AC grid can contribute to time-variant non-linearity, which is undesirable to the development of the data-driven method. A novel real-time scheme is thus proposed to exclude such components from DC fault current. The principle of the scheme is introduced and illustrated with time-domain analysis. The effectiveness is verified by case studies of locating a DC fault in a radial DC network fed by a 3-phase voltage source converter.