A Supervised Learning Approach for Centralized Fault Localization in Smart Microgrids

Abstract

This article proposes a supervised learning approach for centralized localization of faults in microgrids with radial configuration that can operate in grid-connected or islanded mode. The key concept is to identify faults by learning and analyzing the features of voltage and current disturbances detected by a microgrid central protection unit. Two algorithms are developed that determine fault indices and transient features for simulation-based training, classification, and backup protection purposes. The proposed data-driven approach is effective under high penetration of distributed energy resources (DERs), and is robust under microgrid topology variations. Moreover, it requires few sensors (two for radial structure) and is easy to implement, yet provides high degree of reliability. The performance and accuracy of the developed approach are verified through extensive simulations of low-voltage microgrids that incorporate different types of DERs, including wind and solar photovoltaic systems.