Novel XGBoost Classifier Based Relaying Approach with 2 Classes of Protection Zone

Grids and microgrids show different behaviors when configuration switches between grid-connected mode to islanded mode change. Thus, conventional overcurrent relays may not be able to operate appropriately in different modes and face a variety of protection challenges such as relay miscoordination, maloperation, and delayed operation or even failure (blinding) in the relay operations specifically when dealing with ring and meshed grids. This paper proposes protective relays functioning based on a machine learning classifier named XGBoost which replace the O/C relays. XGBoost-Classifier-Based relays are trained with the data derived from the system. They have a logic similar to distance relays but monitor the system based on the features it is trained with, instead of system impedance and uses 3 classes as zones of protection. The XGBoost-Classifier-Based relays can perform standalone without any communication and maintain the constant time interval However, the protection scheme can perform much faster communicate is added with Permissive Underreach Logic. The fault characterization is implemented on a 13.8kV 2-bus ring AC microgrid run on Typhoon-HiL 604. XGBoost-Classifier-Based relay models will be trained on a high-performance computing cluster.