Real-time AI-based Fault Detection and Localization in Power Electronics Dominated Grids

This paper presents a real-time fault detection and classification network for power electronics dominated grids (PEDG). The challenges in detection and localization of faults in active distribution networks are addressed by the proposed approach. The proposed approach is based on a long short-term memory (LSTM) neural network to detect and localize faults based on measurements at the point of common coupling of distributed energy resources (DERs) within the network. The proposed scheme is implementable at the grid-edge in active distribution networks for real-time detection, classification, and localization using DERs as a grid probing tool to enhance the situational awareness of futuristic PEDG. This work includes a detailed theoretical analysis and case study that evaluates the performance of the proposed LSTM-based fault detection and localization in active distribution networks. A comprehensive database is created for the training process and the network operates with optimized hyperparameters. The proposed method is examined for a modified IEEE 14-bus network dominated by DERs. The results demonstrate strong performance and fast (i.e., within one line cycle) fault detection and localization that enhances the situational awareness of the system.