Explainable AI for Fault Detection and Classification in Microgrids

The increasing vulnerability and unpredictability of the power grid, along with challenges in renewable energy integrations, natural disasters, cyberattacks, and limited transmission system investments, necessitate robust algorithms for power system protection. In the given context, this paper proposes a fault detection and classification approach in AC microgrids, considering an explainable artificial intelligence (XAI) model that offers enhanced interpretability, transparency, and debugging capabilities. XAI tools provide information on factors influencing fault detection and classification decision-making. To generate a realistic dataset and to validate the fault detection and classification model, a digital representation of a 50 kW physical microgrid testbed is developed and integrated into a real-time simulation framework using the Opal-RT real-time simulator that enables time-critical simulation and mimics realistic operating conditions. The combination of real-time simulation and XAI provides a robust platform to advance power system protection, facilitating the exploration of fault detection strategies and the evaluation of algorithm performance with deeper insights into decision-making processes, which are justified in this paper in various case studies.