Federated learning-based fault location and identification in hybrid AC/DC distribution systems considering bidirectional power flow

Abstract Modern power systems are evolving toward hybrid AC/DC distribution networks to enhance efficiency, improve renewable energy integration, and support bidirectional power flow. However, these systems present complex challenges for fault detection and localization due to the diversity of AC and DC fault characteristics and their intricate operational behavior. This paper proposes a novel federated learning (FL)-based fault analysis framework that enables privacy-preserving, decentralized training while handling bidirectional energy flow and inverter-based RES integration. The approach introduces a complete solution encompassing fault detection, classification and localization using feature-extracted voltage and current signals. Simulation results under various fault types, network topologies, and operational modes confirm the method’s robustness and real-time accuracy. The proposed framework contributes to intelligent, scalable, and secure fault management in hybrid AC/DC networks.