Hybrid PSO–DNN Framework for Improved Relay Coordination in AC Microgrids under High-Impedance Faults

Traditional protection systems rely on overcurrent relays (OCRs), which are less adaptable to modern microgrids (MGs) with evolving dynamics caused by the increased integration of distributed energy resources (DERs). This paper proposes a modified fault-resistance-based relay characteristic to enhance the operation of directional overcurrent relays (DOCRs) under high-impedance faults (HIFs). The particle swarm optimization (PSO) algorithm is used to determine the optimal relay settings based on normal and fault conditions data from real-time simulations. The PSO also determines the optimal hyperparameter values for an optimal deep neural network (DNN) structure. The DNN model is trained to optimize relay settings across a wide range of fault resistance values while minimizing the number of required simulations. This framework incorporates a continuous learning mechanism to dynamically adapt relay settings in response to changing system conditions, and enhance fault sensitivity, relays’ coordination, and operating speed. It is demonstrated that the proposed method improves the reliability of AC MGs against grid faults by offering an enhanced coordinated protection solution.