Skip to main content
APEC 2026 Main banner

    Microgrid Systems

    T39.2 - Resilient Neural Control for Grid-Forming Microgrids: an Integrated AI-Based Approach to Virtual-Impedance Scheduling and Cyber-Attack Mitigation

    Thursday, March 26, 2026
    10:40 AM - 11:00 AM CT
    Location: 217C
    In this work, we develop an integrated neural control framework for grid-forming inverters (GFMIs) that enables rapid, inverter-level detection and mitigation of cyber-attacks. Each GFMI is equipped with a lightweight bidirectional long short-term memory (Bi-LSTM) network that learns the typical time-based patterns of its local power and voltage signals , enabling accurate prediction of expected virtual impedance (𝑍𝒗) values. A centralized residual network (ResNet) is included to generate 𝑍𝒗 references under nominal conditions, providing coordinated power sharing support. The locally predicted 𝑍𝒗 values are then continuously monitored against the incoming 𝑍𝒗 reference commands. If a notable mismatch is found, the inverter immediately switches to its locally predicted value to keep the system stable.