Hybrid neuro-symbolic learning and reasoning for resilient load restoration in smart microgrids

Timely and accurate load restoration is essential for maintaining operational continuity in microgrids during emergencies such as equipment faults, islanding events, and abrupt demand fluctuations. As microgrids grow in complexity and variability, conventional control schemes often fall short in transparency, scalability, and responsiveness. A hybrid neuro-symbolic control architecture is introduced in this paper, combining the pattern recognition capabilities of neural networks (NNs) with the rule-based consistency of finite state machines (FSMs). This framework enables data-driven recovery decisions while enforcing symbolic validation and physical feasibility through integrated power flow (PF) analysis. The FSM component encodes domain expertise and operational safety logic, serving as a real-time gatekeeper for all proposed actions. That is, the FSM enforces a PF-constrained reachable closure over NN-proposed actions and declares restoration success once a target terminal state is reached. Numerical simulations conducted over a 24 h scenario featuring hourly disturbances demonstrated a restoration success rate exceeding 90%, with critical load fulfillment maintained above 95% in most cases. The average number of actions required per event remained below two, confirming both efficiency and interpretability. By structurally separating heuristic learning and symbolic validation, the system delivers robust performance under dynamic and uncertain conditions. These results highlight the promise of hybrid AI-symbolic approaches in advancing resilient and explainable microgrid control strategies, particularly where safety, adaptability, and real-time operation are equally critical.