Network-Aware Control Barrier Functions for Resilient Microgrids Under Stealthy Drift Attacks

Inverter-dominated microgrids are highly vulnerable to stealthy cyber-physical drift attacks, low-amplitude, slowly varying perturbations that bypass conventional statistical filters to induce voltage degradation and delayed collapse. This paper introduces a resilient, delay-aware supervisory control architecture that acts as an online safety shield at the actuator interface. By jointly modeling nonlinear power-flow interactions and directional communication topologies, we construct physics-informed Control Barrier Functions (CBFs), embedding structural electrical invariants derived from the nodal admittance matrix Ybus. The supervisor directly incorporates heterogeneous, time-varying network delays into its safety constraints and utilizes a threat-adaptive modulation loop driven by spatio-temporal residuals to dynamically scale intervention aggressiveness. Using a Lyapunov-Krasovskii functional, we prove that the closed-loop tracking error is Input-to-State Stable (ISS) under bounded drift and worst-case latencies. High-fidelity simulations on an IEEE 14-bus test feeder demonstrate that the supervisor consistently enforces non-negative safety margins and reduces time-integrated voltage violations. Under coordinated sub-threshold attacks designed to exploit network jitter, the architecture bounds trajectories to physically consistent manifolds and prevents voltage collapse, establishing a scalable cross-layer safety framework for resilient distribution systems.