Stochastic AI-Driven Resilience Framework for Power Grids Considering Communication-Link Outages and Operator Reliability

Modern power systems no longer fail only because of line or generator outages; they also drift into insecure states when the SCADA/EMS links slow down or when the operator cannot react at the required pace. To study this joint effect, we build an AI-supported stochastic resilience framework that treats the communication layer and the human layer as first-class, time-varying elements of the grid. Communication delay and packet–drop behaviour are captured through a multi-level Markov description, while operator performance is estimated through a small cognitive–reliability block that changes with workload and stress. On top of these two sources of uncertainty, a reinforcement-learning controller updates the stabilizing actions so that the system can return to acceptable voltage and frequency bands after faults. Tests on the IEEE 39-bus and 118-bus systems show three tangible gains: damping improves by about 91%, the probability of a blackout event falls by around 42%, and the composite resilience index rises from 0.74 to 0.91 against conventional stochastic assessments. Taken together, the results suggest that resilience estimates become more faithful only when human variability and communication degradation are evaluated in the same loop, which makes the approach suitable for cognitively aware, self-healing grid operation.