Risk-aware AI control framework for resilient distribution grids under extreme events

The increasing penetration of distributed energy resources and the growing frequency of extreme events expose major limitations in conventional distribution-grid control, which is largely designed for nominal conditions and deterministic security criteria. Although artificial intelligence (AI), advanced optimization, and adaptive control have been applied to distribution networks, the literature remains fragmented across risk assessment, prediction, and control execution. This review synthesizes probabilistic resilience modeling, AI-based hazard and outage forecasting, adaptive and hierarchical control, distributed and edge intelligence, and resilience-oriented operational planning. It identifies a central gap: the absence of a closed-loop architecture that translates probabilistic risk and uncertainty into verified operational decisions, control constraints, and restoration actions. To address this gap, the paper proposes the Risk-Aware Control for Extreme Resilience (RACER) framework as an integrative reference architecture that connects risk inference, AI-enabled control, constraint verification, adaptive execution, and human supervision within a layered operational structure. The framework provides a conceptual and operational basis for integrating existing and future methods and establishes a research agenda for the implementation, verification, and validation of risk-aware, AI-enabled grid operation.