Artificial intelligence for stability assessment and early-warning in converter-dominated power systems: A unified review and risk-envelope framework

Converter-dominated power systems with high penetration of inverter-based resources exhibit fast, nonlinear, and multi-timescale dynamics that introduce new stability modes and operational risks beyond the scope of traditional assessment techniques. In recent years, artificial intelligence and machine learning methods have been widely applied to small-signal, transient, voltage, and frequency stability prediction using simulation data and wide-area measurements. Although many studies report high predictive accuracy, the literature remains fragmented across stability categories, data regimes, and model classes, and typically emphasizes point prediction rather than uncertainty, regime transitions, and control-oriented early-warning. This review presents a systematic and literature-rich synthesis of AI-driven stability assessment approaches for converter-dominated power systems, covering learning architectures, feature representations, graph-based methods, and real-time monitoring integration. Key gaps are identified in uncertainty-aware prediction, cross-timescale stability coupling, interpretability for control action, and operational risk quantification. To address these limitations, a novel AI-Based Stability Risk Envelope Framework is proposed, linking predictive models with uncertainty bounds, operating-region tracking, and decision-triggered early-warning mechanisms for dynamics-informed grid operation.