抄録
The new-type power system is rapidly evolving toward a stage characterized by high renewable-energy penetration, deep participation of diverse loads, and the coordinated digitalization of energy resources, while simultaneously facing significant challenges such as reduced system inertia, insufficient voltage support, and increasingly volatile and uncertain power outputs. Against this backdrop, artificial intelligence (AI), with its strengths in high-dimensional system representation, complex nonlinear correlation mining, and real-time optimal decision-making, is becoming a key technological driver across the full chain of power system sensing, forecasting, and regulation. This study provides a comprehensive overview of AI-based technological frameworks on the generation, grid, and load sides, covering machine learning, deep learning, reinforcement learning, and hybrid-intelligence paradigms, and systematically reviews recent advances in renewable-energy output forecasting, load behavior modeling, system situational awareness, economic dispatch, voltage and frequency control, disturbance response, and resilience enhancement. Building on these insights, the study further explores the potential value and emerging trends of digital-twin power grids, physics-informed neural networks (PINNs), explainable AI (XAI), and large model technologies in new-type power system regulation, while identifying key bottlenecks related to data quality and availability, model generalization, computational real-time performance, and engineering implementation. The goal is to establish a structured cognitive framework for AI-enabled power system regulation and to provide theoretical support and directional guidance for future technological innovation, engineering practices, and interdisciplinary integration.
