Redesigning Power Grid Dispatching Automation Systems by Integrating LLMs, KGs, and AI Agents

The rapid integration of high-proportion renewable energy and the escalating complexity of modern power systems have exposed critical limitations in conventional grid dispatching automation systems, particularly in their ability to achieve real-time responsiveness, adaptive coordination, and collaborative decision-making. To address these challenges, this paper presents a novel next-generation dispatching automation framework that synergistically integrates large language models (LLMs), knowledge graphs (KGs), and AI agent technologies. The proposed system harnesses LLMs for advanced natural language processing and semantic reasoning, employs KGs to construct a structured, multidimensional knowledge representation of grid entities and their interdependencies, and deploys AI agents for autonomous decision-making and dynamic operational optimization. Collectively, these components establish a closed-loop intelligent architecture encompassing “perception-cognition-decision-execution”, enhancing system-wide situational awareness and adaptive control. This study proposes a unified integration methodology for LLMs, KGs, and AI agents in power system dispatching with modularized functional implementations designed for scalable deployment. Furthermore, it provides comprehensive multi-scenario analyses to verify operational efficacy. The proposed system offers a transformative technical pathway for advancing intelligent dispatching in next-generation power grids by resolving key bottlenecks in data interoperability, decision transparency, and dynamic adaptability.