Abstract
With the growing integration of renewable energy and electronic devices, power dispatch tasks face increasing complexity due to internal and external factors, such as weather uncertainty, the long-term impact of energy storage operation, and the fluctuations in the electricity market. Recently, the large language model (LLM) Agent has been proposed, incorporating the LLM with human-like capabilities. This promotes building the LLM-based power dispatch agent to address the abovementioned problems. However, the framework of LLM-based power dispatch agents, key capabilities, and associated challenges remain unclear. Therefore, this paper proposes a comprehensive LLM-based Power Dispatch Agent framework, encompassing perception, planning, memory, reflection, and action modules, to tackle real-world tasks. Then, the capabilities and potential applications of LLM-based Power Dispatch Agent among different power dispatch-related tasks are explored. Finally, the challenges of the LLM-based power dispatch agent are discussed in relation to the requirements of power dispatch and current technologies. • Presents a novel Large Language Model-based agent framework for power dispatch operations. • Enables complex power grid control and decision-making through natural language interaction. • Automates multi-step planning and tool invocation for solving critical dispatch tasks. • Validates high performance and robustness across diverse scenarios on standard IEEE test systems.