With the rapid expansion of renewable energy installations and the widespread integration of heterogeneous energy sources

the medium- and long-term mechanisms of power-energy balance in new power systems are undergoing profound transformation. Traditional power supply-demand coordination technologies have become increasingly inadequate. In the context of the 5th paradigm of scientific research

marked by the emergence of “AI for science (AI4S)”

AI technologies

particularly large language models

offer new pathways for an intelligent redesign of power supply-demand coordination architectures. This paper explores the key characteristics of mid-to-long-term power and energy balance in new power systems and identifies the limitations of conventional architectures. It also elucidates the connotation of AI4S-driven supply-demand coordination and proposes a novel architecture framework anchored in three core components: the Power Knowledge Base

Transformation Interface

and Decision Plugin.. Furthermore

it promotes the transition from human-centric decision loops toward machine-autonomous reasoning

realizing a technical roadmap toward a full-process “AI+” pattern. Finally

this paper discusses future research directions to provide insights into achieving reliable power supply and efficient renewable energy utilization under AI empowerment.