综合智慧能源2026，Vol.48Issue(8)：16-28,13.DOI:10.3969/j.issn.2097-0706.2026.08.002
算电协同架构下基于MCP的多智能体协同机制综述
Review of MCP-based multi-agent synergistic mechanisms in computing-electricity coordination architecture
摘要
Abstract
With the large-scale deployment of large language model-driven multi-agent systems in the power sector,computing facility loads exhibit pronounced bursty and tidal fluctuation characteristics,resulting in increasingly prominent conflicts with the secure and stable operation of new power systems.Electricity consumption by China's computing centers reached 196 TW·h in 2025 and is expected to exceed 700 TW·h by 2030,while the computing-electricity coordination market is projected to exceed 1 trillion yuan.Combined with the lack of standardized interaction interfaces for heterogeneous agents,these trends lead to multiple industry challenges,including imbalanced computing resource allocation,impeded renewable energy consumption,and communication silos in cross-domain scheduling.To address the dual-core conflicts of physical resource misallocation and fragmented digital interaction,the latest research progress in China and abroad is systematically reviewed across three dimensions:computing-electricity coordination,multi-agent interaction,and the Model Context Protocol(MCP).At the physical mechanism level,the multidimensional coupled operating mechanism among computing power,electricity,and carbon emissions is deconstructed.Two technical systems,namely power security verification and dynamic carbon footprint quantification,are summarized.The real-time carbon responsibility tracing method based on carbon emission flow theory and the dynamic low-carbon labeling model are described,and the potential for spatiotemporal shifting and cross-domain migration of delay-tolerant computing tasks is revealed.At the communication architecture level,the technical advantages of MCP's three-layer stateful communication architecture and standardized interaction primitives are analyzed,while highlighting the excessive adaptation costs associated with conventional mesh-based integration schemes.At the synergistic mechanism level,the design approach of a two-tier cross-domain synergistic architecture driven by physical constraints is summarized,together with advances in supporting technologies such as computing power routing,intrinsic security protection,and balanced allocation of benefits among multiple parties.Additionally,existing implementation paradigms are summarized in combination with computing hub engineering scenarios,such as the electricity-carbon-computing synergistic operation system of the China Southern Power Grid.A comprehensive assessment indicates that current research still faces technical bottlenecks in agent-state synchronization in weak-grid environments,synergistic scheduling of large-scale heterogeneous clusters,and dynamic access control across trust domains.Future efforts should focus on establishing standardized evaluation systems,developing underlying protocols for cross-domain synergy,and achieving breakthroughs in multi-objective joint optimization algorithms.These directions can provide theoretical support for the standardized and secure implementation of multi-agent computing-electricity coordination in new power systems and promote deeper integration and coordinated development of the digital economy transition and the low-carbon energy transition.关键词
算电协同/算力中心/大语言模型/多智能体系统/模型上下文协议/跨域调度/负载迁移/碳排放流Key words
computing-electricity coordination/computing center/large language model/multi-agent system/model context protocol/cross-domain scheduling/load migration/carbon emission flow分类
能源科技引用本文复制引用
李彬,申楠,徐靖,陈宋宋..算电协同架构下基于MCP的多智能体协同机制综述[J].综合智慧能源,2026,48(8):16-28,13.基金项目
国家电网有限公司科技项目(52060026007C-256-ZN) （52060026007C-256-ZN）
智能电网国家科技重大专项(2026ZD0811203)Science and Technology Project of State Grid Corporation of China(52060026007C-256-ZN) （2026ZD0811203）
National Science and Technology Major Project on Smart Grid(2026ZD0811203) （2026ZD0811203）
