人工智能赋能算电协同的关键技术OA
Key technologies of AI-empowered computing-electricity coordination
随着生成式人工智能和大模型的快速发展,以及算力需求和用电需求的持续增长,算力负荷与电力供给在时间、空间及运行目标上的协调难度不断增加.针对算力系统与电力系统在数据结构、运行周期、调节方式和安全边界等方面的差异,从系统工程视角研究人工智能赋能算电协同的总体架构、运行机制与关键技术.构建由基础支撑、技术引擎、场景应用和治理保障组成的4层总体架构,并以人工智能赋能算电协同运行为核心,形成涵盖状态映射、供需时空匹配、耦合传导、多目标权衡和双向闭环调控的运行机制.在此基础上,构建由多源感知与融合、算电供需预测、算电耦合建模、协同优化决策以及自适应控制与安全执行组成的关键技术链,重点分析时空特征表征、任务与功率映射、模型与学习协同决策、滚动控制及安全约束校验等技术,并进一步归纳跨域数据融合、多尺度决策耦合以及模型泛化与安全验证等关键难点.结合陕西省可再生能源电力消纳和智能算力供给情况开展案例分析,结果表明,陕西可再生能源电力利用基础持续增强,智能算力具有较高的可统筹调度比例且呈现一定空间差异,为算力负荷与新能源供给的时序协同及跨节点算力资源配置提供了现实条件.针对上述难点及区域资源特征,提出完善跨域数据融合与可信共享、构建分层滚动协同决策以及建立策略验证与安全运行机制等对策,以提升算力利用效率、可再生能源电力消纳水平和系统运行韧性.
With the rapid development of generative AI and foundation models,computing demand and the associated electricity demand continue to grow,making the coordination between computing loads and electricity supply increasingly challenging in terms of time,space,and operational objectives.Considering the differences between computing systems and power systems in data structures,operating cycles,regulation mechanisms,and security boundaries,the overall architecture,operational mechanism,and key technologies of AI-empowered computing-electricity coordination are investigated from a system engineering perspective.A four-layer architecture comprising basic support,technology engine,scenario application,and governance assurance is constructed.With AI-empowered computing-electricity coordination as the core,an operational mechanism is established that integrates state mapping,spatiotemporal supply-demand matching,coupling transmission,multi-objective trade-offs,and bidirectional closed-loop regulation.On this basis,a key technology chain is constructed,covering multi-source sensing and fusion,computing-electricity supply-demand prediction,computing-electricity coupling modeling,coordinated optimization and decision-making,and adaptive control and safe execution.Key technologies,including spatiotemporal feature representation,task-to-power mapping,model-and learning-driven coordinated decision-making,rolling control,and safety constraint checking,are analyzed.Furthermore,major challenges involving cross-domain data fusion,multiscale decision coupling,model generalization,and safety verification are summarized.A case study of Shaanxi Province is conducted based on renewable electricity consumption and the supply of intelligent computing capacity.The results show that the foundation for renewable electricity utilization in Shaanxi Province has been continuously strengthened.Intelligent computing resources exhibit a relatively high proportion of coordinately schedulable capacity and certain spatial disparities,providing practical conditions for temporal coordination between computing loads and renewable energy supply,as well as for cross-node allocation of computing resources.In response to these challenges and regional resource characteristics,the following measures are proposed:improving cross-domain data fusion and trusted sharing,establishing hierarchical rolling coordinated decision-making mechanisms,and developing strategy validation and safe operation mechanisms.This will further enhance computing resource utilization efficiency,renewable electricity consumption,and system operational resilience.
马帅印;牛晨阳;栾龙源;庄严;岳本勇
西安邮电大学 计算机学院,西安 710121西安邮电大学 计算机学院,西安 710121陕西省大数据集团有限公司,西安 712000西安超算中心有限公司,西安 710100西安邮电大学 科研处,西安 710121
能源科技
人工智能算电协同算力调度供需时空匹配协同优化可再生能源电力消纳
artificial intelligencecomputing-electricity coordinationcomputing resource schedulingspatiotemporal supply-demand matchingcollaborated optimizationrenewable electricity consumption
《综合智慧能源》 2026 (8)
69-81,13
西安市社会科学规划基金项目(J26JX08)咸阳市科技局项目(L2025-ZDYF-RGZN-003)陕西省社会科学基金项目(2024D054)Xi'an Social Science Planning Fund(J26JX08)Project of Xianyang Municipal Science and Technology Bureau(L2025-ZDYF-RGZN-003)Shaanxi Provincial Social Science Fund Project(2024D054)
评论