综合智慧能源2026，Vol.48Issue(8)：1-15,15.DOI:10.3969/j.issn.2097-0706.2026.08.001
智算中心柔性负荷建模与算电协同调控关键技术及展望
Key technologies and prospects for flexible load modeling and computing-electricity coordinated regulation in intelligent computing centers
摘要
Abstract
To address the dispersed sources of flexibility,complex cross-layer coupling,and the lack of a unified framework for electricity-computing-carbon coordination across multiple intelligent computing parks,a systematic review of research on flexible load modeling and computing-electricity coordinated regulation in intelligent computing centers is conducted,with the cross-layer transmission chain from workloads,computing resources,facility equipment,and grid-connected power as the mainline.The power characteristics,available regulation actions,and recovery-related rebound risks of offline,online,and hybrid workloads are distinguished.The formation mechanisms of computing-side,power-side,and thermal-management-side flexibility are summarized,and the differences in response,duration,and recovery timescales as well as the priority relationships among different flexibility resources are clarified.Then,baseline load modeling,flexibility envelope characterization,and uncertainty-aware prediction methods are reviewed.Typical demand response modes driven by external signals such as time-of-use electricity prices,renewable energy output,and ramping commands are compared,and evaluation indicators including deliverable capacity,energy storage degradation,and quality-of-service costs are summarized from the perspectives of power grids,facilities,and users.Subsequently,the source-load-storage-computing coupling architecture,spatiotemporal workload migration mechanisms,dynamic carbon accounting methods,and coordinated transmission processes of electricity price,renewable energy,and carbon signals in multiple intelligent computing parks are analyzed.Different coordination mechanisms are further compared in terms of driving signals,optimization objectives,and key constraints.The review indicates that the flexibility of intelligent computing centers exhibits cross-layer,multi-timescale,and state-dependent characteristics.The key to reliable participation in grid regulation lies in transforming internal adjustable potential into predictable,deliverable,and verifiable aggregated responses.Specifically,predictability corresponds to capacity and confidence evaluation before dispatch,deliverability refers to the capability of satisfying operational constraints during dispatch,and verifiability represents the performance validation based on unified baselines after dispatch.Together,these three elements establish an engineering closed loop covering prediction,commitment,dispatch,measurement and settlement,and rolling updates.Meanwhile,intelligent computing centers differ from conventional data centers in terms of power density,synchronized communication requirements,and thermal safety boundaries,indicating that some existing conclusions require further validation for intelligent computing workloads.Future key development directions are summarized,including multi-energy coupling,power grid security,coordinated planning and operation,low-carbon value measurement,cross-regional coordination between power grids and computing networks,and AI-native autonomous control.Existing challenges are also identified in engineering-level validation,explicit consideration of adverse effects,and cross-park risk coordination,providing references for exploiting computing load flexibility and developing demand response mechanisms in new power systems.关键词
智算中心/柔性负荷/需求响应/算电协同/低碳电力调度Key words
intelligent computing center/flexible load/demand response/computing-electricity coordination/low-carbon power dispatch分类
能源科技引用本文复制引用
朱继忠,杨晓宁,张迪,王晞罗..智算中心柔性负荷建模与算电协同调控关键技术及展望[J].综合智慧能源,2026,48(8):1-15,15.基金项目
国家自然科学基金项目(U25B6016)National Natural Science Foundation of China(U25B6016) （U25B6016）
