边缘数据中心集群算力-电力-热力协同的两阶段分布鲁棒优化方法

doi:10.19595/j.cnki.1000-6753.tces.251975

摘要
图/表
参考文献
相关文章 (1)

全文: PDF (2839 KB)
输出: BibTeX | EndNote (RIS)

摘要

随着数字经济与新型基础设施建设的深度推进,边缘数据中心围绕城市级区域逐步形成集群化部署态势,受限于单体规模较小且算力设备高度密集,其算力-电力-热力耦合响应更趋快速剧烈。然而,现有运维调控策略普遍存在热环境精细化温控能力薄弱、源-荷双重不确定性处理灵活性欠缺、多能流协同深度不足等问题。针对上述挑战,本文首先建立边缘数据中心集群算力-电力-热力精细化模型,进而基于Willems基本引理以及模型预测控制方法,耦合数据负载处理、外部制冷基础设施运维策略构建机架级热环境隐式控制模型;其次,面向分布式光伏出力以及数据负载到达双重不确定性,基于联合KL散度构建两阶段分布鲁棒优化模型,实现算力-电力-热力全栈耦合调控;进而,考虑大规模优化问题求解效率以及隐私保护需求,设计IC&CG-自适应ADMM算法进行分布式并行求解。仿真表明,基于本文方法可实现机架级温控均方根误差仅为0.39 ℃。相较于传统优化方法,本文所得边缘数据中心集群运行成本中位数以及99%分位尾部风险分别降低4.51%、13.62%,电能使用效率优化至1.22,求解耗时缩短69.68%。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	孙毅
	张凯
	格根敖其
	吴鹏
	朱进

关键词 ：边缘数据中心, 算力-电力-热力协同, 精细化建模, 两阶段分布鲁棒优化

Abstract：

The rapid development of the digital economy and the advancement of new infrastructure have driven explosive growth in the deployment of Edge Data Center (EDC). Consequently, their escalating energy consumption has become a critical bottleneck for the sustainable development of edge computing. Characterized by small scale, high power density, and urban clustering into Edge Data Center Cluster (EDCC), EDCs exhibit tightly coupled and dynamic interactions across the computing power-electricity-heat domains. However, existing operational strategies typically lack precision in thermal environment control, flexibility in managing dual source-load uncertainties, and depth in multi-energy flow coordination.
To address these challenges, this paper first establishes a refined computing power-electricity-heat model for the EDCC. Leveraging Willems' Fundamental Lemma and Model Predictive Control (MPC), we construct an implicit rack-level thermal control model. This model integrates data load processing with the operation of external cooling infrastructure to achieve tight coordination of multi-energy flows. Second, to address the dual uncertainties from distributed photovoltaic (DPV) intermittency and fluctuating data load arrivals, we develop a two-stage Distributionally Robust Optimization (DRO) model. The ambiguity set for this model is constructed using joint Kullback-Leibler (KL) divergence. The day-ahead stage (first-stage) determines the operation of external cooling infrastructure and server on/off states. The intra-day stage (second-stage) optimizes cross-EDC energy sharing, heterogeneous server load allocation, temporal data load shifting, and the scheduling of Computer Room Air Handlers (CRAH) and Uninterruptible Power Supply (UPS) units. Furthermore, to balance computational efficiency and privacy preservation in large-scale optimization, we design an Inexact Column and Constraint Generation (IC&CG) algorithm, nested with an adaptive Alternating Direction Method of Multipliers (ADMM). The IC&CG handles the master problem, while the adaptive ADMM solves subproblems in parallel.
Case studies are conducted on the MATLAB simulation platform and industry-recognized data center thermal environment simulation software. Performance is systematically evaluated across four dimensions: (1)Operational strategy effectiveness, validating the EDCC energy sharing mechanism and the coordinated cooling control strategy; (2)refined thermal control performance, benchmarking against traditional static PUE-based control and lumped-parameter MPC methods; (3)joint uncertainty handling, comparing the proposed method with deterministic optimization, stochastic optimization, robust optimization, and DRO based on independent uncertainty modeling; 4)algorithm performance, comparing the proposed IC&CG-adaptive ADMM against centralized C&CG and standard C&CG-ADMM. Simulation results confirm the core advantages of our method: the rack-level temperature control achieves a Root Mean Square Error (RMSE) of only 0.39°C. Compared to deterministic optimization, our approach reduces the median EDCC operating cost by 4.51% and the 99% quantile tail risk by 13.62%, achieves a cluster-wide Power Usage Effectiveness (PUE) of 1.22, and shortens the solution time by 69.68%.
Based on the above analysis, the main conclusions are as follows: (1) The refined EDCC model and implicit rack-level thermal control effectively unlock the operational regulation potential, optimizing overall energy efficiency. (2) The two-stage DRO model based on joint KL divergence effectively mitigates uncertainties from DPV and data loads; adjusting the KL divergence threshold allows for a dynamic trade-off between robustness and economy. (3) Compared to traditional centralized architectures, the distributed IC&CG-adaptive ADMM algorithm efficiently solves large-scale, multi-energy coupled problems under decentralized information structures, ensuring solution accuracy while significantly reducing computation time.

Key words： Edge data center computing power - electricity - heat coordination fine-grained modeling two-stage distribution robust optimization

收稿日期: 2025-11-18

PACS:

TM715

基金资助:

国家电网公司总部科技项目资助（1400-202457291A-1-1-ZN）

通讯作者: 张凯男,2000年生,博士研究生,研究方向为算力-电力协同等。E-mail：kaizhang0537@163.com

作者简介: 孙毅男,1972年生,教授,博士生导师,研究方向为能源互联网信息通信技术、电力需求侧管理等。E-mail：sy@ncepu.edu.cn

引用本文:

孙毅, 张凯, 格根敖其, 吴鹏, 朱进. 边缘数据中心集群算力-电力-热力协同的两阶段分布鲁棒优化方法[J]. 电工技术学报, 0, (): 37-. Sun Yi, Zhang Kai, Gegen Aoqi, Wu Peng, Zhu Jin. A Two-Stage Decentralized Robust Optimization Method for Computing Power - Electricity - Heat Coordination in Edge Data Center Cluster. Transactions of China Electrotechnical Society, 0, (): 37-.

链接本文:

https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.251975 https://dgjsxb.ces-transaction.com/CN/Y0/V/I/37

[1] 华为技术有限公司. 面向5G的边缘数据中心基础设施[R]. 深圳: 华为技术有限公司, 2019.
[2] Zhang Yingbo, Shan Kui, Li Xiuming, et al.Research and technologies for next-generation high-temperature data centers-state-of-the-arts and future perspectives[J]. Renewable and Sustainable Energy Reviews, 2023, 171: 112991.
[3] Zhu Yiqun, Zhang Quan, Zeng Liping, et al.An advanced control strategy of hybrid cooling system with cold water storage system in data center[J]. Energy, 2024, 291: 130304.
[4] Zhu Yiqun, Zhang Quan, Huang Gongsheng, et al.The influence of uncertainty parameters on the double-layer model predictive control strategy for the collaborative operation of chiller and cold storage tank in data center[J]. Energy, 2024, 313: 133866.
[5] 可思为, 董萍, 马铭宇, 等. 考虑风光荷时空互补的多能源绿色数据中心多目标配置方法[J]. 电力系统保护与控制, 2024, 52(22): 22-33.
Ke Siwei, Dong Ping, Ma Mingyu, et al.A multi-objective allocation method for multi-energy green data centers considering wind, solar and load spatial-temporal complementarity[J]. Power System Protection and Control, 2024, 52(22): 22-33.
[6] 马浩天, 胡俊杰, 童宇轩. 考虑灵活性的数据中心微网两阶段鲁棒规划方法[J]. 中国电机工程学报, 2023, 43(19): 7396-7409.
Ma Haotian, Hu Junjie, Tong Yuxuan.A two-stage robust planning approach for data center microgrids considering flexibility[J]. Proceedings of the CSEE, 2023, 43(19): 7396-7409.
[7] Zhao Jing, Chen Ziyi, Li Haonan, et al.A model predictive control for a multi-chiller system in data center considering whole system energy conservation[J]. Energy & Buildings, 2024, 324: 114919.
[8] 朱介北, 杨顺坡, 俞露杰, 等. 数据中心制冷系统即时消纳可再生能源的建模和变温控制[J]. 电力系统自动化, 2022, 46(20): 13-22.
Zhu Jiebei, Yang Shunpo, Yu Lujie, et al.Modeling and variable temperature control for instant consumption of renewable energy in data center cooling systems[J]. Automation of Electric Power Systems, 2022, 46(20): 13-22.
[9] Lu Ruyuan, Li Xin, Chen Ronghao, et al.An alternative reinforcement learning (ARL) control strategy for data center air-cooled HVAC systems[J]. Energy, 2024, 308: 132977.
[10] Wang Kun, Zhou Qihua, Guo Song, et al.Cluster frameworks for efficient scheduling and resource allocation in data center networks: A survey[J]. IEEE Communications Surveys & Tutorials, 2018, 20(4): 3560-3580.
[11] 文亚凤, 程祖铭, 刘欣雅, 等. 考虑算力-电热耦合的边缘数据中心能量优化管理[J]. 中国电机工程学报, 2026, 46(1): 186-197.
Wen Yafeng, Cheng Zuming, Liu Xinya, et al.Considering the energy optimization management of edge data centers with computing power-electricity-heat coupling[J]. Proceedings of the CSEE, 2026, 46(1): 186-197.
[12] Li Xiaoping, Chen Fuchao, Ruiz Rubén, et al.MapReduce task scheduling in heterogeneous geo-distributed data centers[J]. IEEE Transactions on Services Computing, 2022, 15(6): 3317-3329.
[13] 王继东, 许秋铭, 黄婷, 等. 含共享储能的数据中心微网群分布式优化调度[J]. 电网技术, 2024, 48(8): 3238-3247.
Wang Jidong, Xu Qiuming, Huang Ting, et al.Distributed optimal scheduling of data center microgrids cluster with shared energy storage[J]. Power System Technology, 2024, 48(8): 3238-3247.
[14] Da Silva M D M, Gamatié A, Sassatelli G, et al. Optimization of data and energy migrations in mini data centers for carbon-neutral computing[J]. IEEE Transactions on Sustainable Computing, 2023, 8(1): 68-81.
[15] 张玉莹, 曾博, 周吟雨, 等. 碳减排驱动下的数据中心与配电网交互式集成规划研究[J]. 电工技术学报, 2023, 38(23): 6433-6450.
Zhang Yuying, Zeng Bo, Zhou Yinyu, et al.Research on interactive integration planning of data centers and distribution network driven by carbon emission reduction[J]. Transactions of China Electrotechnical Society, 2023, 38(23): 6433-6450.
[16] 中国电力发展促进会电力与算力协同发展专业委员会, 紫金山实验室, 江苏省未来网络创新研究院, 等.电力与算力协同发展蓝皮书(2025)[R]. 北京: 中国电力发展促进会电力与算力协同发展专业委员会, 2025.
[17] 张献, 丁可浩, 赵黎媛, 等. 计及电动汽车混合充电系统接入的综合能源系统鲁棒优化调度[J]. 电工技术学报, 2025, 40(14): 4446-4459.
Zhang Xian, Ding Kehao, Zhao Liyuan, et al.Robust optimal scheduling of integrated energy system considering electric vehicle hybrid charging system[J]. Transactions of China Electrotechnical Society, 2025, 40(14): 4446-4459.
[18] 李虹, 韩雨萌. 计及新能源不确定性的配电网-多微电网协同优化调度[J]. 电工技术学报, 2025, 40(17): 5571-5588.
Li Hong, Han Yumeng.Coordinated optimization scheduling of distribution networks and multiple microgrids considering the uncertainty of new energy[J]. Transactions of China Electrotechnical Society, 2025, 40(17): 5571-5588.
[19] Wang Zhiying, Wang Yang, Ji Haoran, et al.Distributionally robust planning for data center park considering operational economy and reliability[J]. Energy, 2024, 290: 130185.
[20] Li Weiwei, Qian Tong, Zhang Yin, et al.Distributionally robust chance-constrained planning for regional integrated electricity-heat systems with data centers considering wind power uncertainty[J]. Applied Energy, 2023, 336: 120787.
[21] Ding Zhaohao, Chen Shijie, Sun Yimeng, et al.Data center job scheduling and energy management under uncertain environments[J]. IEEE Transactions on Industry Applications, 2025, 61(4): 5489-5500.
[22] 王波, 王蔚, 马恒瑞, 等. 基于Wasserstein两阶段分布鲁棒的多主体多能微网合作博弈优化调度[J]. 电工技术学报, 2025, 40(17): 5553-5570.
Wang Bo, Wang Wei, Ma Hengrui, et al.Multi-agent multi-energy microgrid cooperative game scheduling based on Wasserstein two stage robust optimization[J]. Transactions of China Electrotechnical Society, 2025, 40(17): 5553-5570.
[23] 王海伦, 丁一凡, 李杨, 等. 计及混合式抽水蓄能改造的清洁微网分布鲁棒容量优化配置[J]. 电工技术学报, 2025, 40(07): 2112-2126.
Wang Hailun, Ding Yifan, Li Yang, et al.Distributionally robust capacity optimization for clean energy microgrid considering pumped-storage retrofitting[J]. Transactions of China Electrotechnical Society, 2025, 40(07): 2112-2126.
[24] Ismail L, Materwala H.Computing server power modeling in a data center[J]. ACM Computing Surveys, 2020, 53(3): 1-34.
[25] 深圳市腾讯计算机系统有限公司, 英特尔(中国)有限公司, 中国电信集团有限公司. 数据中心算力-电力灵活性协同研究[R]. 深圳: 深圳市腾讯计算机系统有限公司, 2023.
[26] Zhao Xinkui, Yin Jianwei, Chen Zuoning, et al.vSpec: Workload-adaptive operating system specialization for virtual machines in cloud computing[J]. Science China Information Sciences, 2016, 59(9): 1-16.
[27] 华为技术有限公司. 计算产品能耗计算器[EB/OL].[2025-08-06]. https://info.support.huawei.com/computing/ftpa/.
[28] Balakumar P, Vinopraba T, Sankar S, et al.Smart hybrid microgrid for effective distributed renewable energy sharing of PV prosumers[J]. Journal of Energy Storage, 2022, 49: 104033.
[29] 张虹,郑广宇,娄娟,等.计及热电负荷不匹配的数据中心余热补偿系统调控策略[J].电力系统自动化, 2025, 49(13): 156-165.
Zhang Hong, Zheng Guangyu, Lou Juan, et al.Regulation strategy of waste heat compensation system for data center considering thermo-electric load mismatch[J]. Automation of Electric Power Systems, 2025, 49(13): 156-165.
[30] 连之伟. 热质交换原理与设备[M]. 北京: 中国建筑工业出版社, 2011.
[31] 曾博, 孟自帅, 董嘉路, 等. 考虑电-水耦合灵活性的数据中心与城市配电-水网协同规划研究[J]. 电工技术学报, 2025, 40(23): 7554-7569.
Zeng Bo, Meng Zishuai, Dong Jialu, et al.Research on synergistic planning of data center and urban electricity-water distribution networks considering electricity-water interdependent flexibility[J]. Transactions of China Electrotechnical Society, 2025, 40(23): 7554-7569.
[32] Isazadeh A, Ziviani D, Claridge D E.Thermal management in legacy air-cooled data centers: An overview and perspectives[J]. Renewable and Sustainable Energy Reviews, 2023, 187: 113707.
[33] Li Xiuming, Zhang Ce, Dong Jiaxiang, et al.Feasibility investigation on a novel rack-level cooling system for energy-saving retrofit of medium-and-small data centers[J]. Applied Thermal Engineering, 2023, 229: 120644.
[34] Li Yiran, Yang Chao, Xia Yuanqing.A robust data-driven model predictive thermal control for rack-based data center[J]. Journal of Building Engineering, 2024, 98: 110877.
[35] Guo Xiao, Che Yanbo, Zheng Zhihao, et al.Multi-timescale optimization scheduling of interconnected data centers based on model predictive control[J]. Frontiers in Energy, 2024, 18(1): 28-41.
[36] Wang Chenchao, Meng Deyuan.Data-based optimization control for learning systems[J]. IEEE Transactions on Circuits and Systems II: Express Briefs, 2023, 70(7): 2560-2564.
[37] Wei Danxiang, Wang Jianzhou, Li Zhiwu, et al.Wind power curve modeling with hybrid copula and grey wolf optimization[J]. IEEE Transactions on Sustainable Energy, 2022, 13(1): 265-276.
[38] Wang Zhongliang, Zhu Hongyu, Zhang Dongdong, et al.Modelling of wind and photovoltaic power output considering dynamic spatio-temporal correlation[J]. Applied Energy, 2023, 352: 121948.
[39] 梁俊文, 林舜江, 刘明波, 等. 主动配电网分布式鲁棒优化调度方法[J]. 电网技术, 2019, 43(4): 1336-1344.
Liang Junwen, Lin Shunjiang, Liu Mingbo, et al.Distributed robust optimal dispatch in active distribution networks[J]. Power System Technology, 2019, 43(4): 1336-1344.