计及算力负载时空特性的算-电-热系统协同优化调度OA
Optimal scheduling of integrated computation-electricity-thermal systems considering the spatiotemporal characteristics of computing workload
电力和算力的协同优化调度,有助于促进可再生能源就地消纳、充分挖掘数据中心用能灵活调配潜力,有效降低数据中心用能成本.现有研究尚未充分量化空间调度模型,且热力侧调度模型过于简化.为提升电力系统与数据中心的运行效率,提出一种计及算力负载时空特性的算-电-热系统优化调度策略.首先,突破将数据中心视为单一节点的传统模式,在考虑算力负载空间迁移时,精细量化光纤通信网络的动态传输损耗与光纤传输约束,并将传输成本纳入目标函数.其次,改变机房产热与制冷设备同步增减功率的既有策略,建立数据中心室内热力学方程,基于阶梯电价机制,利用建筑储冷实现冷量的提前存储与滞后释放.最后,在保证服务质量(QoS)的前提下,以总成本最小为目标,构建算-电-热优化调度模型,分别研究热力侧调度与空间调度的效果.以2个数据中心为研究对象,基于12节点系统,两区域采用差异化峰谷电价,配置不同风光出力曲线及夏季室外温度.通过调度120台服务器,满足批处理与交互式两类算力负载需求,同时通过调度冷热源,满足室内温度及CPU温度约束.仿真结果表明:所提调度策略可实现算力负载的时空调度,全时段内,任务到达后两小时内完成的任务占比得到了提升,且QoS也得到了改善.考虑空间调度时,该策略具备对各类算力负载削峰填谷的能力,可优化区域用电负荷的时空分布,避免任务集中积压,减少大规模服务器频繁启停,从而显著延长硬件使用寿命,并降低维护成本;考虑热力侧调度时,在保障QoS的前提下可显著降低经济成本,为数据中心经济高效运行提供参考.
The coordinated optimization and scheduling of power and computing resources help facilitate the local consumption of renewable energy,fully leverage the flexibility of data center energy usage,and effectively reduce data center energy costs.Existing research has not adequately quantified spatial scheduling models,and thermal-side scheduling models are overly simplified.Therefore,to enhance the operational efficiency of power systems and data centers,an optimal scheduling strategy for integrated computation-electricity-heat systems considering the spatiotemporal characteristics of computing loads was proposed.The conventional approach of treating each data center as a single node was overcome.When spatial migration of computing loads was considered,the dynamic transmission losses and fiber-optic transmission constraints of the optical communication network were finely quantified,and the corresponding transmission costs were incorporated into the objective function.Subsequently,the existing strategy of simultaneously increasing or decreasing the power of heat-generating and cooling equipment was changed.Indoor thermodynamic equations of data centers were established.Based on the tiered electricity pricing mechanism,the advance storage and delayed release of cooling energy were achieved by the thermal storage capacity of the building.With the aim of minimizing total cost while ensuring service quality,an optimal scheduling model for integrated computation-electricity-heat systems was constructed to separately investigate the effects of thermal-side scheduling and spatial scheduling.Two data centers were used as the research objects.Based on a 12-node system,the two regions operated under different peak-valley electricity pricing schemes and featured distinct wind and solar power output curves as well as summer outdoor temperatures.The system met the requirements of both batch and interactive computing loads by scheduling 120 servers,while satisfying indoor temperature and CPU temperature constraints through the scheduling of cooling and heating sources.Simulation results demonstrated that the proposed scheduling strategy achieved spatiotemporal scheduling of computing loads.Across all time periods,the proportion of tasks completed within 0-2 h of arrival increased,and service quality improved.The power balance chart showed a remarkably high degree of matching between source and load.Considering spatial scheduling,the strategy could perform peak shaving and valley filling for various types of computing loads,thereby optimizing the spatiotemporal distribution of regional power loads,avoiding concentrated task accumulation,and reducing the frequency of start-stop cycles of large-scale servers,thereby significantly extending hardware service life and reducing maintenance costs.Considering thermal-side scheduling,the strategy can significantly reduce economic costs while maintaining service quality,offering a reference for the economical and efficient operation of data centers.
苏璇;于宗民;巨云涛;王碧清
北方工业大学 储能科学与工程学院,北京 100144||特大城市韧性电网供电保障技术与应用北京市重点实验室,北京 100144清华四川能源互联网研究院,成都 610213北方工业大学 储能科学与工程学院,北京 100144||特大城市韧性电网供电保障技术与应用北京市重点实验室,北京 100144北方工业大学 储能科学与工程学院,北京 100144||特大城市韧性电网供电保障技术与应用北京市重点实验室,北京 100144
能源科技
算电协同数据中心光纤传输约束冷量优化调度服务质量约束优化调度策略
computing-electricity coordinationdata centerfiber-optic transmission constraintsoptimized scheduling of cooling capacityquality of service constraintsoptimal scheduling strategy
《综合智慧能源》 2026 (9)
107-119,13
国家自然科学基金项目(52237008) National Natural Science Foundation of China(52237008)
评论