首页|期刊导航|综合智慧能源|新疆绿色算力园区多时间尺度低碳调度

新疆绿色算力园区多时间尺度低碳调度OA

Multi-time scale low-carbon dispatch for Xinjiang green computing parks

中文摘要英文摘要

针对新疆新能源基地配套绿色算力园区中风光出力波动、算力任务时限、储能跨时段耦合及外购电碳排放强度变化引起的协同调度问题,提出多时间尺度低碳调度方法.区别于将柔性算力等效为连续可转移负荷的传统处理方式,建立包含任务释放时刻、截止时刻、能量需求、单任务功率上限、聚合算力上限和未完成能量的任务队列模型,用以刻画算力负荷的时间可转移特性.采用随环境温度变化的电能使用效率表征数据中心制冷及辅助能耗,并建立储能荷电状态转移、充放电互斥、公共并网点容量和购电功率爬坡约束,将各项费用与损失统一折算为人民币,构建均值-条件风险价值混合整数线性优化模型.日前层基于未来24 h预测信息形成96个时段的基准计划;日内层每4h更新风光、电价、碳排放因子、储能状态和剩余任务量并滚动优化;实时层以15 min为周期,根据实际状态修正任务、储能和购电功率.以新疆风光基地配套绿色算力园区参数化典型日的数据为例,设置固定任务无储能、任务队列优化、固定任务配置储能和多时间尺度风险协同4类方案.结果表明:相较于固定任务无储能方案,所提方案运行成本下降61.36%,弃电率由9.92%降至0,可再生能源利用率由90.08%提高至100%,计入等值投资后综合成本为38 709.43元,较基准方案下降4.53%;在400组独立样本外场景中,风险方案的成本、碳排放和公共并网点峰值CVaR95分别下降1.62%,3.55%和1.68%,但单一最大峰值略有上升,说明该方法改善的是最不利场景的尾部平均风险,而不能保证每个极端场景均占优.研究结果表明,柔性任务队列与储能协同能够提高新能源就地供能的能力,间接完成碳减排,兼顾园区经济运行、配电网削峰和算力服务质量.

Green computing power parks integrated with renewable energy bases in Xinjiang face coordinated scheduling issues,which are caused by fluctuations in wind and photovoltaic power output,computing task deadlines,inter-temporal coupling of energy storage,and variations in the carbon emission intensity from purchased electricity.To address these issues,a multi-time scale low-carbon scheduling method covering day-ahead,intraday,and real-time operation was proposed.Unlike conventional approaches that represent flexible computing power as continuously shiftable loads,a task queue model was established,incorporating the task release time,deadline,energy demand,upper limit of single-task power,upper limit of aggregate computing power,and unfinished energy,so as to characterize the temporal shiftability of computing loads.The energy usage effectiveness varying with ambient temperature was adopted to represent cooling and auxiliary energy consumption in data centers.Constraints on energy storage state-of-charge transitions,mutual exclusivity of charging and discharging,capacity at the point of common coupling,and purchased-power ramping were established.All costs and losses were converted to Chinese yuan,and a mean-conditional value-at-risk mixed integer linear optimization model was developed.At the day-ahead level,a baseline scheme with 96 time intervals was formed based on the prediction information for the next 24 h.At the intra-day level,wind and photovoltaic power,electricity prices,carbon emission factors,energy storage states,and remaining tasks were updated every 4 h,followed by rolling optimization.At the real-time level,task,energy storage,and purchased power were adjusted every 15 min according to actual operating conditions.A parameterized typical day for a green computing power park associated with a wind and photovoltaic power base in Xinjiang was used as a case study,with four schemes considered:fixed tasks without energy storage,task queue optimization,fixed tasks with energy storage,and multi-timescale risk coordination.The results showed that,compared with the fixed-task scheme without energy storage,the proposed scheme reduced operating cost by 61.36%,decreased the curtailment rate from 9.92%to 0,and increased renewable energy utilization from 90.08%to 100%.After equivalent investment was included,the comprehensive cost was 38 709.43 yuan,representing a 4.53%reduction compared with the baseline scheme.Across 400 independent out-of-sample scenarios,the CVaR95 values of cost,carbon emissions,and peak power at the point of common coupling under the risk-based scheme decreased by 1.62%,3.55%,and 1.68%,respectively.However,the single maximum peak value increased slightly,indicating that the proposed method improves the average tail risk in the most adverse scenarios,but does not guarantee superiority in every extreme scenario.The results indicate that coordination between flexible task queues and energy storage can enhance the capacity for local supply from renewable energy,indirectly achieve carbon emission reduction,and balance the economic operation of the park,peak shaving of the distribution network,and the quality of computing services.

张志豪;王永真;樊小朝;王子龙;史瑞静;孙剑桥;王维庆;何山;程志江;程静

新疆大学 智能科学与技术学院,乌鲁木齐 830017北京理工大学 能源与动力工程系,北京 100081新疆工程学院 能源工程学院,乌鲁木齐 830023上海第二工业大学,上海 201209新疆大学 智能科学与技术学院,乌鲁木齐 830017新疆大学 智能科学与技术学院,乌鲁木齐 830017新疆大学 可再生能源发电与并网技术教育部工程研究中心,乌鲁木齐 830017||新疆大学 电气工程学院,乌鲁木齐 830017新疆大学 电气工程学院,乌鲁木齐 830017新疆大学 智能科学与技术学院,乌鲁木齐 830017||新疆大学 电气工程学院,乌鲁木齐 830017新疆大学 电气工程学院,乌鲁木齐 830017

能源科技

绿色算力园区柔性任务队列多时间尺度调度碳排放因子储能条件风险价值算电协同

green computing parkflexible task queuemulti-time-scale dispatchcarbon emission factorenergy storageconditional value-at-riskcomputing-electricity coordination

《综合智慧能源》 2026 (8)

96-106,11

国家自然科学基金项目(72595835,52266018)National Natural Science Foundation of China(72595835,(52266018)

10.3969/j.issn.2097-0706.2026.08.008

评论