摘要
随着人工智能对算力需求的激增,数据中心(internet data center,IDC)作为数据处理与存储的机构,其能耗需求远超预期,使用新能源是其可持续发展的需要。然而,可再生能源具有出力不确定性,仅依靠数据中心参与需求响应难以实现消纳,可配置储能提高系统灵活性。因此,本工作建立了以规划总成本最优为目标的数据中心与电池储能(battery energy storage,BES)协同规划两阶段鲁棒模型,为防止规划结果过于乐观,引入了储能寿命约束。同时针对在求解所建模型过程中,传统C&CG(column-and-constraint generation)算法存在难以平衡求解速度与精度间关系的问题,本工作提出了一种不精确列和生成约束算法i-C&CG(inexact column-and-constraint generation)进行求解。基于IEEE30节点与IEEE118节点算例系统进行优化解算,仿真结果表明,与仅配置单一储能相比,本工作所提模型储能年等效建设成本下降39785元,数据中心年等效建设成本下降289080元;且本工作所提算法与传统C&CG相比,采用0.18低精度下的i-C&CG,与采用0.16较高精度的C&CG相比较,i-C&CG最多可缩短3632 s的单次迭代求解所需时间,且最终计算结果的相对误差为0.46%,两者收敛间隙与相对最优间隙近似。
The skyrocketing demand for AI computing has fueled a surge in energy consumption within internet data centers(IDCs),surpassing expectations.However,the unpredictable nature of renewable energy poses significant challenges for stable grid operation,which demand response from IDCs alone cannot fully solve.Configurable energy storage enhances flexibility.This paper presents a two-stage robust model for IDCs and battery energy storage(BES)planning,minimizing operational costs amidst wind power uncertainty.A lifespan constraint ensures realistic planning.The traditional C&CG algorithm's speed-accuracy dilemma is tackled with an Inexact C&CG(i-C&CG)algorithm,avoiding exact column/constraint generation.Simulations on IEEE 30-node and 118-node systems highlight the benefits of this approach.The proposed energy storage configuration reduces annual costs by 39785 CNY for storage systems and 289080 CNY for IDCs.The i-C&CG algorithm shortens iteration time by 3632 seconds,achieving a 0.18 precision and 0.46 relative error.These results match the traditional C&CG's convergence and optimality gaps.
作者
王述祯
WANG Shuzhen(Hubei University of Technology,Wuhan 430068,Hubei,China)
出处
《储能科学与技术》
北大核心
2025年第2期671-687,共17页
Energy Storage Science and Technology
关键词
数据中心
储能寿命
不精确列和约束生成算法
internet data centers
energy storage life
inexact column-and-constraint generation
