基于主从博弈的多数据中心与电网协同优化运行方法OA
Stackelberg game-based coordinated optimal operation method for multiple data centers and power grids
随着"东数西算"工程与新型电力系统建设的推进,数据中心已成为电力系统中快速增长的高耗能负荷,其灵活调节能力有望为电网提供低成本的柔性响应资源.然而,现有研究未能充分考虑电网运营商与多个数据中心间的利益博弈及多主体协同中的数据隐私保护问题,制约了理论成果的工程化应用.为此,提出了一种融合主从博弈与自适应交替方向乘子法(ADMM)的双层分布式优化方法.首先,构建了融合IT设备功耗、动态电能利用效率(PUE)冷却模型、排队论任务时延与跨中心迁移的多数据中心精细化灵活性模型,从时间与空间2个维度刻画数据中心的可调节潜力.其次,建立以配电系统运营商为上层领导者、各数据中心为下层跟随者的主从博弈双层优化模型——上层以系统总运行成本最小为目标制定分时激励电价,下层据此自主优化用电方案;利用KKT条件与互补松弛线性化,将双层问题转化为单层混合整数二次规划.在此基础上,设计了ADMM分布式求解框架,并采用基于原始/对偶残差比的自适应惩罚参数策略加速收敛;各主体仅需交换聚合响应量,无需共享任务队列、服务器配置等核心运营数据.基于改进IEEE 33节点系统的仿真结果表明:所提算法经34次迭代收敛,较固定惩罚参数方案提速74.40%,同时避免了对偶残差振荡风险;其系统总运行成本较基线方案降低18.3%,净负荷峰谷差缩小11.2%;与集中式优化相比,总成本仅增加约10.5%.灵敏度分析表明,当数据中心数量从3增至20时,单次迭代计算耗时从0.49 s增至14.46 s,呈近线性增长.所提方法在经济性、收敛性能与隐私保护之间实现了有效平衡:各数据中心仅需上报聚合响应量即可完成优化迭代,无需披露服务器配置、任务队列、储能SOC等内部运行数据,有效保护了各主体数据隐私.
With the advancement of the"East Data,West Computing"project and the accelerated construction of new power systems,data centers have become rapidly growing energy-intensive loads in power systems.Their flexible regulation capability is expected to provide low-cost flexible response resources for the power grid.However,existing research has not fully addressed two critical challenges:the game of interests between the grid operator and multiple data centers,and data privacy protection in multi-agent collaboration,thereby hindering the engineering application of relevant theoretical results.To address these challenges,a bi-level distributed optimization method integrating Stackelberg game theory with an adaptive alternating direction method of multipliers(ADMM)was proposed.A high-fidelity multi-data-center flexibility model was established by integrating IT equipment power consumption,dynamic power usage effectiveness(PUE)cooling models,queueing-theoretic task delay,and cross-data-center workload migration,thereby characterizing the adjustable potential of data centers across both spatial and temporal dimensions.A Stackelberg game-based bi-level optimization model was established,with the distribution system operator(DSO)as the upper-level leader and each data center as the lower-level follower.The upper level aimed to minimize the total system operating cost by setting time-of-use incentive electricity prices,while the lower level autonomously optimized electricity consumption strategies accordingly.The bi-level model was transformed into a single-level mixed-integer quadratic programming by applying Karush-Kuhn-Tucker(KKT)conditions and complementary slackness linearization.On this basis,an ADMM-based distributed solution framework was designed.An adaptive penalty parameter strategy based on the primal to dual residual ratio was adopted to accelerate convergence.Each agent was required only to exchange aggregated response quantities,without sharing task queues,server configurations,or other core operational data.Simulation results based on a modified IEEE 33-bus distribution test system demonstrated that the proposed algorithm converged after 34 iterations,accelerating convergence by 74.40%compared with a fixed penalty parameter scheme,while avoiding the risk of dual residual oscillations.The proposed method reduced total system operating cost by 18.3%and narrowed the net load peak-valley difference by 11.2%compared with the baseline scheme.Compared with the centralized optimization,the proposed method incurred a cost increase of only approximately 10.5%.Sensitivity analysis showed that as the number of data centers increased from 3 to 20,the computation time per iteration increased from 0.49 s to14.46 s,exhibiting a near-linear growth trend.The proposed method achieves an effective balance among economic efficiency,convergence performance,and data privacy protection.Each data center needs to report only aggregated regulation response quantities to complete the optimization iterations,thereby effectively preserving the data privacy of each agent.
张昌林;范鹏飞;韩建沛
辽宁工程技术大学 电气与控制工程学院,辽宁 葫芦岛 125105辽宁工程技术大学 电气与控制工程学院,辽宁 葫芦岛 125105北方工业大学 储能科学与工程学院,北京 100144
能源科技
数据中心主从博弈算电协同交替方向乘子法分布式优化配电系统运营商数据隐私保护
data centerStackelberg gamecomputing-electricity coordinationalternating direction method of multipliersdistributed optimizationdistribution system operatordata privacy protection
《综合智慧能源》 2026 (8)
107-116,10
国家自然科学基金项目(52507116)北京市自然科学基金项目(3254046)National Natural Science Foundation of China(52507116)Beijing Natural Science Foundation(3254046)
评论