基于扩散模型的数据中心虚拟电厂分布鲁棒优化调度策略

Robust Optimal Scheduling Strategy for Data Center Virtual Power Plant Distribution Based on Diffusion Model

摘要: 具有强不确定性的可再生能源并网给包含数据中心的虚拟电厂的安全运行带来了巨大的挑战。为此，本文提出一种基于扩散模型的数据中心虚拟电厂分布鲁棒优化调度模型。首先，为了更准确地刻画光伏的不确定性，采用扩散模型对辐照度–温度联合场景数据进行泛化处理，提高光伏出力数据的准确性和多样性。然后，通过高斯混合聚类模型对光伏出力数据进行削减，并将得到的场景作为分布鲁棒优化集合的初始场景。再次，为进一步挖掘数据中心的需求响应潜力，建立了包含可转移批处理负载的数据中心虚拟电厂的数学模型。最后，为了求解min-max-min分布鲁棒优化问题，通过列和约束生成算法进行求解。仿真结果表明，本文所提出的基于扩散模型的光伏出力场景生成方法能更准确、更有效地刻画光伏的不确定性。此外，本文所提的分布鲁棒优化模型能平衡经济运行和鲁棒性之间的关系。

Abstract: The integration of highly uncertain renewable energy into the grid poses significant challenges to the secure operation of virtual power plants (VPPs) that include data centers. To address this issue, this paper proposes a distributionally robust optimization (DRO) scheduling model for data center VPPs based on a diffusion model. First, to more accurately characterize the uncertainty of photovoltaic (PV) power generation, a diffusion model is employed to generalize irradiance-temperature joint scenario data, enhancing both the accuracy and diversity of PV output data. Then, a Gaussian mixture clustering model is applied to reduce the dimensionality of PV output data, and the resulting scenarios are used as the initial set for DRO. Furthermore, to further explore the demand response potential of data centers, a mathematical model for data center VPPs incorporating transferable batch processing loads is developed. Finally, the min-max-min DRO problem is solved using the column-and-constraint generation algorithm. Simulation results demonstrate that the proposed diffusion model-based PV output scenario generation method in this paper can more accurately and effectively capture the uncertainty of PV. Additionally, the proposed DRO model in this paper can effectively balance the relationship between economic operation and robustness.