Scholarlink

<p>With the rapid expansion of digital services, exemplified by large-scale AI applications like ChatGPT and DeepSeek, data centers’ soaring electricity consumption poses significant threats to grid stability and energy efficiency, while the separation of computation resource allocation and electrical load management further worsens the situation. This study develops a two-stage distributed robust optimization (TDRO) model for computation-electricity co-scheduling in data center integrated energy systems, where the first stage minimizes operational costs based on predictive information, and the second stage constructs ambiguity sets for output errors using data-driven Wasserstein distance. An affine policy is employed to adjust the model, ensuring minimal adjustment costs under the worst-case distribution within the ambiguity sets. The model is further transformed into an equivalent mixed-integer linear programming problem via strong duality principle. Numerical results demonstrate that the proposed approach achieves a 16.6% reduction in total operational costs compared to traditional methods that neglect both renewable uncertainties and load flexibility. Specifically, grid-purchasing costs decrease by 19.5%, fuel costs by 19.0%, and renewable energy curtailment costs by 39.8%. This framework offers a practical tool for sustainable data center operation with enhanced resilience against uncertainties.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Flexible batch loads as “virtual batteries”: distributionally robust optimization for computation-electricity co-scheduling in data center integrated energy system

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Yunshou Mao,
  • Xianan Jiao

摘要

With the rapid expansion of digital services, exemplified by large-scale AI applications like ChatGPT and DeepSeek, data centers’ soaring electricity consumption poses significant threats to grid stability and energy efficiency, while the separation of computation resource allocation and electrical load management further worsens the situation. This study develops a two-stage distributed robust optimization (TDRO) model for computation-electricity co-scheduling in data center integrated energy systems, where the first stage minimizes operational costs based on predictive information, and the second stage constructs ambiguity sets for output errors using data-driven Wasserstein distance. An affine policy is employed to adjust the model, ensuring minimal adjustment costs under the worst-case distribution within the ambiguity sets. The model is further transformed into an equivalent mixed-integer linear programming problem via strong duality principle. Numerical results demonstrate that the proposed approach achieves a 16.6% reduction in total operational costs compared to traditional methods that neglect both renewable uncertainties and load flexibility. Specifically, grid-purchasing costs decrease by 19.5%, fuel costs by 19.0%, and renewable energy curtailment costs by 39.8%. This framework offers a practical tool for sustainable data center operation with enhanced resilience against uncertainties.