Distributionally robust co-optimization of computing workloads and renewable energy uncertainties in geo-distributed data centers considering multi-element influences

TY - JOUR

T1 - Distributionally robust co-optimization of computing workloads and renewable energy uncertainties in geo-distributed data centers considering multi-element influences

AU - Han, Juntao

AU - Tong, Nannan

AU - Lin, Jiayu

AU - Han, Yibo

AU - Wang, Yongzhen

AU - Han, Kai

AU - Li, Yaping

N1 - Publisher Copyright: © 2025 The Author(s)

PY - 2026/1

Y1 - 2026/1

N2 - The paradigm evolution of digital economies, epitomized by artificial intelligence large models, has precipitated an exponential escalation in data centers energy consumption and carbon emissions. However, the collaborative optimization of spatiotemporal load migration and renewable energy uncertainty offers a promising pathway towards low-carbon and economical operations in geo-dispersed data center. Therefore, this study innovatively constructs a spatiotemporal load balancing and energy management collaborative optimization model for geo-distributed data center, integrating renewable energy uncertainty, thermal awareness, and data network transmission. This model is solved by combining the column and constraint generation algorithm with alternating direction method of multipliers (ADMM). Firstly, the generative adversarial network and comprehensive norm constraints are employed to construct renewable energy uncertainties sets and probability distributions. Subsequently, the two-stage distributionally robust optimization integrating renewable energy uncertainty, geographical differences in cooling efficiency, and data network transmission is established. Furthermore, without sharing locally sensitive data, the ADMM algorithm is utilized to optimize the spatiotemporal load balancing and energy management strategies for geo-distributed data center clusters. Case studies demonstrate that, in comparison with scenarios without spatiotemporal load balancing, the proposed approach reduces economic costs, energy purchase, and carbon trading by 13.7%, 26.4%, and 48.4%, respectively, while increasing renewable energy utilization by 2.5%. The study also investigates the impacts of delay-tolerant tasks, geographical differences in cooling efficiency, and data network transmission on system performance. Finally, Nash bargaining is employed to determine computing task transaction pricing and multi-agent benefit allocation, incentivizing data center clusters to participate in workload flexibility regulation.

AB - The paradigm evolution of digital economies, epitomized by artificial intelligence large models, has precipitated an exponential escalation in data centers energy consumption and carbon emissions. However, the collaborative optimization of spatiotemporal load migration and renewable energy uncertainty offers a promising pathway towards low-carbon and economical operations in geo-dispersed data center. Therefore, this study innovatively constructs a spatiotemporal load balancing and energy management collaborative optimization model for geo-distributed data center, integrating renewable energy uncertainty, thermal awareness, and data network transmission. This model is solved by combining the column and constraint generation algorithm with alternating direction method of multipliers (ADMM). Firstly, the generative adversarial network and comprehensive norm constraints are employed to construct renewable energy uncertainties sets and probability distributions. Subsequently, the two-stage distributionally robust optimization integrating renewable energy uncertainty, geographical differences in cooling efficiency, and data network transmission is established. Furthermore, without sharing locally sensitive data, the ADMM algorithm is utilized to optimize the spatiotemporal load balancing and energy management strategies for geo-distributed data center clusters. Case studies demonstrate that, in comparison with scenarios without spatiotemporal load balancing, the proposed approach reduces economic costs, energy purchase, and carbon trading by 13.7%, 26.4%, and 48.4%, respectively, while increasing renewable energy utilization by 2.5%. The study also investigates the impacts of delay-tolerant tasks, geographical differences in cooling efficiency, and data network transmission on system performance. Finally, Nash bargaining is employed to determine computing task transaction pricing and multi-agent benefit allocation, incentivizing data center clusters to participate in workload flexibility regulation.

KW - Geo-distributed data center

KW - Integrated energy system

KW - Renewable energy uncertainty

KW - Spatiotemporal load balancing

KW - Thermal awareness

UR - https://www.scopus.com/pages/publications/105024313604

U2 - 10.1016/j.ecmx.2025.101432

DO - 10.1016/j.ecmx.2025.101432

M3 - Article

AN - SCOPUS:105024313604

SN - 2590-1745

VL - 29

JO - Energy Conversion and Management: X

JF - Energy Conversion and Management: X

M1 - 101432

ER -