Data-driven distributionally robust optimization of low-carbon data center energy systems considering multi-task response and renewable energy uncertainty

The exponential growth in demand for computing power has resulted in a rapid expansion of energy consumption and CO₂ emissions from data centers. Consequently, the full utilization of renewable energy sources is regarded as the most effective strategy for data centers to achieve near-zero carbon emissions. However, due to the mismatch between the intermittency of renewable energy and the time-varying workloads. Data centers still face challenges in integrating renewable energy and exploiting the regulation potential of computing tasks. Therefore, this study proposes a novel multi-featured collaborative optimization framework for low-carbon data center integrated energy systems (DCIES) that integrates task scheduling mechanism, renewable energy uncertainty and hybrid cooling. Firstly, the renewable energy scenario generation is based on the generative adversarial network with gradient penalty. The two-stage distributionally robust optimization model for DCIES based on data-driven uncertainty set is established to address the renewable energy uncertainty. Secondly, this study exploits the flexibility regulation potential of data center by formulating the workload scheduling mechanism for multiple tasks with different server execution times and delay-tolerant times. The results reveal that the DCIES collaborative optimization scheme, which integrates the task scheduling mechanism, renewable energy uncertainty and hybrid cooling, could reduce the total cost and CO₂ emissions by 23.2 % and 28.4 %, respectively, while reducing the renewable energy curtailment by 7.4 %. This multi-featured collaborative optimization of data center computing electricity and thermal provides a novel approach to the low-carbon and sustainable development of data center buildings.