Capacity optimization of data center microgrid coupling computing-power and electricity consumption

The optimization of capacity configuration in data center microgrids, primarily powered by renewable energy, represents a crucial solution for achieving low-carbon computing capabilities. However, the current configuration faces challenges in balance demand matching with cost-effectiveness. These challenges primarily arise from uncertainties in dynamic computational loads and device outputs. Furthermore, factors such as construction costs, operational expenses, and associated revenues impose constraints on these systems. To address these issues, we propose the Electricity-Computing Coupling Model with Data Delay Constraints (ECCM-DDC). By incorporating data delay tolerance constraints into this model, we enhance the efficiency of matching demand and capacity. Initially, we developed an electro-computational coupled power consumption model and an electro-computational coupled cost model applicable to data centers. These models consider batch processing delays and computational power revenue. Building on this foundation, we have created a two-stage robust optimization model aimed at maximizing economic benefits by thoroughly accounting for both depreciation and operational costs of equipment. The model employs the algorithm for its solution, thereby optimizing the identification of the most cost-effective capacity configuration strategy. Finally, the effectiveness of the proposed method is validated through simulation experiments constructed using actual operational data from the National Supercomputing Center Wuxi Qinghai University Branch. The research findings indicate that the proposed capacity allocation method is well-suited for the construction of data center microgrids, effectively balancing cost control with service benefits. Specifically, this method reduces operational costs by 8%, daily equipment investment costs by 16%, and the configured capacity of energy storage devices by 58%. Furthermore, by taking data latency constraints into account, power fluctuations are minimized by 38%.