AlphaDataCenterCooling: A virtual testbed for evaluating operational strategies in data center cooling plants

The significant energy consumption and carbon emissions of data centers, representing 1 % of global energy output and contributing to 2–4 % of global carbon emissions, underscore the urgent need for optimizing energy use and mitigating carbon emissions. To tackle this issue, in this study, we developed an open-sourced versatile virtual testbed named AlphaDataCenterCooling, which was validated using historical operational data from a real data center cooling plant. This virtual testbed is specifically designed to enhance the overall efficiency and sustainability of data center operations by benchmarking different cooling plant control strategies, thereby facilitating the improvement of energy efficiency and the reduction of carbon emissions. The high-fidelity Modelica model of the cooling plant within the developed virtual testbed supports cross-platform (Python, MATLAB) co-simulation. After being validated by replicating three months of continuous historical operational conditions, the model demonstrated a mean absolute percentage error (MAPE) in power prediction of only 7.62 %. We then conducted experiments on the developed testbed to check the AlphaDataCenterCooling's ease of use and high scalability. The results demonstrated that the developed virtual testbed can effectively serve as a digital twin for data center cooling plants and control systems. This capability allows for the testing of control algorithms without the need for field deployment and shows significant potential for applications in fault diagnosis, predictive maintenance, and the development and testing of advanced control algorithms.