Analyzing the saving potential of datacenters via model-based predictive control.

Data centers account for 1 to 1.13% of the world's total electricity consumption and are responsible for 1% of energy-related emissions. The International Energy Agency projects a rapid increase in electricity consumption by datacenters in the U.S., from 200 TWh to 260 TWh by 2026. In China and the European Union, the energy demand for data centers is set to reach 300 TWh and 150 TWh by 2026. Moreover, electricity consumption in countries with growing datacenter markets is increasing rapidly. Therefore, reducing energy consumption is urgently required.Traditional approaches i.e., free cooling, optimizing the airflow management and operating parameters of air conditioning systems, have remained the most sought-after technique to reduce energy consumption. However, in recent years advanced control strategies such as model predictive control (MPC) has proven to be the most promising strategy for efficient control, providing significant energy savings and ensuring safe operation. MPC has been applied to various residential and commercial buildings for different purposes. However, MPC utilizing the thermal mass of the building has not been investigated for datacenters.The traditional approach to predicting the indoor environment of data centers uses computational fluid dynamics simulations. While this method provides accurate results, it is time-consuming and computationally expensive, making it impractical for control studies. Alternatively, a data-driven modeling approach has been proposed for predicting indoor conditions. However, these models often lose their accuracy when encountering the strong nonlinearity and uncertainties of complex systems and require large datasets to ensure robust and generalized performance. Grey-box models that are considered more accurate and robust than traditional approaches have not been discussed for datacenters.Motivated by the problems, a predictive control approach was investigated using MPC in a datacenter with varying temperature bounds depending on the type of IT equipment. The grey-box model structure was utilized in predicting the thermal behavior of the datacenter. MPC simulation was conducted using a typical optimizer, specifically linear programming assuming a linear time-invariant model, to achieve an exact solution. An average saving percentage of 25.3% was achieved using the MPC with grey-box model over ideal feedback control.