Cooling Control of Data Centers Using Linear Quadratic Regulators
Winston García-Gabín, Kateryna Mishchenko, Erik Berglund · 2018
One of the largest contributions to a data center's power usage is its cooling system. One way to decrease the energy usage of a cooling systems by introducing an automatic control adapting the capacity of cooling units is addressed by this paper. Firstly, different configurations of linear quadratic regulators are designed and then evaluated using the nonlinear data center simulator. Design of the controllers is based on the novel idea of regulating the outlet temperature and volumetric airflow of the Computer Room Air Handler units which regulate the cooling power supplied to the servers. As controlled variables for the feedback, maximum and mean temperature of servers were used. Evaluation of controllers performance is based on their estimated energy usage as well as on analysis of the servers temperature profiles with respect to the given data centers temperature threshold value. Numerical experiments include five LQR controllers with different combination of manipulated variables and analyze their performances. Results of the experiments show that the regulator with maximum servers temperature as a feedback and Computer Room Air Handler Output temperature as manipulated variable has the best performance. Controlling the maximum server temperature minimizes hot spots occurrences and allows to operate the cooling system with minimal energy consumption.