Nonlinear Model Predictive Control Based on Nonlinear Autoregressive Exogenous Model for Energy Efficient Heat Removal Module in Data Centers

The cooling of information technology (IT) equipment is the single largest energy overhead in data center operation. The design and strategy of thermal management solution in the data center is crucially important. The Heat Removal Module (HRM) developed by KoolLogix Pte Ltd, which is attached right behind to absorb the direct heat from racks, is a gravity-driven, refrigerant-based and passive heat removal green data centre cooling solution. Besides the innovative thermal solution, the automation control for the operation also plays significant role to obtain the energy efficient. In this paper, we introduce the a multiple inputs and multiple outputs (MIMO) nonlinear model predictive control (NMPC) system based on data-driven nonlinear autoregressive exogenous model (NARX) for HRM in data center. The NARX model is validated with the measurement data for different operational conditions with a relative error less than 5%. The control performance is tested in-silico for different heat power loads. The results indicate that the developed NMPC controller can stably bring the cooling system reaching to a new temperature set-point just in 8.0 minutes, while the conventional Proportional-Integral-Derivative (PID) controller is unable to do so after 75 minutes. Furthermore, the MIMO strategy allows to control the racks individually which is efficient in uneven heat powers scenario.