Impact of uncertainties on the supervisory control performance of a hybrid cooling system in data center

Model-based optimal control (MBOC) is a promising method to reap the potential of the waterside free cooling system in data center. While the theoretical energy savings are impressive, uncertainties in the model, sensors, and actuators could hinder this control strategy's practical application. This study quantified the impact of various uncertainties on the supervisory control method in two steps. First, sensitivity analysis was conducted using the Morris method and the one-at-a-time method to identify the influential impact of uncertain elements. Next, Monte-Carlo simulation was designed to perform an uncertainty study and evaluate the robustness and energy efficiency of the control method. A virtual emulator for simulating the proposed novel hybrid cooling system in data centers in combination with different supervisory control methods was developed. Full-year simulations indicate that the impact of uncertainty on the control performance of the MBOC strategy is greater than that of conventional control strategy; under various uncertainties, the energy consumption and operation mode prediction error rate of the MBOC method were increased by 43.6% and 99.2%, respectively. The research suggests that, if MBOC is adopted for the hybrid cooling system control, more efforts should be placed on reducing the uncertainties from various sources.