The influence of uncertainty parameters on the double-layer model predictive control strategy for the collaborative operation of chiller and cold storage tank in data center

The inefficient operating states and parameter settings are important factors affecting the high energy consumption of data center cooling systems. The double-layer model predictive control (DMPC) strategy is developed to optimize the operating states and parameter settings of cooling system. The theoretical performance of DMPC strategy is impressive. Meanwhile, the control performance is deeply affected by parameter uncertainties in the model, sensors, and actuator in practical application. Therefore, the influence of uncertain parameters on performance of DMPC strategy is quantified, and the key uncertain parameters (including the wet-bulb temperature, approaching temperature of cooling tower, cooling capacity of water tank, COP, water flow rates of cooling water pump and chilled water pump) are identified. The results show that the DMPC strategy can prevent the chillers from operating at very low loads and reduce the 24-h PUE by more than 0.01. However, compared to traditional control strategy, the DMPC strategy is more sensitive to uncertain parameters. Parameter uncertainty leads to 70 % mode prediction error rate of DMPC strategy. In mechanical cooling mode, the DMPC strategy is more sensitive to COP than approaching temperature and wet-bulb temperature. However, in the hybrid cooling mode, the sensitivity of the three parameters is opposite to mechanical cooling mode. In free cooling mode, the DMPC strategy is most sensitive to water pump flow rates. Therefore, in order to reduce the computational and economic costs of hybrid cooling systems using the DMPC strategy in data center, more efforts should be placed on reducing the uncertainties of key parameters in initial investment and modeling.