Model-free Adaptively Predictive Control for Data Center Precision Air Conditioner Based on Improved Particle Swarm Optimization

To achieve accurate control for hot aisle temperature of data center and reduce the energy waste caused by loose temperature management. A model-free adaptively predictive control (MFAPC) method based on improved particle swarm optimization (IPSO) for the precision air conditioner of data center was proposed in this paper. Considering the large space of MFAPC controller parameters and the dynamic complexity of the controlled system in a data center, the inertia weight of the PSO algorithm was modified with variant weight. On the basis of this, the pre-exploration and post-exploration capabilities of PSO were significantly improved to obtain the optimal controller parameters. Due to the existence of cold aisle temperature and airflow limitations in the data center, the control value constraint was transformed into a quadratic planning constraint in this paper. Besides, the optimal parameters for each control step of MFAPC controller were obtained based on IPSO algorithm. By combining the control value constraint and parameter optimization, the MFAPC output of each control step was optimal in the current system state. Finally, the efficiency of the proposed method was validated by a hot aisle temperature prediction model that was built by actual data based on the data center in Beijing. Compared with the conventional MFAPC controller, the proposed IPSO-MFAPC algorithm with control value constraint showed superior performance in terms of overall control error, overshoot, and rapidity. The control result shows that the IPSO-MFAPC method is able to implement accurate control for hot aisle temperature in data centers.