Data-driven Adaptive Control of CRAC in Data Center Based on Online Incremental RVFL

An adaptive control method based on online identification of controller parameters is proposed for the cooling process of Computer Room Air Conditioning(CRAC) in data centers with complex nonlinear and dynamic characteristics. Firstly, the nonlinear system is divided into low-order linear part and high-order unmodeled dynamic term. The random vector functional link network is used to identify the controller parameters. The unmodeled dynamic compensator is introduced into the control strategy to offset the influence of uncertain dynamic term on the closed-loop control system of CRAC-Cabinet system. Then, in order to solve the problem that the identification ability of the offline model is reduced due to the changes of the location layout of the cabinet, the number and placement of electronic devices and other factors in the data center, the online addition of hidden layer nodes is added, and the output weight of the original hidden layer is recalculated and updated when new nodes are added. The simulation results show that the control and tracking ability of the nonlinear adaptive controller with unmodeled dynamic compensation is better than that of the adaptive controller based on linear model, and the experimental results show that the incremental learning network model has faster response speed and higher tracking accuracy than the original model.