MPC-based joint optimization for rack-based cooling data centers: Modeling, performance evaluation, and time-delay characteristic analysis

Abstract

To meet the high heat dissipation requirements of IT equipment, the cooling system requires continuous operation in data centers. However, lack of efficient control strategies poses a dual challenge of high energy consumption and risk of sever downtime due to deviation from thermal environment. This paper proposed an innovative joint optimization method based on model predictive control (MPC-based) for rack-based cooling data centers, which can realize collaborative optimization of cooling system operating parameters and server workload scheduling in rack level. Dynamic heat transfer models were developed to represent the thermal inertia of cooling system and IT equipment, predicting the spatiotemporal variation of thermal environment. The primary objectives are to minimize the energy consumption of the cooling system, and ensure effective control of the thermal environment. The energy management and thermal management performance of the proposed method was comprehensively evaluated through simulation and comparative experiments. The impact of time-delay characteristics on the predictive control performance were first analyzed with different prediction horizons in rack-based cooling data centers. The results show that the proposed joint optimization method has significant advantages in maintaining temperature stability and uniformity, and the energy saving of the cooling system is up to 19.39 %. Further analysis of time-delay characteristics reveals that the thermal inertia of the cooling system and IT equipment affects the control performance. With the extension of the prediction horizon, up to 3.26 % of energy consumption can be further saved for the data center.