Deep Reinforcement Learning based Joint Optimization for Energy-efficient ITE and Cooling Units in Data Centers

Data centers serve as vital infrastructures for modern Internet services worldwide. However, the operation of these extensive facilities comes at a high cost, both in terms of energy consumption and economic expenditure. Unfortunately, previous approaches primarily focused on optimizing the energy consumption of individual subsystems, such as the ITE system or the cooling system, without considering the complex interactions between them. In this paper, we propose a deep reinforcement learning-based approach to enhance the energy efficiency of data centers. Our method simultaneously controls load distribution and cooling set-point temperatures. To achieve this, we develop a comprehensive energy consumption model that accurately captures the energy usage of both the ITE and cooling systems. Furthermore, we introduce a TD3 based reinforcement learning strategy for optimal energy control decisions. We validate our approach through experiments conducted on a simulation platform. The results demonstrate that our method surpasses existing approaches and effectively manages energy savings by controlling both the workload and the cooling system.