Research on optimization of cold source group control strategy in data center

Abstract

Cooling systems play a vital role in maintaining optimal operating conditions in modern data centers (DCs). Efficient control of cold source groups, such as chillers and air handling units, is essential for reducing energy consumption while ensuring temperature stability. This research introduces a novel data-driven approach to optimize the control strategy for cold source groups in DCs by leveraging extensive real-time monitoring data. The control problem is formulated as an energy cost minimization task subject to strict temperature constraints. Addressing these issues, this research proposes an end-to-end group control algorithm based on deep reinforcement learning (DRL). The research suggests a new algorithm called Artificial Gorilla Troops Optimizer-driven Controlled Deep Q-Network (AGTO-CDQN) for dynamically coordinating the operation of multiple cold source units. The research involves collecting both historical and real-time data from DC sensors, including temperature readings, power consumption of cooling units, and server workloads. Experimental results demonstrate that AGTO-CDQN considerably increases the power savings above 15% for IT power consumption, cooling power consumption, total power consumption, and average zone air temperature. These findings highlight the approach’s potential for practical deployment in energy-efficient DC cooling management.