AI-driven cooling optimization in data centers: Reinforcement learning for dynamic workload placement and HVAC control
Year: 2025
Type: article
Abstract: — The rapid expansion of data centers has led to unprecedented energy demands, with cooling systems accounting for a significant portion of overall power consumption. Traditional rule-based methods for workload placement and HVAC (Heating, Ventilation, and Air Conditioning) management often fail to adapt dynamically to fluctuating workloads and thermal profiles, leading to inefficiencies and increased operational costs. This paper proposes an AI-driven framework leveraging reinforcement learning (RL) to jointly optimize workload distribution across servers and fine-tune cooling parameters in real time. By modeling the data center environment as a dynamic system, RL agents learn adaptive policies that minimize power usage effectiveness (PUE) while ensuring service-level agreement (SLA) compliance. Experimental evaluations using simulation-based workload traces demonstrate that the proposed RL-based optimization significantly reduces cooling energy consumption compared to heuristic and static policies, while also improving thermal stability across server racks. The study highlights the potential of hierarchical or multi-agent RL architectures to balance competing objectives such as energy efficiency, workload latency, and operational reliability. This research contributes to sustainable data center management by advancing the integration of intelligent workload scheduling with HVAC control, paving the way for greener large-scale computing infrastructures. Keywords— Reinforcement learning, Data center cooling, Workload placement, HVAC optimization, Energy efficiency, Sustainable computing
Author: Deepak Tomar , Kismat Chhillar , Saurabh Shrivastava , Alok Verma
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7083005727
FWCI:
Cites:
Cited by:
Related to: