Deep Reinforcement Learning for Energy-Efficient Control of Multi-Chiller Systems in Data Centers

In the era of AI and Big Data, the Information and Communications Technology (ICT) sector has witnessed rapid growth, with data centers serving as a critical backbone for processing and storing vast amounts of information. Cooling systems are among the most energy-intensive subsystems in data centers, making their efficient management essential to reduce operational costs and improve sustainability. This study investigates the application of Deep Reinforcement Learning (Deep RL) for the optimal control of chiller systems, which cool a secondary fluid (e.g., a water-glycol solution) to dissipate excess heat. The proposed Deep RL based control strategy aims to minimize energy consumption while ensuring that the required thermal load is maintained under varying operating conditions. A Python-based simulation environment, modeling the key dynamics of a multi-chiller system, is used for agent training and performance evaluation. Simulation results demonstrate that Deep RL has significant potential to enhance the energy efficiency of data center cooling, supporting the development of more sustainable ICT infrastructures.