Dynamic control method for wind-liquid collaborative cooling tail end of data center under load fluctuation

By constructing a computational fluid dynamics simulation model and a deep reinforcement learning framework for data centers, and training a dynamic control agent for air-liquid co-cooling terminals, the problems of adjustment lag and insufficient safety of data center cooling terminals under dynamic loads are solved, achieving real-time intelligent control and energy efficiency improvement.

CN121920243AActive Publication Date: 2026-04-24INSPUR TIANYUAN COMM INFORMATION SYST CO LTD
8 Cites 1 Cited by

Patent Information

Application Number
CN202610369524.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-25
Publication Date
2026-04-24
Estimated Expiration
2046-03-25

AI Technical Summary

Technical Problem

Existing data center cooling control methods suffer from problems such as adjustment lag, low energy efficiency, and insufficient safety when facing dynamic loads. In particular, traditional PID control is difficult to handle multivariable problems, MPC is difficult to accurately predict load fluctuations in real-world environments, and deep reinforcement learning may cause thermal risks in real-world environments. There is a lack of integrated control strategies suitable for air-liquid co-cooling terminals.

Method used

We construct computational fluid dynamics simulation models at the data center and server levels, generate time-series datasets, train time-series agent models, design a deep reinforcement learning framework, combine state space, action space, and reward function to train a dynamic control agent for the air-liquid co-cooling terminal, and achieve real-time control through online iterative optimization.

Benefits of technology

It enables real-time intelligent control of the air-liquid cooling ratio, significantly improving data center energy efficiency, ensuring the thermal safety of servers under dynamic loads, reducing the overall energy consumption of air conditioning fans and pumping systems, and providing absolute thermal safety assurance.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a dynamic control method for a wind-liquid collaborative cooling tail end of a data center under load fluctuation, and relates to the technical field of cooling control of the data center. In order to solve the problem of adjustment lag when a cooling tail end face of a current data center faces a dynamic load, the adopted scheme comprises the following steps: constructing machine room-level and server-level computational fluid mechanics simulation models, and generating a time sequence data set for training an agent model through dynamic simulation; respectively training a machine room-level time sequence agent model and a server-level time sequence agent model based on the time sequence data set; coupling the machine room-level and server-level time sequence agent models and the data center energy consumption model, and constructing an integrated simulation environment; designing a deep reinforcement learning framework; based on an integrated simulation environment and a deep reinforcement learning framework, training a wind-liquid collaborative cooling tail end dynamic control agent; and performing safe deployment on the trained intelligent agent to realize dynamic control of the cooling tail end, and performing online iterative optimization on the time sequence agent model and the intelligent agent according to actually acquired data.
Need to check novelty before this filing date? Find Prior Art