Artificial Neural Network Based Prediction of Temperature and Flow Profile in Data Centers

Physical and Data-Driven Models for Edge Data Center Cooling System

“…A closer look at the simulation time series for the remaining components 1 show that the data-driven model fails to capture the mode-switching behaviour of the chiller unit. These errors, as well as the cooling tower measurement error with the outside air temperature, appear to propagate to the other components and slowly increase the errors for the entire simulation.…”

“…A state-of-the-art artificial neural network (ANN) model that estimates the cold aisle temperature for individual discon- nected points in time reported a MAE of 0.6°C [1]. However, when used to predict longer time series, the performance deteriorated severely, and the estimation error grew to over 5°C in under 300 seconds.…”

“…The state-of-the-art methods in mathematical modeling of data center cooling system are physical modeling, Computational Fluid Dynamics (CFD) modeling [11,20], and recently also data-driven modeling including neural networks [1,6,8]. CFD models target the air flow and heat transfer in data centers, but usually only for the air interactions in the server room.…”

“…Physical models make broad simplifications in describing the dynamics of the cooling system, such as with the chiller and dry cooler, in for example [3,21], or with the chilled-water heat exchanger [18], compact models that additionally include the thermal mass for room, plenum, walls, floor, ceiling and a water storage tank [9], or use networks of thermal nodes to model the internal dynamics of servers [14]. Moreover, neural networks have been used to directly predict the effectiveness of DC cooling systems [6,16], and more rarely the system thermal responses [1].…”

Performance improvement of an industrial control enclosure cooling system

“…And, the ANN model was constructed and trained based on the part of database from CFD simulation. When ANN model combined with appropriate control strategy, can be used for real-time control [19] To improve the performance of the cooling system and the air flow distribution inside the industrial control enclosure, the CFD simulation has been conducted to analysis the air flow distribution inside the enclosure. And, because few studies discuss the heat transfer performance between upper supply vent arrangement and lower supply vent arrangement for enclosure cooling system, two types of commercialized enclosure cooling systems are implemented to assess different layout design of supply air vent and return air vent of the cooling system in the experiment.…”