Data-Driven Temperature Control Strategy for Data Centers Based on Multi-Layer Perceptron Model

Under high load conditions, thermal-control data in a data center changes rapidly, placing stringent requirements on computational timeliness. Although traditional computational fluid dynamics (CFD) methods can accurately simulate temperature fields, their solution process is slow and yields only discrete numerical results, making real-time control difficult to achieve. To address this challenge, this paper proposes a data-driven thermal control strategy based on a multilayer perception (MLP) model, designed to improve computational efficiency while maintaining control accuracy. We construct a simulated thermal-control environment for the data center and generate training data using 6SigmaRoom software. The MLP model quickly learns the nonlinear relationship between cabinet temperatures and control parameters to predict optimal temperature setpoints and adjust baffle angles. Experimental results show that, compared with other machine learning methods, our model significantly increases computation speed and optimizes energy efficiency without sacrificing accuracy. This method provides a highly efficient and scalable optimization framework for large-scale data centers, overcoming the limitations of traditional CFD and offering significant theoretical and practical value.