The two-phase closed thermosyphon is widely used in electronic cooling, energy conversion, and other fields due to its efficient heat transfer characteristics. However, its internal phase-change heat transfer process is complex, involving liquid film evaporation, condensation, and two-phase gas–liquid flow. Traditional experimental methods are limited by measurement accuracy and cost, making it difficult to effectively study key variables such as the temperature field. Computational fluid dynamics (CFD) method can reveal the microscopic flow-heat transfer mechanism, but they consume significant computational resources and require improvements in computational efficiency. To address this, this study proposes a hybrid Transformer and Multilayer Perceptron (TransMLP) algorithm, which innovatively integrates the self-attention mechanism of Transformer and the nonlinear mapping ability of Multilayer Perceptron (MLP) to efficiently predict the temperature field of the two-phase closed thermosyphon. The study first uses high-precision CFD simulation data to construct a comprehensive dataset, trains the TransMLP model, and tests its prediction performance. The results show that the TransMLP model achieves an R2 value above 0.95, close to 1, on all datasets, with a low root mean square error. Overall, its performance outperforms the Transformer, MLP, and Convolutional Neural Network models. This study provides a new paradigm for efficient modeling of two-phase closed thermosyphon, breaking through the cost and efficiency bottleneck of existing methods and laying the foundation for the optimization design and intelligent prediction of heat transfer systems.
Yan et al. (Tue,) studied this question.