Natural convection condensation, with the advantage of high reliability and not requiring complex mechanical drive structures, is broadly used in industrial fields, such as chemical, nuclear power, automotive, etc. This work aims to investigate the heat transfer mechanism and evaluate the performance of natural convection condensation with the artificial neural network (ANN) method, correlation predictions, and the code based on the boundary theory. An empirical correlation was proposed based on the present experimental data with operating conditions in the pressure range of 0.2 MPa -0.6 MPa, subcooled temperature range of 11 K–45 K, and air mass fraction range of 0.0049–0.69. The empirical correlation was validated against a consolidated database, with 91% of the data reproduction falling within the error band of \(\pm\) 30%. An ANN model was put forward with training, validation, and testing using the present experimental data, which yields an error of \(\pm\) 5% in the present test data. When the trained model was utilized to reproduce the additional database, all the data fell within an \(\pm\) 11% error band. Finally, a side-by-side comparison in heat transfer coefficient reproduction was conducted among those rapidly computational methods, and the ANN model turned out to have the best performance.

自然对流冷凝具有可靠性高、不需要复杂的机械驱动结构等优点，被广泛应用于化工、核电、汽车等工业领域。本文旨在研究其传热机理并评估其性能。自然对流凝结采用人工神经网络（ANN）方法、相关预测以及基于边界理论的代码。根据现有实验数据，在压力范围为0.2 MPa -0.6 MPa、过冷温度范围为11 K~45 K、空气质量分数范围为0.0049~0.69的工况下，提出了经验关联式。经验相关性根据统一数据库进行了验证，91% 的数据再现落在\(\pm\) 30% 的误差范围内。提出了使用现有实验数据进行训练、验证和测试的 ANN 模型，在现有测试数据中产生\(\pm\) 5% 的误差。当使用经过训练的模型来重现附加数据库时，所有数据都落在\(\pm\) 11% 误差带内。最后，对这些快速计算方法的传热系数再现进行了并列比较，结果表明，ANN 模型具有最佳性能。