Prediction of Thermal-Hydraulic Parameters in Rod Bundle Assembly Domain Based on Similarity Features

To accurately predict the thermal-hydraulic parameters of reactor core flow domains under high Reynolds numbers and complex geometries, and to enhance the predictive accuracy of neural networks for rapid assessment of core thermal-hydraulic conditions, this study proposes a novel auxiliary prediction method. An analysis of detailed simulation results of rod bundle channels under various operating conditions identified a similarity pattern between the macroscopic thermal-hydraulic parameters of the rod bundle assemblies and the distribution of fine-scale parameters. This pattern was then utilized to construct the input features of the neural network, enabling precise predictions of temperature, pressure, and velocity parameters. The results demonstrate that the developed surrogate model achieves a maximum mean squared error (MSE) of 7.86×10⁻⁴ and a minimum MSE of 1.39×10⁻⁴ on macroscopic parameter test data. For fine-scale parameter test data, the maximum and minimum MSE are 9.39×10⁻³ and 5.20×10⁻⁴, respectively, indicating the model’s high accuracy in predicting core thermal-hydraulic conditions. Moreover, the surrogate model obtains fine-scale thermal-hydraulic parameter fields of the core within 0.504 seconds—an efficiency improvement of 1149 times compared to traditional methods. This provides a powerful technical foundation for developing digital twins of nuclear reactor cores.