A novel hybrid proper orthogonal decomposition and advanced transformer model for rapid prediction of flow and heat transfer in helical cruciform fuel rods

In the fuel assembly design of pressurized water reactors, helical cruciform fuel (HCF) rods have garnered significant attention due to their substantial potential for enhancing thermal-hydraulic performance and achieving higher safety margins. However, the complex geometry of HCFassemblies leads to prohibitively high computational costs for full-scale CFD simulations, severely restricting their application in design optimization. This paper develops a reduced-order model named HyPOD-AdTransformer to achieve efficient prediction of flow and heat transfer characteristics in HCF assemblies. The model incorporates a hybrid reduced-order algorithm (HyPOD), which employs traditional proper orthogonal decomposition (POD) for velocity field processing and proposes explicit boundary POD (EbPOD) for the temperature field to retain critical boundary layer information, thereby improving the applicability of traditional POD for temperature field analysis. Additionally, the AdTransformer is integrated to enhance the model's capability in capturing nonlinear features and global full-field information, improving its ability to represent the complex flow characteristics induced by the HCF helical structure. This reduces prediction errors in the temperature field boundary layer and the mainstream region of the velocity field. Case analysis demonstrates that compared to POD-BPNN (with average errors of 0.29%, 10%, and 1.1% for u, w, and T respectively), the prediction errors of HyPOD-AdTransformer are reduced to 0.00095%, 0.25%, and 0.013%. This model takes boundary conditions as input and directly outputs the full-field distribution without requiring CFD iterations, providing a reliable method for efficient thermal-hydraulic analysis and flow field prediction in complex structures.