High-fidelity computational fluid dynamics simulations can effectively capture transient three-dimensional thermal-fluid phenomena in molten salt reactors (MSRs), but they are computationally expensive and time consuming. The dynamic mode decomposition (DMD) method is used to improve simulation efficiency. The research findings indicate that the DMD method encounters processing difficulties and numerical instability issues when handling a large-scale transient data set. To address these issues, three domain decomposition strategies are proposed: geometry-based, velocity-based, and temperature-based clustering. All three methods effectively improve numerical stability and modeling efficiency, with the velocity-based decomposition showing the best performance.

Based on this method, the number of modes in each subdomain is optimized to construct an efficient and accurate domain-decomposed DMD model. The optimized model can quickly and effectively predict transient three-dimensional temperature and velocity fields in MSRs, with the maximum temperature error under 0.22 K and the relative velocity error within 5%. This result demonstrates that the proposed domain-decomposed DMD method significantly enhances the efficiency and stability of transient prediction, providing an effective method for the fast simulation of transient three-dimensional thermal-fluid behavior in MSRs.