Dynamics of the Molten Salt Fast Reactor Through Parametric Dynamic Mode Decomposition

Data-driven reduced-order modelling techniques are gaining increasing popularity as valuable tools to efficiently simulate the dynamics of complex systems across different operating conditions. In particular, they can be applied to perform multi-query scenario analyses, including sensitivity studies, where multiple evaluations of a model over a specific parametric range are required. Being based on the Reduced-Order Modelling framework, they offer a strong reduction in computational costs whilst maintaining satisfactory accuracy, by constructing a surrogate low-dimensional model from parametric high-fidelity training data. Within the broader framework of non-intrusive reduced-order methods, the Dynamic Mode Decomposition (DMD) represents an easy-to-implement yet efficient method, capable of reconstructing system dynamics from spatial–temporal data while providing an interpretable linear representation of the governing dynamics—a property of particular relevance in nuclear safety applications, where model transparency is essential. Originally developed for nonparametric datasets, its parametric extension (pDMD) accounts for the dependence of system dynamics on a given set of parameters. In this work, two parametric strategies are proposed and compared for the first time on a complex nuclear multiphysics benchmark: the Reduced Operator Interpolation (ROI) and the Reduced Koopman Operator Interpolation (RKOI) approaches, applied to the multiphysics model of the Molten Salt Fast Reactor (MSFR) undergoing an Unprotected Loss of Fuel Flow (ULOFF) accidental transient. The availability of reliable, interpretable, and computationally efficient reduced models opens multiple opportunities across the nuclear engineering workflow, from design phase sensitivity analyses to real-time monitoring and control applications. The proposed framework is general and transferable to other reactor concepts and multiphysics transient scenarios beyond the MSFR. Results demonstrate the accuracy and computational efficiency of both methods, with a detailed comparison of their trade-off, establishing pDMD as a promising tool for parametric nuclear safety analysis.