Investigation on Applicability of Dynamic Mode Decomposition in Burnup Analysis

Abstract In burnup analysis of a reactor, neutron transport calculation and depletion calculation are carried out alternately, bringing a serious computational burden to large-scale problem calculation. In this study, we used a data-driven strategy based on dynamic mode decomposition to extract the modes and characteristics of nuclide density to construct a prediction expression for the evolution of nuclide density. Reducing the burden of burnup calculation can be achieved through direct prediction of the nuclide density in a certain time point in the future based on the dynamic mode decomposition algorithm[1]. Firstly, six snapshots of different size are selected to test the relationship between the size of snapshot and prediction accuracy. The results indicate that when the time scale of the snapshots is the same as the time scale of the period to be predicted, the dynamic mode decomposition algorithm can extract a more comprehensive spatiotemporal evolution model. Secondly, we selected appropriate-sized snapshots to test the impact of different truncated rank on prediction accuracy. The results show that the optimal truncated rank for the assembly-level problems is 11. Thirdly, we explored the effect of the spatial field information on the results of prediction. The results indicate that using snapshots containing spatial information for prediction does not effectively improve accuracy and still incurs a high computational burden. Finally, the applicability of the dynamic mode decomposition algorithm to different fuel assemblies and the computational efficiency of the DMD algorithm are tested. The results show that the prediction deviation of nuclide density for different assemblies is around 1%, while saving 99.87% of the burnup calculation time. The result proves that the dynamic mode decomposition algorithm can efficiently discover the hidden dynamics in burnup analysis and has a considerable application prospect in large-scale burnup analysis.