Development of a reduced order model for severe accident analysis codes by singular value decomposition aiming probabilistic safety margin analysis

Abstract

ABSTRACT Severe accident codes (e.g. MAAP, RELAP, and MELCORE) model various physical phenomena during severe accidents. Many analyses using these codes for safety margin evaluation are impractical due to large computational costs. Surrogate models have an advantage of quickly reproducing multiple results with a low computational cost. In this study, we apply the singular value decomposition to the time-series results of a severe accident code to develop a reduced order modeling (ROM). Using the ROM, the probabilistic safety margin analysis for the station blackout with a total loss of feedwater capabilities at a boiling water reactor is carried out. The dominant parameters to the accident progression are assumed to be the down-time and the recovery-time of the reactor core isolation cooling system, and decay heat. To reduce the number of RELAP5/SCDAPSIM analyses while maintaining the prediction accuracy of ROM, we develop a data sampling method based on adaptive sampling, which selects the new sampling data based on the dissimilarity with the existing training data for ROM. Our ROM can rapidly reproduce the time-series results and can estimate core conditions. By reproducing multiple results by ROM, a time-dependent core damage probability distribution is calculated instead of the direct use of RELAP5/SCDAPSIM.