Applicability of a reduced order model for a safety analysis code to statistical safety analysis

Abstract

ABSTRACT A reduced order model (ROM) for a safety analysis code (RELAP/SCDAPSIM) is developed and is applied to the statistical safety analysis in order to reduce computational load. The ROM that reproduces output of the safety analysis code (e.g., the peak cladding temperature) is constructed using the singular value decomposition and the low-rank approximation for time-series data of a safety analysis code. A sampling method that incorporates dissimilarity among input data is tested to generate training data for the ROM. A loss of coolant accident with multiple failures of safety functions in a BWR is considered as an accident scenario. Statistical results such as the average, the standard deviation, and the 95% cumulative probability value of safety parameters obtained by the ROM and those obtained by the safety analysis code are compared. The reference results obtained by the direct random sampling method with the safety analysis code are within the 95% confidence interval of the results obtained by ROM. Thus the applicability and effectiveness of the ROM to the statistical safety analysis are confirmed through numerical analysis.