An integrated approach to source term uncertainty and sensitivity analyses for nuclear reactor severe accidents

Large-scale computer programs simulate severe accident phenomena and often have a moderate-to-large number of models and input variables. Analytical solutions to uncertainty distributions of interested source terms are impractical, and influential inputs on outputs are hard to discover. Runs of such integral codes for complex severe accidents are generally time-consuming and hence computationally expensive. This article presents an integrated approach to uncertainty and sensitivity analyses for nuclear reactor severe accident source terms, with an example which simulates an accident sequence similar to that occurred at Unit 2 of the Fukushima Daiichi Nuclear Power Plant using an integral code, MELCOR. Monte-Carlo-based uncertainty analysis has been elaborated to investigate the released fractions of representative radionuclides, Cs and CsI. In order to estimate the sensitivity of inputs, which have a substantial influence on the core melt progression and the transportation process of radionuclides, a variance decomposition method is applied. Stochastic process, specifically a Dirichlet process, is applied to construct a surrogate model in sensitivity analysis as a substitute of the code. The surrogate model is cross-validated by comparing with corresponding results of MELCOR. The analysis with the simpler model avoids laborious computational cost/load, so that the importance measures for input factors are obtained successfully.