Application of Bayesian nonparametric models to the uncertainty and sensitivity analysis of source term in a BWR severe accident

Zheng, X. ; Ito, Hiroto; Kawaguchi, Kenji; Tamaki, Hitoshi ; Maruyama, Yu 

An important issue for nuclear severe accident is the source tern uncertainty and sensitivity analysis. Generally, thousands of cases are needed to reach a stable result of sensitivity analysis. Based on the limited data obtained by MELCOR analysis, in which the accident at Unit 2 of the Fukushima Daiichi Nuclear Power Plant is used as an example, an approximate stochastic model has been constructed via Bayesian nonparametrics, specifically, the Dirichlet process. The advantage of a nonparametric model is that any deterministic function between explanatory and response variables is not necessary to be determined. The complexity of model will grow automatically as more actual data is observed. The approximate model saves the computational cost and makes it possible to complement thousands of Monte Carlo computation for uncertainty and sensitivity analysis. Probability density functions of uncertainty analysis by MELCOR and the approximate model are obtained and compared. Two densities show great accordance that proves the good predictive ability of the stochastic model. The appropriateness of the approximate model is further validated by the cross-validation through the comparison with actual MELCOR results. Global sensitivity analysis by Sobol' sensitivity index has been performed with the approximate model. Three input parameters are ranked according to their respective influences on the output uncertainty based on first-order and total effect.