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

A full-scope method is constructed to reveal source term uncertainties and to identify influential inputs during a severe accident at a nuclear power plant (NPP). An integrated severe accident code, MELCOR Vet. 1.8.5, is used as a tool to simulate the accident similar to that occurred at Unit 2 of the Fukushima Daiichi NPP. In order to figure out how much radioactive materials are released from the containment to the environment during the accident, Monte Carlo based uncertainty analysis is performed. Generally, in order to evaluate the influence of uncertain inputs on the output, a large number of code runs are required in the global sensitivity analysis. To avoid the laborious computational cost for the global sensitivity analysis. via MELCOR, a surrogate stochastic model is built using a Bayesian nonparametric approach, Dirichlet process. Probability distributions derived from uncertainty analysis using MELCOR and the stochastic model show good agreement. The appropriateness of the stochastic model is cross-validated through the comparison with MELCOR results. The importance measure of uncertain input variables are calculated according to their influences on the uncertainty distribution as first-order effect and total effect. The validity of the present methodology is demonstrated through an example with three uncertain input variables. (C) 2015 Elsevier Ltd. All rights reserved.