Source term uncertainty analysis: probabilistic approaches and applications to a BWR severe accident

Abstract

A suite of methods has been established to quantitatively estimate uncertainties existed in source term analysis during a nuclear reactor severe accident. The accident sequence occurred at Unit 2 of the Fukushima Daiichi nuclear power plant (NPP) is taken as an example in which it is numerically modeled via the integrated severe accident code MELCOR 1.8.5. This standardized approach mainly consists of four steps: screening analysis, random sampling, numerical computation and verification of uncertainty distributions. First, by using an individually randomized one-factor-at-a-time screening method, a group of variables are preliminarily determined as important uncertain variables. Second, appropriate probability distributions are assigned to all selected variables. Multiple sets of random samples are generated using Latin Hypercube sampling combined with the consideration of rank correlation among input variables. Third, random samples of all selected variables are inputted into MELCOR 1.8.5. Numerical simulation with multiple code runs is implemented. Finally, uncertainty distributions for representative source terms (barium cesium, cesium iodide and tellurium) are obtained and verified. The technique of Bayesian nonparametric density estimation is applied to obtain probability density functions of interested source terms. In order to obtain a reasonable uncertainty distribution, several rounds of Latin Hypercube sampling and computation are conducted. As an alternative method to Wilks sampling criteria, the difference of probability density functions is evaluated through the comparison based on the Kullback-Leibler (KL) divergence. With the subjective judgment of small enough KL divergence, after a certain number of numerical computations, the uncertainty distributions of representative source terms are considered as stable enough as reliable results.