Uncertainty analysis methods for quantification of source terms using a large computer code

Quantification of uncertainties in the source term estimations by a large computer code, such as MELCOR and MAAP, is an essential process of the current probabilistic safety assessments (PSAs). The main objectives of the present study are (1) to investigate the applicability of a combined procedure of the response surface method (RSM) based on input determined from a statistical design and the Latin hypercube sampling (LHS) technique for the uncertainty analysis of CsI release fractions under a hypothetical severe accident sequence of a station blackout at Young-Gwang nuclear power plant using MAAP3.0B code as a benchmark problem; and (2) to propose a new measure of uncertainty importance based on the distributional sensitivity analysis. On the basis of the results obtained in the present work, the RSM is recommended to be used as a principal tool for an overall uncertainty analysis in source term quantifications, while using the LHS in the calculations of standardized regression coefficients (SRC) and standardized rank regression coefficients (SRRC) to determine the subset of the most important input parameters in the final screening step and to check the cumulative distribution functions (cdfs) obtained by RSM. Verification of the response surface model for its sufficient accuracy is a prerequisite for the reliability of the final results obtained by the combined procedure proposed in the present work. In the present study a new measure has been developed to utilize the metric distance obtained from cumulative distribution functions (cdfs). The measure has been evaluated for three different cases of distributions in order to assess the characteristics of the measure: The first case and the second are when the distribution is known as analytical distributions and the other case is when the distribution is unknown. The first case is given by symmetry analytical distributions. The second case consists of two asymmetry distributions of which the skewness is non zero: the lognormal distribution and the two-parameter Weibull distribution. For the last case the empirical distributions generated by the Monte Carlo simulations are used. To study its applicability, the present measure has been applied to two different examples that are widely used in PSA applications. The results have been compared with those of the existing three measures: the standard deviation measure; the information entropy measure; and the bivariate measures. On the basis of the results obtained in the present study, the present approach is a useful measure of the uncertainty importance. The present measure is simple and easy to calculate the uncertainty importance without any prior process for the estimation of distributions. In conclusion, the present method is recommended to be used as a useful tool for the analysis of uncertainty importance