Assessment of two uncertainty quantification methods for source term analysis

Abstract

Two methods of uncertainty quantification have been assessed for application to the source term analysis of Younggwang 3 & 4 using the MAAP code. The two methods of uncertainty analysis, i.e., the response surface method and the Latin hypercube sampling method, produced slightly different results for the empirical cumulative distribution function of the CsI release fraction from containment. In source term analysis, the response surface method has the relative merits in the importance and the sensitivity analyses and by this method, important qualitative information about the relation between the inputs and the output can be obtained. However, the calculated uncertainties of the CsI release fraction were large and the information was insufficient to generate an appropriate response surface. It was difficult to substitute a typical second order response surface model for the original code. Thus, it is found that it is difficult to ascertain the reliability of the response surface method. When the approximate model is not suitable, above advantages of the response surface method do not outweigh the lack of accuracy. The Latin hypercube method is preferred in this case. It is directly and easily implemented and does not require any trial to obtain an approximate model.