(A) study on the improvement of scaling factor determination method using artificial neural network

Abstract

Final disposal of radioactive waste generated from Nuclear Power Plant (NPP) requires the detailed information about the characteristics and the quantities of radionuclides in waste package. Most of these radionuclides are difficult to measure and expensive to assay. Thus it is suggested to the indirect method by which the concentration of a Difficult-to-Measure (DTM) nuclide are determined using the correlations of concentration - it is called the scaling factor - between Easy-to-Measure (Key) nuclides and DTM nuclides with the measured concentration of the Key nuclide. In general, the scaling factor is determined using the log mean average method (LMA) and the regression method. These methods are adequate to apply most corrosion product nuclides. But in case of fission product nuclides and some corrosion product nuclides, the predicted values are not well matched with the measured values. In this study, the ANN method is compared with the conventional SF determination method - the LMA and the regression method - for the improved SF determination. Before these comparisons, the sensitivity analysis for each ANN model is performed to determine the optimum size of hidden layers of ANN models. Moreover, the ensemble model, which combines the ANN model with the regression model, is compared with the original models to evaluate the applicability of the ensemble model in SF determination. It is concluded that the ANN method is superior to the conventional SF determination method in some nuclides and the ensemble model can be used as the supplement of the original models.