Development of nuclear forensic models using kernel regression

Abstract The objective of nuclear forensics is to find out the origin or distribution process of illegal radioactive materials that are out of regulatory control by analyzing some specific characteristics referred to as “signatures”. One such radioactive material is nuclear spent fuel, which is likely to be intercepted in transit and abused. However, numerous samples of spent fuel cannot be obtained in Korea because analyzing irradiated nuclear substances has been limited due to international regulations. Thus, in this paper, spent fuel sample data for nuclear forensics were generated using ORIGEN (ORNL Isotope Generation and Depletion code) based on operational histories of Korean nuclear power plants. This paper focuses on the development of a spent fuel inference model for nuclear forensics to estimate operational histories such as burn up, initial enrichment and cooling time for radioactive materials. The type of measurable nuclides and the accuracy of measured nuclides were assumed to be varied. Therefore, the regression model to estimate operational histories should be precise and robust. An inferential model based on kernel regression (IKR) that can provide estimates based only on data (without the assumptions required in physical models) has been developed in order to predict operational histories. On the other hand, the auto-associative model of kernel regression (AAKR) was used to eliminate some outliers in the input data to enhance the accuracy of the regression.