Bayesian Analysis for the Site-Specific Dose Modeling in Nuclear Power Plant Decommissioning

Abstract

Decommissioning is the process of closing down a facility. In nuclear power plant decommissioning, it must be determined that that any remaining radioactivity at a decommissioned site will not pose unacceptable risk to any member of the public after the release of the site. This is demonstrated by the use of predictive computer models for dose assessment. The objective of this thesis is to demonstrate the methodologies of site-specific dose assessment with the use of Bayesian analysis for nuclear power plant decommissioning. An actual decommissioning plant site is used as a test case for the analyses. A residential farmer scenario was used in the analysis with the two of the most common computer codes for dose assessment, i.e., DandD and RESRAD. By identifying key radionuclides and parameters of importance in dose assessment for the site conceptual model, available data on these parameters was identified (as prior information) from the existing default input data from the computer codes or the national database. The site-specific data were developed using the results of field investigations at the site, historical records at the site, regional database, and the relevant information from the literature. This new data were compared to the prior information with respect to their impacts onboth deterministic and probabilistic dose assessment. Then, the two sets of information were combined by using the method of conjugate-pair for Bayesian updating. Value of information (VOI) analysis was also performed based on the results of dose assessment for different radionuclides and parameters. The results of VOI analysis indicated that the value of site-specific information was very low regarding the decision on site release. This observation was held for both of the computer codes used. Although the value of new information was very low with regards to the decisions on site release, it was also found that the use of site-specific information is very important for the reduction of the predicted dose. This would be particularly true with the DandD code.