Use of metamodels in a probabilistic radiological assessment.

摘要

Others have performed a deterministic performance assessment for the proposed Texas low-level radioactive waste disposal site in Sierra Blanca, Texas. The purpose of this research is to examine methodologies for conducting a probabilistic performance assessment.; The first methodology is a Monte Carlo simulation utilizing Latin Hypercube Sampling on specific computer models based on pre-determined probability distribution functions. The second method uses Monte Carlo Latin Hypercube Sampling with metamodels, which are simplifications of models. The RESRAD (RESidual RADioactive material) code was run as a deterministic method.; A sensitivity analysis of the primary models was used to construct two metamodels, one based on first order linear equations, and the second based on higher order equations. The HELP (Hydrologic Evaluation of Landfill Performance) model to calculate water infiltration amounts and DUST (Disposal Unit Source Term) model to calculate radionuclide releases were used along with several prescribed transport and dose equations from Federal Guidance Reports. Most key variables were described by probability distribution functions. Variable climate scenarios were considered in the evaluation.; Probabilistic results compared favorably with deterministic results. C-14, Cl-36, TC-99 and I-129 were key isotopes considered in both the original deterministic performance assessment and in this study. The original performance assessment had a peak whole body dose equivalent of 1.1 mrem/yr. The RESRAD dose was 6.0 mrem/yr. The metamodel dose using the deterministic performance assessment conditions was 1.3 mrem/yr. The probabilistic results indicated that in 74 of 76 scenarios the 95th percentile doses were below the regulatory dose limit of 25 mrem/yr. When the two exceptions were modified to more accurately reflect realistic conditions, the 95 th percentile doses were below the dose limit. The National Council on Radiation Protection recommends using a 95th percentile in comparison with standards.; Metamodels can be effective screening devices, where parameter and design changes may be quickly analyzed to determine if additional consideration is warranted. They may be run on personal computers with simple spreadsheet software. A good metamodel can be an effective replacement for the model it represents and the result can be significant savings in time, memory and cost as demonstrated by this study.