논문 상세보기
Bayesian Updating in the Nuclide Travel Length for the Deep Borehole Spent Fuel Repository
- 언어ENG
- URLhttps://db.koreascholar.com/Article/Detail/430461
p.173
A GoldSim Total System Performance Assessment has been developed and utilized for assessment of the various conceptual HLW repositories for spent nuclear fuels during last a few decades. Even though, almost all required parameter values associated with the repository system are frequently assumed or sometimes overestimated, they are still far from being highly reliable. Uncertainties nested in nuclide transport modeling around the repository are mainly dominated by these parametric uncertainties aside from intrinsic model uncertainty. Reliable estimate of the parameter values commonly expressed as probability density functions (PDFs) always require a large amount of measured data. Such input distributions are used as input to the probabilistic assessment program through Monte Carlo simulation to quantitatively provide possible uncertainty of the results. However, in most cases, especially in the safety assessment of the repository which is typically related with both long-time span and wide modeling domain, inefficient observed data from the field measurements are common, making conventional probabilistic calculations rather even uncertain. Since Bayesian approach is known to be especially powerful and efficient in the case of lacking of available data measured, such short data could be compensated by coupling with a priori belief, reducing uncertainty. By allowing the a priori knowledge for incorporating insufficient observed data, which include expert’ elicitation, their beliefs and judgment regarding the parameters as well as recent site-specific measurements, based on the Bayes’ theorem, the older parameter distributions, “prior” distribution can be updated to a rather newer and reliable “posterior” distribution. Newer distributions are not necessarily expressed as PDFs for probabilistic calculation. These updates could be done even iteratively as many times as data values are sequentially available, which calls sequential Bayesian updating, making belief of posterior distributions become much higher by reducing parametric uncertainty. To show a possible way to enhance the belief as well as to reduce the uncertainty involved in parameter for the Bayesian scheme, nuclide travel length in the far-field area of a hypothetical deep borehole spent fuel Repository was investigated. The algorithm and module that have been developed and implemented in GSTSPA through current study was shown to work well for all assumed prior, three sequential posterior distributions and likelihoods.
