In the framework an international survey, regulatory approaches and national practices of level 2 probabilistic safety analyses (PSA) are evaluated for seven nations using nuclear energy. Within a PSA, uncertainty analyses are of central importance. However, the common Monte Carlo technique requires extensive computational resources. Therefore, an optimization approach, applying methods of statistical learning, is developed. Based on the set of uncertain input parameters, the approach permits a direct prediction of the end state of an accident sequence, without explicitly modeling the underlying physical process. The prediction is achieved by a decision function, which is derived form a sample of sequences with known end states. With this model, Monte Carlo analyses are possible without CPU time constraints. Further, important drawbacks of conventional optimization methods are avoided, additional phenomenological insights are obtained.
«
In the framework an international survey, regulatory approaches and national practices of level 2 probabilistic safety analyses (PSA) are evaluated for seven nations using nuclear energy. Within a PSA, uncertainty analyses are of central importance. However, the common Monte Carlo technique requires extensive computational resources. Therefore, an optimization approach, applying methods of statistical learning, is developed. Based on the set of uncertain input parameters, the approach permits a...
»