Non-Parametric, Extreme-Value Method for Estimating Bias and Bias Uncertainty in Nuclear Criticality Safety

摘要

Established national standards (ANSI/ANS-8.1 and -8.24) require the quantification of bias, the systematic difference between the actual k and what transport calculations with an associated data library predict, and its uncertainty within a defined area of applicability (AOA), a calculational margin. Many standard approaches require that the bias of the underlying set of relevant benchmark critical experiments be normally distributed, which is sometimes true in practice, but other times not. Alternatively, non-parametric approaches may be used regardless of how the biases in the benchmark experiments are distributed. This summary discusses such an approach using extreme-value methods allowing for the weighting of benchmarks based upon factors such as the degree of relevance toward a particular application; a traditional non-parametric, rank-order method does not allow for such weighting. An approach for such a weighting using sensitivity/uncertainty techniques is discussed. Results are then presented that illustrate how the non-parametric method performs.