ABSTRACT

In this work, a hybrid Monte Carlo/sensitivity-based uncertainty contribution estimation approach is derived and tested. The approach is based on using the statistical samples used in the Monte Carlo uncertainty quantification combined with the subspace analysis to define and solve a set of linear equations to estimate the first-order sensitivities whenever the adjoint and forward sensitivity analyses are not available. The proposed approach empowers the Monte Carlo uncertainty quantification method with a pathway to evaluate and rank the uncertainty contributors efficiently without prior access to the sensitivity coefficients. The proposed approach has been illustrated and verified via an IRT-4 M 8 tubes fuel assembly uncertainty quantification problem where the proposed approach is compared to the sensitivity-based adjoint method. Results indicate that both methods concluded the same key differential contributors to the integral uncertainty with good agreement in the values of the sensitivity coefficients.

摘要

在这项工作中，导出并测试了一种基于混合蒙特卡罗/灵敏度的不确定性贡献估计方法。该方法基于使用蒙特卡罗不确定性量化中使用的统计样本与子空间分析相结合来定义和求解一组线性方程，以便在无法进行伴随和前向灵敏度分析时估计一阶灵敏度。所提出的方法使蒙特卡洛不确定性量化方法具有有效评估和排序不确定性贡献者的途径，而无需事先访问灵敏度系数。所提出的方法已经通过 IRT-4 M 8 管燃料组件不确定性量化问题进行了说明和验证，其中将所提出的方法与基于灵敏度的伴随方法进行了比较。