Abstract

A one-dimensional (1-D) thermal stratification (TS) model was recently developed in our research group to predict the TS phenomenon in pool-type sodium-cooled fast reactors. This paper performs uncertainty quantification (UQ) of the 1-D TS model to evaluate its performance by considering the aleatoric uncertainties that existed in the model parameters and to identify the plausible sources of the epistemic uncertainties. The Latin hypercube sampling–Monte Carlo method (LHS-MC), which is elaborated with an example in this paper to facilitate its understanding and implementation, is used for the UQ process. The advantages of LHS-MC, including both better stability and better accuracy than the conventional random sampling–Monte Carlo method with fewer realizations, are demonstrated in this paper.

In total, 648 temperature measurements acquired from nine experimental transients performed in a university-scale Thermal Stratification Experimental Facility are used to evaluate the performance of the computational 1-D TS model. The UQ result shows that 77.5% of the experimental data can be predicted by the 1-D TS model within uncertainty ranges, which indicates the good performance of the computational model when the aleatoric uncertainties are correctly captured. The rest 22.5% of the experimental data are found located outside of the uncertainty ranges, which reveals the existence of the epistemic uncertainties caused by the lack of understanding of the TS phenomenon and defects in the 1-D model. The simple jet model currently employed by the 1-D TS model is thought to be one of the attributors to these defects.

摘要

我们的研究小组最近开发了一种一维 (1-D) 热分层 (TS) 模型来预测池型钠冷快堆中的 TS 现象。本文对一维 TS 模型进行不确定性量化 (UQ)，通过考虑模型参数中存在的任意不确定性来评估其性能，并识别认知不确定性的可能来源。拉丁超立方体采样-蒙特卡罗方法（LHS-MC），在本文中通过一个例子进行阐述，以促进其理解和实现，用于UQ过程。本文展示了 LHS-MC 的优点，包括比传统的随机抽样蒙特卡罗方法更好的稳定性和更好的准确性，实现更少。

从大学规模的热分层实验设施中执行的九个实验瞬态中获得的 648 个温度测量值总共用于评估计算 1-D TS 模型的性能。UQ 结果表明，在不确定性范围内，一维 TS 模型可以预测 77.5% 的实验数据，这表明在正确捕获任意不确定性时计算模型的良好性能。其余 22.5% 的实验数据位于不确定性范围之外，这揭示了由于对 TS 现象缺乏理解和一维模型缺陷导致的认知不确定性的存在。一维 TS 模型目前采用的简单射流模型被认为是这些缺陷的原因之一。