基于EM算法的热工水力模型的不确定性量化方法研究

常用的最佳估算加不确定性（BEPU）方法是对电厂和程序模型的不确定性来源（输入不确定性）进行量化，并通过不确定性传递得到与安全相关参数（如包壳峰值温度PCT）的双95%不确定度带。然而最佳估算（BE）程序内部的一些物理模型参数作为重要的不确定性来源，在实验中往往不能被直接测量，其不确定性的量化目前主要依赖于专家判断，缺乏客观性。本文基于期望最大化（EM）算法，利用分项实验中可直接测量的响应参数，通过反问题求解推导程序内部模型参数的不确定性概率密度分布，并以再淹没现象为例进行方法应用，得到了程序内部沸腾传热、相间摩擦等模型的不确定度。 

Investigation of Uncertainty Quantification Method of Thermodynamic Models Based on EM Algorithm

The widely used BEPU method is to quantify the important uncertain source (input uncertainty) associated with the power plant and models, and then propagate the uncertainty to obtain the double 95% uncertainty bands of the safety related parameters (such as the peak cladding temperature, PCT). However, as the important sources of uncertainty, some physical model parameters in the best estimation (BE) codes are often difficult to be directly measured in the experiments. The quantification process is mainly dependent on the expert judgment which lacks objectivity. In this paper, based on the Expectation Maximization (EM) algorithm and the response parameters measured directly from the separate effect test, the probability density distribution of the internal model parameters can be derived by solving the inverse problem. Then, the application of the method is performed by taking reflood phenomena as an example. Uncertainties of code internal boiling heat transfer and interfacial friction models are obtained.