基于再淹没现象的RBF神经网络和Kriging的代理模型应用及误差分析

为了将代理模型应用于核工程反应堆热工水力现象中，以再淹没现象为研究对象，构建代理模型，并通过控制实验参数的数量，来化解精度与效率之间的矛盾。首先使用拉丁超立方方法抽取输入样本，并通过RELAP5建模获取输出样本，由MATLAB分别构建RBF神经网络代理模型和Kriging代理模型，对包壳峰值温度（PCT）进行拟合。然后分析比较了2种代理模型的拟合结果，发现2种代理模型的相对误差值均小于0.15%，均适用于再淹没现象；Kriging代理模型的拟合精度高于RBF神经网络代理模型。

Application and Error Analysis of RBF Neural Network and Kriging Surrogate Model Based on Reflood Phenomenon

In order to apply the surrogate model to the thermal hydraulic phenomena of nuclear engineering reactors,this paper takes the reflood phenomenon as the research object,constructs the surrogate model,and solves the contradiction between accuracy and efficiency by controlling the number of experimental parameters.Firstly,the input samples are extracted by Latin hypercube method,and the output samples are obtained by RELAP5 modeling and calculation.The RBF neural network and Kriging surrogate model are constructed by MATLAB to fit the peak temperature of cladding.The fitting results of RBF and Kriging surrogate models are analyzed and compared.It is found that the relative errors of the two surrogate models are less than 0.15%,which can be applied to the reflood phenomenon.The fitting accuracy of Kriging surrogate model is higher than that of RBF neural network surrogate model.