基于卷积神经网络模型的堆芯有效增殖因数预测方法研究

Prediction Method of Effective Multiplication Factor of ReactorCore Based on Convolution Neural Network Model

建立了基于卷积神经网络(convolution neural network,CNN)的深度学习模型,可用于直接预测反应堆堆芯有效增殖因数。对简化的2群压水反应堆(pressurized water reactor,PWR)进行堆芯数值计算,通过核反应堆堆芯数值模拟程序建立了组件截面数据库,得到原始数据,采用CNN模型对归一化的原始数据进行训练,最后得到替代数值计算方法的深度学习模型。此外,为验证CNN模型的预测性能,采用反向传播神经网络(back propagation neural network,BPNN)模型对原始数据进行了训练与测试。模型训练及测试结果分析表明:CNN预测模型性能要明显优于BPNN模型;CNN模型预测堆芯参数的平均相对偏差为0.251%。

In this paper,a deep learning model based on convolution neural network(CNN)is established to directly predict the effective multiplication factor of reactor core.The physical parameters of the simplified two groups of pressurized water reactor(PWR)cores are numerically calculated.The component section database is established through the nuclear reactor core numerical simulation program to obtain the original data.The normalized original data are trained by CNN model,and then the deep learning model instead of the numerical calculation method is obtained.In addition,in order to verify the prediction performance of CNN model,back propagation neural network(BPNN)model is used to train and test the original data.The model training and result analysis show that the performance of CNN prediction model is significantly better than BPNN model,and the average relative deviation of CNN model in predicting core parameters is 0.251%.