Development and validation of a deep learning‐based model for predicting burnup nuclide density
Jichong Lei, Chao Yang, Changan Ren, Wei Li, Chengwei Liu, Aikou Sun, Yukun Li, Zhenping Chen, Tao Yu
Abstract
To address the issue of large inaccuracies in the low-burnup region of aditonal machine learning algorithms for predicting nuclide density, the DRAGON code is used to produce 9600 samples using the nuclide densities of 235U, 239Pu, 241Pu, 137Cs, and 154Nd as prediction parameters. The mean square error (MSE) was used as the loss function for the deep neural network-based nuclide density prediction model. The trained model is used to predict the target nuclides in the test set, and the relative error with the multilayer perceptron model are compared. The prediction results demonstrate that the deep neural network-based prediction model not only overcomes the issue of excessive prediction errors in the low-burnup region of the traditional machine learning algorithm model, but also has lower prediction errors in the medium-burnup and high-burnup regions, demonstrating the feasibility of artificial intelligence in nuclide density prediction.