期刊论文

Song, YM

英文

Annals of Nuclear Energy

0306-4549

2024

201

110415

10.1016/j.anucene.2024.110415

Activation functions play a crucial role in training neural networks. Currently, the most widely used and default activation function in most neural network frameworks is Rectified Linear Unit (ReLU), defined as ReLU(x) = max(0,x), which retains only positive inputs and returns zero for negative values. Although ReLU has simple computation and sparsity characteristics, for input datasets with uneven data distribution, the presence of more negative values often results in a large number of ReLU

本校为第一且通讯机构

Researchers have utilized neural networks as agent models to predict shielded data sets for reactor shielding designs. However, previous models faced challenges related to unbalanced multi-scale characteristics in input and output data sets, resulting in significant generalization errors. Additionally, these models did not consider source term parameters like energy spectrum probability density distribution parameters and angle sine and cosine distribution parameters. To address these challenges, this study proposes an improved neural network model with the Swish-NNTA learning scheme. This lea...