现代应用物理2025，Vol.16Issue(1)：105-114,132,11.DOI:10.12061/j.issn.2095-6223.202412049
单体相移深度神经网络学习共振截面的网络构建与训练
Network Construction and Training for Learning Resonance Cross Sections by Single Phase-Shift Deep Neural Networks
摘要
Abstract
The recently proposed single phase-shift deep neural network(SPDNN),characterized by its compact network size and remarkable learning precision,has emerged as the first viable deep learning solution for the fitting and evaluation of complex neutron resonance cross sections.During the learning phase of the SPDNN for resonance cross sections,a variety of factors play a crucial role in shaping the effectiveness of the network training,its efficiency,and the generalization ability of the resulting model.These factors include the spectral range and bandwidth of the resonance cross sections,which together determine the size of the phase-shift layer of the network,the number of hidden layers,the number of neurons per layer,the activation function,the loss function,the number of training iterations,and the pre-processing techniques applied to the training data.To further enhance the practical applicability of SPDNN in resonance cross section research,a thorough investigation of the influence of these factors on the network fitting performance is conducted.Through this rigorous analysis,optimal network configurations and training methods tailored for SPDNN in the context of resonance cross section research are identified,thereby advancing its deployment in nuclear data science applications.关键词
单体相移神经网络/中子共振截面/核数据评价/神经网络训练/频谱Key words
single phase-shift deep neural network/neutron resonance cross section/nuclear data evaluation/neural-network training/spectrum分类
信息技术与安全科学引用本文复制引用
胡泽华,应阳君..单体相移深度神经网络学习共振截面的网络构建与训练[J].现代应用物理,2025,16(1):105-114,132,11.基金项目
核数据重点实验室基金资助项目(JCKY2022201C155) （JCKY2022201C155）
