质子诱发核散裂反应中产物截面的贝叶斯神经网络预测方法  

Bayesian neural network prediction methods for fragment cross sections in proton-induced spallation reactions

作　　者：彭丹 魏慧玲[1] 普洁[1] 程凯旋 王玉廷 魏啸宝 陈茜茜 马春旺[1] PENG Dan;WEI HuiLing;PU Jie;CHENG KaiXuan;WANG YuTing;WEI XiaoBao;CHEN XiXi;MA ChunWang(Institute of Particle and Nuclear Physics,School of Physics,Henan Normal University,Xinxiang 453007,China)

机构地区：[1]河南师范大学物理学院,粒子物理与原子核物理研究所,新乡453007

出　　处：《中国科学：物理学、力学、天文学》2022年第5期111-118,共8页Scientia Sinica Physica,Mechanica & Astronomica

基　　金：国家自然科学基金(编号:11975091);河南省高校科技创新团队(编号:21IRTSTHN011)资助项目。

摘　　要：核的散裂反应是由几十MeV以上能量的轻粒子诱发的一类剧烈核反应,可以产生种类繁多的放射性核素.散裂反应中的余核产生截面数据是核物理和核技术应用领域的关键基础数据,但由于反应涉及的体系从小到大,反应能量跨越3–4个数量级,以及产生余核的电荷和质量数范围很宽,整体上现有的多种理论模型对余核产生截面的预测精度有待进一步提高.本文以机器学习方法中的贝叶斯神经网络方法为例,建立对质子诱发核散裂反应中余核产生截面的贝叶斯直接学习方法和物理模型引导的间接学习方法,提高对质子诱发核散裂反应中余核产生截面的预测范围和预测精度.Nuclear spallation reactions can be induced by light particles with the incident energy greater than 10 Me V resulting in the production of various radioactive nuclides.The cross sections of fragment creation and fragment production are the fundamental and basic data,respectively for nuclear physics and nuclear applications in a wide variety of fields.Due to the complexity in spallation systems varying from light to heavy,incident energy ranging from MeV to GeV,fragment charge and mass numbers,the current models should be upgraded for increased precision.Using the Bayesian neural network(BNN),machine learning approaches have been established to predict the fragment cross sections in the proton-induced nuclear spallation reactions.The direct BNN method,together with the physical model guided BNN+SPACS and BNN+s EPAX methods,has been shown to improve the precision of fragment cross sections.

关 键 词：机器学习 贝叶斯神经网络 散裂反应 余核反应截面 

分 类 号：O571.415[理学—粒子物理与原子核物理] TP181[理学—物理]