Antalya Bilim University, Vocational School of Health Services, Department of Medical Imaging Techniques, Antalya, Turkey.
Appl Radiat Isot. 2021 Apr;170:109584. doi: 10.1016/j.apradiso.2020.109584. Epub 2021 Jan 9.
The aim of this study is to develop an accurate artificial neural network algorithm for the cross-section of (n,p) reactions at 14.5 ∓0.5 MeV neutron energy which is important to developing materials for fusion reactor design. The experimental data used at artificial Neural network calculations have been taken from the Experimental Nuclear Reaction Data (EXFOR) database. Bayesian algorithm has been used at training section of artificial neural network. Regression (R) values of artificial neural network calculations have been found as 0.99363, 0.98574 and 0.99257 for training, testing and all process respectively. In addition to artificial neural network calculations, TALYS 1.95 nuclear reaction code has been used to reproduce (n,p) reactions at 14.5 ∓0.5 MeV. Two-component exciton model and Constant Temperature Fermi Gas Model have been used as pre-equilibrium and level density models respectively. Mean square errors of our calculations have been found 48.51 and 495.06 for artificial neural network and TALYS 1.95 respectively. Artificial Neural network estimations have been compared and analyzed with the TALYS 1.95 calculations and the experimental data taken from EXFOR database.
本研究的目的是开发一种针对 14.5±0.5 MeV 中子能量的 (n,p) 反应截面的精确人工神经网络算法，这对于开发聚变反应堆设计用材料非常重要。人工神经网络计算中使用的实验数据取自实验核反应数据 (EXFOR) 数据库。贝叶斯算法被用于人工神经网络的训练部分。人工神经网络计算的回归 (R) 值分别为 0.99363、0.98574 和 0.99257，用于训练、测试和整个过程。除了人工神经网络计算外，还使用 TALYS 1.95 核反应代码来再现 14.5±0.5 MeV 的 (n,p) 反应。双组分激子模型和恒温和费米气体模型分别用作预平衡和能级密度模型。我们的计算的均方误差分别为人工神经网络的 48.51 和 TALYS 1.95 的 495.06。人工神经网络的估计值与 TALYS 1.95 的计算值和取自 EXFOR 数据库的实验数据进行了比较和分析。
