Research On ¹⁰B(n,?)~7Li And ¹²C(n,n)¹²C Reaction Based On Bayesian Neural Network Approach

In light nuclei,¹⁰B and ¹²C play an important role in nuclear medicine,neutron radiation and other fields.¹⁰B has strong radiation protection and neutron absorption ability,while ¹²C is one of the most abundant nuclides in all biological organisms.However,due to the relative complexity of light nuclear reactions,the theoretical research on ¹⁰B(n,?)~7Li and ¹²C(n,n)¹²C reactions is very slow at home and abroad.Seeking new theoretical models or calculation methods to study the emitted particle information of this kind of light nuclear reactions has important application value to academic significance.At present,Bayesian neural network(BNN)method stands out among many algorithms because of its strong prediction ability and the uncertainty of prediction value.It has great application potential.In this paper,the measurement data of differential cross section in ¹⁰B(n,?)~7Li reaction under 1e V?2.4 Me V incident energy are studied by Bayesian neural network method.The incident energy and angle are used as inputs for machine learning.Because the incident energy range is up to 8 orders of magnitude,the incident energy is logarithmically processed in this work in order to reduce the accuracy error of the measurement data.On the basis of verification,the predicted values of differential cross section of any incident energy and any exit angle are given.At the same time,the ¹⁰B(n,?)~7Li reaction cross section in the range of 1e V?2.5 Me V is obtained by angle integration,which is in good agreement with the recent experimental data measured in China,and also in good agreement with the internationally famous ENDF/B?VIII.0 and JEFF3.3 evaluation database.The results show that the BNN method can reproduce the differential cross section and reproduce the cross section in ¹⁰B(n,?)~7Li reaction at the same time.Based on the above discussion,the BNN method is further extended to ¹²C(n,n)¹²C reaction with multiple formants.Firstly,taking the incident energy and angle as the two inputs of the BNN method,2453 sets of experimental data of elastic scattering differential cross section under all 170 energy points in the incident energy region in the range of 0.05?28.15Me V are learned by machine.On the basis of verification,the predicted value of differential cross section of any incident energy and any exit angle in the 30 Me V incident energy region has also been given.At the same time,the predicted value of elastic scattering cross section is consistent with the overall trend of experimental measurement,but there is a large deviation near the formant.Further analysis shows that the traditional BNN algorithm has shortcomings in the study of data with multiple formant structures.Therefore,this work attempts to add the energy level structure effect of composite nucleus(including the energy size and energy level width of each energy level)as a new characteristic quantity of the BNN method machine learning.The results can not only reproduce the experimental values of elastic differential cross section,but also obtain the theoretical values of elastic scattering cross section,which is in good agreement with the experimental data.The results show that the BNN method with energy level structure effect can improve the theoretical calculation results well and provide a good reference for the further study of light nuclear reaction data.