Research On Spectrum Reconstruction Technic For Water-Pumping-Injection Multi-sphere Neutron Spectrometer

Whether in the spatial region outside the dense atmosphere of the Earth or in the various types of neutron radiation fields on the surface,to obtain the neutron dose,the neutron energy spectrum must be obtained first,and the multi-sphere neutron spectrometer as one of the neutron energy spectrum measuring instruments.Because of its simple operation,isotropic and other characteristics are widely used.The traditional multi-sphere neutron spectrum measuring instrument can only measure about 20 MeV,and the number of detecting spheres is large.The overall instrument is abnormally heavy,and have "Sphere-sphere interference","poor consistency" and other issues,so this paper aims to develop a simple,high energy range and portable neutron energy spectrum measuring device and its unfolding method.In this paper,funded by the National Natural Science Foundation of China,"Research on the New Wide-Range-Energy Neutron Spectrum Detector for Multi-Homocentric Spheres by Pumping and Injecting Water"(No.41774120),Sichuan Key Research and Development Project “Research on Water-Pumping-Injection Multi-sphere Neutron Energy Spectrum Measurement System”(No.2019YFG0430)and the sub-question of Technical Basic Research Project of National Defense Industry Science and Technology Bureau "Space Radiation Neutron Measurement and Dose Calibration Technique"(JSJL2014404B001)-"Spatial Radiation Neutron Spectrum Analysis Technology",the design of a water-pumping-injection multi-sphere neutron spectrometer and its neural network unfolding method were carried out:(?)Firstly,study the influence on the air inside and outside the detector,changing the influence of moderating materials,different detector structures and different auxiliary metal materials on the response function(especially quantitatively study the response function of different detector structures and different auxiliary metal materials on high energy parts).Secondly,by using a combination of non-uniform gradient moderating structure and sandwich structure,a set ofwater-pumping-injection multi-sphere neutron spectrometer,which water is used as the main neutron moderating material,and lead is used as an auxiliary material,and water is injected or pumped through concentric spherical shells of different layers to change the moderating thickness,is designed.The new detector solves the problems of "sphere-sphere interference" and "poor consistency".Finally,the energy-fluence response function of 32 water injection conditions of the water-pumping-injection multi-sphere neutron spectrometer was studied by FLUKA Monte Carlo simulation program.(?)Through the study of the dimension reduction method of the water-pumping-injection multi-sphere neutron spectrometer reading matrix,the response function optimization of the water-pumping-injection multi-sphere neutron spectrometer is established.The IAEA No.403 report data(including 251 neutron spectra of reactors,accelerators,cosmic rays,etc.)compresses the reading matrix from 32×251 to 32×2 by the t-SNE method,ie,dimensional reduction from 251 to 2.It realizes the visual judgment of the similarity of the "detection sphere",so as to classify and screen,and compares the PCA method and the SVD method,and verifies the dimensionality reduction effect of the t-SNE method for the reading matrix of the water-pumping-injection multi-sphere neutron spectrometer is better.(?)Established a neural network unfolding method for the water-pumping-injection multi-sphere neutron spectrometer,and realized the analysis of the count value detected by the water-pumping-injection multi-sphere neutron spectrometer,and determined the optimal hidden layer node number and the best training function algorithm in the water-pumping-injection multi-sphere neutron spectrometer unfolding application;in order to solve noise problem of the instrument electronics and other parts in the actual water-pumping-injection multi-sphere neutron spectrometer counting measurement.The existing current noise,Gaussian noise and other problems have been denoised by the neural network unfolding method.The noise reduction processing results show that the neural network unfolding performance generated by the noise reduction training is stable,indicating that the noise reduction training can resist the counting to a certain extent.Due to the noise generated by electronics or other factors,it has certain practicability to analyze the count value of the water-pumping-injection multi-sphere neutron spectrometer.The main innovations of this paper:(?)A novel water-pumping-injection multi-sphere neutron spectrometer was designed,and the response function matrix of its 32 water injection conditions was studied by FLUKA simulation.(?)A multi-sphere neutron spectrometer classification method based on t-SNE algorithm was established.32 kinds of water injection conditions(detection spheres)were classified according to the water-pumping-injection multi-sphere neutron spectrometer.The classification results showed that t-SNE has a good effect on the response function matrix of the water-pumping-injection multi-sphere neutron spectrometer.Compared with the PCA and SVD classification methods,the difference by t-SNE classification methods between the detected spheres can be clearly displayed.(?)A neural network unfolding method for the water-pumping-injection multi-sphere neutron spectrometer with water injection is established,which can reconstruct the energy spectrum measured by the water-pumping-injection multi-sphere neutron spectrometer.