Singular value decomposition and artificial neutral network for analyzing bonner sphere data

The objective of this study was to build an effective and reliable method based on the artificial neural network (ANN) model for unfolding neutron spectrum. The number of counts measured by 15 Bonner spheres and 281 neutron spectra were selected as the database. After singular value decomposition was used to determine the relationship between Bonner spheres, 11 Bonner spheres were chosen as input descriptors. The three-layer feedforward neural networks (11-5-1) were employed to predict the spectrum in each energy bin. Using information entropy theory and the results of the ANN calculations, the sensitivity of each sphere to the entropy of the spectrum was quantitatively analyzed. The spectra results were compared with the results obtained using the maximum entropy method (MEM). The averaged root mean-square-error (MSE) of the MEM output and the desired spectra was 0.012; the averaged MSE of the ANN calculations was 0.006. The MSE results indicate that the 11-5-1 ANN models are able to accurately and reliably predict neutron spectra. The ANN model developed in this study to unfold neutron spectra from the counts measured by 11 Bonner spheres provides an alternative method for unfolding spectrum. The singular value decomposition is an effective method for the analysis of data obtained from Bonner spheres and the neutron spectra.