Artificial Neural Networks Modeling Evolved Genetically, a New Approach Applied in Neutron Spectrometry and Dosimetry Research Areas

Recently, the use of the artificial neural networks technology has been applied with success in the research area of nuclear sciences, mainly in the neutron spectrometry and dosimetry domains, however, the structure (net topology), as well as the learning parameters of the neural networks, are factors that contribute in a significant way in the networks performance. It has been observed that the researchers in the nuclear sciences area carry out the selection of the network parameters through the trial and error technique, which produces poor artificial neural networks with low generalization capacity and poor performance. It has been observed that the use of the evolutionary algorithms, seen as search and optimization approaches, it has allowed to be possible to evolve and to optimize different properties of artificial neural networks, such as the proper synaptic weight initialization, the optimum selection of the network architecture or the selection of the training algorithms. The aim of the present work is focused in analyzing the intersection of the artificial neural networks and the evolutionary algorithms, analyzing like it is that the evolutionary algorithms can be used to help in the design processes and training of an artificial neural network, in such a way that the neural network designed is able to unfold in an efficient way neutron spectra and to calculate equivalent doses, starting only from the count rates obtained from a Bonner spheres spectrometric system.