Neural network for local atomic structure analysis

In this paper, we consider the possibility of studying the local atomic structure of a substance by analyzing extended fine structures of electron energy loss spectra (EXELFS - Extended Electron Energy Loss Fine Structure) using neural network technologies to determine the structure parameters: partial coordination numbers, chemical bond lengths and thermal dispersion parameters. The experimental data of EXELFS spectra of one-component systems obtained in the transmission mode were chosen as model objects. At the first stage of the work, the interatomic distance was chosen as the parameter to be determined, since minimal processing is required to obtain this characteristic from the initial experimental data. Various learning methods and types of neural networks are considered: supervised learning, unsupervised learning and reinforcement learning. The analysis of the work of researchers in the field of spectroscopy, solving various problems using machine learning methods, is carried out. The results of the operation of an additional module capable of analyzing the spectra of one-component systems are shown.