Study of resilience of neural network solution of inverse problem based on integration of optical spectroscopic methods to noise in data

For many methods of optical spectroscopy, there is no analytical and/or direct numerical solution for the problem of determination of concentrations of each component in multi-component solutions by spectra. Therefore, recently, the application of machine learning methods to solve these spectroscopic inverse problems (IP) has been actively investigated. In our previous studies, it was suggested to use an ensemble of optical spectroscopy methods to increase the accuracy of the solution obtained by machine learning methods. Joint use of Raman spectroscopy and optical absorption spectroscopy methods to determine the concentrations of heavy metal ions in water using neural networks was considered. In this paper, we investigate the resilience of the considered IP to noise in data. The task was set to find out whether the joint use of these two types of spectroscopy can improve resilience of the solution to noise in input data of the considered IP in comparison with the case of using each of these types of spectroscopy separately. As possible alternative ways to increase the resilience of the neural network solution of this problem, the previously studied methods of group determination of parameters were considered. The main result is similar to that of the previous studies: combination of a “strong” method with a much “weaker” one does not allow one to increase the results of the “strong” method alone. This regards not only the error of the IP solution, but also its resilience to noise in the input data.