Estimation of the concentrations of harmful impurities in water using laser Raman spectroscopy and machine learning methods
K.A. Laptinskiy1, S.A. Burikov1,2, A.M. Vervald2, A.A. Guskov1,2, I.V. Plastinin1, O.E. Sarmanova1, L.S. Utegenova2, T.A. Dolenko1,2
1Lomonosov Moscow State University, Skobeltsyn Institute of Nuclear Physics, Moscow, Russia
2Lomonosov Moscow State University, Faculty of Physics, Moscow, Russia
Keywords: Raman spectroscopy, wastewater diagnostics, machine learning method, spectra preprocessing, artificial neural network
Abstract
In the course of technological development of society, the problem of violation of the ecological state of the environment inevitably arises, in particular in water reservoirs. In order to proper respond to the changes in the concentrations of various pollutants in natural water reservoirs, it is necessary to develop an express remote method. The creation of such a method is possible on the basis of Raman spectroscopy. However, during its development, a large number of various difficulties arise, in particular regarding the method of preprocessing the obtained data. The paper presents the results of using machine learning methods to develop a remote method for determining the type and concentration of dissolved ions in aqueous media from Raman spectra. The use of artificial neural networks made it possible to identify and simultaneously determine the concentration of each of the eight ions (Zn2+, Cu2+, Li+, Fe3+, Ni2+, NH4+, SO42, and NO3-) in a multicomponent aqueous mixture with errors, which meet the needs of environmental monitoring of natural and waste waters. A significant influence of the method of preprocessing Raman spectra on the result of solving the inverse spectroscopic problem is discovered. The results can be used for solution of the multiparameter inverse problem of qualitative and quantitative determination of ions in water.
|
