Proceedings Article | 13 April 2020
KEYWORDS: Ions, Discrete wavelet transforms, Absorption, Feature extraction, Principal component analysis, Inverse problems, Absorption spectroscopy, Artificial neural networks
With the rapid development of industries such as mining activities, metal plating, pesticides and fertilizers manufacture, petroleum refining etc. heavy metal pollution has become a global threat. Unlike organic contaminants heavy metal ions are not biodegradable and tend to accumulate in soil, water and living organisms once being released in the environment. Heavy metal ions accumulated in an organism exceeding their critical level cause various diseases and disorders. That is why it is so vital to monitor the presence and concentrations of heavy metal ions in wastewaters. In this study we propose a new approach to solution of inverse multiparametric problem – determination of concentrations of 5 types of ions (Co2+, Ni2+, Cu2+, SO42-, NO3-) employing Raman and absorption spectroscopy, via adaptive data analysis methods. Such approach beneficially differs from conventional chemical analysis since it allows simultaneous and non-invasive monitoring of all the ions present in the aqueous solution compound in real mode. Aqueous solutions of CuSO4, NiSO4, CoSO4, Cu(NO3)2, Ni(NO3)2, Co(NO3)2 salts were studied. The database consisting of 3854 absorption spectra and 3823 Raman spectra of these solutions was obtained. Adaptive data analysis methods comprise various data-driven methods which don’t require any apriori knowledge about the object of the study. This peculiarity beneficially distinguishes such methods from conventional modal-based approach especially concerning multicomponent aqueous solutions of heavy-metal ions where non-linear interactions take place. In this study artificial neural networks of various architectures as well as partial least squares regression were used in order to solve inverse multiparametric problem of optical spectroscopy. Results of the application of all the methods are discussed in the report. This study has been performed at the expense of the grant of Russian Science Foundation (project No. 19-11-00333, K.A.L., S.A.B., I.V.I, T.A.D., S.A.D. – conducting experiment, processing of spectra, discussion) and was supported by Foundation for the Advancement of Theoretical Physics and Mathematics “BASIS” (project No. 19-2-6-6-1, O.E.S. – application of artificial neural networks).
