This study explores the training of a neural network to determine the mineral composition of complex mineral associations formed in the thermal fields of Kamchatka, which are associated with modern geothermal systems. Typical secondary minerals of productive reservoirs are key to characterizing the stages and conditions of the development of modern geothermal systems, the exploration of which is crucial for the advancement of geothermal energy. These typical secondary minerals include kaolinite, montmorillonite, quartz, opal, minerals of the alunite and jarosite groups, gypsum, calcite, and zeolite minerals. All of them possess individual infrared spectra and can be interpreted individually by an operator. However, in complex mixtures, the absorption bands overlap. To recognize the qualitative mineral composition in such cases, machine learning was applied. We utilized convolutional neural networks, and the training dataset contained approximately one thousand labeled spectra, which enabled the achievement of high accuracy in determining the mineral composition using the neural network.
Sergeeva et al. (Thu,) studied this question.