Machine learning models study for assessing the effect of coal mining and energy enterprises on ecosystems

Introduction. The development of coal - fired mining and energy enterprises leads to a significant increase in ecosystem pollution. Emissions from coal - fired power plants pose a health hazard to humans, since when coal is burned, pollutants are released from the chimney into the atmosphere and carried by the wind over significant distances. These emissions contain elements and compounds such as Hg, Pb, N2O, SO2, solid particles and various heavy metals, many of which belong to hazard classes I and II of harmful substances. Since soil is a specific indicator of the state of the environment, studying the role of anthropogenic substances in the components of the natural environment, their impact on living organisms, and migration along food chains is an urgent task. The new methods and mathematical models of monitoring being developed make a certain contribution to the development of the theory of anthropogenic soil pollution caused by emissions from coal - fired thermal power plants. The aim of the research is to improve the efficiency of assessing the impact of industrial pollution on ecosystems by predicting the gross content of metals in the soil cover. Materials and methods. The modeling methods described in this article are a continuation of the research carried out at the Southern Federal University under the supervision of T.M. Minkina. Taking into account the analysis of the results of monitoring studies, mathematical modeling methods were applied, including classical approaches and modern methods of data mining, such as ensemble and neural network methods of machine learning. The subject of the study in this paper is the methods of forecasting the gross content of heavy metals in soils adjacent to the Novocherkassk State District Power Plant. The object of the study is coal and energy complex enterprises that emit pollutants into the atmosphere during the implementation of the technological process, which are then carried by the wind over significant distances. Results. A number of empirical studies were conducted that meet the objective of developing models that most accurately approximate experimental data. As a result, models were obtained for predicting the gross content of heavy metals in soils at 28 sites located in the vicinity of the Novocherkassk State District Power Plant. Discussion. The obtained machine learning models allow simulating the value of the gross content of heavy metals in the soil cover for a year ahead with an error not exceeding 12% on average for each metal. Regarding the formation of the feature space, the most significant factors were identified: site number, distance from the pollution source, direction from the pollution source, retrospective data on the gross content of metals. Also, during the study, it was found that grouping metals by the scale of gross content in soils improves the modeling result. It was proven that the use of a panel data set, compared to individual samples for each group of metals, is a more effective forecasting method. In addition, the models described in the study take into account the situation in which it is possible to obtain a forecast for a site where pollution levels have not been previously monitored (without retrospective data). At the same time, adding soil property data to the model did not affect the increase in the quality of the forecast models. Conclusion. During the research, machine learning models were created to assess the impact of coal mining and energy enterprises, which made it possible to obtain new results for modeling the gross content of heavy metals in soils. A number of hypotheses were tested regarding the formation of the feature space, the training data set, and the assessment of the prospects for using forecasting methods (classical and intelligent). A comparative analysis of the developed models was performed using the proposed optimization algorithms and recommendations were formed for choosing the most preferable one based on the assessment of the quality of the modeling results. Resume. The analysis of the obtained results showed that the technogenic impact on ecosystems by enterprises in the coal and energy industry is significant and, using the created models, can be subject to control and modeling of the content of heavy metals in the soil cover. Suggestions for practical application and directions for future research. The developed models and methods using machine learning algorithms should be used in monitoring technogenic pollution of ecosystems not only by coal power plants, but also by other mining enterprises. For industrial enterprises where aerosol particles are distributed during technological processes, the results of this work are applicable provided that the relevant data are entered into the software and the influencing parameters and factors of a geographical and climatic nature are adjusted.