Mapas auto-organizáveis aplicados à análise de poluentes atmosféricos

Abstract

Air pollution is a problem that is increasingly present in our society due to the growing development of countries. In the study of air pollutants, multivariate statistical methods are commonly used, however machine learning has proved to be a great alternative, presenting techniques capable of dealing with highly complex problems, such as air pollution. In this work, the machine learning technique, Self-Organizing Maps (SOM), was used to explore and analyze data on atmospheric pollutants and meteorological parameters from an air quality monitoring network, with stations located in the city of Salvador - Bahia. SOM offers several resources capable of making the study of data more comprehensive, which were used for the development of an individual and mutual analysis on the stations, being also compared with a principal component analysis. From the application of the SOM, it was possible to identify correlations between the pollutants of the studied stations, highlighting mainly the similarity between the pollutants NO, NO2 and CO, as well as the similarity of the MP10, SO2 and O3 with the meteorological parameters in each station. Based on the arrangement of neurons and the formation of the clusters performed, the SOM allowed the identification of characteristics related to the set of samples, enabling a study on the formation of clusters, distributions, concentrations of atmospheric pollutants and meteorological and how each monitoring station is related to its formation/contribution, enabling a more comprehensive analysis than that offered in traditional methods. The results obtained showed how useful and effective SOM can be in air pollution problems, highlighting different possibilities in the form of data analysis that can be explored for the development of new researches.