Previsão de ozônio troposférico na região metropolitana do Rio de Janeiro com base em técnicas de imputação de dados faltantes e calibração multivariada

• Ano de defesa: 2022
• Autor(a) principal: Oliveira, Rafael Corrêa Gama de
• Outros Autores: [e-mail]
• Orientador(a): Corrêa, Sergio Machado
• Banca de defesa: Guarieiro, Lílian Lefol Nani , Alvim, Débora Souza , Baptista, Leonardo , Martins, Eduardo Monteiro
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade do Estado do Rio de Janeiro
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Ambiental
• Departamento: Centro de Tecnologia e Ciências::Faculdade de Engenharia
• País: Brasil
• Palavras-chave em Português: Engenharia ambiental; Ar - Controle de qualidade; Ar - Poluição; Ozônio atmosférico; Aprendizado do computador
• Palavras-chave em Inglês: Environmental engineering; Air - Quality Control; Air - Pollution; Atmospheric ozone; Computer Learning
• Área do conhecimento CNPq: ENGENHARIAS::ENGENHARIA SANITARIA::SANEAMENTO AMBIENTAL::CONTROLE DA POLUICAO
• Link de acesso: http://www.bdtd.uerj.br/handle/1/18175

Tropospheric ozone has a negative impact on the environment and, consequently, on public health. Due to this problem, this work proposes to study statistical models using machine learning methodology to predict tropospheric ozone at the Metropolitan Region of Rio de Janeiro. Multivariate calibration techniques based on Least Squares Regression, Random Forests, Support Vector Machines and Artificial Neural Network combined with the MissForest imputation algorithm were applied to understand the interaction between ozone with nitrogen oxides, carbon monoxide, windy velocity, solar radiation, temperature, relative humidity, among other pollutants, whose data were collected at the Metropolitan Region of Rio de Janeiro at four air quality stations of different profiles, in the localities of Adalgisa Nery, Porto das Caixas, INEA Laboratory and Vila São Luiz, between 2014 and 2018. These techniques provide an easy and feasible way of modeling and analyzing atmospheric pollutants, which can be used and combined with other statistical methods for model validation. The results showed that the chemometric techniques Random Forests, Support Vector Machines and Artificial Neural Network can be used to the modeling and prediction of tropospheric ozone concentrations, with a calculated determination coefficient up to 0.92. The root mean squared error of prediction, ranges between 4.17 and 22.45 µg m-3, depending on the air quality monitoring stations and season.