Predição das emissões fugitivas de gás metano na camada de cobertura final em aterro sanitário a partir das redes neurais artificiais

• Ano de defesa: 2023
• Autor(a) principal: Barreto, Carlos Alberto Alves
• Orientador(a): Paiva, William de
• Banca de defesa: Luiz, Marcia Ramos , Vieira, Fernando Fernandes , Santos, Gerson Marques Dos , Neto, Cláudio Luis de Araújo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso embargado
• Idioma: por
• Instituição de defesa: Universidade Estadual da Paraíba
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência e Tecnologia Ambiental - PPGCTA
• Departamento: Centro de Ciências e Tecnologia - CCT; Pró-Reitoria de Pós-Graduação e Pesquisa - PRPGP
• País: BR
• Palavras-chave em Português: Aterro sanitário; Gás metano; Redes neurais artificiais; Variações meteorológicas
• Palavras-chave em Inglês: landfill; methane gas; artificial neural network; meteorological variations
• Área do conhecimento CNPq: ENGENHARIA SANITARIA
• Link de acesso: https://repositorio.uepb.edu.br/handle/123456789/72134

The quantification of methane gas flow rates through the final cover layer in a landfill can vary over time and space, so the tools used to predict methane gas behavior are essential. The present study aimed to determine the efficiency of the final cover layer in a landfill through in-situ experimental data, environmental variables, and geotechnical factors for predicting fugitive methane emissions using an artificial neural network predictive model. The experimental field of research was the Sanitary Landfill located in the semi-arid region of northeastern Brazil. The determinations of gas flows through the cover layer were carried out using the static flux chamber test methodology in combination with measurements of pressure, gas concentrations, internal temperature, external temperature, and wind speed. Statistical treatments, Descriptive Data Analysis, Principal Component Analysis, Multiple Linear Regression, and finally the application of Artificial Neural Networks were performed. The synthetic database was designed using the multiple linear regression model, in which 1000 randomized and statistically validated values were generated. This study revealed that the best model obtained to predict methane gas emission occurred in the input layer with ten neurons, in the hidden layer with twenty neurons and in the output layer with one neuron, with the activation function of the hidden layer as Tansig and the activation function of the output layer as Purelin, following the architecture (10-20-1), with a Mean Absolute Error (MAE) of 0.001, Determination Coefficient (R2) of 1.00, Efficiency Coefficient (E) of 1.00, Root Mean Square Error (RMSE) with a value of 0.001, and Normalized Root Mean Square Error (NRMS) of 9.3E-0.8, obtaining the best training algorithm as Bayesian Regularization Backpropagation. When observing the predictive scale, methane flow values outside those recommended by standards were presented, with a maximum value of 492.73 g.m-2.dia-1. The prediction of the results showed that the modeling of the artificial neural network can effectively predict the methane flow through the final landfill cover layer, which is important for determining its efficiency. The efficiency demonstrates the need to establish parameters so that the aspects observed for the cover layer are efficient in keeping methane gas emissions within the criteria determined by international standards, bringing improvement to both environmental and financial scale aspects for these sites.