Predição por redes neurais artificiais da captura de CO2 por adsorção em carvão ativado

• Ano de defesa: 2024
• Autor(a) principal: Venturella, Suzan Roberta Tombini
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Francisco Beltrao; Brasil; Programa de Pós-Graduação em Engenharia Ambiental: Análise e Tecnologia Ambiental; UTFPR
• Palavras-chave em Português: Adsorção; Carbono ativado; Redes neurais (Computação); Aquecimento global; Mudanças climáticas; Adsorption; Carbon, Activated; Neural networks (Computer science); Global warming; Climatic changes; CNPQ::ENGENHARIAS::ENGENHARIA SANITARIA; Engenharia Sanitária
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/33962

In 2022, CO2 levels in the atmosphere reached the mark of 420 ppm and became a concerning matter among the scientific community. Climate change may cause, in an indirect way, adverse effects on people’s health due to changes in air pollution, spread of diseases, food insecurity and others. Thus, the investments foreseen in the Paris Agreement are essential to reducing the impacts caused by climate change, and it also gives the countries involved the opportunity to develop new technologies to mitigate CO2 emissions. As the rise in CO2 emissions is considered as the main cause for climate change, carbon capture becomes an alternative of these technologies, and a technique that can be used is gas adsorption on solid surfaces. Efforts are being made to apply computational techniques in industry and academia due to the cost and high time demand required for conducting experiments related to adsorption. Given these conditions and the challenges associated with using traditional computational methods to solve engineering problems, simpler and more versatile approaches have been explored. To study the uptake of CO2 on activated carbon under different experimental conditions, this work used artificial neural networks (ANN) feed with experimental data for pressure, temperature, superficial area of the adsorbent, CO2 and CH4 uptake and their molar fractions. For training the network it was used of the Levenberg-Marquardt algorithm and Matlab© software. The data was distributed as 70% for training, 15% for validation, 15% for testing and trained for 10, 15, and 20 layers. The network comprises the tangent activation function in the hidden layers and the linear function in the output layer. The mean squared error and the correlation coefficient were used to analyze the outputs obtained through the network test. The best result was obtained for the 20-layer training presenting MSE equivalent of 3.80x10-3 and an R2 of 0.98347 for the training data, 0.98328 for the validation data, 0.97365 for the test data, and it shows that the neural artificial network is capable of predicting the CO2 adsorption on activated carbon. Therefore, the uptake of CO2 through artificial neural networks makes it possible to study the uptake of CO2 on activated carbon. Uniting the experimental method to the machine learning method may be a tool capable of allowing the study for the capture of other greenhouse gases and reduce costs and time invested in laboratory experiments.