Comparison of the Prediction of Anaerobic Digestion Through Different Architectures of Neural Networks

• Autor(a) principal: Diogenes A.N.
• Data de Publicação: 2023
• Outros Autores: Sugai-Guerios M.H., Pykosz, Leandro Correa
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da Udesc
• dARK ID: ark:/33523/0013000003cw6
• Texto Completo: https://repositorio.udesc.br/handle/UDESC/2485

© IFIP International Federation for Information Processing 2023.In this work, multilayer perceptron and convolutional neural networks were trained in order to model the flow of biogas, from real data obtained from a treatment plant that uses anaerobic reactors for the treatment of domestic sewage using as input variables the physical- chemicals: sewage flow (L/s), COD (mg/L), TSS (mg/L) and SSV (mg/L). Based on these parameters, several simulations were performed in order to predict the biogas flow (Nm3/h). The simulations took place to identify which network would present the best performance, both used the same training criteria and the average of the results obtained with a 4–8-1, respectively were a regression coefficient R2 = 0.90 for the multilayer perceptron and for the convolutional R2 = 0.93 and an MSE error rate tending to zero.