Inteligência artificial para modelar o processo chuva vs. vazão natural afluente ao reservatório de Sobradinho

• Autor(a) principal: Alexandre, José Pedro Lima
• Data de Publicação: 2022
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFS
• Texto Completo: https://ri.ufs.br/jspui/handle/riufs/18551
• topic: Engenharia civil; Ensino superior (UFS); Modelagem chuva-vazão; Represa de Sobradinho, BA; Redes neurais artificiais (RNA); Reservatório (Sobradinho, BA); Rio São Francisco; Vazões naturais; Hidrologia; Rainfall-runoff modeling; Artificial neural networks; Natural flows; ENGENHARIAS::ENGENHARIA CIVIL::ENGENHARIA HIDRAULICA::HIDROLOGIA

Brazil has a large amount of available water. However, part of the Northeast region suffers from a lack of this resource. The regularization of the flow regime of the São Francisco river with the construction of the Sobradinho reservoir contributed to the reduction of the abundant floods in the downstream region. This work presents the application of artificial neural networks (ANN) and the k-nearest neighbors algorithm (KNN) for modeling the rainfall-runoff process considering the natural flow to the Sobradinho reservoir. Precipitation data were collected from the HidroWeb portal and natural inflow data from the Câmara de Comercialização de Energia Elétrica (CCEE) portal. The set of data were divided into calibration (70%) and validation (30%), at random. Simulations were performed by using the Weka machine learning software, and four formulations were tested for monthly analysis. The goodness of fit of the results are shown by means of the Nash-Sutcliffe coefficient. The initial objective was to model the runoff-runoff process in Sobradinho in order to predict the total inflow of the next year based on the flows of past years. Several models were tested, with several configurations of attributes, however, the results were all unsatisfactory for the four formulations of the annual analysis with flows. For this reason, it was decided to use the monthly analysis, with rainfall and flow data. Thus, good and very good results were obtained for the four formulations, in both models investigated, ANN and KNN. In modeling, the formulation with rainfall-runoff attributes from three previous periods showed the best results for ANN and the formulation with only precipitation attributes showed the best results for KNN, with efficiency indices and classification of very good.