Predição de nível de rios da Região Hidrográfica do Rio Paraguai utilizando algoritmos de Aprendizado de Máquina

• Ano de defesa: 2022
• Autor(a) principal: Rogério Alves dos Santos Antoniassi
• Orientador(a): Renato Porfirio Ishii
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Predição; Aprendizado de Máquina; Nível de Rio
• Link de acesso: https://repositorio.ufms.br/handle/123456789/4642

The Pantanal is part of the Paraguay River Hydrographic Region (RH-Paraguay) and has, as a characteristic, the flooding of a portion of its area at certain times of the year. This flood generates a series of socioeconomic problems for the population living in its surroundings. These problems can be mitigated when their occurrence is predicted in advance. In this sense, this work investigates the application of Machine Learning (ML) techniques for the prediction of river levels in RH-Paraguay, using daily data from upstream stations to predict the level of downstream stations. From this perspective, modeling with ML techniques proves to be effective in predictions, as similar works appear in the related literature. In this way, a data source with daily level values is used, and a sample of three stations is selected. Then, the time lag between the flood peak of one station and another was removed, searching the ideal hyperparameters for the Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU) and Bidirectional Long Short Term Memory (BiLSTM) networks and submitted to the training process. Subsequently, the models with the best results of each algorithm were selected, which were compared with the Regression technique currently used. The results show that the three models tested can be used for the prediction task, in which the three present improvements in relation to the current model. The model with the GRU algorithm stood out for presenting the lowest error rates and for being 23.84% more accurate than the Regression model, while LSTM and BiLSTM are, respectively, 18.09% and 19.16% more accurate than the Regression model. The LSTM and BiLSTM models are closer to the real value in the peaks of maximum and minimum levels, when compared to the Regression and the GRU.