• Login
    View Item 
    •   DSpace Home
    • BIBLIOTECA DIGITAL: Teses & Dissertações
    • 40001016002P6 Programa de Pós-Graduação em Ciências Geodésicas
    • Dissertações
    • View Item
    •   DSpace Home
    • BIBLIOTECA DIGITAL: Teses & Dissertações
    • 40001016002P6 Programa de Pós-Graduação em Ciências Geodésicas
    • Dissertações
    • View Item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    Estimativa de cotas e vazões de curto prazo em bacias hidrográficas usando redes ConvLSTM

    Thumbnail
    View/Open
    R - D - ALEX BRUNO KRAEMER.pdf (4.087Mb)
    Date
    2022
    Author
    Kraemer, Alex Bruno, 1993-
    Metadata
    Show full item record
    Subject
    Hidrologia - Modelos
    Vazante
    Geodésia
    xmlui.dri2xhtml.METS-1.0.item-type
    Dissertação Digital
    Abstract
    Resumo: Nos últimos anos, o aprendizado de máquina (AM) é uma das linhas de pesquisa que tem se fortalecido na resolução de problemas cada vez mais complexos, estes problemas demandam de ferramentas computacionais sofisticadas e autônomas, reduzindo a necessidade de intervenção humana. Embora o advento de AM ofereça inúmeros benefícios, ainda há desafios em seu uso para a modelagem hidrológica, visto que, de acordo com HERATH et al. (2020), até o momento, nenhum modelo hidrológico consegue ter um desempenho igualmente satisfatório em toda a dinâmica dos fatores ambientais presentes em uma bacia hidrográfica. Atualmente, as inundações constituem o principal desastre natural no mundo e ocorrem durante eventos de alta pluviosidade aliado as características do relevo e uso do solo. Para tentar resolver este problema, neste trabalho é proposto o treinamento de um modelo de rede neural artificial (RNA) denominado ConvLSTM, empregando a abordagem distribuída. Para tal, foi necessário estabelecer os dados que influenciam diretamente no escoamento superficial d’água, sendo: (1) cota fluvial, (2) relevo, (3) Curve Number (CN) e (4) precipitação. Em seguida, processou-se os dados para se adequarem ao padrão de entrada da ConvLSTM, a qual utiliza o formato matricial. Para o treinamento do modelo, utilizou-se a plataforma Google Colab com processamento em nuvem, linguagem python 3 e as bibliotecas keras. Realizaram-se três experimentos: no primeiro, utilizou-se o modelo ConvLSTM, sem dados sintéticos; no segundo, acrescentaram-se dados sintéticos de chuva nas amostras de treino; no terceiro, empregaram-se dados sintéticos de cotas fluviais. Confrontando os resultados da estação de referência, observou-se que o acréscimo de dados sintéticos nas amostras de treino melhorou a capacidade do modelo em predizer os valores de cotas fluviais para a estação de referência.
     
    Abstract: In recent years, machine learning (ML) is one of the research fields that has grown in importance for the resolution of increasingly complex problems. Such problems call for sophisticated and autonomous computational tools, thereby reducing human intervention. Even though the advent of ML brings countless benefits, there are many challenges concerning their use in hydrological modeling, as according to HERATH et al. (2020), currently, no hydrological model can perform equally well across the dynamic of all environmental factors present in a watershed. Floods are currently the world's major natural disaster that occurs during high rainfall associated with relief and land use characteristics. To address this issue, this paper proposes the training of an artificial neural network (ANN) model entitled ConvLSTM, with a distributed approach. To achieve this goal, it was necessary to establish the water surface runoff data, which are: (1) fluvial height; (2) relief; (3) Curve Number (CN); and (4) precipitation. Then, the data was prepared to fit the ConvLSTM input pattern, which employs the matrix format. Google Colab platform with cloud processing, python 3 language, and Keras libraries were used to implement the model training. Three experiments were carried out: in the first one, the ConvLSTM model was deployed without synthetic data; in the second one, synthetic rainfall data was added to the training samples; and finally, synthetic river level data was employed. By comparing the results at the reference station, one could observe that the addition of synthetic data in the training samples enhanced the model's power to predict the fluvial height of the reference station.
     
    URI
    https://hdl.handle.net/1884/76575
    Collections
    • Dissertações [268]

    DSpace software copyright © 2002-2016  DuraSpace
    Contact Us | Send Feedback
    Theme by 
    Atmire NV
     

     

    Browse

    All of DSpaceCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsxmlui.ArtifactBrowser.Navigation.browse_typeThis CollectionBy Issue DateAuthorsTitlesSubjectsxmlui.ArtifactBrowser.Navigation.browse_type

    My Account

    LoginRegister

    Statistics

    View Usage Statistics

    DSpace software copyright © 2002-2016  DuraSpace
    Contact Us | Send Feedback
    Theme by 
    Atmire NV