Modelos robustos de previsão de vazão baseados em Wavelets e redes neurais artificiais

• Ano de defesa: 2019
• Autor(a) principal: Freire, Paula Karenina de Macedo Machado
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Civil e Ambiental; Programa de Pós-Graduação em Engenharia Civil e Ambiental; UFPB
• Palavras-chave em Português: Engenharia Civil e Ambiental; Previsão de vazão; Redes Neurais Artificiais (RNA); Transformada Wavelet Discreta (TWD); Civil and Environmental Engineering; Flow forecast; Artificial Neural Networks (ANN); Discret Wavelet Transform (DWT); CNPQ::ENGENHARIAS::ENGENHARIA CIVIL
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/26404

The economic development could be related to the quantity and quality of its water resources. Proper management of these resources is able to minimize the effects of various natural phenomena, such as droughts that directly affect the energy sector, which have recently led to blackouts in Brazil. Forecasting models are commonly used to provide better planning and operation of water supply systems. However, procedures that consider the short- to long-term forecast are still scarce in the literature, especially in the northeastern Brazil. In this way, the aim of this work was the development of robust streamflow forecasting models for different horizons (daily, monthly and annual). Eighteen models were developed, six for daily forecasting (RNAdQ→Q, RNAdP→Q, RNAdPQ→Q, WRNAdQ→Q, WRNAdP→Q e WRNAdPQ→Q); six for monthly forecasting (RNAmQ→Q, RNAmP→Q, RNAmPQ→Q, WRNAmQ→Q, WRNAmP→Q e WRNAmPQ→Q); and six for anual forecasting (RNAaQ→Q, RNAaP→Q, RNAaPQ→Q, WRNAaQ→Q, WRNAaP→Q e WRNAaPQ→Q), which were tested for streamflow forecasting to the Xingó reservoir. The efficiency of these models were tested and compared to the efficiency of the traditional models based on artificial neural networks (RNAdX, RNAmX and RNAaX), which are models that use antecedent streamflow data into Xingó reservoir to forecast future streamflow in the same reservoir. The efficiency was also tested and compared to the efficiency of the hybrid wavelet-artificial neural network models (WRNAdX, WRNAmX and WRNAaX), which are models that use antecedent streamflow data treated through the wavelet transform to forecast future streamflow in the same reservoir. The results showed that the hybrid models (WRNA) were superior to those using only artificial neural networks (RNA). Specifically, for daily forecastings, the best model was the WRNAdQ→Q hybrid model, which uses the transformed streamflows from the upstream reservoirs, as well as the combination of these streamflows, as input into the model for the streamflow forecasting in Xingó reservoir. For monthly and annual forecastings, the ideal models were the WRNAmPQ→Q and WRNAaPQ→Q hybrid models, respectively, which use the transformed precipitation over the Três Marias reservoir basin and the transformed streamflow into Xingó reservoir as input for each model to forecast the streamflow into Xingó reservoir.