Uma abordagem para identificar viabilidade de um local para implantação de ultrafiltração de água de chuva com a utilização de Deep Learning

• Ano de defesa: 2021
• Autor(a) principal: Pereira, Elias Morais
• Orientador(a): Rabello, Roberto dos Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade de Passo Fundo
• Programa de Pós-Graduação: Programa de Pós-Graduação em Computação Aplicada
• Departamento: Instituto de Ciências Exatas e Geociências – ICEG
• País: Brasil
• Palavras-chave em Português: Redes neurais (Computação); Precipitação (Meteorologia); Água - Purificação - Filtração; Aprendizado do computador; Chuvas
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://tede.upf.br:8080/jspui/handle/tede/2111

Neural networks, which are inspired by the concept of biological neurons, are commonly used in many applications, including in the field of weather forecasting. The solar irradiation at a specific location can help predict the amount of electricity that will be generated through solar panels, and an accurate prediction can help calculate the size of the system. In the same line, hydrometeorological measurements are one of the most challenging tasks in nature, in which precipitation has become the most significant and technical factor, where neural network approaches provide promising results to help in the decision-making process for precipitation forecasting. In this context, the present work establishes an approach for predicting precipitation and solar irradiance using deep learning models and in choosing a suitable site for installing a rainwater ultrafiltration system using photovoltaics. These models were applied to predict precipitation and solar irradiation for the next six months, according to the last month of historical data collected. Linear and machine learning models were tested and compared with the deep learning models in order to draw a line of tests and determine the RMSE of each model. The skillfull model chosen and employed in this work was the long short-term memory (LSTM). This model had an RMSE of 42.53 for precipitation and 0.45 for solar irradiation. It was superior to other deep learning models, such as MLP and ConvNet. Normalization and average combining calculations using three weights (0.7, 1.0, and 1.3) were used to corroborate the model’s viability. A web application was developed to present the results. According to the experiment performed, the approach proved to be adequate, and can serve as a decision making tool in calculations of the size of a photovoltaic system, predicted amount of rainwater, as well as for a rainwater ultrafiltration system.