Abordagens inteligentes para estimar a produção de energia eólica

• Ano de defesa: 2018
• Autor(a) principal: Gebin, Luis Gustavo Gutierrez
• Orientador(a): Salgado, Ricardo Menezes
• Banca de defesa: Gonzaga, Flávio Barbieri, Ohishi, Takaaki
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Alfenas
• Programa de Pós-Graduação: Programa de Pós-Graduação em Estatística Aplicada e Biometria
• Departamento: Instituto de Ciências Exatas
• País: Brasil
• Palavras-chave em Português: Energia Eólica; Modelos Estatísticos
• Área do conhecimento CNPq: PROBABILIDADE E ESTATISTICA::PROBABILIDADE E ESTATISTICA APLICADAS
• Link de acesso: https://repositorio.unifal-mg.edu.br/handle/123456789/1287

One of the major concerns of the twentieth century is to reconcile economic and social development with environmental preservation. Thus, the energy sector becomes the focus of study, since the part of emission of power gases is derived from the electric generation. As a consequence, renewable and clean energy sources, which do not cause harm to nature, such as a wind energy production, have gained prominence in Brazil and the world. Once again, the production of wind energy is an energy that can be taken out of the wind, man does not have total control over his generation, which makes it desirable that there is some confidence as to his electric potential. The justification for the development of these forecasting methods could be further extended by the energy potential of production and efficient in other years in terms of energy distribution. Besides that, you can have more security to decrease the thermal complementation, because the thermoelectric turbines are used in Brazil in seasonal periods, when hydroelectric production is low. However, an analysis of wind data is not a trivial task due to the existence of exogenous variables that can affect production. Seen this, the work aims to make the prediction of a week on an hourly scale of wind power production in the four seasons of the year with the intelligent models (Random Forest and XGBoost) and a ARIMA model, after that, apply in intelligent models a selection features, besides proposing a new model based on ensemble. From the results obtained, it can be seen that the Random Forest was the model most benefited by the selection feature, while the XGBoost, even without a selection, managed to perform interestingly, given its approach method. The ARIMA model, even without perfect fit, is a bit inferior to the smart models. As for the ensemble model, it is perceived that it was superior to the most disadvantageous models, especially in relation to RMSE. It is worth emphasizing that the ensemble model can improve in the prediction, mainly, in the extreme data, in which all the individual models overestimated.