Técnicas de previsão do recurso solar integradas a partir da teoria do portfólio

• Ano de defesa: 2020
• Autor(a) principal: Lima, Marcello Anderson Ferreira Batista
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Engenharia elétrica; Energia renovável; Teoria da previsão; Energia solar - Previsão; Teoria do portifólio; Solar resource forecast; Portfolio theory
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/52896

In the last decades, several forecasting methods have been implemented in order to improve the forecast of intermittent energy resources. Motivated by the growth in the use of Photovoltaic (PV) plants, this thesis develops a technique called PrevTP, based on the integration of forecasting techniques using the Portfolio Theory (TP). TP is a tool used in the financial sector so that the risk of loss of investments is reduced through the diversification of assets. For solar predictability, TP, through PrevTP, is adapted with a focus on reducing forecasting errors. PrevTP takes advantage of diversified forecast assets, that is, when one of the assets obtains forecast errors, the other asset compensates for the error through pre-defined weightings based on the methodology used. As PrevTP predictive assets, this thesis uses the structures of the learning techniques: Multilayer Perceptron (MLP) Backpropagation; Radial Basis Function (RBF); Support Vector Regression (SVR); and Deep Learning (DL). The technique developed is applied to two sites with different solar irradiation conditions: Fortaleza, Brazil, and Algeciras, Spain. Using the 4 techniques, it is possible to perceive the reduction of forecasting errors through integration. The stages of application of PrevTP are: data collection, prediction of solar irradiance through the assets individually, study of forecast errors, data processing of forecast errors by PrevTP, final definition of asset weightings by PrevTP and, finally, verification of forecasting errors with the 4 techniques integrated by the proposed methodology in comparison with the individual assets. The results obtained show that Mean Absolute Percentage Error (MAPE) for predictions using PrevTP is 4.52% in Brazil and 5.36% in Spain. In both cases, PrevTP results are better than other techniques used alone, with MAPE values between 6.08% to 8.53%, which characterizes PrevTP as a tool with positive impacts for the management of solar energy