Planejamento da geração de energia fotovoltaica usando inteligência artificial em suas diferentes topologias de instalação

• Ano de defesa: 2024
• Autor(a) principal: Piotrowski, Leonardo Jonas
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Elétrica; UFSM; Programa de Pós-Graduação em Engenharia Elétrica; Centro de Tecnologia
• Palavras-chave em Português: Inteligência artificial; Geração de energia fotovoltaica; Degradação da potência de saída fotovoltaica; Aprendizado profundo de máquina; Topologias de instalação fotovoltaica; Artificial intelligence; Photovoltaic power generation forecast; Photovoltaic output power degradation; Deep learning; Photovoltaic installation topologies; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/33907

Solar energy is used worldwide to generate electricity with photovoltaic (PV) panels and is one of the fastest growing alternatives that contributes to meet the global energy demand. Among some of the factors that make this type of electricity generation an excellent alternative is the fact that it is a renewable, noise-free and universal source. There are several types of installation (topologies) for photovoltaic modules, such as using solar tracking, cooling and the traditional fixed topology that can be commonly used in photovoltaic plants installed on the ground and/or on rooftops. In all related projects, it is important to carry out an initial planning that adequately meets the demand for electricity and the return on investments involved, regardless of the installation topology. Among the various factors, and one of the most important, is the forecast of the energy generation that the installation will be able to provide. In this sense, this study addresses planning that consists of applying artificial intelligence (AI) considering the prediction of solar irradiance, durability and performance of modules in different installation topologies. Knowing that the degradation of the panel output power can assume different values, it needs to be estimated and accounted for in order to adequately model the prediction of the generated energy, providing greater reliability and availability of photovoltaic generators. In this thesis, AI contributed to a better planning of PV installations in two aspects. On the one hand, deep machine learning was applied in the training of a neural network to make it capable of predicting solar irradiance and, on the other hand, fuzzy logic was applied to find the degradation of the panel installed under different topologies. In short, this modeling consisted of collecting a history of solar irradiation data, processing this data and inserting it into a recurrent neural network (RNN) with long and short-term memory (LSTM) to perform the prediction. Fuzzy logic together with technical information about the modules and the type of topology used was applied to find out the degradation and, consequently, the prediction of the electrical output power. Several computational simulations were performed to adjust the developed model. The financial aspects of the PV topologies studied here were accounted for to verify the economic viability. The results obtained showed that both topologies are viable and, although the solar tracking topology results in greater degradation in relation to the others, it still compensates for the greater amount of electricity generated in the analyzed period. The prediction of solar irradiance reached a satisfactory accuracy compared to other methods listed in this thesis. In addition, the predicted solar irradiance was validated with real data obtained for the city of Santa Maria, RS, Brazil. A reduction in prediction error associated with a better understanding of PV panel durability contributed to a more reliable planning of different installation topologies.