Estimativa da energia elétrica gerada por um sistema fotovoltaico utilizando redes neurais artificiais

• Ano de defesa: 2018
• Autor(a) principal: Bianchi, Ermete Cauduro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Mato Grosso; Brasil; Faculdade de Arquitetura, Engenharia e Tecnologia (FAET); UFMT CUC - Cuiabá; Programa de Pós-Graduação em Engenharia de Edificações e Ambiental
• Palavras-chave em Português: Redes neurais artificiais; Backpropagation; Geração de energia elétrica fotovoltaica; CNPQ::ENGENHARIAS; Estimation; Artificial neural networks; Generation of photovoltaic electric energy
• Link de acesso: http://ri.ufmt.br/handle/1/2711

This work presents the studies realized to estimate the electric energy by a photovoltaic system connected to the grid, using artificial neural networks as a tool – RNA. To this end, eight different cases were implemented, with the inputs of the artificial neural network as the environmental variables. The answer of the artificial neural network is the estimated photovoltaic electric energy generation. The applied model of artificial neural network and the training are functions available in the software MATLAB. To be able to make estimates, RNA needs to be trained to "learn" how to respond to the environment. In the training of the artificial neural network were used the data of climatic variables: ambient temperature, relative humidity, wind speed and solar irradiation. Besides the climatic variables, the temperature of the photovoltaic panel was also used. The structure of the artificial neural network used was feedfoward backpropagation with Levenberg Marquardt learning algorithm, with two hidden layers, the first layer using logsig activation function and the second tansig, each with four neurons. The estimations of the generation of photovoltaic electric energy were carried out with environmental data of the meteorological station of the FAET, in the own place of the study, and of the automatic meteorological station of INMET, that is located around five kilometers of the place of study. To verify the performance of each estimated case, comparisons were made considering data at five minute and one hour intervals for the environmental data from the FAET station. Comparisons between estimates with data from the two stations at one hour intervals. Comparisons were also made including or not the panel temperature as input from the neural network. As a main result we have that the estimates are better when the data with an interval of one hour and with the data of the meteorological station of the FAET are used. It was also found that in most of the estimated cases, those who considered the surface temperature of the panel achieved better results.