Previsão da produção de uma usina fotovoltaica usando redes neurais artificiais
Carregando...
Data
2019-11-05
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Goiás
Resumo
The solar or photovoltaic power generation forecasting is a very complex subject, since it involves the development of complex statistical models. The present work seeks to circumvent the complexity of statistical models through the use of a recurrent neural network type, thus bringing a different dynamic to the problem. Long Short-Term Memory-LSTM networks were used to forecast at horizons of 15 minutes, one hour and one day ahead. The windowing technique was used together with the LSTM network to make the prediction, thus avoiding the use of linear regression given the stochastic nature of the problem addressed. Power generation and climate data were collected from 6:00 to18:00 from the photovoltaic plant and weather station, respectively installed at the B block of the School of Electrical, Mechanical and Computer Engineering (EMC) of the Federal University of Goiás (UFG). Three different LSTM neural network models were implemented and trained with the data from the three time horizons (15 minutes ahead, one hour ahead and one day ahead), thus selecting the models that presented the smallest error, between the real value and the predicted value.
Descrição
Palavras-chave
Previsão de geração fotovoltaica, Redes neurais recorrentes, Long Short-Term Memory, Técnica da janela, Photovoltaic generation forecasting, Long Short-Term Memory - LSTMRecurrent neural networks, Window technique
Citação
SANTOS, Silvio Silva dos. Previsão da produção de uma usina fotovoltaica usando redes neurais artificiais. 2019. 63 f. Trabalho de Conclusão de Curso (Graduação) – Escola de Engenharia Elétrica, Mecânica e de Computação, Universidade Federal de Goiás, Goiânia, 2019.