Aplica??o de redes neurais na predi??o de disponibilidade de recursos energ?ticos solares no Munic?pio de Serop?dica

• Ano de defesa: 2019
• Autor(a) principal: Araujo, Erylaine Reis Rubim Moreira
• Orientador(a): Silva, Robson Mariano da
• Banca de defesa: Delgado, Angel Ramon Sanchez, Coutinho, Elu? Ramos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal Rural do Rio de Janeiro
• Programa de Pós-Graduação: Programa de P?s-Gradua??o em Modelagem Matem?tica e Computacional
• Departamento: Instituto de Ci?ncias Exatas
• País: Brasil
• Palavras-chave em Português: radia??o solar; redes neurais artificiais; regress?o linear m?ltipla
• Palavras-chave em Inglês: solar radiation; artificial neural networks; multiple linear regression
• Área do conhecimento CNPq: Matem?tica
• Link de acesso: https://tede.ufrrj.br/jspui/handle/jspui/5279

The present study aims to develop and evaluate a methodology for estimating solar radiation in the city of Serop?dica, located in the state of Rio de Janeiro. The objective of this work is to evaluate if the Artificial Neural Networks (ANNs) model is appropriate for this purpose. For that, were obtained hourly data corresponding to the period from May 1, 2017 to January 31, 2019 of the National Meteorological Institute (INMET) through the Agricultural Ecology station located in the region of study. Firstly, it was assessed whether there was a need to use all the data provided by the station. The following experiments were performed by varying the number of neurons in the hidden layer, training networks composed of one and two inner layers. Different statistical parameters were used to evaluate the performance of the models (r, MAE, RMSE, D, R2, C and skill). At each stage of the work, the ANN models were compared with the Multiple Linear Regression (MLR) to verify which method would be satisfactory. As a result, it was possible to analyze that there is no need to use all the variables made available by the Agricultural Ecology station. Analyzing the average of the 50 simulations performed, it was possible to verify that the RNA with the architecture of a hidden layer presented more accurate results than the others, with an average confidence index (D) of 88% and an average determination coefficient (R2) of 85%. Even though they were superior, the RNA models did not show significant gains compared to RLM models. Thus, it was possible to conclude that ANNs are an adequate tools to estimate the incidence of solar radiation.