Previsão da velocidade dos ventos por redes neurais artificiais e arima de box & jenkins

• Ano de defesa: 2013
• Autor(a) principal: Rossi, Daniela Jardin [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Redes neurais (Computação); Previsão com Metodologia de Bonx-Jenkis; Ventos - Velocidade; Box-Jenkis forecasting
• Link de acesso: http://hdl.handle.net/11449/111121

This work presents a hybrid methodology using MLP (multilayer perceptron) neural network and statistical ARIMA of Box and Jenkins to predict wind time series. The ARIMA models predict the time series data that are nonlinear, and the error obtained is introduced in the MLP by backpropagation training, forming the hybrid additive model. Tests were done with two data sets. Test 1 uses data from Ilha Solteira, SP region developed by an Agrometheorogical project of UNESP Campus of Ilha Solteira. Test 2 the data are from a project developed by INPE (National Space Research Institute) which provides solar and wind data in adequate high to energy generation in several regions of Brazil. Results are analyzed by MAPE (mean absolute percent error) and are in accordance with those found in the specialized literature