André Luiz França Batista, Thelma Safadi, Willian Soares Lacerda
Proceeding of IX Congresso Brasileiro de Redes Neurais/Inteligência Computacional, 2009. v. 1.
Publication year: 2009

Abstract

The present work investigates two different methodologies to perform prediction of time series of fluvial flows. Models from the Box & Jenkins method were constructed, as well as models based on the technique of Artificial Neural Networks. The proposed models were used to predict future values of the historical series of natural discharges from Rio Grande. The time series data were collected in the control section at the fluviometric station of Madre de Deus de Minas, MG. Subsequently a comparative analysis was made between both techniques used in the prognosis of the time series. The results obtained in the comparison show that each methodology can be adjusted appropriately to the set of observations under study, however each technique has advantages and disadvantages.

Keywords

Artificial neural networks, Agrometeorology, Times series forecasting.