Barros, Monica; Castro, Reinaldo; Gutierrez, Enrique
DEE
Modelling and forecasting highly volatile time series is a complex task, where linear models are not adequate. An example of such series is given by the electricity spot prices in Brazil. In this article we apply neural networks to forecast monthly spot prices up to six months in advance in the Northeast region of Brazil. The input variables considered are Stored Energy, Natural Inflow Energy and past values of spot prices. Two models are presented, each comprising six different network structures. Each network, in turn, specializes in a given forecast horizon. The forecasting performance of both models will be compared and it will be shown that the model which includes all input variables hás the best performance.
