MODELO PREVISOR PARA SÉRIES DE TEMPO BASEADO EM REDES NEURAIS ARTIFICIAIS E DECOMPOSIÇÃO DE MODO EMPÍRICO

Abstract

 In this article, a comparative was made from results of the multi-step forecasts ahead (recursive strategy) of monthly flows collected at post 266-Itaipu. Forecasts from three methods were compared: ARIMA, Artificial Neural Networks (ANN) Feedforward and a hybrid method formed by Empirical Mode Decomposition (EMD), ANN Feedforward and Multiple Linear Regression (MLR). The hybrid method, called EMD-ANN-MLR,  presented minor forecasting errors to the individual ARIMA and ANN. The Mean Absolute Percentage Error (MAPE) obtained for a forecast twelve steps ahead using the ARIMA, ANN and EMD-ANN-MLR methods were 7.5%, 14.8% and 20.6%, respectively.