Renewable energy, ARIMA; RNA, Holt-Winters, Hybrid modeling.
This work aims to conduct monthly average forecasts of wind speed in regions of northeastern Brazil. For this, we used the following time series models, Auto Regressive Integrated Moving Average (ARIMA) and Holt-Winters (HW), which are commonly used in the literature, and computational artificial intelligence using Artificial Neural Networks (ANN). Still two hybrid models were tested, the first results from the combination of ARIMA and ANN models, which are similar studies in the literature, and the second it is an attempt to innovative hybrid model, which is a result of the combination of HW models with RNA. The results show that for most study sites the hybrid model HW with RNA, performed better compared to others. An example of this, in terms of statistical errors in Fortaleza and Natal were found lower values of MAPE variable in the order of 3.80% and 2.85% respectively. Yet the time series provided by this model showed better adjustments to the data of the observed time series, ie, showing greater similarity between maximum and minimum wind speed comparing the two series. This work may have utility as a possible tool for the exploitation of wind power in various locations, setting up in an attempt to more guarantees to decision makers in the installation of new wind farms, with a view, the possibility of predictions intensity of local wind speed, ie to meet the wind regime in the future.