Proposal for Prediction of Wind Speed through Hybrid Modeling Elaborated from ARIMAX and ANN Models

Abstract In this paper, a hybrid model capable of predicting wind speed (monthly and hourly averages) with good accuracy in regions of the Brazilian Northeast. This model is elaborated from the combination of two models, the Auto-Regressive Integrated Moving Average with eXogenous inputs (ARIMAX) and Artificial Neural Networks (ANN). The choice for these models was motivated by the fact that they can incorporate both linear (ARIMAX) and nonlinear (ANN) features commonly found in time series. The hybrid model correlates pressure, temperature and precipitation with wind speed in order to consider important local meteorological characteristics. It is possible to verify the efficiency of the hybrid model to provide good adjustments to the observed data of the winds speeds, being this affirmation based on the values found by means of accuracy measurements, with average percentage error of approximately 5.0%, and coefficient value of Nash-Sutcliffe efficiency of 0.96. These results confirm the existence of precision for the winds speeds predicted following the profile of their observations, in particular it is possible to identify similarities between both time series (in terms of maximum and minimum values), thus showing the capacity of the model in represent characteristics of seasonality.