Redes neurais artificiais para previsão de velocidade do vento: estudo de caso para Maracanaú – CE

Abstract

This paper presents two methodologies for hourly wind speed forecasting using Artificial Neural Networks (ANN). The estimation considered a wind speed database colectedfrom the town of Maracanaú – CE. Two Neural Networkslike FTDNN (Focused Time Delay Neural Network) are considered in this work. A network was trained only with data series of wind speed and other network was trained using as input data, together with data of wind speed, direction information, roughness and the air temperature. Historical data were collected during the period of one year. Performance measures of the system (correlation coefficient- r, mean square error - RMSE and relative bias - VR) were applied. The results generated by the ANN sindicate that the applications have a good potential for wind speed forecasting.Based on the performance measures, it wasobserved that, despite the similar results in both applications, there was an improvement with the inclusion of meteorological parameters, suggesting the applicability of the methodology for predicting wind speed and its use in feasibility studies and wind potential, and its importance in reducing the impacts of variability of wind speed.