Previsão horária do Vento utilizando o modelo WRF para geração de energia eólica na região Nordeste do Brasil

Abstract

Wind energy requires accurate forecasts for adequate integration into the electric grid system. To this purpose, regional numerical models such as WRF have been increasingly used to predict wind power of wind farms.This dissertation focuses on the forecast and validation of wind speed and production of two wind farms using WRF in hight resolution. By the observational analysis, it was observed that spring is the season of 2017 that presents the highest average monthly wind speeds, with maximum peak in the month of September, while lower speeds vary during the March-April-May quarter. The diurnal period between 10:00 am and 5:00 pm local time is the one with the most frequently wind speed ≥ 10 m.s-1 . Wind direction varies from east to south, with predominance of east and southeast directions, in about 80% of the time. During the quarters that correspond to summer and fall, Weibull distributions are more concentrated around 6.5 and 7 m.s-1 , indicating that in this period of the year there is a greater probability of occurrence of lower average speeds, implying in lower productivity for wind energy. The opposite is observed in winter and spring, with the exception of June, where the distributions are more concentrated around speeds ≥ 8 m.s-1 and there is a higher probability of higher average speeds to occur, showing that this was the best period of the year 2017 for wind power generation. With respect to numerical modeling, results show that WRF yields good operational forecasts of wind energy for the study region with the appropriate combination of physical schemes. The BouLAC local closure scheme experiment showed better predictions than the hybrid ACM2 scheme for both study periods (April and September). The best mean monthly absolute error obtained was 1.16 ms-1 for wind speed and 12.6% for wind generation in April, and 1.19 ms-1 and 13.3%, respectively, in September. Model performance considering the 24 h x 48 h forecast horizon was similar, although each one performs better for each experiment under certain weather conditions.