Aplicação de ferramentas de inteligência computacional para previsão da velocidade do vento

Abstract

Many countries have devoted a lot of attention in the use of renewable energy sources to ensure the growing of energy demand and sustainable development. Projections by the Global Wind Energy Council (GWEC) indicate that by 2015 the worldwide capacity of installed wind power will reach 449 GW, almost the double in 2010. The wind generation is characterized by the variability in production and the restricted control. The wind power forecast is essential for a reliable, safety and economic operation of power system. This paper presents and analyse the predicted wind velocity of Artificial Neural Networks (FTDNN, Elman, Jordan, NARX) to the forecasts horizons of 1h, 6h, 12h and 24h. Other Computacional Intelligence tools (PSO & GA) are used to improve the ANN performance. It was used as reference the results obtained through the Persistence model. It was used a time series with 45,658 measurement of wind speed, which 80% were selected for the training phase and 20% for validation purposes. As criteria for evaluate the performance of ANN were considered the error methods: MAE, RMSE and MAPE. The results shows that all the ANN ha similar results for 1h forecasts and better than the results from Persistence model. The use of PSO as a training tool results in better forecasts than the ones from backpropagation training.