Previsão da velocidade do vento utilizando redes neurais artificiais e modelos autorregressivos

Abstract

The economic and environmental benefits of wind power generation made wind energy one of the most promising sources for electric power generation in Brazil. However, the uncertainty associated with wind data, the source of this generation, generally cannot be overlooked. Thus, the data must be accurately evaluated to effectively reduce the risks of wind generation in energy system operations, therefore it motivates the development of forecasting techniques that take advantage of measurements in almost real time, collected from geographically distributed instruments. In this approach, forecasting methods, based on artificial neural networks and linear autoregressive models are compared, aiming at the very short term horizon. A new approach is being proposed in this work, a linear autoregressive model and an artificial neural network for the probabilistic forecast of wind speed in the very short term. The proposed set approach was extensively evaluated, using real data from five anemometric stations installed in the metropolitan region of Curitiba. The results demonstrate that the uncertainties in the wind data can be reliably predicted and that a competitive performance is obtained.