Probability Distribution Functions for Short-Term Wind Power Forecasting

Abstract

Wind energy estimation is pivotal to ensure grid side management and optimal dispatch of wind power. Wind speed distribution for given wind site can be modeled using various probability distribution functions (PDF) like Weibull, Gamma and Log-normal distribution functions. PDFs like Weibull and Log-normal do not fit the real-time wind speed scenarios. In this paper we analyze the PDFs for short-term wind power forecasting for a low wind speed regime based on combined wind speed and wind direction PDF. Short-term wind power forecasting based on \(\varepsilon \)-Support Vector Regression (SVR) and Artificial Neural Network (ANN) was carried for three wind farm sites in Massachusetts. The forecasting results were tested for Mixture density Weibull and Lindley PDFs and in terms of Root mean squared error, Weibull PDF outperformed Lindley PDF.