Wind power forecasting: Comparing two statistical signal processing algorithms

Wind power forecasting (WPF) has turned into a substantial tool for limiting the negative impact of wind power intermittency on power system. In this paper, we study and compare two different WPF algorithms: classical autoregressive model (AR), as a base case method, and kernel density estimation (KDE) with minimum mean square error estimator (MMSE). Using the data history of a wind farm in Colorado, these two algorithms are implemented in MATLAB and used to produce 24 hours ahead predictions of wind power time series of the said wind farm. The results obtained from the two methods are then compared from various perspectives (precision, applicability, etc.). The comparisons show that although AR-based wind power prediction has slightly less error than KDE, AR-based prediction does not produce probability density function (PDF) of wind speed/power, while KDE does. PDF of wind speed/power is an important parameter for estimating the required reserve allocation in economic dispatch studies.