Model optimization for very-short-term wind power forecasting using Hilbert-Huang Transform

Abstract

Wind power forecasting is one of the most effective solutions of large scale wind power integration. However, the data which contains wind speed and wind power output is the non-stationary stochastic signal to be taken as a time-energy series, and the data varies largely in every frequency component. It creates forecasting errors when treating wind speed or wind power. In addition, the very short-term wind power forecasting, which the time frame is several minutes ahead, can help balance the supply and need of grid, facilitating the steady and healthy operation of power market. Facing the wind data feature and forecasting importance, this paper proposes a novel approach using autoregressive model (AR) and Hilbert-Huang Transform (HHT) to improve the accuracy of wind power forecasting. The input data is decomposed into several components, in every of which, the data satisfies the linear characteristic better than the one before transformation. Case study from a wind farm in Texas, U.S.A shows that the proposed approach performs better than the traditional linear forecasting method.