Short-term Wind Power Forecasting Based on Variable Time Window and Feature Extraction

Abstract

With the rapid development of wind power, the impact on the grid of ramping events of large scale highly-concentrated wind power generation becomes serious. The intermittence and variability of wind speed definitely causes of the fluctuation and predication accuracy of wind power. This paper proposes a combined model based on variable time window and feature extraction in wind speed for short-term wind power prediction. At first, a multifractal spectrum is applied to investigate wind speed characterizations. Then on the basis of the wind fluctuation definition, an abstracting feature extraction approach is proposed by use of a sliding variable time window algorithm which could self-adaptively adjust the size of time window width. The historical data is classified according to the fluctuation events abstracting results. Different prediction models are developed by selecting the specific parameters after analyzing the fluctuation events characteristics. The presented method employs the spectrum analysis to correct the power error, aiming at the complexity and multiformity of output wind power in different time period. Finally, case studies are carried out to verify and evaluate the availability of the proposed model. Results shows that the short-term forecasting accuracy of wind power has been improved in various wind situations.