Short-term Wind Power Forecasting Through Combined DWT-SOM and MEEFIS Method

Abstract

For optimal utilization of wind energy resources, wind power forecasting (WPF) is critical in balancing the electricity supply and demand in a power distribution network. Continuously forecasting wind power generation enables one country or region to operate its grid smoothly around the clock. In this article, a combined method is developed for the short-term WPF, which is comprised of discrete wavelet transform (DWT), self-organizing map (SOM) and multi-layer ensemble evolving fuzzy inference system (MEEFIS). Firstly, the row wind power sequences in the time domain are decomposed into several components in the frequency domain by using DWT, which are further selected as the features for wind power clustering. Then the SOM algorithm is adopted to cluster row wind power data and divide it into different data groups with features. Afterwards, the MEEFIS model is used to predict components in all types of data group, and the prediction results of components of various groups are overlapped to generate the final wind power forecasting value. Finally, the simulation analysis is done with actual wind farm power data in one region. The results of experiments demonstrate that this method can effectively improve the wind power forecasting accuracy and brings potential applications.