Wind Power Short-term Forecasting Based on Empirical Wavelet Transform and Multiple Kernel Learning

We propose a combination forecasting method based on empirical wavelet transform (EWT) and multiple kernel learning (MKL) for short-term wind power. We decompose time series data by using the EWT method and build different forecasting models according to a sub-sequence of multi-component signals, which are implemented using the SimpleMKL algorithm, the MKL-wrapper algorithm and the MKL-chunking algorithm. The forecasting results are then superimposed. To verify the effectiveness of the proposed method, we apply the EWT-MKL method to a wind farm in different seasons for single-and multi-step direct forecasting of short-term wind power. We also apply the proposed method to the instance with the actual wind speed data for indirect forecasting of the short-term wind power from the NREL laboratory, which is also implemented by the SVM and the wavelet SVM methods in the same condition. Experimental results show that the proposed EWT-MKL method with different algorithms has high forecasting accuracy and the corresponding combined models have good generalization, thereby showing the effectiveness of the proposed method.