Ultra-short-term wind power prediction based on variational mode decomposition and clustering method

Wind power ultra-short-term prediction plays a key role in day scheduling of power systems and cross-provincial power trading. However, the randomness and non-stationary nature of wind power, and the mismatch between the training data and the predicted data have become the main obstacles for the improvement of prediction accuracy. This paper proposes a prediction algorithm based on variational mode decomposition (VMD) and clustering. Firstly, the original non-stationary sequence is transformed into several relatively stationary modes utilizing VMD. Secondly, key features can be conveniently selected for each mode, which has a central frequency. Then the k-means clustering method is used to cluster each mode into several typical patterns based on correlation coefficient based distance definition. Finally, the method of “offline training, online matching” is adopted to perform training and prediction process. The actual data of a wind farm in Jilin Province was used to verify the results. The results show that the proposed method can improve the accuracy of ultra-short-term in multi-step prediction of wind power.