Short-term wind power forecasting based on BOMLS K-means similar hours Clustering method

Abstract

With the continuous development of wind power in the world, the accuracy and stability of wind speed and power prediction are very important for the economic operation and safety management of power system. The proposed method consists of hybrid clustering method and ensemble learning method. Hybrid clustering method combines the advantages of equal-length and unequal-length segmentation. Sample segments with the same trend can be clustered regardless of whether they are equal in length. The selection of segmentation points can be optimized by Bayesian algorithm, which can accelerate the K-Means clustering speed and efficiency. Ensemble learning can combine multiple weak learners into a strong learner. Finally, the data from a wind farm in southern China show that the wind power forecasting method based on the above method is more effective than the single learning model method.