Short Term Wind Speed Prediction Based On Wind Speed Patterns

As the climate changes and the energy crisis caused by excessive consumption of fossil fuels have become more and more serious,all countries in the world are devoted to the research and development of renewable energy and clean energy.Wind energy is considered as a sustainable source of power generation and one of the most important green energies,wind speed is the main factor that affects wind energy.Many intelligent methods have been proposed for wind speed prediction.At present,considering the different variation characteristics of wind speed in different seasons or periods,some scholars extract wind speed patterns by clustering and construct wind speed prediction models for different wind speed patterns.However,it is difficult to regard a type of features as an independent overall feature set in the process of classical clustering,therefore,we propose a wind speed pattern mining method based on multi-view for short-term wind speed prediction.Specifically,firstly,the multi-view feature is extracted,we extract the features which are consist of the wind speed statistics,the wind speed fluctuation information and the wind speed trend information in an artificial way,in addition,we try to use the deep learning to extract multi-view features;Secondly,in order to maintain the independence and completeness of the wind speed variation features from different views during the clustering process,the new feature representation is obtained by non-negative matrix decomposition of different features,and a number of new features be merged for getting a final complete feature space.Finally,based on the final features,the wind speed is divided into several different clusters,and the wind speed is predicted separately.The experimental results show that the multi-view clustering method can effectively extract different wind speed patterns and build wind speed prediction models for different clusters effectively to improve the wind speed prediction accuracy.