Spatial correlation of wind speed is helpful to improving its prediction quality

especially when there are sudden changes of wind speed. A new method for ultra-short term wind speed prediction based on the idea of “offline modeling by classification

and online feature matching for model selection” is proposed. By analyzing time series among historical data

the time segments having spatial correlation in different wind farms are identified. The time segments of wind speed in the current time window are divided into sample subsets with different evolution patterns according to the features of time series and other external conditions. Prediction models and corresponding parameters for different patterns are optimized offline based on their sample subsets

respectively. While for online application

prediction models and the corresponding parameters are selected by feature matching

according to evolution patterns and other external conditions in the current time window. Finally

a case study using actual historical data is presented to validate the effectiveness of the proposed method. This work is supported by the State Key Program of National Natural Science Foundation of China(No. 61533010)

NSFC-NRCT(Sino-Thai)Cooperation Research Project(No. 51561145011)and State Grid Corporation of China.