A Multi-Source Wind Speed Fusion Method for Wind Power Prediction Based on <tex>$k$</tex>NN-SVR

Abstract

Wind power now is taken as an essential alternative to burning fossil fuels for humans and begin to be used in our daily life. However, the wind power is hard to predict since the wind speed is a highly fluctuating resource. This paper presents a multiple wind speeds fusion method to improve the accuracy of short-term wind power prediction based on a $k$ - nearest neighbors-based support vector regression ( $k$ NN-SVR) model. First, a $k$ NN algorithm is devised to select the historical wind speed points that is nearest to the prediction point from the historic data set. Then, a SVR model is built to obtain the fusion wind speed based on three independent numerical weather predictions at the nearest historic prediction point, which improves the accuracy of the wind speed prediction. Finally, the fusion wind speed is employed to predict the wind power by using an back propagation neural network (BPNN) model. Experimental results that are on the basis of the data from an actual wind farm in Shanxi Province, China, show the efficiency of the fusion method.