Short-term Wind Power Prediction Based on Principal Component Analysis and Spectral Clustering

The large-scale construction of wind farms leads to evident increase of wind power penetration. To ensure secure and stable operation of power grid and the accommodation of wind power, it is necessary to predict the wind power. To cope with the defect of too high data dimension in traditional prediction method, a prediction method, in which the data dimension reduction was performed was based on principal component analysis (abbr. PCA) and spectral clustering (abbr. SC), was proposed. Firstly, on the basis of PCA the principal component of power sequence of each generating unit in the wind farm was extracted to implement the dimension reduction of power sample information and the predicted object. Secondly, considering the fluctuation characteristic of wind speed and spatial distribution characteristics of each generating unit, the spectral clustering of wind speed information was conducted to realize further dimension reduction of sample data. Finally, based on the principal component information of wind power and the result of wind speed clustering an Elmer neural network-based wind power principal component prediction model was established, and by means of the inverse transformation the power prediction result of each generating unit in the wind farm was finally obtained. By use of actual data from a certain offshore wind farm in Nantong, Jiangsu Province the established method was verified. Verification results show that using the proposed method the predicted accuracy of wind power can be improved.