Dimension-Reduction and Reconstruction of Multi-dimension Spatial Wind Power Data Based on Optimal RBF Kernel Principal Component Analysis

In order to deal with the curse of dimension caused by diversification, complication and elaboration of wind power data, this paper proposes a nonlinear dimension-reduction and reconstruction method based on the principal of optimal RBF kernel component analysis (ORBF-KPCA), so as to impose accurate burrow the intrinsic features of massive wind power data. Concerning the problem of difficulty in selection of kernel parameters in the traditional KPCA method, this paper searches the optimum kernel parameter to minimize the difference between the dimension-reduced results of raw data and homogenous data by means of a cross-validation method. Then a multi-dimensional scale (MDS) technique based on k-nearest neighbors is used for preimage the reconstruction method. An actual example shows the proposed method is better than the common linear dimension-reduction methods in reliability and continuity of results, and can not only improve forecasting efficiency to avoid the curse of dimension, but also achieve higher prediction accuracy.