Application of kernel principal component regression in mid-term load forecasting

After investigating the mechanism of KPCR and its relation to least square method, this paper proposes mid-term load forecasting with KPCR, where error propagation exists. There are strong nolinear relations between influence factors in mid-term load forecasting. The simulation experiments are done with KPCR, good prediction accuracy is obtained. The paper puts emphasis on the matters about training sample selection, data disposition, and model parameter selection, etc.. Some explanations are presented on experiment results. In the experiment with simple model, the principal component number p can be set in feature space by way of direct searching.