Eigenload Model Based Yearly Electric Load Demand Evaluation and Forecasting for State Grid Corporation of China

This study addresses the yearly load movement evaluation and forecasting based on the Principal Component Analysis (PCA). A called Eigenload model for describing the annual load movement by employing PCA is introduced. Principal orthogonal eigenvectors of covariance matrix of the load data called "Eigenloads" are used to build subspace for the load movement evaluation. Each load movement is projected onto subspace of the eigenspace and described by a linear combination of the "Eigenloads". Incorporated with the polynomial curve fitting algorithm to estimate its subsequent representation weights with respect to the " Eigenloads" generated by previous load movements, the Eigenload model is extended to forecasting subsequent movement of the load demand. The proposed method is applied to experiments of annual load demand evaluation for State Grid Corporation of China (SG)and its five branches in 2004-2006, and forecasting for their annual load demands in 2007. Experimental results agreeing well with their actual load demands show workability of the proposed model. Result analysis indicates that the Eigenload model is outperforming the classical autoregressive (AR) model on the forecast tasks.