Load forecasting for multiple sites: development of an expert system-based technique

Most papers on short-term load forecasting (STLF) emphasize the use of a certain technique rather than the methodology used to arrive at that technique. Moreover, few, if any, of the techniques reported in the literature have looked at the electric power load from a site-independent viewpoint. A newly developed site-independent technique for STLF is analyzed in this paper. The load is modeled using a parameterized rule base and a parameter database. The extraction and representation of knowledge about the load are discussed. A site-independent expert system-based technique is tested using data from different sites around the United States including Alabama, Georgia, Florida, Massachusetts, Texas, Virginia and Washington. The mean absolute percentage errors for 24-h ahead load forecast range from 1.23 to 3.35% over all hours in all seasons. The expert system itself was implemented in UNIX/C, and the solution time per 1 day of forecast is around 4 s.