An expert system for short-term load forecasting

Time series approaches, as conventional algorithms in the short-term load forecasting of electric power systems yield considerably high accuracy under the circumstance of no abrupt disturbances such as sharp changes of weather and in special holiday, or, if the results of these algorithms are manually adjusted by experienced operators in order that they can be further used. An expert system for short-term load forecasting is proposed by the authors. The time series approach is combined with heuristic inference and a composition method is developed to forecast the hourly load up to 48 hours in advance. The influence factors such as weather, holidays, season and dispatch control are considered. The structure of the expert system, programmed in PROLOG language, is described by the authors. This expert system is tested by actual power system load data of 210 days. The average forecasting errors of these tests are discussed by the authors