Implementation of short-term load forecasting expert systems in a real environment

This paper presents the Short-term Load Forecasting Expert System (LoFES) which was developed and implemented far adopting artificial intelligence (Al) techniques in short-term load forecasting and supporting the human forecaster. This system had been developed from 1993 to 1995 in Korea Electric Power Corp. (KEPCO), and it has been operated in a real environment since 1996. LoFES is intended to provide user-oriented features with a Graphical User Interface(GUI), and all the forecasting procedures of this system are carried out with a GUI. The forecasting methods of the system include an exponential smoothing method, a multiple regression method, and a neural network based method for ordinary day load forecasting; and fuzzy inference method for special day load forecasting. In the actual operation in 1996, this system provided good forecasting accuracy with the mean absolute percentage error below 1.6%, and it outperformed the conventional method used in KEPCO and effectively supported the human forecaster in forecasting process.