Using data and knowledge: an architecture for implementing load forecasting systems

Conventional load forecasting systems used in the power, water and gas industries are normally based on mathematical modeling techniques. These systems are often reasonably accurate in predicting the loads for normal days when no special events take place, such as public holidays, abnormal weather, and some unexpected social events. However, these numeric forecasting systems have a limited amount of success in predicting the loads for abnormal days when some special or irregular events occur. This has led to the application of expert systems in which qualitative simulators are used in place of analytic schemes in the forecasting systems. Although these systems can handle special events well, their performance is usually not as good as conventional numeric algorithms when predicting the loads for normal days.<
>