Formulation and analysis of a rule-based short-term load forecasting algorithm

The formulation of rules for the rule base and the application of such rules are discussed. The classification of the load forecast parameters into weather-sensitive and nonweather-sensitive categories is described. The rationale underlying the development of rules for both the one-day and seven-day forecast is presented. This exercise leads to the identification and estimation of parameters relating load, weather variables, day types, and seasons. Sample rules that are the product of identifiable statistical relationships and expert knowledge are examined. A self-learning process is described which shows how rules governing the electric utility load can be updated. Results from both the one-day and seven-day forecast algorithms are presented, where the seven-day forecast is generated using both accurate and predicted weather information. The monthly average load forecast errors range between 2.97% and 10.71% for the seven-day forecasts. For the one-day forecasts, the average seasonal errors range between 1.03% and 1.53%.