Weather-normalized intermediate term capacity forecasting: a procedural innovation

The authors present the theoretical foundations of a procedural innovation for weather normalization of hourly loads. They describe an alternative statistical approach that will produce efficient estimators in the presence of complex serial correlation in the model's error structure E.J. Hannan's efficient estimator (1970), a form of frequency-domain regression analysis, was used to develop efficient hourly load/weather equations. From then-normalized load database, monthly peaks were extracted. Box-Jenkins (B-J) analysis was performed on these monthly peaks for production of 24-months intermediate forecasts. The results of these forecasts were compared with B-J forecasts prepared from actual monthly peaks. Of the 12 forecasts compared, the average monthly error for 11 was reduced by 50% or more. For 9 of these forecasts, the error reduction was 66% or more.< >