Electric utilities use load forecasts for different purposes including trading, system operation, infrastructure planning, and demand-side management. Demand forecasting for electric utilities is different from that for other industries such as retail and airline passengers or incoming calls of a contact center, because electricity is a special type of good that cannot be stored to meet the demand when there is a need. Thus electric utilities’ operations depend heavily on accurate load forecasts to balance the demand and supply in an almost real-time fashion.
Demand forecasting for holidays is even more challenging because of the limited sample size of historical holiday data and the varying load profiles across different holidays or across the same holiday in different years. These challenges result in higher forecast errors for holidays, than for regular days. This paper investigates practical techniques for holiday forecasting: modelling holidays as weekends, modelling holidays using holiday dummy variables, two-stage method in which the second stage models residuals.