For lead times of more than 6 h, calendar data and meteorological data play an important role and need to be included in the model [25]. Huang et al. [17] use a time series model with an exogenous variable, temperature, to forecast hourly load one day and one week ahead. Soares and Medeiros [18] use several calendar data (day of the week, holiday, working day after holiday, working day before holiday, bridge day, etc.) to forecast the electricity demand in the southeast of Brazil up to seven days ahead. Mirasgedisetal. [26] include calendar data, temperature and humidity to forecast daily and monthly electricity demand up to one year ahead.The factors used in demand forecasts depend on the forecasting horizon. For very short-term demand forecasts, i.e., periods of 6 h or less, it is sufficient to use only historical data of the time series [13].
For instance, Huang and Shih [15] use a univariate time series model to forecast short-term electricity demand. Taylor [16] uses an exponential smoothing model to forecast up to one day ahead.