Forecasting Hourly Electricity Demand in Egypt: A Double Seasonality Approach

This study applies double seasonal Holt-Winter method, Double seasonal
autoregressive integrated moving average (DSARIMA) model and Artificial Neural
Networks (ANNs) in forecasting the Egyptian electricity demand series. Double
seasonal Holt-Winter method, DSRIMA model and ANNs are commonly used for
forecasting electricity in many countries. Electricity demand series in many countries
has daily and weekly seasonal cycles. A daily seasonal cycle appears from the
similarity of the electricity demand from one day to the next. Similarly, a weekly
seasonal cycle appears from comparing the electricity demand on a certain day of
different weeks. Double seasonal Holt-Winter method, DSRIMA model and ANNs
are proposed by many researchers to capture theses two seasonal cycles and
consequently offered accurate forecasts.
In this study, we investigate these three forecasting methods in forecasting hourly
electricity demand in Egypt to arrive at the best forecasting method. The previous
forecasting methods are applied on hourly Egyptian data set. The mean absolute
deviation (MAD), the mean absolute percentage error (MAPE), the mean square error
(MSE) and the root mean square error (RMSE) are used to evaluate the forecasting
accuracy of these methods. The results show the superiority of double seasonal Holt-
Winters method in forecasting the Egyptian electricity demand. DSARIMA model
and ANNs are competitive to each other. DSARIMA model gives accurate forecasts
than ANNs up to two weeks forecast ahead, while ANNs are more accurate for
forecast time horizon longer than two weeks.