USING THE MULTIPLICATIVE DOUBLE SEASONAL HOLT-WINTERS METHOD TO FORECAST SHORT-TERM ELECTRICITY DEMAND

Balancing electricity demand and supply is a challenging task for the National Grid. Accurate short-term electricity demand forecasts are of vital importance to ensure that electricity can be supplied according to demand without wasting financial and energy resources. Time series data on electricity demand in England and Wales, collected at half-hourly intervals, reveals both a within-day and a within-week seasonal cycle. In this paper, two new univariate short-term forecasting methods accommodating two seasonal cycles are proposed to model electricity demand: double seasonal Holt-Winters exponential smoothing and singular spectrum analysis. The methods are compared to a benchmark seasonal ARIMA model in terms of forecasting performance as measured by the mean absolute percentage error. The double seasonal Holt-Winters method with a simple autoregressive model fitted to its residuals outperforms the benchmark model; its mean absolute percentage errors are significantly smaller than those of the seasonal ARIMA model. In contrast, the method of singular spectrum analysis outperforms neither the benchmark nor the seasonal ARIMA model.