Short-term electricity demand forecasting in the time domain and in the frequency domain

This paper compares the forecast accuracy of different models that explicitely accomodate seasonalities in the time domain with methods in the frequency domain. The forecast accuarcy of the methods are evaluated by their MAPE’s using a time series of half-hourly electricity demand for England and Wales. The methods in the time domain include a well-specified multiplicative seasonal ARIMA model, standard Holt-Winters method, double seasonal HoltWinters method both with and without residual auto-correlation correction. The methods in the frequency domain introduced in this paper are Fourier series that are modified to accomodate stochastic dependence. The last class of models in the frequency domain is shown to perform almost as good as the methods in the time domain and is promising in terms of accuracy and adaptability.