Spectral decomposition and extrapolation of variations in electricity loading

The possibility of using a frequency domain approach to forecasting short, medium and long term electricity consumption data is discussed in the paper. In particular, a new super-resolution spectral estimation technique allows for the decomposition of the variations in such data about selected trends and extrapolation beyond the limits of the original data. It is shown that for short- and medium-term data, the frequency components present appear harmonically related and in phase, but a stationarity problem exists for the long-term data, preventing useful forecasts being made. However, the technique does appear to be a possible alternative to the more traditional time-domain approaches.