Medium range forecasting for power system load by fast Fourier transform

Abstract

Medium range forecasts of daily electric power demand, spanning one to two weeks are required for preparation of ~hort time maintemmce schedule of unit auxiliaries and peaking stations. Forecast of daily load for seven days by the available multiplicative SARIMA model. suffers from divergent error levels of multistep forecasts and to take all the specific features of any particular day, the order of such a model is prohibitively large. The present paper demonstrate an altogether different approach to the time frame analysis of stochastic processes the periodic nature of electric power system load demand spanning 24 and 168 hours has been gainfully exploited by grouping the data in seven subgroups chaructcrising each week-day separately and then each day's data is transformed into frequency spectrum. The amplitudes of the spectrum have then been time-series modelled and forecasts are made by inverse Fourier transform.