Electric Energy Monthly Demand Forecasting by Spectral Analysis

Abstract

Medium-term load forecasting is a useful tool for the maintenance planning of grids and as a market research of electric energy. In this work medium-term load forecasting methods are developed, the most forgotten time scaling process in the load forecasting bibliography. These methods will be applied to the peninsular Spanish monthly energy consumption. Methods traditionally employed with this objective are based on regression, statistical techniques (mainly Box-Jenkins ARIMA), and also with neural networks, fuzzy logic or expert systems. Most of them need the use of nonelectric variables, mainly climatic or economic ones, which strongly influence electric energy demand. These variables, of cyclic nature, provide a periodic behaviour to the energy consumption time series. This work presents a study of this periodic behaviour by means of spectral analysis, with the identification and interpretation of the dominant frequencies. A forecasting method for future values of electric energy demand will be then presented, which is based on a simple regression technique combined with neural networks. It does not take into account any climatic or economic variables, because only periodic behaviour of the time series is considered. Acceptable results are reached, with percentage errors lower than 5 % in most cases.