Electric energy monthly demand forecasting by spectral analysis

Abstract

Medium-term load forecasting is a useful tool forthe maintenance planning of grids and as a market research ofelectric energy. In this work medium-term load forecastingmethods are developed, the most forgotten time scaling processin the load forecasting bibliography. These methods will beapplied to the peninsular Spanish monthly energy consumption.Methods traditionally employed with this objective are based onregression, 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, mainlyclimatic or economic ones, which strongly influence electricenergy demand. These variables, of cyclic nature, provide aperiodic behaviour to the energy consumption time series. Thiswork presents a study of this periodic behaviour by means ofspectral analysis, with the identification and interpretation ofthe dominant frequencies. A forecasting method for futurevalues of electric energy demand will be then presented, whichis based on a simple regression technique combined with neuralnetworks. It does not take into account any climatic oreconomic variables, because only periodic behaviour of thetime series is considered. Acceptable results are reached, withpercentage errors lower than 5 % in most cases.