Long-term forecasting: a critical review of direct-trend extrapolation methods

Abstract

A treatment is given of probabilistic autonomous models, i.e. statistical models of direct-trend extrapolation for long-term electrical energy demand forecasting. A methodology of mathematical modelling for global forecasting based on regression analysis is presented. A critical survey of the most common regression models is given, as well as the original algorithms and results of the class of combined exponentially-polynomial regression models. Data of increasing electricity consumption for a number of Yugoslavia urban areas (populations from several thousand up to 1.5 million) are processed, in order to draw conclusions about the applicability of the regression models considered. The results of Yugoslav demand forecast up to the year 2000 are presented. Although the processed data are typical of the conditions in developing countries, satisfactory agreement has been observed with the behaviour of forecasting models applied to the data for West Germany, a typical developed country.