Application and comparison of adaptive methods for the long term prediction of the electrical energy consumption in Greece

This paper tackles the problem of the long-term prediction of the electrical energy consumption with three different prediction approaches using real data. The first method combines the well-established multimodel partitioning filter (MMPF) implementing extended Kalman filters (EKF) with Support Vector Machines (SVM), the second is a hybrid method of MMPF and Genetic Algorithms (G.A) and the last method implements an artificial multilayer layer feed-forward neural network (ANN). The accuracy of a long-term forecasting (considering a time interval greater than one year) is of great importance, since it contributes to the planning and expansion of a country’s electric power system reliably and economically. Various factors affecting long term prediction such as the existing installed capacity, the annual average ambient temperature and humidity, the annual electric energy consumption per capita, the energy consumption per person and the gross domestic product (GDP) were considered in all three approaches. The results indicate that all methods are reliable, however the combination of MMPF and SVM provides a more accurate long-term load forecasting in terms of absolute percentage error. Therefore the system administrator based on its forecasts will able to use the available resources for planning the construction of new generation facilities and also expanding the transmission line grid using a cost effective plan.