Disaggregating time series on multiple criteria for robust forecasting: The case of long-term electricity demand in Greece

Abstract

Electricity demand forecasting is an essential process in the operation and planning procedures of power systems that considerably influences the decisions of utility providers. Main aim of this paper is, first, to examine the relationship between a time series and influential multiple criteria, and, second, to provide long-term electricity demand forecasts in Greece. An original disaggregation or ordinal regression analysis methodological framework is outlined to optimally assess a robust additive value model which is as consistent as possible with a given time series. The accuracy and stability of this modeling approach is guaranteed through the calculation of statistical error measures and robustness analysis indices, respectively. For the case of Greece, the additive value forecasting model is inferred from data related to the training period 1999–2013. The proposed method has been applied for the forecasting of the annual total net electricity demand in the Greek interconnected power system during the following testing period 2014–2016. The model implies that the level of economic growth, represented by the national gross domestic product, imposes the greatest influence on the electricity demand followed by the energy efficiency progress and the weather conditions in the country. The ordinal regression models perform considerably better than the multiple linear (least-squares) regression model, in terms of prediction reliability, resulting into a minimum MAPE equal to 0.74%. The exact method has been also applied for the extraction of electricity demand projections till 2027 based on alternative economic growth scenarios, indicating a constant increase of the electricity demand in Greece.