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

Highlights

• The influential factors of electricity demand evolution in Greece are analyzed.

• Economy and energy efficiency impose the greatest impact on electricity demand.

• Αn ordinal regression method for robust electricity demand forecasting is proposed.

• Both ordinal and multiple linear least-squares regression models have been applied.

• The ordinal regression models provide the most accurate forecasts.

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.

Introduction

The philosophy of disaggregation or ordinal regression analysis in multiple criteria decision aid (MCDA) is to infer preference models from given preferential structures and to address decision aiding activities through operational models within this framework (Jacquet-Lagreze and Siskos, 1982, Jacquet-Lagrèze and Siskos, 2001, Siskos, 1985, Siskos, Grigoroudis and Matsatsinis, 2016). In most of these methods, such as the UTA-type methods, an additive value model is inferred by disaggregating preference statements of ordinal nature, such as ranking of reference actions, pairwise comparisons of reference actions, etc. in such a way that the value model is as much consistent as possible with the given preferences.

The models presented in this paper disaggregate quantitative measures by estimating additive value functions that are optimally consistent with the available data. In this line of disaggregation, Kettani, Oral, and Siskos (1998) presented an estimation model to describe the behavior of real estate markets. The model is based on certain actual real estate market data, as well as, on the perceptions of real estate agents who are active in the market. The parameters that describe the behavior of the real estate market are estimated, through the exact model, by using mathematical programming tools within a multiple criteria analysis framework, same as in UTA-type methods. The usefulness and applicability of this approach is empirically tested via the implementation of a related case study according to the data of the city of Edmonton, Alberta, Canada.

In a different context, Grigoroudis and Siskos, 2002, Grigoroudis and Siskos, 2010) developed the MUlticriteria Satisfaction Analysis (MUSA) method, a multicriteria preference disaggregation approach that provides quantitative measures of the customers’ satisfaction, considering the qualitative form of the customers’ judgments. The main objective of the MUSA method is the aggregation of individual judgments into a collective value function, assuming that the customers’ global satisfaction depends on a set of criteria or variables representing service/product characteristic dimensions.

All the above approaches are trying to preserve the robustness of the models they infer from given preferential data. Greco, Slowinski, Figueira, and Mousseau (2010) proposed a general methodological framework, namely the Robust Ordinal Regression (ROR), which can be implemented synergistically to the disaggregation methods. The exact technique aims at enhancing the robustness of the estimated results and is based on the principle that the decisions and proposals emerge after considering all those parameters that are compatible with the preferences of the decision makers (DMs).

The ordinal regression models proposed in this paper are developed in order to analyze the electricity demand time series in Greece. During the last eight years, Greece is facing a sovereign debt crisis that resulted into negative impacts on the national GDP, including significant reduction of the personal consumption expenditures and the business investments in the national economy. This unfavorable economic environment has also adversely influenced electricity demand in the country. Consequently, the application of an electricity demand forecasting model, which takes into consideration the complex environment and the multi-level criteria, is essential for sustainable energy planning in the country.

Main aim of this paper is, first, to examine the relationship between the annual electricity demand and the influential socioeconomic parameters/criteria, including, among others, the GDP, the energy efficiency progress, the weather conditions and the electricity price and, second, to provide long-term electricity demand forecasts for the case of Greece. For the purposes of current study, a novel multicriteria modeling approach is proposed for robust long-term electricity demand forecasting. The initial inspiration of the whole methodological approach has been derived from the customer satisfaction evaluation modeling technology introduced by Grigoroudis and Siskos, 2002, Grigoroudis and Siskos, 2010) for which special linear programming techniques are developed. In this context, an original ordinal regression analysis method is outlined and two disaggregation forecasting models, as alternative formulations of forecasting, are proposed in order to optimally assess a robust additive value model which is as consistent as possible with the annual electricity demand data of the previous periods.

For the case of Greece, the additive value forecasting model is inferred from data related to the training period (years 1999–2013). The accuracy and stability of this multicriteria methodology is evaluated through the calculation of relevant statistical error indicators and robustness analysis indices, respectively. For completeness of current research, the model outcomes are also compared with the ones extracted from the utilization of the multiple linear (least-squares) regression model. The multiple linear regression model has been selected for the needs of current model's validation as it constitutes a widely utilized and accurate method for electricity and energy (in general) demand forecasting, supported by long and well-established theory and methodologies (Bianco, Manca and Nardini, 2009, Bianco, Manca and Nardini, 2013, Ekonomou, 2010, Mohamed and Bodger, 2005, Kankal, Akpınar, Kömürcü and Özşahin, 2011, Yumurtaci and Asmaz, 2004).

The proposed robust multicriteria 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 (years 2014–2016). According to the forecasting model with the greatest accuracy, additional electricity demand projections have been quantified for the upcoming period 2017–2027. Moreover, alternative scenarios of the electricity demand evolution are presented via a related sensitivity analysis of the GDP growth in the country.

The paper is organized as follows: Section 2 presents the two proposed disaggregation forecasting models. Section 3 outlines the original electricity demand forecasting modeling approach for the case of Greece while Section 4 outlines the results obtained from the application of the two proposed ordinal regression models and the multiple linear regression model to the Greek interconnected power system along with a related discussion. Finally, general conclusions and new research avenues are provided in Section 5.