A comprehensive low-risk and cost parallel hybrid method for electricity load forecasting

Abstract

The accuracy and risk of electricity load forecasting are the most critical features, which play a significant role in efficient management, future economic planning, and decision making by financial and operational decision-makers of generation and distribution powers. It is the main reason for providing more comprehensive models to increase the accuracy and reduce electrical forecasting risk, despite numerous existing models. On the other hand, the comprehensiveness of modeling is the principal source of efficiency in the forecasting models; because the modeling's completeness has a non-strict positive and negative relationship with the accuracy and risk, respectively. However, the literature indicates that developing more comprehensive, or equivalently more accurate, and more reliable forecasting models is yet often a problematic task, especially in electrical systems. Due to their complexity, ambiguity, and multiple mixed structures, electrical markets and systems are the most challenging markets and systems for time series forecasting. Literature indicates that electrical systems' data often simultaneously contain linear, nonlinear, seasonal, nonseasonal, certain, and uncertain patterns. Therefore, a comprehensive forecasting model in such systems must simultaneously model all of these patterns and structures. In this paper, a comprehensive low-risk, low-cost parallel hybrid model is proposed for electricity load forecasting. The main distinguishing of the proposed model is the comprehensiveness of modeling. In the proposed model, different patterns and structures of underlying data, i.e., seasonal and nonseasonal, linear and nonlinear, and fuzzy and nonfuzzy, are simultaneously modeled. In addition, in the proposed model, a direct global optimal weighting approach is proposed in order to combine components. The proposed weighting model is direct; so, its computational cost will not be greater than other weighting models. Moreover, it is a global optimal method; thus, it can be generally guaranteed that its performance will not be worse than all other weighting models. Moreover, the proposed model uses a parallel hybridization of components; thus, it can decrease the risk of using inappropriate models, as well as the risk of obtained results. Empirical electricity load forecasting results indicate that the proposed model can yield more accurate results with low risk and low cost than its components and some other single and hybrid models.