A Hierarchical Parametric and Non-Parametric Forecasting Source Models with Uncertainties: 10 Years Ahead Prediction of Sources for Electric Energy Production

Abstract

AbstractLong-term accurate forecasting of the various sources for the electric energy production is challenging due to unmodelled dynamics and unexpected uncertainties. This paper develops non-parametric source models with higher-order polynomial bases to forecast the 16 sources utilized for the electric energy production. These models are optimized with the modified iterative neural networks and batch least squares, and their prediction performances are compared. In addition, for the first time in the literature, this paper quantifies the unseen uncertainties like the drought years and watery years affecting especially the hydropower and natural gas-based electric energy productions. These uncertainties are incorporated into the parametric imported-local source models whose unknown parameters are optimized with a modified constrained particle swarm optimization algorithm. These models are trained by using the real data for Türkiye, and the results are analysed extensively. Finally, 10 years ahead estimates of the 16 imported-local sources for the energy production have been obtained with the developed models.