Forecasting Turkey's Primary Energy Demand Based on Fuzzy Auto-Regressive Distributed Lag Models With Symmetric and Non-Symmetric Triangular Coefficients

Abstract

This study aims to guide policymakers in allocating resources and planning for the future by consistently estimating energy data trends. Because of the complexity and uncertainty of energy demand behavior and many influencing factors, we decide to take advantage of a fuzzy regression model to determine the actual relationships in the energy demand system and provide an accurate forecast of energy demand. For this purpose, because of energy demand drivers, fuzzy possibilistic approaches with symmetric and non-symmetric triangular coefficients are integrated with the autoregressive distributed lag (ARDL) model, each in a time-series format with feedback mechanisms inside. After regularizing the L1 (Lasso regression) and L2 (ridge regression) metrics to minimize the overfitting problem, the optimal fuzzy-ARDL model is obtained. Turkey's primary energy consumption is projected based on the best model by benchmarking the static and dynamic possibilistic fuzzy regression models according to their training and test values.