Harnessing Diverse Data for Hourly Electricity Demand Forecasting in Turkiye: Comparative Analysis of AI-Based Methods

Abstract

Accurate forecasting of national electricity demand is vital for grid reliability and market efficiency. This study aims to identify the most effective dataset and modeling approach for predicting electricity demand across Turkey. A comprehensive, data-driven framework is developed by integrating market, financial, outage, and consumption-related variables. Various statistical, machine learning, deep learning, and transformer-based models are evaluated. Among them, the CatBoost model, trained with selected and engineered features, delivers the best results with a MAE of 739 MWh, outperforming the EXIST market baseline. The findings highlight the importance of robust feature selection and confirm the value of data-driven methods for enhancing operational decision-making in the power sector.