Regime-aware day-ahead electricity consumption forecasting for Türkiye: A meta-learning-based ensemble approach

Date

2026-6

Author

Savaş, Alp Demir

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

73
views

0
downloads

Cite This

Day-ahead electricity consumption forecasting is very important for power system operation, market planning, and reserve scheduling. This thesis focuses on developing accurate day-ahead electricity consumption profiles for Türkiye. The main aim is to forecast the hourly consumption profile for the next day under real operating conditions. For this purpose, hourly national consumption data together with weather and calendar variables were used for the 2019–2025 period. During historical model development, observed meteorological variables were used to construct and analyze weather-related predictors, whereas day-ahead forecasts for the target periods were generated using forecasted weather inputs rather than realized next-day weather. A major focus of the study was the effect of regime changes caused by weekends, public holidays, and periods such as Ramadan and Eid, since electricity consumption on these days usually follows a different pattern compared to ordinary days. To this end, a fully reproducible forecasting pipeline was developed, including data preparation, feature generation, chronological validation, hyperparameter optimization, and final testing Performance evaluations showed that both tree-based and hybrid models provided strong results. The best overall performance was achieved by the final regime-aware framework, in which a Ridge and LightGBM residual-based meta-learner combines the strongest candidate forecasts together with calendar, weather, and regime-related information. In the 2025 test period, the meta-learner achieved a mean absolute percentage error (MAPE) of 1.57%, with 1.42% for non-holidays and 4.09% for holidays.

Subject Keywords

Day-ahead forecasting, Electricity consumption forecasting, Meta-learning, Holiday effects, Türkiye

URI

https://hdl.handle.net/11511/119454

Collections

Graduate School of Informatics, Thesis