Szczegóły publikacji
Opis bibliograficzny
Forecasting sectoral electricity consumption in the selected European countries: machine learning versus deep learning models / Atif Maqbool KHAN, Artur WYRWA // Energy and AI [Dokument elektroniczny]. — Czasopismo elektroniczne ; ISSN 2666-5468 . — 2026 — vol. 25 art. no. 100826, s. 1-23. — Bibliogr. s. 20-23, Abstr. — Publikacja dostępna online od: 2026-06-23
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 169323 |
|---|---|
| Data dodania do BaDAP | 2026-09-11 |
| Tekst źródłowy | URL |
| DOI | 10.1016/j.egyai.2026.100826 |
| Rok publikacji | 2026 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Creative Commons | |
| Czasopismo/seria | Energy and AI |
Abstract
Sector-level electricity demand forecasting remains underexplored in multi-country energy systems despite its importance for grid planning, electrification, and decarbonization. This study develops a comparative forecasting framework for 266 annual sector-country time series covering 14 sectors in 19 European countries from 1995 to 2023. It contributes to benchmarking statistical, machine learning, and deep learning models under a unified empirical setting and assesses model robustness across heterogeneous sectors and national energy systems. Five approaches—ARIMA, SVM, Random Forest, LSTM, and CNN–BiLSTM–AR— were evaluated. Forecasting performance varies markedly across sectors and countries; Random Forest achieves the lowest forecasting errors in most sector-country cases and the best overall accuracy (MAPE = 8.31%), followed by ARIMA (MAPE = 9.84%), while LSTM and SVM show weaker average performances. Diebold–Mariano tests indicate that Random Forest, LSTM, and SVM outperform ARIMA in most comparisons. The results show that ensemble-based nonlinear methods, especially random forests, provide a robust approach to long-term sectoral electricity demand forecasting in Europe, with implications for grid planning, electrification, and energy system optimization.
