Parameter‐Efficient Adaptation of GPT‐Style Foundation Models for Power Grid Load Forecasting
Electronics Letters · 2026-01 · 62 (1)
DOI: 10.1049/ell2.70612 ↗Kurz gefasst
Fachartikel in der Zeitschrift Electronics Letters, online erschienen am 19.05.2026. Verfasst von Lina Sheng und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 14 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
This letter proposes a novel power grid time‐series forecasting approach that combines foundation models with parameter‐efficient LoRA fine‐tuning, specifically for power grid load forecasting. We provide a concise characterization of power‐grid load temporal structure with periodicity and long‐range dependence, and leverage it to motivate parameter‐efficient adaptation of a GPT‐style time‐series foundation model for day‐ahead forecasting. Furthermore, two adaptation strategies are proposed within a GPT‐style framework, including a partial unfreezing mechanism that selectively updates the top Transformer layers and a LoRA‐based refinement approach that injects low‐rank adapters into attention and feed‐forward sublayers. Finally, the comprehensive experiments are conducted on the AEP_hourly dataset, demonstrating that the proposed approach consistently outperforms traditional baselines and frozen or partially fine‐tuned variants while using significantly fewer parameters and computational resources.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Sheng, L., Xu, Y., Zhang, X., & Wei, M. (2026). Parameter‐Efficient Adaptation of GPT‐Style Foundation Models for Power Grid Load Forecasting. Electronics Letters, 62(1). https://doi.org/10.1049/ell2.70612
Vancouver
Sheng L, Xu Y, Zhang X, Wei M. Parameter‐Efficient Adaptation of GPT‐Style Foundation Models for Power Grid Load Forecasting. Electronics Letters. 2026;62(1). doi:10.1049/ell2.70612
BibTeX
@article{Sheng2026,
author = {Sheng, Lina and Xu, Yao and Zhang, Xiaoyu and Wei, Mengli},
title = {Parameter‐Efficient Adaptation of GPT‐Style Foundation Models for Power Grid Load Forecasting},
journal = {Electronics Letters},
year = {2026},
volume = {62},
number = {1},
doi = {10.1049/ell2.70612}
}
EUVIMED