SFDRLF: Spatiotemporal Feature Decoupling and Dynamic Fusion Model for Power Regional Load Forecasting
IEEE Access · 2026 · 14 · 97999-98013
DOI: 10.1109/access.2026.3705736 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hongbo Li und 5 weiteren Autor:innen. Die Arbeit nennt 24 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Science and Technology Project of State Grid Corporation of China, titled “Research on Construction and Application Technology of Intelligent Multimodal Measur….
Abstract (Originalsprache)
Regional power loads in power grids exhibit complex spatiotemporal coupling characteristics. Simply relying on historical time-series data for load forecasting cannot achieve high accuracy. This paper proposes a Spatiotemporal Feature Decoupling and Dynamic Fusion Model for Power Regional Load Forecasting (SFDRLF) to improve the accuracy of short-term regional load forecasting. The SFDRLF model adopts a dual-branch encoder architecture to decouple and extract dynamic temporal patterns and static spatial attributes. Through spatiotemporal dynamic graph reasoning, the model adaptively constructs dynamic spatial relationships based on the real-time operational status of nodes, and further integrates these relationships with prior static physical topology via a gated fusion mechanism. A relation-aware head is then utilized to efficiently aggregate information on the fused graph, ultimately enabling accurate regional load forecasting. Experiments conducted on the Australian Electricity and UK Power Networks datasets demonstrate that SFDRLF significantly outperforms state-of-the-art baseline models. Specifically, on the UK Power Networks dataset, SFDRLF achieves an $R^{2}$ of 97.25%, with an MAE of 980.40, an RMSE of 1650.85, and a MAPE of 18.35%. The proposed model effectively captures the spatiotemporal linkage characteristics of regional power loads, significantly enhancing the accuracy and robustness of power load forecasting.
Förderung
- Science and Technology Project of State Grid Corporation of China, titled “Research on Construction and Application Technology of Intelligent Multimodal Measur…
Zitieren
APA
Li, H., Han, L., Ma, X., Yuan, L., Lv, M., & Wei, Z. (2026). SFDRLF: Spatiotemporal Feature Decoupling and Dynamic Fusion Model for Power Regional Load Forecasting. IEEE Access, 14, 97999-98013. https://doi.org/10.1109/access.2026.3705736
Vancouver
Li H, Han L, Ma X, Yuan L, Lv M, Wei Z. SFDRLF: Spatiotemporal Feature Decoupling and Dynamic Fusion Model for Power Regional Load Forecasting. IEEE Access. 2026;14:97999-98013. doi:10.1109/access.2026.3705736
BibTeX
@article{Li2026,
author = {Li, Hongbo and Han, Liangyu and Ma, Xiaoqin and Yuan, Lufeng and Lv, Mingjia and Wei, Zhenhua},
title = {SFDRLF: Spatiotemporal Feature Decoupling and Dynamic Fusion Model for Power Regional Load Forecasting},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {97999--98013},
doi = {10.1109/access.2026.3705736}
}
EUVIMED