EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Spatial Attention or Future Conditioning? Transformers for Regional Wind Power Forecasting

Samaneh A. Mofrad, Wei Wu, Leyao Liu, Xiangting Chi, Matthew Piggot, Simon Warder

Nordic Machine Intelligence · 02.05.2026 · 6 (2) · 18-26

DOI: 10.5617/nmi.12925 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Nordic Machine Intelligence, online erschienen am 02.05.2026. Verfasst von Samaneh A. Mofrad und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We evaluate two Transformer-based architectures, CLCS- Transformer (spatial-temporal) and EDFormer (future- conditioned), for two-day-ahead wind power forecasting across Norwegian bidding zones (2020 to 2025). All models are benchmarked against a strong baseline (XGBoost) to assess whether architectural complexity yields consistent performance gains. The CLCS-Transformer explicitly models regional spatial dependencies and aggregates temporal information across multiple scales, targeting spatio-temporal heterogeneity in wind generation. EDFormer adopts an encoder–decoder design that jointly models temporal and cross-variable interactions by conditioning on both histor- ical and future exogenous covariates (weather forecasts). Results show that transformer models achieve performance comparable to XGBoost overall, but that (i) conditioning on future exogenous covariates improves forecasts across several zones, and (ii) the CLCS spatial–multi-scale design yields consistent gains in regions with heterogeneous or sparse farm distributions. These findings highlight how model design and feature conditioning interact with geographic characteristics to influence forecasting skill.

Zitieren

APA

Mofrad, S. A., Wu, W., Liu, L., Chi, X., Piggot, M., & Warder, S. (2026). Spatial Attention or Future Conditioning? Transformers for Regional Wind Power Forecasting. Nordic Machine Intelligence, 6(2), 18-26. https://doi.org/10.5617/nmi.12925

Vancouver

Mofrad SA, Wu W, Liu L, Chi X, Piggot M, Warder S. Spatial Attention or Future Conditioning? Transformers for Regional Wind Power Forecasting. Nordic Machine Intelligence. 2026;6(2):18-26. doi:10.5617/nmi.12925

BibTeX
@article{Mofrad2026,
  author = {Mofrad, Samaneh A. and Wu, Wei and Liu, Leyao and Chi, Xiangting and Piggot, Matthew and Warder, Simon},
  title = {Spatial Attention or Future Conditioning? Transformers for Regional Wind Power Forecasting},
  journal = {Nordic Machine Intelligence},
  year = {2026},
  volume = {6},
  number = {2},
  pages = {18--26},
  doi = {10.5617/nmi.12925}
}
Kostenlos starten →