EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Monthly Peak-Demand Forecasting and Conditional Long-Horizon Scenarios with a Tri-Model Ensemble

Carlos Barrera-Singaña, María Paz Comech, Hugo Arcos

Forecasting · 14.11.2026 · 8 (6) · 109

DOI: 10.3390/forecast8060109 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Forecasting, online erschienen am 14.11.2026. Verfasst von Carlos Barrera-Singaña und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Long-horizon peak-demand projections combine statistical uncertainty with assumptions about electrification and new connections. This study separates those sources of variation. A statistical baseline combines long short-term memory, LightGBM and temporal fusion transformer models trained directly on monthly peak levels. Weights are estimated from complete recursive 24-month forecast paths, and a dependence-aware bootstrap of log-level errors generates 5000 statistical trajectories. Net physical contributions from vehicles, heating and new projects are then added under explicitly defined structural scenarios. The case studies use 132 monthly peak-demand observations for each country, covering 2015–2025. For the terminal 2024–2025 block, the ensemble yields root mean square errors of 1541.5 and 4215.4 MW for Spain and France, respectively. In Spain, LightGBM achieves a slightly lower error than the ensemble, whereas in France, the ensemble outperforms the three individual learners but not the strongest ETS benchmark. The statistical baseline achieves P5–P95 coverage of 95.83% in Spain and 83.33% in France. Under the reference turnover scenario, annual P50 increases from 41.473 to 57.255 GW in Spain and from 82.591 to 107.582 GW in France between 2026 and 2045. Vehicle renewal, heating substitution and peak coincidence materially affect these projections. The output is a set of conditional annual peak-demand tables for subsequent generation and transmission expansion studies. Neither the structural scenarios nor their quantiles constitute an empirically validated twenty-year prediction interval.

Zitieren

APA

Barrera-Singaña, C., Comech, M. P., & Arcos, H. (2026). Monthly Peak-Demand Forecasting and Conditional Long-Horizon Scenarios with a Tri-Model Ensemble. Forecasting, 8(6), 109. https://doi.org/10.3390/forecast8060109

Vancouver

Barrera-Singaña C, Comech MP, Arcos H. Monthly Peak-Demand Forecasting and Conditional Long-Horizon Scenarios with a Tri-Model Ensemble. Forecasting. 2026;8(6):109. doi:10.3390/forecast8060109

BibTeX
@article{BarreraSingaa2026,
  author = {Barrera-Singaña, Carlos and Comech, María Paz and Arcos, Hugo},
  title = {Monthly Peak-Demand Forecasting and Conditional Long-Horizon Scenarios with a Tri-Model Ensemble},
  journal = {Forecasting},
  year = {2026},
  volume = {8},
  number = {6},
  pages = {109},
  doi = {10.3390/forecast8060109}
}
Kostenlos starten →