Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

WindDragon: Enhancing wind power forecasting with Automated Deep Learning

Julie Keisler, Etienne Le Naour

TL;DR

The paper tackles the challenge of short-term wind power forecasting for grid operation by leveraging spatial information from Numerical Weather Prediction wind speed maps. It introduces WindDragon, an Automated Deep Learning framework built on the DRAGON AutoDL platform, which regresses 2D wind speed maps to wind power forecasts for horizons $h \in \{1,\dots,6\}$ hours at national scale. By adapting DRAGON to a two-graph architecture and employing a steady-state evolutionary search with a global multi-region loss, WindDragon automatically discovers architectures that outperform traditional CNN, ViT, and XGBoost baselines, especially when using full wind speed maps rather than summary statistics. Across 12 French regions (2018–2020 data), WindDragon delivers measurable gains, including 22 MW (6%) over CNN and, for full-map experiments, 31 MW (7.4%) and 47 MW (11.5%) improvements over mean-based XGB, corresponding to an annual impact of about 193 GWh, underscoring the method’s potential for enhancing wind power integration into power systems.

Abstract

Achieving net zero carbon emissions by 2050 requires the integration of increasing amounts of wind power into power grids. This energy source poses a challenge to system operators due to its variability and uncertainty. Therefore, accurate forecasting of wind power is critical for grid operation and system balancing. This paper presents an innovative approach to short-term (1 to 6 hour horizon) windpower forecasting at a national level. The method leverages Automated Deep Learning combined with Numerical Weather Predictions wind speed maps to accurately forecast wind power.

WindDragon: Enhancing wind power forecasting with Automated Deep Learning

TL;DR

The paper tackles the challenge of short-term wind power forecasting for grid operation by leveraging spatial information from Numerical Weather Prediction wind speed maps. It introduces WindDragon, an Automated Deep Learning framework built on the DRAGON AutoDL platform, which regresses 2D wind speed maps to wind power forecasts for horizons hours at national scale. By adapting DRAGON to a two-graph architecture and employing a steady-state evolutionary search with a global multi-region loss, WindDragon automatically discovers architectures that outperform traditional CNN, ViT, and XGBoost baselines, especially when using full wind speed maps rather than summary statistics. Across 12 French regions (2018–2020 data), WindDragon delivers measurable gains, including 22 MW (6%) over CNN and, for full-map experiments, 31 MW (7.4%) and 47 MW (11.5%) improvements over mean-based XGB, corresponding to an annual impact of about 193 GWh, underscoring the method’s potential for enhancing wind power integration into power systems.

Abstract

Achieving net zero carbon emissions by 2050 requires the integration of increasing amounts of wind power into power grids. This energy source poses a challenge to system operators due to its variability and uncertainty. Therefore, accurate forecasting of wind power is critical for grid operation and system balancing. This paper presents an innovative approach to short-term (1 to 6 hour horizon) windpower forecasting at a national level. The method leverages Automated Deep Learning combined with Numerical Weather Predictions wind speed maps to accurately forecast wind power.
Paper Structure (19 sections, 8 figures, 2 tables, 1 algorithm)

This paper contains 19 sections, 8 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. WindDragon: a framework for regression on wind speed maps
  3. The DRAGON framework.
  4. WindDragon: adapting the DRAGON framework for wind power forecasting.
  5. Experiments
  6. Datasets.
  7. Data preparation.
  8. Results.
  9. Conclusion and impact statement
  10. WINDDRAGON
  11. Search algorithm.
  12. Results.
  13. Baselines details
  14. Convolutional Neural Network (CNN).
  15. Vision Transformer (ViT).
  16. ...and 4 more sections

Figures (8)

  • Figure 1: Global scheme for wind power forecasting. Every 6 hours, the NWP model produces hourly forecasts. Each map is processed independently by the regressor which maps the grid to the wind power corresponding to the same timestamp.
  • Figure 2: WindDragon’s meta model for wind power forecasting
  • Figure 3: Data preparation for the region Auvergne-Rhône-Alpes. The wind farms are represented in red. The first image shows the distribution of wind farms across the administrative region.
  • Figure 4: Wind power forecasts for a week in January 2020. The figure displays the ground truth as dotted lines, and the forecasts from the two top-performing models, WindDragon and the CNN.
  • Figure 5: Dragon automatically found architecture applied on the Grand Est region.
  • ...and 3 more figures