1. What is the WindAI Data Challenge?
The WindAI Data Challenge is a competition to develop more accurate and robust models for forecasting wind-power production in Norway, two days in advance. This is crucial for grid stability and efficient energy management.
2. Who can participate?
The competition is open to everyone—individuals or teams—world-wide. Computational resources from the Norwegian AI Cloud are available only to participants with a Norwegian registration number or teams that include at least one Norwegian resident. Statnett employees may join but are not eligible for the monetary prize.
3. What is the objective of the challenge?
Build a model that predicts the total wind-power production for each Norwegian bidding zone (NO1 – NO4) two days ahead, at an hourly resolution for 24 hours.
4. What data is provided?
- Weather Forecasts (MET MEPS): 15-member ensemble forecasts for each wind park—wind speed, direction, temperature, relative humidity, pressure, gusts—up to 65 h ahead.
- Weather Now-casts (MET): Near real-time “zero-hour” weather for each wind park (temperature, pressure, humidity, precipitation, wind speed & direction).
- Observed Wind Power (Statnett): Historical wind-power production aggregated per bidding zone.
- Wind-Power Metadata (Statnett): Wind-park details: bidding zone, maximum capacity, and commissioning date.
- REMIT UMM (NordPool — optional): Public unavailability messages relating to maintenance or outages.
5. How do I access the data?
Download the datasets and a descriptive Jupyter notebook from the competition website or via the Google Drive link provided there.
6. Can I use external data?
Yes—any public data helpful for forecasting is welcome, for example:
- Climate and geographical data
- Other weather forecasts or local-conditions data
- Energy-market information (e.g., production in other European regions)
- REMIT UMM unavailability messages
The data must be accessible to Statnett in a real-world production setting.
7. What is a bidding zone?
A bidding zone is a geographical area in the power market that shares the same spot price for electricity.
8. Is the wind-power data provided at the individual wind-park level?
No. Observed wind-power data are aggregated at bidding-zone level. You do, however, get weather data per wind park and must aggregate your forecast to the bidding-zone total.
9. What programming language can I use?
You may use any programming language.
10. How will the models be evaluated?
Primary metric: Root Mean Square Error (RMSE). The top-10 teams will also be judged on:
- Accuracy (65 %): RMSE during a two-week live-forecasting period
- Trustworthiness & Explainability (20 %): Model clarity, uncertainty estimation, benchmarking, and prediction explainability
- Implementation & Presentation (10 %): Ease of use, retraining feasibility, clarity, and Statnett integration
- Robustness to Changes (5 %): Adaptability to wind-park capacity changes and climate change
- Computational Requirements: Used as a tie-breaker if needed
11. What is the submission format?
- September 22–26 & September 29–October 3: Daily CSV forecasts (workdays) containing hourly predictions for each bidding zone.
- Final submission (September 20):
- 5-minute video explaining your approach
- 2–3 page mini-paper on methods & results
- Zip archive with code & a “how-to-run” notebook
- PDF version of the notebook
12. What computational resources are available?
Norwegian residents—or teams with a Norwegian resident—may request virtual-machine access via the Norwegian AI Cloud. A webinar in August will demonstrate usage. Contact Kushtrim Visoka for details.
13. What are the important dates?
- May 5, 2025 — Competition launch & data release
- July 1, 2025 — Registration deadline
- August 2025 — HPC workshop
- September 20, 2025 — Code, paper & video deadline
- Sept 22–26 & Sept 29–Oct 3, 2025 — Two-week live forecasting period (one missed submission allowed)
- Nov 26–27, 2025 — Award ceremony at Nordic AI Meet, Norrköping, Sweden
14. Where can I find more information?
Visit the competition website for details, downloads, and updates, or email kushtrim.visoka@nora.ai.
15. What are the prizes?
The winning team receives 50 000 NOK and an opportunity to showcase its solution to Statnett.
16. Will there be a publication opportunity?
Mini-papers will be considered for a joint peer-reviewed publication on the competition results. Outstanding contributions may be invited to submit a full paper to the Nordic Machine Intelligence Journal.
17. Where do I register?
Register on the competition website. Provide a primary contact and team list if registering as a team, and indicate whether you require Norwegian AI Cloud resources.
18. What if I have further questions?
Contact the organizers through the website or via kushtrim.visoka@nora.ai.
19. Will there be Q&A sessions?
Yes. A digital Q&A session will be held on September 1, possibly together with physical workshops. Details about the August HPC training webinar will be emailed to registered teams and posted on the website.
20. Should participants train only one model for all four zones, or separate models for each zone?
Both approaches are acceptable. A single model may be developed for all bidding zones, or separate models can be trained per zone.
21. Where can the exact locations of the wind farms be found?
Wind-farm locations (coordinates and metadata) are available from the Norwegian Water Resources and Energy Directorate (NVE) map service: https://temakart.nve.no/tema/vindkraftverk.
22. Is it sufficient to share a GitHub link for the Python code?
No. Submissions must include a zip archive of the code together with a PDF version of the notebook. GitHub links alone are not sufficient, as submissions must be fixed at the deadline and evaluators require a stable version for review.