10 August 2026
AI helps predict wind power production at Dragaliden
Researchers and project partners at Luleå University of Technology have completed the first phase of an AI-based wind power forecasting study within the Dragaliden subproject, part of TAM Tools for Wind Power under the MARTINA project.
The work shows that artificial intelligence and machine learning can predict short-term wind power production with high accuracy, using historical wind farm data and operational information from multiple wind power sites.
Wind power is one of the most important sources of renewable energy, but it also brings a practical challenge. The wind changes constantly. For wind farm operators, this means that power production can rise or fall quickly, sometimes within minutes. Better forecasts can therefore support planning, maintenance, grid stability, and more efficient use of renewable energy.
In the Dragaliden study, the team analyzed several years of wind farm data, including measurements collected at short time intervals. The goal was to predict how much power the wind farm would produce in the near future. Instead of relying only on traditional statistical models, the team trained and compared several artificial intelligence and machine learning models.
The results from the first phase are very encouraging. Several models were able to closely follow actual power production, and the prediction errors were low. These models can learn patterns from historical data and combine many factors, such as wind speed, time of day, seasonal behavior, and turbine-related information, to make reliable predictions.
For people working with wind power, the value of such models is practical. A good forecast can help answer simple but important questions, such as: How much power are we likely to produce in the next interval? Is the production pattern normal? Are there signs that the turbines are behaving differently than expected? Can we plan operations and maintenance more intelligently?
A key strength of the project is that the models were tested carefully on unseen data. This means that the AI models were evaluated not only on data they had already seen during training but also on later time periods. This gives a more realistic picture of how the models may behave in real operational settings.
The work also illustrates how AI can support the green transition in northern Sweden and beyond. By combining domain knowledge from wind power with modern data-driven methods, the project contributes to smarter renewable energy systems. The findings show that AI need not be seen as a distant or abstract technology. In this case, it serves as a practical tool to help operators understand and predict wind farm behavior.
The work has been carried out through close collaboration between project partners and researchers. The team includes Jens Sperens, COB, Dragaliden Vind AB; the Luleå University of Technology team consisting of Björn Backe, Petter Kyösti, and Rajkumar Saini, and Sumit Rakesh; and interns Sanyam Kathed, Hith Rahil Nidhan, and Mridul Maheshwari.
Published:
Updated: