KI4Wind

At a glance

• To date, wind turbine control has focused on individual turbines at the local level. As a result, significant optimization potential - such as that offered by controlling the turbines while accounting for their mutual influence within the wind farm - remains untapped. 
• In the KI4Wind project, the partners are developing an AI-optimized method for controlling the operation of wind turbines in wind farms, which extends the turbines’ service life while simultaneously increasing energy yield.
• Among other things, Fraunhofer IWES is responsible for providing synthetic simulation data, which, in combination with field data, serves as the training basis for the ML algorithm.

The challenge

In addition to the expansion of renewable energies, increasing the efficiency of existing wind farms is of particular importance for achieving climate neutrality. However, modern wind turbines have so far been optimized primarily on a local basis. The complex interaction of multiple turbines within a wind farm has been largely ignored, even though a comprehensive analysis offers significant optimization potential: loads can be reduced, thereby extending the turbines’ service life. Additionally, the farm efficiency - and thus the electricity yield - can be increased.

However, the dynamic behavior in wind farms with numerous wind turbines is highly complex. Finding an optimum balance between loads and power output is a challenging task in flow simulation. Machine learning can offer promising approaches here, but so far these have primarily been used for the predictive detection of faults and damage in individual wind turbines. To the best of the project consortium’s knowledge, AI-supported control of operating parameters to optimize turbine performance and component lifespan has not yet been implemented in the field.

The solution

This is where the KI4Wind research project comes in. This project investigates the use of machine learning methods to identify specific operating and load conditions of wind turbines in order to reduce physical loads while simultaneously optimizing electrical power output.

To achieve this goal, synthetic simulation data is combined with series of experiments using measurement data from strain sensors on the rotor blade as well as other turbine data. This combined data is used to train an AI-based agent for operational management, which is then capable of identifying optimal operating conditions and controlling them. The developed models are intended to be integrated as modules into local wind turbine control systems or deployed centrally in a cloud-based optimization approach.

Fraunhofer IWES contributes its many years of expertise in the areas of wind turbine simulation, automation technology, and real-time systems for realistic testing within its own test infrastructure to the project. The real-time-capable simulation platform developed by IWES for the holistic simulation of wind farms serves as a test platform for the methods developed in this project. Representative simulation data for wind farm operation, as well as for offline development and validation, will be provided. Additionally, as part of the project, IWES is developing algorithms for AI-based wind speed estimation and the replication of simulation results.

Added Value

Innovative AI solutions for optimizing the operating parameters of wind turbines and wind farms represent a quantum leap in plant control: On the one hand, annual energy production is maximized; on the other hand, the overall service life of the plants is extended beyond typical permit periods through load-reduced operation. Extending the service life by just one year could save 5.28 million tons of CO₂ in Germany. In the medium term, the project strengthens the competitiveness of German companies in the wind energy sector.

Link to the final report
https://doi.org/10.34657/28955