Joint project IntelliWind
Intelligent models for self-optimizing load reduction in wind turbine
Key Info
Basic Information
- Duration:
- 01.05.2022 to 31.10.2024
- Acronym:
- IntelliWind
- Group:
- Drive Systems
- Funding:
- BMBF
Motivation
As a core component of the energy transition, the construction and operation of wind turbines (WTs) has been strongly promoted in Germany in recent years. The development-driving factor here is the cost-of-energy (COE), i.e. the ratio of total costs to annual energy supply. To reduce the COE, the rated power and thus also hub heights and rotor diameters of modern wind turbines are constantly increasing, which also poses a challenge to the requirements for structural integrity in terms of maximum service life. The use of load-reducing control strategies is therefore the subject of current research activities in order to enable a reduction of the load amplitudes in the rotor and tower and thus material savings by means of electronic control interventions.
Modern control and estimation methods based on a mathematical model have the advantage over classical, single-loop control loops of explicitly using knowledge about the plant behavior to reduce the dynamic loads on the rotor and tower by means of control interventions. The achievable control quality and thus also the achievable reduction of the mechanical loads are primarily limited by the accuracy of the mathematical model of the WT used. It should be noted that the computing time of current programmable logic controllers (PLCs) is limited, so that the model must be strongly reduced to the relevant dynamics. Investigations on a real WT have shown that the nominal models have uncertainties which, if not taken into account, not only significantly reduce the advantages of model-based control methods, but also endanger the stability properties of the control loop and limit the accuracy of the load estimation.
The initial question of this project is therefore to what extent a data-driven, artificial intelligence (AI)-based identification of partial models and their online adaptation in load estimation and control reduces the discrepancy between simulation and reality for these model-based approaches.
Project Goals and Methods
The overall goal of the project is to use AI-based regression to identify the system behavior of WTs in a data-driven manner and to investigate its usability in model-based control procedures compared to physical modeling. The following sub-goals are derived from this:
- Derive a data catalog and corresponding specifications for data acquisition and evaluation with respect to data-driven modeling of WTs.
- Automated training for an AI-based grey-box model with real measured data based on Support Vector Machines (SVM), Artificial Neural Networks (ANN), Gaussian processes ...
- Extension of the static model for automatic online adaptation during operation to minimize model errors
- Utilization of the AI-based model in load estimation and control to improve accuracy.
Innovations and Perspectives
The novelty of the approach described lies in the enhancement of the state-of-the-art system model by AI methods. The fact that the complex and computationally intensive submodels are identified on the basis of data, while the basic physically motivated structure is retained, increases the acceptance of the industry for the new type of modeling.
Furthermore, the adaptation of the AI model to the constantly changing operating conditions should be specifically mentioned as a novelty. Thus, the control results can be optimized at runtime and aging phenomena of the WT can be explicitly considered.
Finally, the modeling approach offers the attractiveness of being easily transferable to different plants due to the identical structure of WTs across plants, provided that an adequate data catalog is available or can be generated.
Project partner |
---|