Physics-Informed Deep Koopman Framework for Control-Oriented Wind Turbine Modeling
Lists
Only you see this
Abstract
The complex nonlinearity of wind turbine dynamics presents a challenge for control-oriented wind turbine modeling. The recently emerging deep learning and Koopman operator theory-based modeling frameworks offer a solution for this challenge. However, in the context of wind turbine modeling, these frameworks require a large training dataset, which contains massive wind turbine operational time series. Unfortunately, such a dataset is difficult to collect in practice. To address this issue, by leveraging Koopman operator theory and physics-informed deep learning techniques, this paper proposes a physics-informed deep Koopman framework. The salient feature of this framework is that the underlying physical laws behind wind turbine dynamics are incorporated into its training process. By doing so, this framework can establish a high-dimensional global linear wind turbine dynamic model in a physics-informed manner, without the need for the aforementioned dataset. Besides, a damping term is introduced in this framework to ensure the accuracy and stability of the established model. Simulation results verify the accuracy of the established high-dimensional global linear wind turbine dynamic model as well as its performance in model predictive control.
BibTeX
@misc{tian2026physics,
author = {Runze Tian},
title = {{Physics-Informed Deep Koopman Framework for Control-Oriented Wind Turbine Modeling}},
year = {2026},
url = {https://openreview.net/forum?id=yfCBJr3LCc},
howpublished = {OpenReview},
}Loading related content...