Physics-Embedded Deep Koopman Modeling for Wind Turbine Fatigue Load Reduction Control

By reducing fatigue loads, the operation and maintenance costs of wind farm can be effectively decreased. However, the dynamics of wind turbine, especially those related to fatigue loads, exhibit complex nonlinearity. This presents a challenge for establishing an effective control scheme to achieve fatigue load reduction. To address this issue, in this article, by leveraging the Koopman operator theory and deep learning techniques, a physics-embedded deep learning modeling framework is proposed. Using this framework, a high-dimensional global linear wind turbine dynamic model, which can accurately describe the dynamics of fatigue load-related states, is obtained. The salient feature of this framework is that the physical mechanisms of fatigue load-related states are embedded in a manner consistent with Koopman operator theory to improve the accuracy and generalization ability of the obtained model. Based on the obtained model, a coordinated model predictive control scheme is proposed. Meanwhile, a novel fatigue load reduction-oriented cost function is introduced in this scheme. By doing so, the active power output and pitch angle of each wind turbine are coordinated for fatigue load reduction and active power command tracking. Simulation results verify the effectiveness of proposed modeling framework and control scheme.