Hybrid meta-optimized GNN network to optimize pitch angle and active power of wind turbines for reducing fatigue load

Fatigue load of wind turbines (WTs) has a significant adverse impact on their service life. This paper proposes an innovative hybrid meta-optimized GNN network (HMG-Net) to optimize pitch angle and active power of WTs for reducing fatigue load in wake-affected wind farms (WFs). Firstly, to learn from historical operational data, the Dilated Kolmogorov Convolutional Neural Network (DKCNN), Kolmogorov-Arnold Network (KAN), and Model-Agnostic Meta-Learning (MAML) are integrated the combined modeling framework. As the typical machine learning methods, this approach eliminates the need for high-dimensional mathematical models of WTs while achieving dual-objective optimization of tower and drive train loads reduction. By leveraging MAML, the framework mitigates information silos and ensures robustness across varying operational conditions, avoiding over-specialization. Additionally, a multi-order Graph Neural Network (GNN) enhances the expressive capabilities of WT nodes and efficiently models coupling relationships among WTs without requiring complex WTs group dynamics. Comprehensive simulations validate the proposed HMG-Net under diverse conditions, demonstrating its effectiveness and adaptability, with a wake-affected WF, is simulated in MATLAB.