A multi-fidelity framework for power prediction of wind farm under yaw misalignment

Collective yaw control is a promising approach for wind farm flow control. The investigation approaches exhibit varying levels of fidelity. High-fidelity numerical simulations offer accurate representations but are computationally expensive, while low-fidelity analytical models provide rapid calculations with reduced accuracy. Confronted with this dilemma, the multi-fidelity surrogate model emerges as a compelling solution. In this paper, we introduce a multi-fidelity framework based on the co-Kriging algorithm to efficiently predict the wind farm power under yaw misalignment. Two surrogate models are compared, the single-fidelity Kriging (SFK) model and the multi-fidelity co-Kriging (MFK) model. Both models provide desired accuracy, with MFK model outperforming SFK model. For one-dimensional and three-dimensional cases, the MFK model significantly reduces the demanding high-fidelity samples, while remains accuracy of 96.5% and 93.9%, respectively. We further investigate the influence of low-fidelity data sources on MFK model, including the Gauss-Curl Hybrid (GCH) wake model, Gaussian wake model, and Jensen wake model. The MFK-GCH model shows better prediction accuracy, indicating that low-fidelity data with more physical information benefits the model. Furthermore, sensitivity analysis is given to ensure reliable and consistent results. The multi-fidelity framework enables efficient and accurate power prediction, which lays the foundation of yaw optimization in large-scale wind farms.