Robust data-driven wake steering for wind farm power maximisation
File(s)
Author(s)
Gori, Filippo
Type
Thesis
Abstract
Maximising the power output of wind farms is essential to reduce the cost of wind energy and meet growing renewable energy targets. Wake interactions between turbines typically cause annual power losses of 10–30 %, motivating the development of control strategies such as wake steering, in which upstream turbines are intentionally yawed to redirect their wakes away from downstream turbines. While wake steering has shown strong potential to mitigate wake losses, its performance remains highly sensitive to modelling and algorithmic choices, challenging its reliable deployment at scale. The novelty of this work first lies in identifying and establishing optimisation sensitivity as a fundamental barrier to robust wake steering. Specifically, a systematic investigation quantifies how wake model and optimiser choices, and atmospheric conditions, influence optimal yaw settings and predicted power gains. Using hierarchical test cases from a two-turbine to the 80-turbine Horns Rev wind farm, the analysis reveals strong dependencies and multimodal optimisation landscapes, which obstruct the identification of consistent, physically meaningful solutions. Mitigation strategies based on global search methods and constraint formulation reduce variability, yielding more coherent and practicable yaw strategies. Building on this foundation, the thesis introduces a Gaussian process-based yaw-dependent parameter tuning framework that dynamically calibrates analytical wake models as a function of yaw misalignment
by integrating higher-fidelity measurements. Extensions through dimensionality reduction and generalisation techniques enable scalability to large wind farms and adaptability across atmospheric conditions. Applications to small turbine arrays and the Horns Rev layout show that the method allows low-fidelity models to reproduce the effects of unmodelled wake physics with high accuracy while maintaining the computational efficiency required for closed-loop control. Collectively, these contributions provide new insights and practical methodologies that advance the capabilities of wake steering frameworks, addressing key literature gaps and strengthening the foundations for robust and data-enhanced wind farm control.
by integrating higher-fidelity measurements. Extensions through dimensionality reduction and generalisation techniques enable scalability to large wind farms and adaptability across atmospheric conditions. Applications to small turbine arrays and the Horns Rev layout show that the method allows low-fidelity models to reproduce the effects of unmodelled wake physics with high accuracy while maintaining the computational efficiency required for closed-loop control. Collectively, these contributions provide new insights and practical methodologies that advance the capabilities of wake steering frameworks, addressing key literature gaps and strengthening the foundations for robust and data-enhanced wind farm control.
Version
Open Access
Date Issued
2025-11-03
Date Awarded
2026-03-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Laizet, Sylvain
Wynn, Andrew
Publisher Department
Department of Aeronautics
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)