Multi-fidelity Bayesian Optimisation of Wind Farm Wake Steering using Wake Models and Large Eddy Simulations

Area of Science:

• Renewable Energy Engineering
• Computational Fluid Dynamics
• Aerodynamics

Background:

• Wind turbine wakes reduce efficiency and increase fatigue in wind farms.
• Wake steering strategies can enhance total wind farm power output.
• Analytical wake models are commonly used but have limitations.

Purpose of the Study:

• To compare the effectiveness of analytical wake models and high-fidelity large eddy simulations (LES) for optimizing wind farm yaw configurations.
• To introduce a multi-fidelity Bayesian optimization framework to reduce computational costs associated with LES.

Main Methods:

• Utilized large eddy simulations (LES) for higher-fidelity wind farm flow field modeling.
• Developed a multi-fidelity Bayesian optimization framework with surrogate models.
• Employed a multi-fidelity acquisition function to guide optimization iterations.

Main Results:

• Large eddy simulations identified more optimal yaw configurations for full wind farms compared to analytical models.
• The multi-fidelity Bayesian optimization framework achieved comparable results to single-fidelity optimization with fewer expensive LES runs.
• Significant reduction in computational cost was demonstrated for achieving optimal wind farm power improvements.

Conclusions:

• High-fidelity LES capture complex wake interactions more effectively for yaw optimization.
• The proposed multi-fidelity Bayesian optimization framework offers a computationally efficient approach to wind farm optimization.
• This method enables achieving substantial wind farm power gains at a reduced computational expense.