Helping mitigate climate change through efficient reinforcement learning-based wind farm flow control

Improving wind farm efficiency is critical for reducing greenhouse gas emissions and scaling renewable energies. One effective approach to increase a wind farm's power output is wake steering, where certain turbines are intentionally misaligned with the wind to enhance downstream airflow and reduce wake losses. However, designing robust, large-scale wake steering controllers remains challenging due to uncertain and time-varying wind conditions. We propose an attention-based reinforcement learning architecture and a carefully designed reward shaping methodology to develop more efficient wake steering controllers. Using a steady-state, low-fidelity simulator, we show that our approach increases energy capture relative to strong baselines, illustrating how machine learning can directly improve renewable energy generation and contribute to climate change mitigation.