In a 27 October preprint, reinforcement learning chooses turbine angles and selective power cuts to balance electricity against a protected area’s sound level.[1]
Turning redirects a turbine’s wake, the flow reaching machines behind it. The control problem connects several turbines: a loss of output at one machine can be offset by improved conditions for another.[2]
Six turbines were simulated. A controller trained in FLORIS was transferred without retraining to the more detailed FAST.Farm model.[1]
Turbulent FAST.Farm test
| Control | Farm power loss | Time above 40 dB |
|---|---|---|
| Uniform power reduction | 18.1% | 30.5%Source 1 |
| Noise Steering | 7.4% | 8.3%Source 1 |
Uniform-reduction settings also transferred unchanged after FLORIS tuning. The test used one turbulent wind profile and direction; 40 dB was its chosen limit, and neither controller always met it.[1]
Quieter where?
Sound at a particular location and a turbine’s total noise output are different measurements. A separate field experiment, published in Wind Energy Science on 4 September, makes that distinction tangible. Thomas Duc and Arthur Finez placed 24 sound meters around a commercial 2.2 MW turbine and tested its orientation from 20 degrees one way to 20 degrees the other.[2]
Turning rotated the pattern of louder and quieter directions. The estimated overall sound power nevertheless rose by roughly 0.6 dB(A). A sensor can enter a quieter part of the pattern even when total emitted noise has increased. This was a different turbine and experiment; it did not test the AI controller.[2]
Sources & further reading
- Noise Steering: Exploiting Wake Steering and Acoustic Directivity for Noise-Constrained Wind Farm Control via Reinforcement Learning arXiv · 27 Oct 2026 · Accessed 28 Oct 2026
- Experimental investigation of the effect of wake steering on the noise emission of a commercial wind turbine Wind Energy Science · 4 Sep 2026 · Accessed 28 Oct 2026