01
The wind moves first.
Can a turbine set its blade for the wind that is coming?
02
77,760 readings.
Three simulated 250 kW turbines, every 10 minutes for six months. The paper used hourly data.
03
A simple model misses by a third.
Linear regression: 33.95% error on the blade angle.
04
Predict. Then correct the miss.
XGBoost predicts the angle. A second model learns only from its mistakes.
05
33.95% → 1.11%.
Error down 96.7%.
06
+21% energy.
In the published paper: 841.54 → 1,019.41 kWh on a sample day.
07
Google did it at scale.
DeepMind predicted 700 MW of wind 36 hours ahead. Its value rose about 20%.
Sources: Google DeepMind, 2019 ↗
08
Next: real wind.
Simulated data is clean. A working turbine is the real test.