Wind power forecasting
A two-layer LSTM (PyTorch) predicting turbine power 10 minutes ahead from 8 SCADA features.
R² = 0.96 on the held-out test set.
Short-term wind power forecasts support grid operation and renewable integration. This project predicts a turbine’s power output 10 minutes ahead.
Approach
- Inputs: 8 SCADA features (wind speed and direction, external and internal temperature, nacelle direction, blade pitch angles).
- Model: two stacked LSTM layers (128 hidden units each) with dropout, trained in PyTorch on 18-step (3-hour) sliding windows.
- Evaluation: separate validation, tuning and held-out test sets, reported with MSE, MAE, RMSE and R².
Result
R² = 0.96 on the held-out test set.