Ancillary Service Prices
We point a trained Random Forest Regression model to the DA price forecasts. This assume past relationship between DA and different ancillary products will continue to hold. However, with more battery supply entering these markets we expect the price levels to drop over time.
A Random Forest Regression Model was trained on historical prices. A 80-20 Train-Test split was used alongside following features for predicting each ancillary service price:
- DA (EUR/MWh)
- DA_lag1 (DA price lagged by 1 hour)
- Hour
- Day of the Week
- Month
We can see that the model performs well when compared with actual prices:
Updated 6 months ago