FrankKr
commented
1 year ago Commits + analysis: https://github.com/alliander-opensource/AIFES/commit/7af617ccc7d5adc90e28f68548277c5f14b34b21
Results: (closed due to data): https://alliander.sharepoint.com/:f:/r/teams/PortfolioSO/Gedeelde%20documenten/Collaboration%20Allianer%20-%20TenneT%20T-Prognose/Data%20and%20results/Results/%2334_Single_vs_multiple_horizons_training?csf=1&web=1&e=wQjOOd
Why/What: The current forecasting methodology trains a single model to predict at continuous lead times. This a quite different from what traditionally happens, where a single model is trained for each lead time. Analyse how this strategy compares in terms of accuracy to a single model per horizon
Do it as a Proof Of Concept, to show how this setup can be used to perform (academic) analysis.