Loading the latest forecast…
Loading the latest forecast…
Thu, 19 Nov 2026 · forecast 3 days ahead
All hours in Germany-Luxembourg local time (CET).
| Hour (local) | Forecast €/MWh |
|---|---|
| 00:00 | 187.4 |
| 01:00 | 183.8 |
| 02:00 | 177.9 |
| 03:00 | 174.3 |
| 04:00 | 175.3 |
| 05:00 | 180.4 |
| 06:00 | 209.5 |
| 07:00 | 248.3 |
| 08:00 | 244.5 |
| 09:00 | 221.6 |
| 10:00 | 195.5 |
| 11:00cheapest 4h | 178.0 |
| 12:00cheapest 4h | 174.7 |
| 13:00cheapest 4h | 169.2 |
| 14:00cheapest 4h | 185.2 |
| 15:00 | 192.6 |
| 16:00 | 220.1 |
| 17:00 | 249.3 |
| 18:00 | 254.6 |
| 19:00 | 239.0 |
| 20:00 | 223.0 |
| 21:00 | 204.7 |
| 22:00 | 192.8 |
| 23:00 | 180.6 |
published livePredicted at 13:49 CET on 16 Nov, 3 days before delivery.
The grid operator's day-ahead forecast: demand, and the wind and solar that meets part of it. What is left over has to come from gas, hydro and other dispatchable plants. The grid operator had not published this yet when the D+3 forecast was made, so the model did not see it. It is shown as the physical record of the day, not as an input.
A small input at this horizon, in single-digit percentages of model attention, and the physical driver behind the D+3 forecast.
MAE is the average hourly miss in EUR/MWh. Naive repeats the same hour last week. Improvement is how much smaller our miss is. The figures above score the model serving now on its own days; the chart keeps every day, with earlier models' days faded. Its backtest, the same model trained only on data from before 7 Sept and scored on 7 Sept to 21 Oct: €28.2 against €64.8 naive, 57% smaller miss. This backtest is not a live result.
Share of the trained model's normalized feature importance, grouped by concept, at each horizon. More on the features.