Where our electricity
comes together.

These are experimental predictions for Denmark’s two electricity regions. A fixed ensemble combines tree models and pretrained time-series models using Energinet’s electricity readings and ECMWF weather forecasts supplied by Open-Meteo. Demand includes system losses.

Updates are attempted every 30 minutes. The models predict hourly averages and normally issue a new outlook when a new complete source hour is available. Every forecast keeps its original issue time and values, so later observations can be compared fairly.

Missing recent readings can delay an update. At most three older missing input hours per week may be filled using earlier observations; the panel reports when this happens. Larger gaps keep the last valid outlook visible with a delay notice.

The historical benchmark covered 13 October 2026 forecast dates per region, at one daily issue time with complete recent inputs. Live refreshes and occasional filled inputs are a broader operating test. The live errors above compare saved forecasts with observations arriving after issue; they are separate from that benchmark.

Prediction ranges target 80% coverage for individual hours, not the whole 48-hour path. Rolling calibration uses only completed past forecasts and observations. The range audit reused 13 October dates and does not establish guaranteed or year-round coverage. Solar ranges are shown only in daylight: overnight source readings failed the coverage checks. A range is saved when issued and is never added retrospectively to an old forecast.

Demand explanations passed reference-sample stability checks for the tree component. Wind and solar explanations failed and are not published. The explanation does not represent the full ensemble or establish causal effects.

Read the benchmark and methodology ↗ · Energinet · Weather by Open-Meteo / ECMWF