About Unagi
Unagi is a free, open-source dashboard that forecasts day-ahead electricity prices for Sweden and publishes how accurate those forecasts turn out to be. It reads public data and translates it into one readable view — nothing here is behind a login.
What we predict
- All four Swedish bidding zones: SE1, SE2, SE3, SE4.
- Daily-average and hourly prices from tomorrow out to seven days ahead (d+1 to d+7).
- An 80% prediction interval around each forecast.
Accuracy
We publish our accuracy live rather than describe it. For each zone we show the trailing 28-day mean absolute error (MAE) next to a same-weekday-average baseline, and how often the actual price landed inside the 80% interval.
These numbers are visible on the forecast itself, so the forecast and its track record sit side by side.
Method
Prices are predicted with LightGBM (gradient-boosted trees), retrained daily on recent history in a walk-forward setup. The 80% interval comes from conformalized quantile regression (CQR). Horizons beyond d+1 are produced recursively.
Main inputs: forecast demand, wind and weather, hydro reservoir levels, German (DE) day-ahead prices, calendar effects, nuclear availability and recent price history.
Everything is open source on GitHub.
Data sources
- ENTSO-E — Generation mix, cross-border flows and hydro reservoir levels for SE1–SE4.
- SMHI / Open-Meteo — Wind, temperature and solar/radiation — forecasts and history.
- eSett — Nordic imbalance settlement prices.
- Riksbank — EUR/SEK reference rates for converting market data.
- Skatteverket — Electricity tax and VAT (moms) rates for the cost breakdown.
- Energimarknadsinspektionen (Ei) — Network-tariff and retailer reference data.
Public API & MCP
A public API and an MCP server (for asking Unagi questions from an AI assistant) are planned but not released yet. If you have a concrete use case — a Home Assistant sensor, a research project — email hello@unagieel.net and tell us what you need.
Contact
Questions, corrections or data-source suggestions are welcome at hello@unagieel.net, or open an issue on GitHub.