Hourly Price Calendar

Wind Power Outlook

Forecast History

How the Forecast Works

The forecasts on this site are generated with an XGBoost-based machine learning pricing model (gradient-boosted decision trees) that estimates hourly electricity prices from weather, generation, and market features. Each hour is evaluated independently based on its expected conditions, which lets the model react to changes in wind, imports, generation availability, and seasonal demand.

The model is trained on historical market data starting from 2023 and produces a price estimate by learning patterns that have historically correlated with price outcomes.

Key Inputs Considered by the Model

• Wind conditions across the Nordics/Baltics (local stations + regional wind fields)
• Temperature forecasts capturing demand and seasonality
• Generation availability (including nuclear where applicable)
• Cross-border transmission flows for the forecast area
• Solar radiation as a proxy for solar output
• Calendar effects such as year, weekday, hour, and holidays

Wind Forecasting

For all areas, recent ENTSO-E actuals are used to anchor the wind picture. Forward-looking wind forecasts are built from weather station measurements combined with historical power production and installed capacity data. The exception is Finland's first 72-hour horizon, where Fingrid's operational wind power forecast is used directly.

Model Logic and Interpretation

The model learns statistical relationships between explanatory variables and price outcomes. High wind or mild temperatures are typically associated with lower prices, while low wind, cold weather, or reduced import capacity tend to push prices higher.

Geographic Scope and Development

The current implementation covers Finland (FI), Sweden (SE1-SE4), Norway (NO1-NO5), Denmark (DK1-DK2), and Germany (DE). As the scope grows, the model will incorporate additional regional weather, generation, and transmission data.

Limitations of the Forecast

The model only considers variables included in its training data. It does not directly account for unexpected geopolitical events, sudden market interventions, or rare structural shocks with limited historical precedent. Treat forecasts as data-driven estimates, not exact future outcomes.