Wind power forecasting
A wind power forecast predicts how much electricity a wind farm will feed into the grid, hour by hour. Frontwatt forecasts every turbine separately from over 100 weather scenarios and reports the result as a range, from P10 to P90, instead of a single number.
Updated
Why wind output is hard to predict
A turbine’s output does not follow wind speed in a straight line. Between cut-in and rated speed, power rises roughly with the cube of wind speed, so a 10% error in wind speed can turn into a 30% error in power. Above rated speed the output is flat, and above cut-out speed the turbine stops altogether.
Weather also moves. When a front passes, wind speed and direction can change within an hour or two, and a wind farm can drop from near full output to a fraction of it. These ramps are where most forecast error, and most imbalance cost, comes from. The hard question is rarely how windy tomorrow will be. It is when exactly the change will arrive.
How Frontwatt forecasts a wind farm
- Weather scenarios at every turbine. We read the latest runs of AI-based and physics-based global ensemble models, including the ECMWF IFS and AIFS ensembles, at the position of each turbine. Each ensemble member is one plausible version of tomorrow.
- Wind at hub height. Models give wind at fixed heights such as 100 metres. We bring it to each turbine’s hub height and correct for air density, which changes with temperature and pressure.
- Each turbine’s own power curve. The wind of every scenario runs through the power curve of the turbine model at that position, with cut-out behaviour, park losses and the capacity limit of the grid connection.
- Totals scenario by scenario. Turbines are added up within each scenario, so the farm total keeps the correlation between turbines that see the same weather.
- Calibration on your meter. When metered history is available, a machine-learning layer corrects systematic errors, such as a turbine that underperforms its curve or a site that is windier than the model grid. New calibration models learn only from data that existed when each past forecast was made.
- Your limits. Planned maintenance and curtailment are applied last, so no scenario exceeds what the plant can deliver.
The result is a set of over 100 hourly output scenarios. From them we take the median, P50, and the ranges P25–P75 and P10–P90. What these numbers mean.
In short: turbine-level physics gives a sound forecast from the first day. Calibration on metered output makes it sharper as history builds up.
What you need to provide
| Input | Why it matters | Needed |
|---|---|---|
| Plant location and installed capacity | Selects the weather and caps the output | Yes |
| Turbine models, hub heights and positions | Power curves and site-specific wind | Strongly recommended |
| Hourly metered output | Calibration and scoring | Recommended |
| Planned outages and curtailment | Keeps the forecast within real limits | When they occur |
In Türkiye, EPİAŞ publishes the hourly output of licensed plants, so calibration can start before you send us any data.
What you get
- Hourly P10, P25, P50, P75 and P90 for the rest of today, tomorrow and the day after.
- A new forecast each time a new weather model run is published.
- The day-ahead schedule with the lowest expected imbalance cost.
- Daily accuracy scorecards against your metered output.