Today, most branches of the energy business rely on an exact forecast of the future wind power input. Power plant scheduling, direct marketing, power trading and grid operations can only be carried out when an exact, reliable and permanently available prediction of wind power exists for the coming hours and days. As a general rule, high forecast accuracy is dependant on an optimal combination of wind farm SCADA data representing current production and forecast data from weather services.

In many practical situations, however, the online availability of SCADA data is problematic as it frequently arrives delayed at the forecasting system, does not have the necessary time resolution or needs correcting or other post-processing steps.

Prediction Quality

In ideal cases, in which SCADA data are both online and of high-quality, statistical prediction models generally produce the greatest degree of forecast accuracy. In all other cases, physical modelling approaches are generally superior as they are more robust when it comes to problematic measured data.

In developing the wind power prediction model OSHybrid, Overspeed has achieved a new hybrid model which combines the advantages of both physical and statistical modelling. The model's core is the description of a given wind farm and its surroundings with physical parameters such as power curves, terrain roughness and orography. The output of this model is corrected with a statistical model which takes the historic time series of the wind farm power into account. The optimization of the model is adaptive, for example monthly. As soon as the model detects larger deviations between model parameters and past tuning parameters, these changes are no longer integrated into the online forecast, but are rather first assessed by an expert and corrected if necessary.



This procedure insures that all advantages of physical and statistical approaches are combined:

  • OSHybrid delivers good forecasts for newly installed wind farms without historic measured data.
    Purely statistical models do not work without SCADA data, so predictions for a new farm are only available after several weeks or even months.
  • OSHybrid is robust when it comes to missing or qualitively poorly measured power data.
    As its core is based on physical modelling and the manual checking of statistical parameters, the model is not led astray through poor data.
  • OSHybrid is ideal for applications with offline SCADA data.
    As the statistical model tuning takes place regularly but offline, it is particularly well-suited to applications in which SCADA data is only available offline.
  • OSHybrid is well-suited for SCADA data which contains downregulation.
    Large wind farms in weak power grids are frequently restricted in power for certain periods of time. These times can be excluded from tuning periods thus avoiding a falsification of the results.
  • OSHybrid produces good results for wind farms under construction.
    Such wind farms constantly vary in installed power as new turbines are added. Forecasts are optionally optimized dynamically through automatic tuning.
  • OSHybrid is good for application-specific tuning.
    Optimization occurs according to the applied business model. For example, a forecast used in power trading differs greatly from a forecast of extreme values. OSHybrid offers various optimization criteria which can be implemented according to the situation at hand.


Flyer OSHybrid Wind Power Prediction Model

Flyer Wind and Solar Power Predictions