Wind Energy

Non-linear methods for forecasting load and energy supply in wind energy systems

Significant short-term changes in wind speed are a particularly severe threat to the mechanical stability of wind energy plants and to reliable energy supply in local energy networks. The aim of this project, run by the "Max-Planck-Gesellschaft zur Förderung der Wissenschaften e.V. (MPG)" is to develop methods that deliver short-term forecasts of wind speed and produced electrical power. 

 

Summary

The project saw non-linear phase space methods used to control wind power systems for the first time. The short-term forecasts of wind speed and produced electrical power are used to cut the mechanical load on rotor blades (particularly for squalls), optimise the use of wind/diesel systems and reduce short-term fluctuations in the power supplied. This enables the service life of wind power plants to be extended, and the energy yield and the quality of the supplied power to be increased.

The partners initially studied a scalar, linear model for predicting the ground wind speed using different time series for wind speed and wind turbine power output. The forecasts were then compared and improved using non-linear methods developed. As additional spatial information, such as wind speed at other locations, can have a significant influence on forecasting capability, the linear and non-linear methods were expanded to multivariate time series, i.e. multiple data sets recorded simultaneously. The final step was a comparative study of the forecasting capability of these phase space methods.

Results:

  • The non-linear phase space method developed is significantly superior to conventional linear models, particularly when forecasting wind squalls in which the wind speed can rise sharply within a few seconds.
  • The mean forecasting error for the wind turbine power forecast is up to 10% less when using non-linear phase space models than with linear models.
  • Multivariate methods deliver significantly improved forecasts compared to scalar methods. For multivariate forecasting, non-linear methods can also be superior to corresponding linear methods on average.

More Project Informations

Project number:  0329848

Project period:  1999 - 2002

Project region:  Germany (Saxony)

Project contact:

Herr Prof. Dr. Kantz

View Publication

Source: German National Library of Science and Technology Hannover (TIB)