Environmental Engineering Reference
In-Depth Information
Figure 2. Error decomposition example generated with 1-year of data from the western part
of Denmark.
implementation (e.g. [33], [34], [35]), some effort has been made in recent years by adopt-
ing traditional wind power prediction tools to ensemble data from the MSEPS system in
research projects and studies (e.g. [36], [37], [38]).
6.
Aspects of the Forecasting Error
There is still a prevailing opinion in the wind energy community that the wind power
prediction error is primarily generated by wrong weather forecasts (e.g. [11], [39]). From
a meteorological perspective, this is a statement that may cause misunderstanding, because
part of the error is due a complex mixture of weather related errors. The weather forecast
process itself can only be blamed for the linear error growth with forecast length. We have
therefore conducted an error decomposition in order to quantify the different error sources
with a large ensemble of MSEPS weather forecasts. Traditionally, increased spatial weather
prediction model resolution has been said to provide better forecasts (e.g. [40], [41]), but
the shortest model waves may anti-correlate with the truth and cause double punishment in
the verification (e.g. [32],[42]) and thereby additional model error. An ensemble prediction
approach is another way to improve forecasts with fewer anti-correlation hours and the
possibility to predict and understand forecast errors.
6.1.
Wind Power Error Decomposition
In order to understand the forecasting error in wind power, we have carried out a de-
composition of the influencing components that have let to misunderstanding in the past.
The analysis to do this includes forecasts generated in 6 hour frequency. This means that
Search WWH ::




Custom Search