Wind power forecasting - Wind Power Integration: Connection and System Operational Aspects - page 228

in all error measures over all look-ahead times if the results for the 15 wind farms
are compared to the single wind farm for the new reference persistence model, as
shown in Figure 6.7. The improvement is significant at the shorter look-ahead
times, dropping off as look-ahead time increases.
The ease with which more advanced wind power forecasting models can
improve on basic persistence at longer look-ahead times can give a false sense of
their performance quality. This new reference model is a more challenging
benchmark against which to measure the performance of more advanced wind
power forecasting models.
Another tool which is widely used for exploratory analysis is the plot of the
error distribution. The error distribution for the basic persistence forecast model at
look-ahead times of 1 and 6 hours, for the same six months of test data, is shown in
Figure 6.8 for the single wind farm and for the 15 wind farms. The bins are 0.05 pu
of installed power. In each graph the sharpness and skewness of the distribution can
be observed, as can the presence and extent of large forecast errors which are, of
course, very significant to power system operators. It is clear that the error dis-
tribution is sharper for short look-ahead times and for aggregation of wind farms.
The improvement obtained by using an advanced forecast model over a
reference forecast model can be quantified using the expression

EC ref ð k Þ EC ð k Þ
EC ref ð k Þ

Imp ref ; EC ð k Þ¼

where EC is the evaluation criterion which can be either MAE, RMSE or SDE.
Using the test data again for the two cases of (i) a single wind farm and (ii) 15
geographically dispersed wind farms, the improvement in the RMSE evaluation
criterion of the new reference model over the basic persistence model is plotted in

0.6

0.5

0.4

0.3

0.2

MAE
RMSE
SDE

0.1

0

0 6 12 18 24 30 36 42 48

Look-ahead time (h)

Figure 6.7 Improvement in new reference persistence error measures due to

aggregation of wind farm output