Wind power forecasting - Wind Power Integration: Connection and System Operational Aspects - page 226

look-ahead times. As n goes to infinity this model tends to the global average P ð t Þ ,
which is the average of all available wind power measurements at time t . This
global average can also be used as a reference model.
A new reference model has been proposed by Nielsen et al. (1998) which
combines the advantage of the basic persistence and the moving average forecast
models. It is given by

P NR ð t þ k j t Þ¼a k P ð t Þþð 1 a k Þ P ð t Þ
where subscript NR stands for new reference and a k is the correlation coefficient
between P ( t ) and P ( t þ k ). This model requires the analysi s of a training set of
measured data to calculate the required statistical quantities P ð t Þ and a k .
An example of a time series of measured power data in per unit of the installed
capacity for a single wind farm and for 15 geographically dispersed wind farms
over a period of one year is shown in Figure 6.5. Each time series is divided into
two six-monthly sets - a training set and a test set. It is clear that the time series of
the 15 wind farms is smoother than the single wind farm and also that the rated
output is never reached in the case of the 15 wind farms (see Section 5.3.2).

Single wind farm

Training set

Test set

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

(a) 1-Jan 29-Jan 26-Feb 26-Mar 23-Apr 21-May 18-Jun 16-Jul 13-Aug 10-Sep 8-Oct 5-Nov 3-Dec 31-Dec

15 geographically dispersed wind farms

Training set

Test set

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

(b) 1-Jan 29-Jan 26-Feb 26-Mar 23-Apr 21-May 18-Jun 16-Jul 13-Aug 10-Sep 8-Oct 5-Nov 3-Dec 31-Dec

Figure 6.5 Time series of measured power at (a) a single wind farm and at

(b) 15 geographically dispersed wind farms divided into training and
test sets