Environmental Engineering Reference
In-Depth Information
Persistence
Prediktor
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
0
6
12
18
24
30
36
42
48
0
6
12
18
24
30
36
42
48
Look-ahead time (h)
Look-ahead time (h)
Figure 6.16
Error measure versus look-ahead time for individual wind farms for
persistence and for Prediktor
country at the time. There is clearly a wide geographic spread over the country. The
wind farms were divided into regional groupings labelled 'nw' (north-west), 'w'
(west) and 'sw' (south-west). The wind farm capacity in each of these groupings is
also indicated in Figure 6.15.
The only measured power data available were the metered 15-minute time-
resolution energy data from each of the wind farms. This data set did not contain
operational data on the availability of wind farms or indeed the availability of
individual wind turbines, nor did it contain wind speed or wind direction data from
the wind farms. As a consequence, Prediktor could not take the operational status of
the wind farms or the wind turbines into account and operated on the basis of an
assumed 100 per cent availability.
The results presented below are for the period October through to December
2001, with a training period for the MOS corrections covering February through to
September 2001. In Figure 6.16 the SDEs for individual wind farms are plotted
against look-ahead times for persistence and Prediktor. Prediktor performed better
than persistence for all look-ahead times except those less than 6 hours. As can be
seen for Prediktor, the SDEs lie in the range 18-33 per cent, whereas the corre-
sponding SDEs for persistence can be up to 52 per cent.
The beneficial effect on forecast error of both persistence and Prediktor models
due to the aggregation of wind farm power output from geographically dispersed
wind farms is shown in Figure 6.17. SDEs for persistence and Prediktor are plotted
against look-ahead time for a single wind farm (the line) in the north-west region,
for the 'nw' group of wind farms (the white column) and for all the wind farms (the
black column).
The effect of aggregation of wind power output is explored further in
Figure 6.18, which shows six distributions of errors for the Prediktor model, three
on the left for the 6-hour look-ahead time and three on the right for the 12-hour
look-ahead time. The top graphs show the distribution for a single wind farm in the
north-west region, the middle graphs show the distribution for all wind farms in the
north-west region and the bottom graphs show the distributions for all wind farms.
Search WWH ::




Custom Search