Operation of power systems - Wind Power Integration: Connection and System Operational Aspects - page 185

online data improves forecasting, particularly for short-term horizons (2-3 hours),
this should be balanced against the additional costs involved. In Ireland the Mor-
eCARE evaluation programme obtained measurements from 11 geographically dis-
persed wind farms, which were then used to create a countrywide, 48 hour forecast in
1 hour steps (Barry and Smith, 2005). Regular HIRLAM (high resolution limited
area model) data are provided by the national meteorological office, but this intro-
duces a 4.5 hour (3.5 hours in winter) time lag. In Spain, systems such as Prediktor,
'Casandra' and 'Sipreolico' have been applied (Giebel et al. , 2003), although from
January 2006 there has also been a requirement for all wind farms to provide a
24 hour forecast of production, 30 hours in advance of the operational day. At pre-
sent, the vast majority of existing wind farms are located onshore. With expansion
offshore, wind forecasting will also be required to include these new installations. It
is generally assumed that offshore forecasting should be easier than onshore,
although wind/wave interactions may complicate matters, and the benefits of error
aggregation will be largely lost. The pan-European ANEMOS project, for example,
has an aim of providing a short-term forecast of wind power production up to 2 days
ahead, including the impact of offshore developments (Kariniotakis et al. , 2006a).
Wind forecasting clearly brings benefits to a power system, in lessening the
impact of high wind penetration, and hence the associated cost, while increasing
the penetration limit at which wind generation can be safely tolerated. This needs
to be balanced against the financial outlay for the forecasting system itself, the
associated staff training and software maintenance costs, and the requirement to
collect and store production data from operational wind farms. The question then
arises as to who should pay for this wind forecasting service, and over what time
horizon predictions should be made. The answer depends on the particular elec-
tricity market arrangement. In California, for example, it was agreed that wind
developers will create a (2 hour) wind forecasting system but allow the independent
system operator (ISO) to operate it. If the wind farm owners pay a forecasting fee,
and schedule according to the forecast, then all generation imbalance penalties are
cancelled (Asmus, 2003). In Germany it is the sole responsibility of the TSOs
(and DSOs) to balance wind power, and so they provide their own forecasting tools.
The day-ahead market closes at 3 pm on the preceding day, requiring a prediction
horizon of 33 hours. Eastern Denmark participates in the Nordpool Elspot
day-ahead market, which closes at noon on the preceding day, requiring a predic-
tion horizon of 36 hours (Holttinen, 2005c). The England and Wales NETA spot
price market had a gate closure time of 1 hour. Deviations from the agreed power
schedule result in penalties being imposed through the mechanism of the balancing
market (Johnson and Tleis, 2005). It was, therefore, in the wind farm operators'
interest to invest in a forecasting tool. In April 2005 NETA was extended to include
Scotland, forming BETTA (British electricity trading and transmission arrange-
ments). Nordpool also operates a 1 hour market, Elbas, which can be used by
parties from Sweden, Finland and Denmark.
It follows from the above that the requirements for secondary (and tertiary)
control depend greatly on the market closure time. An increasing delay between the
wind forecast (and demand forecast) and the actual time of production will