Operation of power systems - Wind Power Integration: Connection and System Operational Aspects - page 182

farm output, can improve on persistence methods for similar time periods. How-
ever, for time periods beyond 3-4 hours, approaches employing forecast data from
national meteorological office numerical weather prediction (NWP) models offer
significant improvements in forecasting ability. The NWP output often feeds a
separate model that generates site-specific wind speed and power estimates.
Two approaches are commonly adopted here: physical equations and relationships
are utilised to estimate wind farm power output; or statistical models, sometimes
backed up with online measurements for short-term predictions, are applied. Sub-
sequent up-scaling of these results provides a prediction of the production for
an entire region. State of the art approaches are able to achieve an average pre-
diction error of 8-10 per cent, relative to installed capacity, over a 12 hour period,
with some degradation in performance for longer time horizons. In comparison,
average load forecasting errors are of the order of 1-2 per cent, and are much less
dependent on the time horizon. For system operation it is the combined wind and
load forecasting error that is important, since this will determine the additional
regulating requirements placed on the power system as a whole.
Since the wind forecasting errors should be largely independent of the demand
forecasting errors, the total error should be less than the sum of the individual
errors. Figure 5.28 illustrates the 36 hour ahead wind speed forecast for a particular
site in Ireland in hourly steps for three distinct days, allowing comparison between
the actual and predicted values. In general, the wind speed predictions are seen to
be reasonable, with little fall-off in performance for longer time horizons.
Figure 5.28b illustrates a fairly unusual day: when looking 10 hours ahead the
actual wind resource is significantly greater than predicted. For the Ireland case,
this has probably arisen due to the inherent difficulties associated with predicting
low pressure systems. So, for example, wind behaviour at the forecast site can be
profoundly affected by minor deviations in the expected path of the low pressure
system, deepening or shallowing of the weather system and/or unpredicted changes
in its speed. Figure 5.12 earlier illustrated the effects of a low pressure system
crossing Ireland, so errors associated with the direction and timing of its passage
can be directly translated into large errors in forecast wind power production. Such
difficulties are likely to be common. In Denmark, for example, assuming an
installed base of 2,400 MW of wind generation, a deviation of 1 m/s in average
wind speed has been estimated to result in a 320 MW variation in wind power
production (Bach, 2005).
In most cases, wind-forecasting errors arise from timing significant weather
fronts incorrectly: a 4-6 hour phasing error can be seen in Figure 5.28a. Since the
passing of such fronts can be associated with changes in wind speed, large power
errors can occur. Over time these phasing errors tend to cancel out, so that wind
energy forecasting can be very good, compared with wind power forecasting. As
discussed in Section 5.4, this should encourage greater use of energy storage and
demand-side management. Consequently, a system operator may be concerned
about a large forecast variation in wind power later in the day. Uncertainty about
the timing of such an event will affect commitment/de-commitment instructions
to conventional plant. Anticipation of extreme storm conditions will also cause