Wind power forecasting - Wind Power Integration: Connection and System Operational Aspects - page 244

totalling 500 MW (Lang et al. , 2006b), with promising results. In the case of the
single wind farm the normalised MAE was 11.4 per cent and the SDE 15.7 per cent,
while for the total wind power connected to the TSO the normalised MAE was
6.2 per cent and the SDE was 8.3 per cent. All statistics refer to 24-48 hour forecasts
for each day's 00:00 UTC model run.

6.6 Conclusions

While considerable progress has been made in wind power forecasting in the last
decade or so, in terms of better understanding of the processes involved and higher
accuracy of the forecasts, there is still plenty of scope for improvement. Clearly
forecasts for the wind power production of whole regions or supply areas are more
accurate than forecasts for particular wind farms due to the smoothing effect of
geographic dispersion of wind power capacity. However, with the trend towards
large wind farms, and particularly large offshore wind farms, there is a growing
requirement for accurate forecasts for individual wind farms. There are consider-
able differences in the accuracy that can be achieved for wind power forecasts for
wind farms located in very complex terrain compared to open flat terrain. Current
research on meso-scale modelling aims to address this problem. The development
of offshore wind power also presents difficult challenges for wind power fore-
casting. The uncertainty attached to a forecast is a key concern, particularly when
looking to the participation of wind power in electricity markets, as discussed
further in Chapter 7 (Wind Power and Electricity Markets). The area of ensemble
forecasting shows potential for further progress. The other area of interest will
be the integration of wind power forecasting tools into the energy management
systems of TSOs.