Wind farms are a divisive issue both in this country and
abroad. Their supporters will tell you that they are a clean and increasingly
efficient way of harnessing a potentially limitless source of energy. Sceptics,
the new environment secretary Owen Patterson among them, will argue that they
are unsightly, noisy and ineffective. In arguments between these two sides the
science of how we work out the efficiency of wind farms is often ignored. The
simple fact is that wind energy can never be incorporated large scale into a
county’s power supply unless there is a reliable method of forecasting the
wind. Without this there is no way of working out how much power a set of wind
turbines will generate, making it virtually impossible to plan for their use.
In this paper the fully range of wind speed prediction
methods are reviewed. It is highlighted
that there are both Physical modelling methods, which use the physical
characteristics of a given site to predict future wind speed and Statistical
models that run almost entirely on previously observed data. The paper also
goes on to review a number of new Artificial Intelligence based models.
Most physical models are mathematical NWPs or Numerical
Weather Prediction models. They use a wide range of input data including the
orographic characteristics of a site, the ‘roughness’ of terrain, average
pressure and temperature and potential obstacles in the area. This data is used
to help predict wind speeds at a particular site. More advanced physical models
also include subsidiary programmes that can model the effect of obstacles in
more detail (WAsP programs) or even take into account the effect of turbine
shadowing (PARK programs). Obviously then physical models require a large
amount of data of be gathered before they are run and often the data they
generate needs to be analyses further. The paper recommends that for short term
forecasting of wind speed accurate evaluation models are also needed in order
to give reliable results.
There are a much wider range of statistical models, the
majority of which are based upon the input of historical wind speed data and
the identification of patterns and trends by computer programmes. These range
from more simple Autroregressive models (AR) to more complex Autoregresssive
moving average models (ARMA), each with their own limitations and advantages.
For example the paper cites a study that has shown ARMA models to have 95%
accuracy on both long and short scales of prediction but only when using
2-yearts of previous wind speed data. There are also spatial correlation models
which aim to increase the accuracy of predictions by using data from nearby sites
as well as the location of wind turbines. These models have been shown to be
very effective on flat terrain but almost useless when trying to predict wind
speed over complex topography.
The paper also mentions artificial intelligence based models
that are a much more recent development in wind speed forecasting. A number of examples
of such models are presented in the paper but there is little consensus on
their effectiveness.
The overall trends seem to be that NWP based physical models
perform well over large spatial scales and long time periods. Statistical
models are often more effective over very short-term temporal scales and
certain AI models only appear accurate when there is a very large amount of
historical data to compute. There is however no ‘silver bullet’ model that can
accurately predict wind speed regardless of location or time scale. As the
paper rightly points out there is much more study in this area needed if wind
power is ever to become a viable renewable energy source and it is most likely
that a synthesis of a number of different models will prove the most effective
in forecasting wind speed in the future.
(The full paper, ‘Ma Lei, Luan Shiyan, Jiang Chuanwen, Liu Hongling, Zhang Yan,
A review on the forecasting of wind speed and generated power,
Renewable and Sustainable Energy Reviews,
Volume 13, Issue 4, May 2009, Pages 915-920’ is available here)
