Short term wind speed forecasting with evolved neural networks

Concerns about climate change, energy security and the
volatility of the price of fossil fuels has led to an increased
demand for renewable energy. With wind turbines being one
of the most mature renewable energy technologies available,
the global use of wind power has been growing at over 20%
annually, with further adoption to be expected. As a result
of the inherent variability of the wind in combination with
the increased uptake, demand for accurate wind forecasting,
over a wide range of time scales has also increased.
We report early work as part of the EU FP7 project
’ORIGIN’, which will exploit wind speed forecasting, and
implement and evaluate smart-meter based energy management
in 300 households in three ecovillages across Europe.
The ORIGIN system will capitalise on automated weatherstation
data (available cheaply) to inform predictions of the
wind-turbine generated power that may be available in short
term future time windows. Accurate and reliable wind-speed
forecasting is essential in this enterprise.
A range of different methods for wind forecasting have
been developed, ranging from relatively simple time series
analysis to the use of a combination of global weather forecasting,
computational fluid dynamics and machine learning
methods. Here we focus on the application of neural
networks, without (for the time being) the use of numerical
weather predictions or expensive physical modelling methods.
While work of this nature has been performed before,
using past wind speeds to make predictions into the future,
here we explore the use of additional recent meteorological data to improve on short-term forecasting. Specifically,
we employ evolved networks and explore many configurations
to assess the merits of using additional features such
as cloud cover, temperature and pressure, to predict future
wind speed.