Reducing wind energy forecast error with a hybrid ensemble prediction
methodology
Abstract
As a commodity, wind energy is typically traded in ex-ante time frames
and is dependent on forecasts to secure physical positions up to 36
hours after gate closure for trading in Day-Ahead markets. Wind energy
is traded in discrete quantities, however it is generated from an
intermittent and variable resource. Deterministic forecasts are
preferred for energy trading as the most compatible solution to provide
a defined forecast quantity. However, deterministic forecasts cannot
capture the stochastic nature of the underlying power source and are
therefore sub-optimal. Ensemble based forecasts have the potential to
reduce forecast error by accounting for uncertainties not captured in
deterministic models. However, ensemble forecasts are not always
available at the vertical levels at which wind turbines operate.
Therefore, a method is needed to apply ensemble information to turbine
hub heights for energy forecasting purposes. This paper presents a novel
machine learning based method that translates the perturbations from a
localised Numerical Weather Prediction model’s 10m wind speed component
to an ensemble energy forecast at 100m. The extrapolated ensemble based
forecast has improved the forecast accuracy by 9% when compared to the
deterministic output. The findings will have important implications for future energy trading, transmission system operation and meteorological forecasting.