Estimating Long-Term Annual Energy Production of a Large Offshore Wind Farm from Short Large-Eddy Simulations: Methods and Validation with a 10-year LES Run

Atmospheric large-eddy simulation (LES), a computational fluid-dynamics technique that resolves turbulence in the atmospheric boundary layer, is increasingly used for wind resource assessment (WRA), by including wind turbine parametrizations and using external weather data as initial- and boundary conditions. The large computational costs of doing such a 'real-weather' LES, however, limits length of the simulation to < 1 year; whereas long-term, multi-year, mean power production values are of high interest to many parties in the wind energy sector. To address this need, this work presents several methods to estimate long-term mean power production/annual energy production and wind from a < 1 year LES run, by applying Bayes' theorem on short-term LES output and long-term ERA5 reanalysis data.
A 10 year LES run of a hypothetical large offshore wind farm is performed in order to validate these 'long-term correction' methods, in three scenarios of increasing complexity. First, long-term correction of 365 consecutive days gives estimates of long-term mean power with a mean absolute error of 0.35 %, and 95th percentile of the absolute error within 0.8 % of the long-term mean, reducing the uncertainty by an order or magnitude. Second, in the scenario when the simulation period is not fixed, using several simple day selection techniques to select the simulation period can reduce the error further. Then, only around 200 days are needed to arrive at the same error values. The results indicate that long-term correction is insensitive to the particulars of the day selection methods, but that including a diverse set of days from different years and seasons is essential. Third, a method to also include wind observations in the long-term correction is presented and tested. This requires an additional 'free stream' LES run without active turbines, and gives estimates of long-term power and wind that are corrected for a potential LES bias. Although validation of this final approach is difficult in the employed modelling strategy, it gives valuable insights, and fits within the common WRA practice of combining models and observations.
The presented techniques are based on physical arguments, computationally cheap, and simple to implement; and as such could be a useful extension to the diverse set of modelling, observational, and statistical techniques used in WRA.