Challenges in Modeling Wind Power Generation Based on Available Weather Data

This paper presents the results on an initial effort to model a power output of small wind turbine (WT) in the situation when neither WT power curve (PC) nor actual winds are known. For small isolated locations, like Isle of Eigg, Scotland, considered in this study data, gathering might not be a trivial issue, especially weather data. In this case, mesoscale numerical weather modeling becomes the only source of weather data. These models require a heavy computational process and are only capable of utilizing about 50km grid. Obviously, modeled data may not be accurate and definitely does not account for a local terrain. Nevertheless, it is these data that were used in this paper in attempt to model an output of a 6kW WT using a parametric PC fitting and shallow neural network fitting. The paper discussed all the challenges, shows advantages and disadvantages of each of the two approaches and ends with recommendations on how to improve model accuracy to account for a local terrain based on the difference between the actual and modeled power output.