Assessment of Parametric and KDE Statistical Models for Wind Turbine Energy Yield

Precise estimation of available wind speed at a certain location is of paramount importance when planning and designing power distribution networks. Parametric Families of Probability Distributions (PFPDs) have been popular choices in the literature. These, however, apply certain assumptions that are too restrictive and may often end up in wrong model identification. To mitigate these limitations, Nonparametric Family of Probability Distributions (NPFPDs), such as the Kernel Density Estimation (KDE), have been considered in recent power systems studies. In this article, three popular PFPDs (Rayleigh, Weibull and Gaussian), and two recently suggested PFPDs for power systems studies (Birnbaum-Saunders and Nakagami-m) are assessed and compared with the KDE approach using Ruleof-Thumb (ROT) for bandwidth estimation. The resulting probabilistic wind speeds are then used to estimate the productivity of the wind turbine by means of energy yield. Results show that PFPDs produced high error values; leading to inaccurate decisions when conducting power systems planning studies. In contrast, the KDE-ROT estimator achieved the lowest estimated absolute error. As such, nonparametric approaches present better candidates for accurately estimating wind turbine energy.