Estimating wind speed probability distribution by diffusion-based kernel density method

Abstract

Accurate estimation of wind speed probability distributions is a challenging task in wind power planning and operation. Different from the commonly used parametric methods which consist of selecting a suitable parametric model and estimating the parameters, this paper presents an improved non-parametric method to estimate wind speed probability distributions. Based on the diffusion partial differential equation in finite domain, this method accounts for both bandwidth selection and boundary correction of kernel density estimation. Preprocessing techniques are designed to handle data with different recording manners to produce smooth probability density functions. Probability densities of specific grid points are obtained by inverse discrete cosine transformation and are further used to calculate assessment indices of wind resources. The method has been tested to estimate probability densities of parametric distributions and actual wind speed data measured in different places. Simulation results show that the proposed approach is of practical value in fitting wind speed distribution models.