Deep assessment of wind speed distribution models: A case study of four sites in Algeria

Abstract

The aim of this study is to assess the accuracy of different probability functions for modeling wind speed distribution at four locations, distributed over Algeria, to minimize the uncertainty in wind resource estimates. Despite mixture models perform better results, their complexity induced us to use in this work eight distributions with a maximum of three parameters, namely Weibull, Gamma, Inverse Gaussian, Log Normal, Gumbel, GEV, Nakagami and Generalized Logistic distribution to model the wind speed, fitted with four parameter estimation methods. In addition to the methods of moments and the maximum likelihood which are commonly used, the power density method and the L-moments method are developed and utilized for the first time in wind resource assessment field, to estimate the parameters of most distributions used in this work. Moreover, two goodness-of-fit tests based on the coefficient of determination and the root mean square error, are conducted in order to select good fitting probability distribution functions. According to the coefficient of determination and the root mean square error, the GEV and Gamma are the most appropriate, compared to the others used distributions. Furthermore, the L-moments method is the most effective one, among the used parameter estimators, followed by the maximum likelihood method. On the other hand, in term of power density error, different results were found, where the Power Density Method gave the best results with the Gamma, Inverse Gaussian and Log Normal distributions. Otherwise, owing to the difference in the wind characteristics for each studied site, it can be stated that to minimize the uncertainty in wind resource estimates, it is important to determine the method that gives the best parameters for each distribution.