A feasibility study of using a best group fitting method to determine wind data probability distribution.

Probability distributions of wind data in a short time scale, such as hourly, are critically important for the risk assessment in relation to wind energy, optimal route scheduling for commercial airlines and unmanned aerial vehicles. It is known that the probability distributions of wind direction, wind speed, and wind gust have a large variety—ranging from Weibull to Lognormal distributions. A systematic method is needed to determine which distribution best fits a given wind dataset in a very efficient manner. This paper attempts to provide such a method. It is the best group fitting (BGF) method that efficiently fits the hourly wind data to a large number of distributions simultaneously, in contrast to the traditional one-by-one data fitting. The BGF method utilizes a Python package named Fitter which uses Scipy package, tries 80 distributions simultaneously, and checks what is the most probable distribution. We then apply the BGF method for the hourly data of u-wind, v-wind, wind speed, and wind gust in four geologically different locations, including the Alta Wind Energy Center, the Rocky Mountain region, the Colorado Plateau, and the San Diego urban district. The performance of the commonly used distributions and a wider selection of distributions are ranked for these locations at different times of a day. Our datasets and results show that, in addition to the traditional distributions such as Weilbull distribution and Lognormal distribution, the Burr distribution class stands out among the wind distributions and even outperforms the most used distributions in the literature. Our fitting results to the operational wind data have demonstrated that it is feasible to use BGF to make an efficient group fitting, which may imply that BGF can be used to search for the best probability distribution from different hourly weather datasets. [ABSTRACT FROM AUTHOR]