Bayesian infinite mixture models for wind speed distribution estimation

Wind energy, as a clean, environment-friendly, and inexhaustible renewable energy, has attracted significant attention, and wind speed distribution plays an important role in its development. Currently, various finite mixture distributions have been proposed to fit the empirical wind speed distribution. However, they have two disadvantages. First, different distribution components result in different fitting performances, and it is difficult to determine the optimal type of single distributions to construct a mixture model. Second, the number of distribution components has a significant effect on the final performance of the mixture model. It is well known that a mixture of Gaussian distributions (MoG) can fit any complex continuous distribution. To overcome the abovementioned disadvantages, based on stick-breaking construction, this study extends the original MoG to two different types of infinite mixtures of Gaussian distributions (IMoG) with different prior parameter distributions. These distributions are called IMoG-I and IMoG-II. The proposed models not only have superior fitting performance but also determine the optimal number of Gaussian components automatically. The results show that both IMoG-I and IMoG-II perform better than single, heterogeneous, and homogeneous mixture distributions. In most cases, the coefficient of determination-based criterion can reach its maximum of 1. In addition, the optimal number of Gaussian components varies with different datasets. When modeling different wind speed distributions in this study, they are all less than 10. The proposed wind speed distribution estimation models are good choices for unknown areas with complex wind regimes.