Unsupervised Classification of Wind Speed Directions Based on Functional Discriminative Latent Mixture Model

Classification of wind speed pattern is useful in several fields, such as studying the long-term effects on the environment and estimating wind energy potential. Nowadays, analyzing high-dimensional functional data is an essential topic of modern statistics and learning systems which are widely used in different applications. While there have been numerous studies on modeling and clustering wind speed data using classical methods, functional methods are limited. The classical multivariate classification methods of wind speed data naturally depend on a finite number of characteristics that do not include all the available information in the data domain. As a result of these limitations, in this work, we focus on classifying wind speed curves using a functional mixture model-based clustering. This approach has powerful and effective classification tools. In the present research, the functional discriminative mixture (FDM) model has been adapted and applied to cluster wind speed curves in the Taiz region over the last four decades. The parameters of the FDM model have been estimated using the functional Fisher EM-based algorithm. The proposed method has been successfully performed and demonstrated meaningful results for wind speed profiles and their temporal variations. The current findings show that the proposed functional method is the best-suited model to cluster functional wind speed data at different aspects. Compared to traditional multivariate approaches, functional results provide with additional insights for understanding the dynamic behavior of wind speed curves.