Dimension Reduction Stacking for Deep Solar Wind Clustering

In-situ observations of solar wind plasma exhibit statistical differences according to their coronal origins. These in-situ conditions are a direct result of various processes such as ionization and acceleration occur in the inner corona. Machine learning methods have been successful in characterizing solar wind in-situ observations using unsupervised deep clustering and dimensionality reduction techniques, but it remains unclear as to how solar wind data embedding and downstream clustering could be improved while providing better interpretability in machine learning process. In this study, we explore the impact of distance metrics on solar wind in-situ data clustering. We evaluate the metric performance by applying it to dimension-reduction-stacking and deep clustering techniques and comparing it with state-of-the-art methods using solar wind in-situ measurements. Our work demonstrates the potential for customized distance metrics to improve the interpretability and performance of deep clustering approaches applied in solar wind in-situ observations.