Depth-Based Clustering for Multivariate Time Series with Applications in Wind Energy

• Activity Number: 192 - Contributed Poster Presentations: Section on Statistics and the Environment
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Laura L. Tupper*
• Companies: Williams College
• Keywords: spatio-temporal data; multivariate time series; clustering; depth statistics; similarity measures

Abstract

We explore the unsupervised classification of spatio-temporal data, with an application in wind power generation. Clustering can be used as a data-driven way to characterize the behavior of a dataset, revealing patterns and allowing the selection of representatives to characterize the space, but it relies on the selection of a useful measure of similarity between observations. Previous work has used the band distance, a depth-based distance metric for high-dimensional vector data, to cluster time series of wind speeds. We now extend the depth-based approach to take into account wind speeds at multiple locations, and compare the results of simple concatenation, dimension reduction of the multivariate data using spatial techniques, and a similarity measure based on the full matrix of times and locations.