k-ShapeStream: Probabilistic Streaming Clustering for Electric Grid Events
Mohini Bariya, Alexandra von Meier, John Paparrizos, Michael J. Franklin
Abstract
We present k-ShapeStream, a clustering method for streaming time-series data. In addition to the algorithmic novelty, the method represents a highly practical approach for electric grid data analytics, requiring no model assumptions or ground truth information, running sustainably on ever growing datasets, and providing intuitive and insightful results to grid operators. We demonstrate the effectiveness of k-ShapeStream using several months of real synchrophasor data from an operational distribution network in California. Through two case studies on (i) transformer tap changes; and (ii) voltage sags, we illustrate how k-ShapeStream assists in identifying and analyzing recurring grid events, a critical task for decision making in electric grids.