Streaming Analytics for Power Systems: A Tale of Two Time Scales
summary
Uses GPS-synced PMU streams (60 Hz) and 4 TB of real grid data to run PCA on streaming PMU windows, finding the top 3–4 components capture >99.5% of variance and flag anomalies within a few samples (orders of magnitude faster than single-channel methods). For localization, applies robust PCA (nuclear norm + l1) solved by augmented Lagrangian; in 68-bus resonance experiments 40 of 44 cases pinpoint the source or narrowly constrain the search space.
AI-written summary · about