Power Grid Faults Classification via Low-Rank Tensor Modeling

With the increased extent of synchrophasor-based wide area monitoring systems in modern large-scale power systems, more data-limited classification of fault events in smart grids becomes possible. Recent works have identified and ex-ploited the low-rank property of synchrophasor data and power system dynamics for fault type classification. We develop a new methodology for modeling particular types of power grid faults from a tensor perspective, taking advantage of this well-known low-rank property, and classifying observed faults into categories by comparing to these low-rank representations. Using simulated data from the IEEE 39-bus system, we demonstrate that this approach is robust to limited temporal observation and sparse embedding of measurement units on the grid.