Synchrophasor-Based Event Detection, Classification, and Localization Using Koopman, Transient Energy Matrix, Best Worth Method, and Dynamic Graph

The integration of high-resolution data from phasor measurement units (PMUs) in the power grid operation provides an opportunity for enhanced situational awareness and possible decision support to the system operator. Nevertheless, the effective extraction of valuable information from the PMU data stream requires a data analytics tool. Existing techniques rely on feature selection, labeled data and typically do not consider the nonlinear, dynamic, and stochastic nature of the power system. In this paper, a novel online algorithm is developed for event detection, localization, and classification using synchrophasor data considering system dynamics. A new multi-stage technique includes: i) Prony Analysis is used to detect dynamic events/ disturbance window, ii) temporal and spatial signature of event is extracted by fitting a linear model to the PMU measurements for identified event/ disturbance based on Koopman mode decomposition, iii) the transient energy matrix (TEM) of the event is constructed with the eigenanalysis of the fitted linear modes, iv) physics-based classification is done using TEM without need for any labeled data, v) Best Worth Method (BWM) is utilized to incorporate expert opinion for event classification, as necessary, and vi) the dynamic graph-theoretic technique is developed and used for localization based on TEM, and the physical event is confirmed based on spatial correlation of measurements. The developed algorithm is validated using multiple events such as faults and variations/switching of load/ generator/capacitor/distributed energy resources for the IEEE 14 and 39 Bus test systems as well as real PMU industrial data.