An Observability-based Approach to Extract Spatiotemporal Patterns from Power System Koopman Mode Analysis

Abstract

Koopman mode analysis has shown considerable promise for the analysis and characterization of global behavior of power system transient processes recorded using wide-area sensors. In this paper, a framework for feature extraction and mode decomposition of spatiotemporal data based on the Koopman operator is presented. A physical interpretation of the Koopman modes as columns of a matrix of observability measures is derived, and criteria for selecting a reduced set of Koopman modes are proposed. This approach allows to selectively isolate and quantify the dominant physical mechanisms underlying the recorded power system data, and it can be used for wide-area monitoring and assessment. Simulation results on both simulated and measured data show that the method can accurately identify the dominant spatial patterns or shapes and temporal patterns associated with specific system behavior.