Forced Oscillation Source Location via Multivariate Time Series Classification

TL;DR

This paper presents a machine learning approach using multivariate time series analysis to accurately locate the source of forced oscillations in power systems, enhancing grid stability and security.

Contribution

It introduces a novel method combining Mahalanobis distance and dynamic time warping for source localization using synchrophasor data, with high accuracy demonstrated in simulations.

Findings

High accuracy in source localization in simulated power systems

Effective use of Mahalanobis distance and dynamic time warping for MTS comparison

Method applicable for online disturbance source identification

Abstract

Precisely locating low-frequency oscillation sources is the prerequisite of suppressing sustained oscillation, which is an essential guarantee for the secure and stable operation of power grids. Using synchrophasor measurements, a machine learning method is proposed to locate the source of forced oscillation in power systems. Rotor angle and active power of each power plant are utilized to construct multivariate time series (MTS). Applying Mahalanobis distance metric and dynamic time warping, the distance between MTS with different phases or lengths can be appropriately measured. The obtained distance metric, representing characteristics during the transient phase of forced oscillation under different disturbance sources, is used for offline classifier training and online matching to locate the disturbance source. Simulation results using the four-machine two-area system and IEEE 39-bus system indicate that the proposed location method can identify the power system forced oscillation source online with high accuracy.