A Clustering-Based Approach for Estimation of Low Frequency Oscillations in Power Systems

Abstract

This paper presents a clustering technique to rank the dominant modes of low frequency oscillations in power systems under model order uncertainty of the signal. The rotational invariance property of the signal subspace is utilized to obtain the modes of oscillations. Rotational invariance of the signal under multiple shifts is considered to obtain the preliminary estimates of the modes. Subsequently, the modes are clustered to obtain the dominant modes that belong to the signal subspace. A signal subspace participation index (SSPI) is used to rank the modes according to their dominance in signal subspace. Thereafter, the convex combination of the preliminary modes, estimated from the multiple shifts of the signal subspace, is used to obtain the final mode estimates. The simulation results are presented for signal generated using the modes obtained from a two-area system. Additionally, the methodology is also tested on field data obtained from the PMU installed at a generating station in India.