Clustering of Power System Oscillatory Modes Using DBSCAN Technique

Measurement-based power system modal analysis is useful to operators in that alarms can be issued when the damping of some natural oscillatory modes becomes unacceptably low or when forced oscillations suddenly appear in a power system. Considering the fact that mitigating the mechanism of each mode or forced oscillation requires a specific remedial action, the mode estimates should be properly clustered and identified before such corrective actions can be decided. In this paper, the problem of clustering of estimates from an ambient oscillation monitoring algorithm into groupings representative of different system modes is addressed for the first time. The well-known DBSCAN method is applied for clustering the estimates. Archived results from three days of oscillation monitoring implementation in a real system, which were obtained from Fast Frequency Domain Decomposition (FFDD) modal analysis method in the Réseau de Transport d'Électricité (RTE) power system, are utilized for evaluation of the methodology.