Improvement of System Identification Using N4sid and Dbscan Clustering For Monitoring of Electromechanical Oscillations

Poorly damped electromechanical modes may lead to multiple stability issues or islanding of the grid. Therefore, it is important for power grid operators to be able to observe and monitor critical modes. In this article, an algorithm based on system identification with N4SID and clustering of eigenvalues with DBSCAN was developed for online monitoring. N4SID traditionally needs the model order as input which was avoided with the usage of clustering on intervals of model orders. This also benefits the estimates, as true modes are clustered and noises are filtered. The algorithm was tested on simulated and real PMU data, and the results were promising. The algorithm quickly finds new modes in the system after a topology change with relatively good accuracy. It also finds modes in the real PMU data coherent with previous studies. Initial results favour rotor speed and rotor angle as input signals to the algorithm.