Identification of Dominant Modes in Power System Using Unsupervised Learning Approaches

The modernized power grid is being pushed to become more interconnected as the demand for electric power rises. It causes a loss of inertia in the power system, resulting in more severe disruptions. Quickly identifying the low frequency oscillatory modes and associated characteristics will allow the power system operator to respond to a specific occurrence without wasting time. This paper provides a comparative review of two unsupervised learning techniques approaches for an electromechanical mode shape estimation. This paper aims to generate an alarm in the control room, which prompts the controls to act when damping is weaker in the system. The proposed method is studied using PMU data extracted from a Kundur two-area system at various disturbance conditions. As a Performance comparison, accuracy and computational time are verified for both techniques. DBSCAN clustering method shows superior accuracy and viability than other clustering methods.