A Robust Spectral Clustering Method Based on PMU Measurements for Coherent Areas Identification

The paper deals with the separation of power system into coherent areas; this is a relevant issue for managing the network in both normal operating conditions and during anomalous events. In particular, the attention is focused on partitioning the power system in such a way as to group together frequency signals, measured by means of phasor measurement units (PMU), exhibiting similar oscillatory behavior after the occurrence of a fault or disturbance. Unfortunately, the increasingly massive presence of renewable energy sources is undermining the clustering methods defined so far, requiring new solutions to the problem. To overcome the considered drawbacks, the authors propose hereinafter to ( ${i}$ ) improve the grouping capabilities of an iterative spectral clustering method thanks to the definition of new parameters for similarity estimation (Modified Bray Curtis index) and cluster thresholding (weighted Fiedler value) as well as (ii) enhance its robustness with respect to both measurement noise and uncertainty affecting the PMUs by means of a deep test procedure. To this aim, particular attention is paid in the design and assessment stage to the definition of both filtering algorithm and measurement parameters (e.g., the length of the analysis window). Once defined these parameters, the method is capable of 100% correctly separating transmission network sections oscillating with similar trends in a number of tests conducted on simulated and actual signals, so highlighting the promising performance of the method highlighting its reliability and efficacy in different test conditions.