Power system coherency assessment by the affinity propagation algorithm and distance correlation

Abstract

This paper assesses the coherency in power systems employing the affinity propagation (AP) algorithm with different distance metrics and quality measurements. This assessment allows determining the appropriate metric to cluster a frequency dataset that possesses coherent patterns. Thanks to the AP method does not require initialization for the number of clusters, its convergence characteristics guaranteed by the optimization process, and its capacity for using different distance metrics as input, the AP is adopted to identify and distinguish such coherent patterns that embody the collective motion of an operative area in a power system. The AP method is a data-driven method that uses an affinity matrix as input, i.e., the square matrix computed with pairwise distances. Since the distance function significantly impacts the quality of the resulting clustering, this contribution evaluates three different distance metrics, distance correlation, and the results are compared using four cluster indexes. The data collection is constituted by a set of frequency signals and the representative objects are the nodes identified as the center of each operative area. This contribution presents experimental results using simulated signals with added noise and real event signals captured by 94 PMUs. We found that our proposed strategy achieves highly competitive results for identifying coherent generator and non-generator buses in large-scale power systems.