Discovering Clusters in Power Networks from Orthogonal Structure of Spectral Embedding

This paper presents an integrated approach to partition similarity graphs, the task that arises in various contexts in power system studies. The approach is based on orthogonal transformation of row-normalized eigenvectors obtained from spectral clustering to closely fit the axes of the canonical coordinate system. We select the number of clusters as the number of eigenvectors that allows the best alignment with the canonical coordinate axes, which is a more informative approach than the popular spectral eigengap heuristic. We show a link between the two relevant methods from the literature and on their basis construct a robust and time-efficient algorithm for eigenvector alignment. Furthermore, a graph partitioning algorithm based on the use of aligned eigenvector columns is proposed, and its efficiency is evaluated by comparison with three other methods. Lastly, the proposed integrated approach is applied to the adaptive reconfiguration of secondary voltage control (SVC) helping to achieve demonstrable improvements in control performance.