A post-processing methodology for robust spectral embedded clustering of power networks

Abstract

Partitioning of electric networks into zones or areas is a procedure that has numerous applications in power system planning, operation and control. Spectral clustering based approaches are among the most favoured ones to solve the partitioning problem. Applications of spectral clustering include definition of control zones, analysis of connectivity structure of power networks, intentional controlled islanding, design of sectionalising strategies, and visualisation. Although spectral clustering is a state-of-the-art family of methods with numerous extensions, some practical issues can arise when applying it to large-scale power networks. While spectral clustering becomes significantly more robust to outliers when combined with a robust post-processing method like k-medoids, the connectedness of the resulting partitioning cannot be guaranteed. This paper proposes a greedy algorithm to solve the connectedness issues inherent to many robust post-processing methods. Furthermore, it is proposed to utilise a label propagation based heuristic to improve the quality of the final partitions. The test results evaluate the steps of the methodology on a large-scale 1354-bus PEGASE test network.