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

Slow Coherency Identification and Power System Dynamic Model Reduction by Using Orthogonal Structure of Electromechanical Eigenvectors

“…As discussed in Section V-B, it is known that the NPCC 48-machine test system can be well decomposed into 9 areas based on its 9 slowest electromechanical modes. Our previous case study in Section V-B comes to the same conclusion, but it additionally identifies alternative area structures consisting of 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,33,34,35 , 36 2 32, 37, 38, 39, 40, 41, 42 3 43, 44, 45, 46, 47, 48 3 and 6 areas (see Figure 2). The present case study illustrates the validity of decomposing the NPCC system into 6 areas by using our grouping algorithm in Section IV.…”

“…Due to NP-completeness of (3), greedy machine assignments described in Section IV-B cannot be guaranteed to converge close to the global optimum of (3). Because of these complications, the solutions obtained by the methods from Sections IV-A and IV-B may noticeably benefit from graph cut refinement [28]. The following algorithm similar to the one from Section IV-B has been successfully applied to improve the Ncut M values: 1) Evaluate the generic expansions (4) of the input machine groups C 1 , .…”

Generator grouping cutset determination based on tree construction and constrained spectral clustering

“…The higher number of misplaced generators with methods [11, 14] is partly caused by the excessive number of connected components that can often occur as the result of partitioning. While the minor connected components could be reconnected using various heuristics [2, 12], such an end result would be a product of the initial partitioning algorithm and a proper connectivity heuristic that should also respect the bus grouping constraints.…”

“…in [10] (in general, this problem is known as k‐way cut in the literature). The partitioning resulting from spectral clustering can often be noticeably improved by using various post‐processing algorithms [11], with the two relevant refinement algorithms suitable for both active power balance improvement and increase in islands’ electrical cohesion detailed in [12].…”

Spectral MST-Based Graph Outlier Detection with Application to Clustering of Power Networks

“…This problem is usually tackled by applying a robust post-processing method on top of spectral clustering [3] or by removing the outliers in advance [6]. The first option limits the set of methods to produce the final partitioning, and severe outliers may still be able to impair the partitioning result [7]. Thus it is strongly recommended to filter out the severe outliers before applying spectral clustering [6].…”