An adaptive clustering algorithm with high performance computing application to power system transient stability simulation

This paper presents an adaptive clustering algorithm based on power system network topology, initial power flow and given architecture. The sizes of the small cliques are derived using multi-constraint and multi-objective graph partitioning theory. The vertices of the graph represent units of computation, and the edges encode data dependencies. Tests for a 39-bus network, a 1056-bus network, a 3872-bus network and a 10188-bus network are reported, and the results show that the cluster-based partitioning produces smaller hyper-edge cut size and higher speedup than the traditional direct partitioning. An application to 3872-bus network shows that by the improved tree-based partitioning the speedup is improved by up to 51% compared to the traditional tree-based approach. Moreover, the results show that the partitioning results of improved tree-based partitioning algorithm and improved graph-based one are comparable. These results suggest that adaptive architecture- aware clustering algorithm can be combined with heterogeneous and changing computing resources.