Optimal clustering for efficient computations of contingency effects in large regional power systems

The goal of this paper is to determine optimal clustering in large power networks for efficient contingency screening. A decentralized algorithm for “DC” contingency screening based on Diakoptics is revisited first. It has been shown that this algorithm is much more computationally efficient compared to the existing Distribution Factor Matrix methods for a pre-specified clustering. This paper will address how to establish the best clustering and quantify how much more efficient that clustering is compared to a pre-specified one. The optimality is defined in terms of computational complexity and necessary communication among the clusters. The optimal clustering requires the minimum balanced computational effort across the clusters with the minimum amount of information exchange. Optimal clustering will be illustrated on a sparsely connected RTS-96 bus system and a densely connected NPCC 36-bus system.