Controlled islanding of power networks based on graph reduction and spectral clustering

Intentional controlled islanding aims to split the power system into self-sustainable islands after a severe disturbance, but prior the uncontrolled network separation. Given its nature (i.e. last resort for blackout prevention), this emergency control technique must be adopted as quickly as possible. This paper proposes a computationally efficient method based on graph reduction and spectral clustering. The paper contributes by describing important details of the graph reduction process in the context of controlled islanding and by the formalisation of this process. Furthermore, it demonstrates how to adopt embedded graphs to enhance the Multiway Spectral Clustering graph partitioning. Finally, it is shown how to explicitly incorporate important cannot-link constrains between coherent generator groups into the islanding problem. The proposed method is detailed using the IEEE 39-bus test case. To evaluate the algorithm performance, the method is applied to realistically-sized PEGASE test networks