Controlled Islanding of Power Systems Based on Structural Deep Clustering Network

Controlled islanding is widely acknowledged as an effective countermeasure to prevent power systems from widespread blackouts against severe disturbances such as cascading outages. However, it is challenging to identify the proper cut-set of transmission lines for network splitting adaptively during real-time operation. To address this problem, this paper proposes a structural deep clustering network (SDCN)-based approach for controlled islanding in power transmission networks to prevent transient instability and widespread blackout. With buses to be nodes and branches to be edges, the graph representation of power network is proposed as the input features for the SDCN model. To ensure that all the islands can operate securely after network splitting, the cross-entropy loss that is related with the classification of the generator buses is introduced into the loss function of the SDCN model so that the generators in each island will be coherent. The procedure of the proposed SDCN-based approach for controlled islanding is then discussed in detail. Case studies on the IEEE 9-bus and 118-bus systems are presented to demonstrate the effectiveness and the scalability of the proposed SDCN-based approach.