Preventive controlled islanding against multiple probabilistic cyber threats in power grids

While digitalization and integration of communication networks play a vital role in modern power system operation, they also expose the power grid to potential cyber attacks that can cause significant societal disruptions. Considerable effort is devoted to the early detection of cyber intrusions. However, the detection of an adversary does not guarantee that attacks can be neutralized before physical disturbances occur. This work proposes a preventive controlled islanding method to enhance the power grid resilience against multiple anticipated attacks with known probabilities. The proposed approach is formulated as a graph-constrained coalition formation problem, focusing on grouping electrical nodes rather than directly selecting lines to disconnect, and determining islanding solutions using exact and adapted Monte Carlo tree search (MCTS) algorithms. Case studies demonstrate that both the locations and probabilities of attacks influence the optimal islanding strategy. Furthermore, the proposed formulation allows the optimal number of islands to emerge from the solution rather than being a preset parameter. The results also show that MCTS can identify (near-)optimal solutions with substantially less computation time than the exact algorithm.