Power system controlled islanding is an effective protection measure to avoid power system blackouts and system breakdown. It is necessary to determine the islanding scheme promptly and accurately. Based on generator coherency constraint

a searching method for the optimal islanding scheme is proposed. First

an improved Laplacian Eigenmap algorithm is used to identify the coherent generators according to the real-time dynamic post-fault trajectories obtained by wide area measurement system (WAMS). The network simplified graph with reduced searching space is then obtained by aggregating coherency generators using coherent generator virtual nodes. The extended Laplace partition strategy is applied to obtain the preliminary islanding surfaces. At last

by considering power flow directions

a variable neighborhood heuristic searching algorithm is used to search for the optimal islanding surfaces with the minimal net power imbalance within the islands. Simulations on the New England 39-bus system validate the effectiveness and flexibility of the approach proposed. The differences of coherent behavior by different fault modes are fully considered. This ensures the controlled islanding strategy with high selectivity and robustness.