Study on an optimal controlled islanding scheme based on active semi-supervised spectral clustering algorithm

Controlled islanding is a timely process which can split the power system into electrical islands in place based on the real-time operating condition and fault information. The key to avoid the collapse of the system is how to locate the out-of-step splitting sections rapidly and accurately. However, with the larger system scale, the computation complexity of optimal controlled islanding is increasing geometric exponentially which has to be faced with NP-hard problem. This paper proposes an active semi-supervised clustering algorithm (ASSC) to solve the problem. Firstly, the model for optimal controlled islanding of power systems is constructed with the minimal composite power-flow disruption as the objective function and generator coherency got from the active learning strategy as the supervised constraint which is suited to be clustered. Secondly, the solving process can be transformed into the relaxation solution of constraint spectral clustering aimed at partition of the static graph. Finally, the optimal controlled islanding schedule can be selected based on the improved K-means algorithm. The computation efficiency can be improved without losing the power grid possible solution. Moreover, the accuracy, effectiveness and rapidity are verified in the simulations of IEEE 118 standard system and Sichuan Province and Chongqing City Power Grid.