Controlled splitting is a kind of emergency control measures with real-time dynamic information in a given power system appropriately employed. Implementation of timely and appropriate controlled splitting for a given power system under large disturbance can block the propagation of faults

and system collapse possibly caused by malfunctioning of protection devices can be avoided to some extent

and hence rapid restoration of the whole power system can be attained. Given this background

a controlled splitting method is proposed by quickly searching splitting sections and optimally adjusting the operation strategies of isolated islands. Firstly

an improved version is presented for the spectral clustering algorithm developed in the field of machine learning

and a constrained spectral clustering algorithm proposed to account for the coherent constraints of generators. The problem of searching splitting sections is then transformed into the one of finding generalized eigenvalues. Furthermore

in order to avoid the drawbacks of the traditional k-medoids algorithm that the clustering result is sensitive to the specified initial cluster centers and the search procedure is inefficient

an improved k-medoids algorithm is proposed and employed with the constrained spectral clustering algorithm to search the optimal splitting sections. In each attained island

if any security constraint is violated

adjustments of generation outputs as well as load demands can be carried out to ensure the security. Finally

the IEEE 118-bus test system and the actual power system are employed to demonstrate the feasibility and effectiveness of the proposed method.