When a serious cascade of system failures leads to instability, the rapid search of the cohomology cluster is the prerequisite for quench control to suppress the oscillation. In view of the non-stationary and nonlinear characteristics of the power angle signal after the generator is seriously disturbed and the need to classify the clusters based on the experience, this paper proposes a new method of unsupervised identification for clusters based on t-distributed Stochastic Neighbor Embedding (t-SNE) method. The generator power angle signal is used as source data in the wide-area measurement environment, which is modeled and mapped into a two-dimensional subspace through the t-SNE method. The similarity of generator operating characteristics after disturbance is measured by the degree of aggregation between mapping points in two-dimensional coordinates. Subsequently, an unsupervised clustering of generator sets is performed using affinity propagation algorithms. The research shows that the proposed method is simple in principle and easy to solve practical problems. Based on the measured data for calculation and analysis, the influence of model parameters on the clustering can be avoided. Compared with the traditional method of clustering, the effectiveness and speed of the proposed method can be verified through the system simulation of 73 generators in Hunan Province network in 2014. This work is supported by National Natural Science Foundation of China (No. 51377016).