Artificial neural network and machine learning are gradually used in the identification of power system’s self-organized criticality (SOC). Currently, traditional methods on identification of SOC such as OPA model take time or just give opinion on the trending of SOC’s evolution. And most of the artificial neural network methods are based on labeled samples regardless of information provided by unlabeled samples. Identification method of power system’s self-organized critical state based on semi supervised learning-radial basis function neural network is proposed in this article. This method can both use unlabeled samples to improve learning performance and has optimal approximation and global optimization of traditional RBF neural network. Consequently it takes less computation time and get higher rate of correctness, which are concerned most in online identification. This method is verified correct and superior by simulation analysis on gird of Chinese western area. And it can provide theoretical and technical basis for real time prevention of large-scale blackouts.