Abstract: The complex interaction of nonlinear loads, power electronic equipment and network factors have led to a diversification and mutual coupling of power quality issues. The various operating modes of power users further increase the uncertainty of power quality, presenting it with a probabilistic distribution. In the light of the limitations of traditional statistical methods in processing and analysing large-scale power quality data, an improved Possible fuzzy C-means（PFCM） clustering algorithm is proposed, and by introducing the covariance matrix and entropy weight method for preprocessing of original data, the accuracy and robustness of clustering analysis are effectively improved. Firstly, build a simulation model for the improved IEEE 33-node system and generate power quality characteristic data in a scenario-based manner. Then the distribution characteristic information of power quality in the power grid is excavated through three different algorithms. Finally, by comparing the identification results of the three algorithms, the effectiveness and superiority of the improved PFCM algorithm in the identification of power quality disturbances are verified.