There are problems of low detection accuracy and high complexity of traditional power big data anomaly detection methods. Thus a power big data anomaly detection method is proposed, one which combines the possibility fuzzy C-means algorithm with the improved Particle Swarm Optimization (PSO) algorithm. The improved PSO algorithm and the redefined clustering effective function are used to optimize the initial centers and number of the possibilistic fuzzy C-means algorithm. Through the simulation, the proposed algorithm is compared with the algorithm before improvement. The superiority of the proposed algorithm is verified. The experimental results show that the algorithm can accurately realize power big data outlier detection, and the error detection rate is reduced from 0.36% to 0.05%. This work is supported by the China Southern Power Grid Guangdong Provincial Science and Technology Project (No. 0002200000041529).