It is of great significance to accurately identify the type of disturbance signal to analyze and control the power quality. In this paper, a new method of power quality disturbance identification based on matching pursuit optimized by particle swarm optimization(PSO-MP) and RBF neural network is proposed. Firstly, in order to let the residue signal to better reflect the different disturbance signal difference, the method using the fundamental atomic library to extract the fundamental; Then, the MP algorithm is optimized by PSO to reduce the calculation amount, which combined with discrete Gabor atom libraries can accurately extract atomic parameters of residual disturbance signal by sparse decomposition; Finally, using RBF neural network identify disturbance signals by features, which is the mean and standard deviation of the atomic parameter and projection of residual signal on the atom. Simulation examples show that the proposed method can effectively identify several common power quality disturbances with a small amount of computation and good anti noise performance.