Aiming at the drawbacks of traditional dictionary learning methods, such as single sample signal and poor reconstruction effect, a new approach of subdictionary concatenate learning（SDCL）with label information was proposed to identify the power quality disturbances (PQD) signal. Firstly, the different types of testing and training of the PQD signal samples are dimension reduced feature extraction with principal component analysis (PCA), add the label information to train samples, then the different categories of power quality samples are trained into redundant subdictionary and concatenated into structured dictionary. Finally, using dictionary learning algorithm to optimize the structured dictionary and the object class is determined through minimizing the redundant error. The simulation results show that the recognition effect of SDCL method is better than that of SVM and SRC, and has good antinoise robustness, and the multiple PQD identification rate reaches above 91.4% in the noisy circumstance with the signal to noise ratio above 20 dB.