In a power system, the characteristic selection criteria for power quality disturbance classification are not uniform. Generalization ability is weak, and the classification effect and efficiency need to be improved. In order to solve these problems, first a multi-layer extreme learning machine auto-encoder is used to optimize the input weights and extract the characteristics of electric power quality disturbances. Secondly a multi-label classification algorithm based on ranking is used to consider the correlation between labels and to classify various power quality disturbances. Combining the two algorithms, a multi-label classification model based on multi-level extreme learning machine is designed, and the optimal network structure of multi-level extreme learning machine and the optimal classification threshold of multi-label classification are obtained. The experimental results show that the proposed method can be applied to the classification of single and compound disturbances of power quality and improve the classification effect and efficiency with higher classification accuracy and excellent robustness and anti-noise ability. This work is supported by National Natural Science Foundation of China (No. 51275084) and 2020 Science and Technology Research Program of Henan Province (No. 202102210134).