The invention discloses a power quality disturbance signal classification method based on DBN-ELM, and belongs to the technical field of signal classification. According to the method, a deep belief network DBN is used to carry out feature extraction on a disturbance signal. When the DBN is built, the learning rate of each RBM in the DBN is optimized by using an improved quantum particle swarm algorithm, and then the extracted features are classified by using an extreme learning machine ELM to obtain the category of the power quality disturbance signal. The characteristics of deep learning arefully utilized, the classification accuracy and noise immunity are improved, and single disturbance and composite disturbance can be accurately classified.