Abstract: Targeting the problem of power quality disturbance feature redundancy and low classification accuracy in complex grid environment,this paper uses a multilayer convolutional neural network to obtain the low-dimensional to high-dimensional feature information of power quality disturbance signals layer by layer,introduces a feature attention mechanism to construct multi-feature fusion layer,eliminating the feature redundancy and improving the attention to the key features of the disturbance signals. In addition,the local and global features of the disturbance signals are extracted,and the generalization ability of the model is enhanced to improve the classification accuracy of the disturbance. Based on this,the power quality disturbance recognition method based on multi-feature fusion attention network is proposed. The simulation results show that the proposed method can not only effectively identify the power quality disturbances under single disturbance and compound disturbance,but also effectively overcome the influence of noise interference on the model. The performance of this method in feature identification is remarkably superior to that of the mainstream disturbance classification methods,and the model has stronger anti-noise performance.