Abstract: As the new power system is being built with new energy as its main body,the number of power electronic devices connected to the grid is increasing day by day. The resultant power quality disturbances have become increasingly complex and severe,with multiple types of power quality disturbances occurring simultaneously, leading to a decrease in the recognition accuracy of traditional algorithms. To address this issue, this paper proposes a composite power quality disturbance identification algorithm based on Markov transition field and EfficientNet. Firstly,the Markov transition field is used to visualize and map power quality disturbance signals into two-dimensional feature images;secondly,the image data is processed by EfficientNet Convolutional neural network to realize the feature extraction of disturbance signal;Finally,neural architecture search is used to automatically adjust the super parameters of Convolutional neural network for network training,and a classification and recognition model of power quality disturbances is established. The simulation results show that the proposed method can accurately and efficiently extract disturbance signal features, and has good classification performance for composite power quality disturbances and strong noise resistance.