Abstract: In order to gradually realize the new power system, a large number of power electronic components have been put into use in the power grid. As a result, the power quality problem becomes more and more serious, which is mainly manifested in the compound of power quality disturbances (PQDs) and the decrease in the applicability of traditional identification algorithms. To solve this problem, the visual trajectory circle technology was proposed, which could transform the one-dimensional disturbance signals into the two-dimensional trajectory circle images with obvious shape characteristics. These images were input to the depth residual network (ResNet) for autonomous feature learning and classification. Firstly, Hilbert transform (HT) was applied to the complex power quality disturbances to get the envelope sequence based on sampling time. Then, taking the amplitude as the polar diameter and the instantaneous phase corresponding to the polar angle, the trajectory circle images were obtained in polar coordinates. At last, they were input to the optimal model of the ResNet18 to learn how to classify PQDs. In order to verify the effectiveness of the proposed algorithm, the classifications of PQDs were conducted through in the simulation and experiment in this paper. The results show that the proposed method cannot only better extract the complex PQDs feature, but also effectively overcome the shortcomings of traditional methods, such as difficulties in poor anti-noise performance and etc. The purpose of high precision PQDs classification can be achieved.