Topology change will change the fault signal characteristics. The traditional fault traveling wave localization method for distribution networks is based on the fixed topology design. Through the single feature information of the time or frequency domain, the fault is located, and the localization accuracy is low under the topology change conditions. For this reason, a fault localization method based on a graph attention network is proposed. First, the distribution characteristics of fault traveling wave in time and frequency domains are quantitatively analyzed, and it is found that it is difficult to effectively distinguish different fault locations with a single time or frequency domain information, so the panoramic information representation of fault traveling wave based on wavelet transform is proposed. Then, the measurement points and overhead lines are taken as the nodes and edges of the graph. The panoramic information of traveling waves is used as the node features to construct the graph data, so as to establish the fault localization method based on the graph attention network and locate the fault of the distribution network by mining the correlation between the node features, the information of the network topology, and the fault location, thus enhancing the adaptability of the method to topological changes. Simulation results show that the method has a high localization accuracy of 98.8%. It is not affected by transition resistance, noise, and other factors and has a strong adaptive ability to topology changes.