Abstract: Fast and accurate power system disturbance localization is an important prerequisite for power system active control. Wide-area measurement systems provides a data basis for event localization. Besides, the continuum dynamics model provides model support. Combining PMU measurement data and continuum model, a joint model and data-driven based event localization method is proposed in this paper. The disturbance is detected by singular spectrum analysis and cumulative sum of squares index. Furthermore, the partial PMU measurement information is used to form a sub-topology. The two-dimensional spatiotemporal features are constructed via the measurement data and the propagation model in the sub-topology. Finally, according to the correlation between spatiotemporal features, a double compound triplet network based on metric learning is proposed, which realizes fast event localization using less measurement information, and provides a basis for active control strategies. Based on the IEEE39-bus system, the effectiveness and robustness of the proposed method are verified.