Abstract: Fast and accurate transient stability analysis of the power system is of great significance to the safe and stable operation of the power system. With the increasing complexity and diversity of equipment components in the modern power system, the nonlinear characteristics of the system are increasingly strengthened. As a traditional transient stability analysis method of the power system, the time-domain simulation is too time-consuming. To solve this problem, a transient stability analysis method based on the spatio-temporal graph convolutional network（STGCN） is proposed, which combines short-term simulation and neural network prediction to reduce the time required for transient stability analysis and can be applied in a variety of simulation analysis scenarios. By modeling the transient stability analysis as a sample space mapping problem with the proposed method, a data-driven method is adopted to train the neural network model and establish a mapping from the spatial structure and temporal power flow data of power grid in the transient process to the transient stability. The model simultaneously extracts the characteristics of the spatial structure and temporal power flow of power grid before, during, and after the failure in the transient process to realize the fast and accurate judgment of the transient stability of the system. Compared with the traditional transient stability analysis method,the proposed method only needs to complete a short-time simulation analysis, which improves the analysis efficiency. Compared with other machine learning models, the STGCN model simultaneously discovers the spatial and temporal characteristics of the transient process of the power system, introduces more prior knowledge about stability, and has better characteristic extracting ability and analysis performance. Test results based on the New England 39-bus system verify the feasibility, effectiveness and superiority of the proposed method.