Fast and accurate transient stability assessment (TSA) is necessary to ensure the safe operation of power system. However, the number of power components such as buses and generators in the actual power grid is huge. It brings severe challenges to transient assessment. A data-driven power system transient evaluation model is proposed in this paper. Firstly, the principal component analysis (PCA) is used to reduce the feature dimension of time series data, which improves the running speed of the system. Secondly, a convolutional neural network (CNN) model is established for stability prediction, and the loss function is used as the model training target. Finally, it is shown in the IEEE-39 node system arithmetic that the proposed TSA method could achieve stability assessment accurately and has superiority in response time. The method is expected to advance the application of deep learning in online evaluation of power systems.