Abstract: In order to address the limitation of existing data-driven methods for transient assessment in power systems, which heavily rely on extensive datasets and lack interpretability, this paper embeds physical knowledge into traditional data-driven methods and proposes a power system transient stability assessment method based on physics informed convolutional neural network. The proposed method considers large-scale wind power grid-connected power systems and embeds the transient stability physics equations of the power system into the neural network loss function, which allows for the direct approximation of the physics processes through the neural network, ensuring that the prediction results adhere to the physical laws of the power system and enhancing the reliability and interpretability of transient stability assessment. Due to both data-driven and knowledge-driven characteristic, the proposed method reduces the dependence on extensive data sets while maintaining robustness and generalization capability. In addition, the proposed method addresses the challenge of using topological data as direct inputs to neural networks by performing feature extraction and dimensionality reduction through a convolutional neural network. The experimental results on IEEE 9-bus and IEEE 39-bus test systems with wind turbines demonstrate that the proposed method outperforms existing approaches in terms of accuracy, computational efficiency, and generalization ability.