Abstract: Transmission grid one-line diagrams are widely used to visualize the information of power systems. However, it is often necessary to design proprietary algorithms to generate one-line diagrams of different styles to meet various presentation demands. The main reason iwas that existing automatic generation methods cannot could not efficiently adapt to the change in the application scenario. This paper proposes proposed an automated approach to generating transmission grid one-line diagrams. It can could adapt to a broader range of layout requirements and allows allowed rapid regeneration of layouts after changes in demand. To begin with, a mathematical model of the layout of transmission grid one-line diagrams is was established in this paper. The factors that affect legibility and the quantitative metrics for evaluating the layout quality are were then analyzed. After that, a deep learning model based on graph autoencoders is was built to learn the patterns of transmission grid one-line diagrams. The learned patterns are were used to generate an optimized layout that could meets the given requirements of the evaluation metrics. Finally, the proposed method is was implemented on an actual provincial-level transmission grid. The experimental results show that the generated layout can meet the expectation of the selected scenarios, and is superior to previous algorithms in terms of generation time and layout quality. Thus, the capability of the method in quickly adapting to different application demands is verified.