Abstract: 　　This paper introduces a method that combines rule and data-driven approaches for the automatic planning and layout generation of exhibition hall spaces. Firstly, the designated space is divided based on exhibition hall design rules and optimized through beautification and connection methods. Next, the visual saliency of spaces is calculated using spatial relationships, and a priority queue is utilized to select appropriate positions and placement methods for exhibits. Then, by using rule-generated data as a dataset and incorporating semantic prediction along with a topological prediction network, diverse spatial structures are generated. Finally, a predictive model is employed to determine the optimal salient positions of exhibits in descending order of importance, enabling the automated generation of space layouts. Experimental results demonstrate the method is able to adapt to different exhibition halls and quantities of paintings, generating exhibition hall space layouts with excellent visual effects.