The spatial layout of underground protective buildings faces strong coupling among protection requirements, functional relationships, and geological conditions, while conventional experience-driven design struggles to maintain stability under complex constraints. This study presents an intelligent layout method based on generative design to support automatic generation and iterative optimization of spatial configurations. Multidimensional constraints are encoded through a unified tensor representation, enabling integrated reasoning across protection, adjacency, and spatial feasibility. A generative framework combining variational autoencoding, reinforcement learning–based spatial adjustment, and evolutionary optimization is constructed to form a closed loop between layout generation and performance evaluation. Protective pressure fields, spatial relationships, and circulation organization are embedded directly into the evolutionary process, allowing layouts to converge toward coherent and stable configurations. The proposed method establishes a systematic approach for intelligent spatial organization of underground protective buildings under high-dimensional constraints.
Huang et al. (Thu,) studied this question.