Intelligent design of shear wall layout based on conditional flow matching

Data-driven generative methods have begun to reshape preliminary design of reinforced concrete (RC) shear wall structures. While existing models such as GANs, GNNs, and diffusion models have advanced intelligent layout generation, their applications remain constrained by adversarial training instability in GANs, complex graph construction in GNNs, and inefficient iterative sampling in diffusion models. Furthermore, existing image-based generative architectures lack specialized mechanisms for explicitly modeling the elongated geometric characteristics of shear walls and incorporating deterministic engineering priors, such as elevator shaft layouts, during generation. To address these limitations, this paper proposes an efficient shear wall layout generation method based on Conditional Flow Matching (CFM). By fitting a probability flow vector field and solving the ordinary differential equation (ODE), CFM circumvents the time-consuming denoising process, enabling rapid generation. Specifically, a strip attention mechanism is introduced at the bottleneck of a U-Net architecture to enhance the perception of long-axial geometric morphologies. Meanwhile, the design basic seismic acceleration acts as a global condition parameter to regulate the shear wall ratio. Crucially, a mixed source distribution is formulated to explicitly inject engineering priors at the start of the ODE integration process. Experiments demonstrate that this approach achieves rapid, high-fidelity generation and guarantees the structural enclosure of critical areas, providing a novel pathway for the intelligent and controllable design of shear wall layouts.