Intelligence Assessment Methods for Field-Level Plans Generation of Large and Medium-sized Stadiums Based on Deep Learning

Due to the complexity of functional flow organization, stadium design involves a high workload when using traditional methods. Recently, generative AI has shown significant potential in architectural design. A generative-evaluative collaborative generation method for Large and Medium-sized Stadium field-level plan is proposed, using Stable Diffusion for generation and adapted-Convolutional Neural Networks for assessment. The latter is introduced as an assessment model in Stepwise Generation. By assessing the generation results’ functional topology and area ratio, the model sorts out results with higher similarity to outstanding case studies, indicating a greater potential for further design, then feeds them into the subsequent generation step. The combination of two deep learning models establishes an effective human-computer interaction mechanism between ‘generative’ and ‘critical’ AI, enhancing the scientificity of design decisions, and offering a technical path for intelligent design of complex public buildings.