【作者】 李承垣； 闫思怡； 丁佳； 梁凝瑞； 曹雨； 任雨箫； 许蓁；

【Author】 Li Chengyuan;Yan Siyi;Ding Jia;Liang Ningru;Cao Yu;Ren Yuxiao;Xu Zhen;School of Architecture,Tianjin University;

【机构】 天津大学建筑学院；

【摘要】 生成式人工智能的快速发展,为建筑设计带来新的可能性。然而,现有模型存在训练数据分类不科学、设计逻辑理解不足及筛选优化流程不完善等问题。本文提出基于生成式AI与Grasshopper联动的建筑设计流程,高效生成大学校园总平面图。结果显示,Res Net50在功能分布图特征提取中表现优越,Lo RA模型在校园布局生成中适应性强,类别预测准确率为74.75%。Grasshopper平台的遗传算法能显著优化方案合理性。更多还原

【Abstract】 The rapid development of generative artificial intelligence has opened new possibilities for architectural design.However,existing models often suffer from issues such as unscientific training data classification,insufficient understanding of design logic,and incomplete filtering and optimization processes.This study proposes an integrated architectural design workflow based on generative AI and Grasshopper for efficient generation of university campus master plans.The results demonstrate that Res Net50 performs well in feature extraction from functional distribution maps,while the Lo RA model shows strong adaptability in campus layout generation,achieving a classification accuracy of 74.75%.The genetic algorithm implemented on the Grasshopper platform significantly enhances the rationality of the generated schemes.更多还原