【作者】 杨柳；

【导师】 王扬； 周祥；

【作者基本信息】 华南理工大学 ， 建筑学（专业学位）， 2019， 硕士

【摘要】 论文目标为探索人工智能深度学习(Deep Learning)在青年公寓户型设计中的跨界结合应用,及使用后评价(Post occupancy Evaluation,POE)和交互式进化算法(Interactive Evolutionary Computation,IEC)介入深度学习框架的技术路径,意在掌握青年公寓设计经验,减少重复劳动,提高设计输出效率,构建自动生成青年公寓户型平面图的算法框架。论文首先采用文献资料研究的方法,宏观把控整体逻辑框架,对国内外青年公寓和深度学习应用进行文献综述,搜集互联网、文献书籍实例,学习设计策略和方法并整理高质量数据库和图库;接下来自下而上地对青年公寓租户生活现状、体验及需求进行考察,建立可落地的长租公寓研究框架,进行青年长租者需求洞察、分类并生成用户画像,绘制长租用户体验旅程图,分析长租公寓现状痛点及机会点,并为深度学习自动生成内容质量的主观评价体系做好基础的数据调研工作;然后结合前面的桌面研究和实地调研展开青年公寓设计策略与方法的解析,提出总体设计原则,结合上一章内容明确空间使用者的分类方式和依据,对不同户型分类配置分别展开设计方法的论述,为利用深度学习自动生成户内平面图做好理论和技术资料的准备;最后了解和盘点常用生成式对抗网络,架构程序的整体逻辑,选择合适框架进行实验并记录过程,对生成成果进行主客观评价等,基于现有技术和未来发展趋势,合理展望并布局未来发展方向。实验结果证明:1.可以利用人工智能深度学习高效地自动生成青年公寓户型平面图;2.基于使用后评价和用户体验角度展开大量调研和访谈,可以做到真正回应青年群体使用痛点与需求;3.随着软硬件技术的提升,该自动生成程序和评价模型会互相博弈升级,愈发贴近真实工作中的需求,提升工作效率。更多还原

【Abstract】 The primary goal of this paper is to explore the cross-border application of Deep Learning in youth apartment design and the technology route of POE and IEC involved in algorithmic framework.The aim is to master the design experience of youth apartment,reduce duplication of basic work,improve the efficiency of design output,and construct an algorithm framework for automatic generation of youth apartment plane.Firstly,the paper uses the method of literature research and macro-control of the overall logical framework to review the literature of youth apartments and Deep Learning applications at home and abroad,collect examples of Internet and literature books,learn design strategies and methods,and collate high-quality databases;then,investigating the current living conditions experiences and needs of youth apartment from bottom to top.The research framework is to provide insight into the needs of young renters,classify and generate Persona,Customer Journey Map,analyze the pain points and opportunities of youth apartment,and do basic data research for the subjective evaluation system of content quality automatically generated by Deep Learning;and then launch the Youth Gong based on the previous desktop research and field research.Combining with the analysis of design strategies and methods,the general design principles are put forward.Combining with the content of the previous chapter,the classification methods and basis of space users are clarified,and the design methods for different types of households are discussed separately,so as to prepare theoretical and technical data for the automatic generation of intra-household plan by Deep Learning.Finally,analyze and compare common Generative Confrontation Networks,choosing the appropriate framework to carry out experiments and record the process,evaluating the results objectively and subjectively,and based on the existing technology and future development trend,reasonably looking forward to and laying out the future development direction.The experimental results show that: 1.Deep Learning can be used to automatically generate the flat plan of youth apartments;2.Based on post-use evaluation and user experience,a large number of investigations and interviews can truly respond to the pain points and needs of youth groups;3.With the improvement of software and hardware technology,the Generator and Discriminator will gradatim upgrade and become more close to real needs,improve work efficiency.更多还原