Abstract: "Case-based design" (CBD) is a design method developed in the 1980s. Unlike the first generation of rule-based design methods used in the 1960s and 1970s, "CBD" simulates the cognitive behaviors of people based on past empirical knowledge to address new challenges. First, CBD establishes a case database by systematically extracting historical case features. Next, it searches and matches appropriate solutions using structured features when facing fresh design problems. CBD proponents argue that addressing complicated design problems without optimal solutions is more suitable.
　　This paper reviews the development process of "CBD". Importantly, it shows the great potential of combining "CBD" with the latest artificial intelligence (AI) technology and summarizes the problems encountered while integrating the newest methods with the AI technology: 1) On an urban scale, most of the existing methods are "integral generation", which cannot effectively generate the repair and filling of specific types of urban forms; 2) Most of the present methods lack further evaluation and screening of the generation results; 3) It is challenging to mark data and operate with the current methods, which are also unfriendly to designers.
　　DeepCity, an AI-assisted design system, is presented in this study to address these issues. The primary features of this system are: 1) automatically recognizing and analyzing different types of urban morphology based on an image clustering algorithm and making the AI model learn specific kind of morphological modes, thus realizing the texture darning of specific urban morphological types; 2) The AI model is trained to output performance indicators according to urban morphological image features. The model can directly and quickly evaluate the physical performance indicators of the generated results and assist designers in screening, modifying, and deepening schemes; 3) realize the full-process automation from case scanning, data annotation, model training and generation, and design evaluation to data vectoring by combining the Python program and the grasshopper (GH) platform, which designers widely use. This action reduces the user threshold and time cost of designers.
　　The working process of cooperation between DeepCity and designers was elaborated from design cognition, design generation, and design evaluation based on a case study on urban texture darning surrounding the ancient city of Nantou in Shenzhen. The findings of DeepCity in urban morphological clustering evaluation and fast generative morphological thermal environment prediction were evaluated using quantitative analysis. DeepCity can effectively assist designers in conducting typological analysis on urban morphology and generating design prototypes with historical context features. Moreover, it can make faster evaluations of the physical properties of design schemes. Finally, the application potentials and shortcomings of the AI-assisted "CBD" system were discussed and summarized. The results are expected to enlighten field researchers and further promote AI-assisted design practices.

Key words: artificial intelligence, generative design, urban morphology, case-based design, generative adversarial networks

摘要： 梳理了“基于案例设计”的发展历程，针对最新的与人工智能结合的基于案例设计系统，数据标注困难，操作难度大，且大多数城市生成的研究仅关注于整体式生成，对于同一城市不同的肌理并未进行细致区分等问题，提出了一个与Grasshopper平台结合的人工智能辅助设计系统DeepCity.该系统从设计认知、设计生成、设计评估三个方面，为设计师提供基于图像聚类的城市形态类型学分析、基于生成对抗网络的指定城市形态自动织补与室外热环境的快速评估。最后，讨论总结了人工智能辅助基于案例设计的应用潜力与不足。

关键词: 人工智能, 生成式设计, 城市形态, 基于案例设计, 生成对抗网络