基于检索增强生成（RAG）技术的上海云宇星空 大模型知识库建设探索
Exploration of Knowledge Base Construction for Shanghai Cloud Cosmos Large Model Based on Retrieval-Augmented Generation (RAG) Technology
徐 玮
上海市数字城市规划研究中心 副主任，高级工程师，硕士 上海市量子城市空间智能创新重点实验室（筹） 副主任，大模型方向（语料建设）带头人 85367093@qq.com
摘要: 为提升超大城市空间治理精细度，上海启动量子城市工作并积极探索研发规划资源行业基座大模型——云宇星空。以 云宇星空大模型为研究对象，从检索增强生成（RAG）技术应用的全流程总结规划资源行业模型的创新要点。面对超 大城市空间治理的多元需求，云宇星空大模型建立五步法知识工程建设流程；通过专题领域知识图谱建设和材料类型 分类、规范标签体系等开展高质量数据集建设；语料生产和模型检索增强生成环节，以知识图谱叠加多路召回的方式， 提升模型召回率和精准率；并依托语料动态管理平台实现语料从收集到生产入库及动态扩容的闭环管理。后续拟在模 型空间智能能力深度拓展、知识图谱与RAG融合、多智能体协同的RAG架构等方面进一步提升完善。
Abstract: To improve the precision of spatial governance in megacities, Shanghai has launched the Quantum City initiative and is actively researching and developing a foundational industry model for planning and resources—the Cloud Cosmos. Taking the Cloud Cosmos model as the research object, this paper summarizes the innovative features of the planning and resources industry model from the perspective of the full-process application of RAG technology. In response to the diverse needs of megacity spatial governance, the Cloud Cosmos model has established a five-step knowledge engineering process. It has developed high-quality datasets through the construction of domain-specific knowledge graphs, material type classification, and standardized tagging systems. In the stages of corpus production and model RAG, a combination of knowledge graphs and multi-path recall is employed to improve the model’s recall rate and precision. Additionally, a dynamic corpus management system enables closed-loop management of the corpus from collection and production to database entry and dynamic expansion. Future work is planned to further enhance and refine the model in areas such as deep expansion of spatial intelligence capabilities, deep integration of knowledge graphs and RAG, and multi-agent collaborative RAG architectures.
关键词：检索增强生成；云宇星空大模型；知识库建设；空间智能；规划资源行业
Keyword: Retrieval-Augmented Generation (RAG); the Cloud Cosmos Large Model; knowledge base construction; spatial intelligence; planning and resources industry
中图分类号：TU984
文献标识码: A
刘雪颖，云静，李博，等. 基于大型语言模型的检
索增强生成综述[J]. 计算机工程与应用，2025，
61（13）：1-25.
Liu Xueying, Yun Jing, Li Bo, et al. Survey on
retrieval-augmented generation based on large
language models[J]. Computer Engineering and
Applications, 2025, 61(13): 1-25.
任海玉，刘建平，王健，等. 基于大语言模型的智
能问答系统研究综述[J]. 计算机工程与应用，
2025，61（7）：1-24.
Ren Haiyu, Liu Jianping, Wang Jian, et al. Review
of intelligent question answering systems based on
large language models[J]. Computer Engineering
and Applications, 2025, 61(7): 1-24.
上海市规划和自然资源局. “云宇星空大模
型”——上海量子城市空间智能建设之规划资
源行业大模型[EB/OL] （2025-12-21）[2026-06-
07]. https://mp.weixin.qq.com/s/ZRpzwKrcMhTEEBayiN11Q.
Shanghai Municipal Bureau of Planning and
Natural Resources. “The Cloud Cosmos Large
Model”: domain foundational large model for
the planning and resources industry in Shanghai
Quantum City spatial intelligence construction[EB/
OL]. (2025-12-21) [2026-06-07]. https://mp.weixin.
qq.com/s/ZRpzwKrcMhT-EEBayiN11Q.
上海市人民政府. 上海市国民经济和社会发展
第十五个五年规划纲要[Z]. 2026.
Shanghai Municipal People’s Government. The
15th Five-Year Plan for national economic and
social development of Shanghai[Z]. 2026.
上海规划资源. 量子无界、智治新篇，一文读
懂上海量子城市空间智能建设行动[EB/OL].
（2025-12-21）[2026-06-07]. https://mp.weixin.
qq.com/s/08g9LBvUoUpWki_Lo9saFQ.
Shanghai Guihuaziyuan WeChat Account. Quantum
boundless, new chapter of smart governance: a
guide to Shanghai Quantum City spatial intelligence construction action[EB/OL]. (2025-12-21)
[2026-06-07]. https://mp.weixin.qq.com/s/08g9
LBvUoUpWki_Lo9saFQ.
卓梁雨，万振龙，马超凡. RAG及AI编码技术在
海关知识库平台中的应用与优化[J]. 中国口岸
科学技术，2026，8（1）：10-15.
Zhuo Liangyu, Wan Zhenlong, Ma Chaofan.
Application and optimization of RAG and AI
coding technology in customs knowledge base
platform[J]. China Port Science and Technology,
2026, 8(1): 10-15.
Gao Y F, Xiong Y, Gao X Y, et al. Retrievalaugmented generation for large language models:
a survey[PP/OL]. V1. (2023-12-18) [2026-06-07].
https://doi.org/10.48550/arXiv.2312.10997.
籍欣萌，昝红英，崔婷婷，等. 大模型在垂直领
域应用的现状与挑战[J]. 计算机工程与应用，
2025，61（12）：1-11.
Ji Xinmeng, Zan Hongying, Cui Tingting, et al.
Current status and challenges of large models
in vertical domain applications[J]. Computer
Engineering and Applications, 2025, 61(12): 1-11.
Lewis P, Perez E, Piktus A, et al. Retrievalaugmented generation for knowledge-intensive
NLP tasks[PP/OL]. V4. (2021-04-12) [2026-06-
07]. https://doi.org/10.48550/arXiv.2005.11401.
Zhao T, Wallace E, Feng S, et al. Calibrate before
use: improving few-shot performance of language
models[C]//Proceedings of the International Conference on Machine Learning, 2021: 12697-12706.
Cheng X, Luo D, Chen X, et al. Lift yourself up:
retrieval augmented text generation with selfmemory[C]//Advances in Neural Information
Processing Systems, 2024.
Gao L, Madaan A, Zhou S, et al. PAL: programaided language models[C]//Proceedings of the
International Conference on Machine Learning,
2023: 10764-10799.
Huang L, Yu W J, Ma W T, et al. A survey on
hallucination in large language models: principles,
taxonomy, challenges, and open questions[J].
ACM Transactions on Information Systems, 2025,
43(2): 1-55.
张玉鑫. 打造量子城市 构建数字化赋能超大城
市治理新模式[J]. 中国测绘，2025（5）：7-10.
Zhang Yuxin. Building a Quantum City to construct
a new digital empowerment model for mega-urban
governance[J]. China Surveying and Mapping,
2025(5): 7-10.
程南炜，徐见微，王进. 面向城市规划设计的多
模态RAG知识库系统构建方法研究[J]. 中国信
息界，2026（3）：85-89.
Cheng Nanwei, Xu Jianwei, Wang Jin. Research
on the construction method of multimodal RAG
knowledge base system for urban planning and
design[J]. China Information Times, 2026(3): 85-89.
