空间数据智能大模型研究

王少华1*，谢幸2，李勇3，郭旦怀4，才智5，刘瑜6，乐阳7，潘晓8，陆锋9，吴华意10，桂志鹏10，丁治明11，郑渤龙12，张富峥13，秦涛2，王静远14，陶闯15，陈正超1，卢浩16，李家艺10，陈红阳17，乐鹏10，禹文豪18，姚尧18，孙磊磊14，张勇5，陈龙彪19，杜小平20，李响21，张雪英22，秦昆10，宫兆亚6，董卫华23，孟小峰24* 1中国科学院空天信息创新研究院；2微软亚洲研究院；3清华大学；4北京化工大学；5北京工业大学；6北京大学；7深圳大学；8石家庄铁道大学；9中国科学院地理科学与资源研究所；10武汉大学；11中国科学院软件研究所；12华中科技大学；13快手自然语言处理中心和音频中心；14北京航空航天大学；15维智科技；16北京超图软件股份有限公司；17之江实验室；18中国地质大学（武汉）；19厦门大学；20中国科学院数字地球重点实验室；21华东师范大学；22南京师范大学；23北京师范大学；24中国人民大学E-mail for correspondence:孟 小 峰 （xfmeng@ruc.edu.cn）,王 少 华 （wangshaohua@aircas.ac.cn） 摘要：本报告是由ACM SIGSPATIAL中国分会组织撰写，是《国家空间数据智能年度发展报告（2022）》和《空间数据智能与城市元宇宙白皮书（2023）》的延续和拓展，并在第五届空间数据智能学术会议（SpatialDI 2024）上正式发布。本报告以空间数据智能大模型为核心，聚焦空间数据智能大模型原理、方法和应用前沿，对空间数据智能大模型的定义、发展历程、现状与趋势、面临挑战等议题进行深入阐述，对空间数据智能大模型的关键技术以及在城市、空天遥感、地理、交通等应用场景做了系统性阐述，同时整理总结了现阶段空间数据智能大模型在城市、多模态、遥感、智慧交通、资源环境等主题的最新应用案例，最后对空间数据智能大模型的发展前景进行了总结与展望。空间数据智能大模型方面，本报告对空间数据智能大模型这一核心概念的背景和定义做了阐述，并深入讨论了空间数据智能大模型的三阶段发展历程，分析了空间数据智能大模型的研究现状和发展趋势；在此基础上，本报告提出了空间数据智能大模型现今所面临的三个重大挑战。空间数据智能专题大模型方面，本报告围绕现阶段空间数据智能大模型的研究状况，梳理了空间数据智能大模型城市、空天遥感、地理、交通等四大专题领域研究进展。空间数据智能大模型关键技术方面，本报告系统介绍了空间数据智能大模型的关键技术、特点优势、研究现状、未来发展等核心信息，涉及时空大数据平台、分布式计算、3D虚拟现实、空间分析与可视化等大模型的基础性能，以及地理空间智能计算、深度学习、大数据高性能处理、地理知识图谱、地理智能多情景模拟等大模型的复杂空间综合性能，解析上述关键技术在空间数据智能大模型的位置和作用。空间数据智能大模型应用方面，本报告系统梳理空间数据智能大模型的最新应用案例，横跨城市、多模态数据处理、遥感智能计算、智慧交通、资源环境等五大领域，着眼于未来空间数据处理分析场景的发展变化，展望空间数据智能大模型的三个发展趋势，为未来空间数据智能大模型在产、学、研多方面发展提供了参考。本报告为促进AGI时代空间数据智能大模型的发展及其在城市、空天遥感、地理、交通等领域的应用，同时推动地理信息科学、计算机科学等领域在空间数据智能大模型交叉研究方面的理论、技术与应用的学术交流，解决空间数据智能大模型产业发展面临的重大挑战和瓶颈问题指明了方向。 关键词：空间数据智能大模型；智能计算；AGI；GeoAI；多模态 the continuation and expansion oftheNational Spatial Data Intelligence Annual DevelopmentReport(2022)andthe Spatial Data Intelligence and Urban Metaverse White Paper (2023). It wasofficially released at the 5th Spatial Data Intelligence Academic Conference (SpatialDI2024).This report takes the spatial data intelligent large model as the core, focuses on the principles,methods and application frontiers of the spatial data intelligent large model, and provides an in-depth explanation of the definition, development process, current situation and trends, challengesand other issues of the spatial data intelligent large model. The key technologies of large dataintelligent models and their application scenarios in cities, air and space remote sensing,geography, transportation, etc. are systematically elaborated. At the same time, the currentapplication scenarios of large spatial data intelligent models in cities, multi-modal, remotesensing, smart transportation, etc. are summarized. The latest application cases on topics such asresources and environment, and finally the development prospects of spatial data intelligent largemodels are summarized and prospected.This report elaborates on the background and definitionof the core concept of spatial data intelligent large models, and deeply discusses the three-stagedevelopment process of spatial data intelligent large models, and analyzes the spatial dataintelligent large models. Research status and development trends; on this basis, this reportproposes three major challenges faced by large spatial data intelligent models today.This reportfocuses on the current research status of spatial data intelligent large-scale models and sorts outthe research progress in four major thematic areas of spatial data intelligent large-scale models:cities, air and space remote sensing, geography, and transportation.This report systematicallyintroducesthe key technologies,characteristics and advantages,research status,futuredevelopment and other core information of spatial data intelligent large models, involvingspatiotemporal big data platforms, distributed computing, 3D virtual reality, space The basicperformance of large models such as analysis and visualization, as well as the complex spatialcomprehensive performance of large models such as geospatial intelligent computing, deeplearning,high-performance processing of big data,geographical knowledge graphs,andgeographical intelligent multi-scenario simulation, analyze the application of the above keytechnologies in spatial data The location and role of smart large models.This report systematicallysorts out the latest application cases of large-scale spatial data intelligent models, spanning fivemajor fields including cities, multi-modal data processing, remote sensing intelligent computing,smart transportation, and resources and environment, focusing on the future of spatial data. Itdeals with the development and changes of analysis scenarios and looks forward to the threedevelopment trends of spatial data intelligent large models, which provides a reference for thefuture development of spatial data intelligent large models in industry, academia, and research.This report aims to promote the development of large spatial data intelligent models in the AGIera and their applications in urban, air and space remote sensing, geography, transportation andother fields, and to promote cross-research on large spatial data intelligent models in the fields ofge