基于改进DBSCAN空间聚类算法的北京市人工智能产业集聚格局研究  

Artificial intelligence industrial agglomerations in Beijing:An spatial pattern study based on improved DBSCAN algorithm

作　　者：张平[1] 范文慧[1] 贾婧 刘义[1] Zhang Ping;Fan Wenhui;Jia Jing;Liu Yi(Department of Automation,Tsinghua University,Beijing 100084,China;College of History,Culture and Tourism,Liaocheng University,Liaocheng 252059,Shandong,China)

机构地区：[1]清华大学自动化系,北京100084 [2]聊城大学历史文化与旅游学院,山东聊城252059

出　　处：《地理科学》2024年第2期238-247,共10页Scientia Geographica Sinica

基　　金：国家重点研发计划项目(2021YFF0900801)资助。

摘　　要：企业作为产业的重要主体,其发展直接表征着产业的发展,企业的空间格局对产业的发展及资源配置具有重要的引导意义。本文基于北京市工商注册在业的人工智能企业数据,提取企业注册地址并转化为地理位置信息,通过改进有噪声的应用背景下的基于密度的空间聚类算法(DBSCAN),分析北京市细粒度层级下人工智能产业集聚在全市域的空间分布格局特征。在DBSCAN算法改进方面,首先调整Minpts参数为企业注册资本总额与企业数量2个维度,企业数量大于5家且注册资本总额大于一定数额,为形成产业集聚区的2个必要条件;其次提取簇内位于边界的企业位置点作为集聚区地理边界点,将边界点连线并绘制形成人工智能产业集聚区。本文重点分析了企业注册资本和地理聚合半径对人工智能产业集聚区形成的影响,同时采用核密度估计法作为参照验证,表明改进DBSCAN方法具有精确刻画产业集聚区地理边界和确定不同规模产业集聚区的优势。通过分析得知,北京市人工智能产业集聚具有明显的中心分布特征,集中在城六区,呈现“两大龙头带动,北京市全域遍地开花”的分布情况,海淀区、朝阳区处于人工智能集聚程度高水平,相较其他区域,人工智能产业发展遥遥领先;西城区、东城区、丰台区、昌平区处于集聚程度较高水平;通州区、大兴区、平谷区、密云区、石景山区、房山区、门头沟区、怀柔区、顺义区处于集聚程度中等水平;延庆区集聚程度较低。通过改进DBSCAN算法精确定位出中关村区域、上地西二旗区域、五道口区域、望京区域、国贸区域、亦庄经开区等人工智能产业集聚区。进一步探究发现,海淀区的人工智能科研人才优势,朝阳区的信息技术领域企业基础,是两区人工智能发展突出的直接原因。延庆区等郊区远离北京市中心城区,产业资源匮乏,同时由于区As enterprises are key players in industries,the enterprise development determines the industry development.The spatial pattern of the enterprises has important guiding significance for the development of the industry and its resource allocation.This paper analyzes the spatial distribution characteristics of fine-grained artificial intelligence(AI)industrial agglomerations in Beijing by using the improved Density-Based Spatial Clustering of Applications with Noise(DBSCAN)algorithm to deal with the geographic location information of AI enterprises registered in the city.In terms of improving DBSCAN algorithm,in the first instance the Minpts parameters are adjusted to two dimensions,the total registered capital of enterprises and the number of enterprises.The number of enterprises is more than 5 and the total registered capital is greater than a certain amount,which are two necessary conditions for the formation of industrial agglomeration area.Then,the enterprises located at the boundary in the cluster are chosen as the geographical boundary points which are connected to form the geographical boundary of the AI industry agglomeration area.This paper focuses on the impact of the total registered capital of enterprises and cluster radius on AI industrial agglomerations.Meanwhile,the kernel density estimation method is used to verify that the improved DBSCAN has the advantage of accurately depicting industrial agglomeration areas and identifying industrial agglomeration areas of different sizes.According to the analysis,AI industrial agglomerations in Beijing are concentrated in six districts of the urban center,showing the distribution of‘Driven by the two leading urban districts and spread all over Beijing’.Among the sixteen districts of Beijing,Haidian District and Chaoyang District are at the highest level,where the development of the AI industry is far ahead,followed by Xicheng District,Dongcheng District,Fengtai District and Changping District.Tongzhou District,Daxing District,Pinggu District,Miyun District

关 键 词：人工智能 产业集聚 改进DBSCAN算法 核密度估计 GIS 北京 

分 类 号：F129.9[经济管理—世界经济]