基于空间聚类方法的京津冀城市群多层级空间结构研究

在京津冀协同发展战略背景下,以建设京津冀世界级城市群为引领,遵循城市发展规律,优化城市空间布局,明确京津冀城市群等级结构及其空间特征具有重要意义。本文以京津冀城市群156个区县为研究对象,从经济中心性、交通中心性、信息中心性、人口中心性4个角度,利用4种空间聚类方法进行5个等级的聚类分析,并基于克氏中心地理论对京津冀城市群等级划分结果进行空间结构分析。结果显示,自组织特征映射神经网络算法(SOM)较适合京津冀城市群的等级划分;京津冀城市群正从以北京城区为单核心的圈层空间结构向3条带型空间结构转变,其中京津都市发展走廊发育成熟,沿海都市发展带也初具规模,而包括雄安新区在内的京石都市发展带正在孕育。

Multi-level spatial structure analysis of urban agglomeration in the Beijing-Tianjin-Hebei region based on spatial clustering algorithms

With the main focus of regional competition shifting from cities to urban agglomerations, it is important to analyze the spatial structure and direction of coordinated development in urban agglomerations. This is especially significant in the Beijing-Tianjin-Hebei urban agglomeration, which is a core urban agglomeration in China. Machine learning algorithms are relatively new methods for addressing geographical problems. Clustering method, as unsupervised learning, is useful for classifying geographical units without the need for priori knowledge. Using data from 156 counties in the Beijing-Tianjin-Hebei urban agglomeration, this study applied four clustering algorithms: the K-means, density-based spatial clustering of applications with noise (DBSCAN), Chameleon, and self-organizing map (SOM) methods, for classifying counties and districts in the Beijing-Tianjin-Hebei urban agglomeration from the perspectives of economic centrality, traffic centrality, information centrality, and population centrality. GDP of the counties in 2014 was used to represent economic centrality; density of road networks in counties and attraction factor, calculated by the unsold train tickets in different time periods of the year, represent traffic centrality; Sina Weibo check-in data were used to represent information centrality; and county/district population represents population centrality. The result classifies the urban agglomeration into several levels. Respectively, K-means algorithm classifies counties into five levels; DBSCAN algorithm classifies counties into six levels; Chameleon algorithm classifies counties into six levels; and SOM algorithm classifies counties into five levels. SOM is the most applicable algorithm for the division of the urban agglomeration because the structure of counties is stable. This study further analyzed the spatial structure of the urban agglomeration with the central place theory, which points out that an agglomeration should contain certain number of counties in every level. The result of the SOM algorithm matches the central place theory. This research shows that there were remarkable gaps between different levels of the urban agglomeration. The central area of Beijing, as the core of the region, has strong radiation effect on the surrounding areas, but its functions are shared by the nearby counties. Moreover, the second and third level central cities distribute evenly and play an important role in regional development.