基于局部聚类的网络Voronoi图生成方法研究

提出一种将网络约束下的Voronoi和空间聚类相结合的方法,通过构造局部的聚类分析方法对网络边进行加权,根据实际的点过程性质可以把权重定义为加权或者乘权,进行标准化后与道路段本身长度融合进行计算,依此生成网络Voronoi图,以期理解城市街道的空间特性。以武汉市江汉区为例,对城市网格管理系统产生的城市事件进行算法验证,结果表明,该方法提供了一种灵活的网络约束下的服务区域划分工具,可用于基于网络空间点过程影响下的服务区划分,也可用于系统性地定量刻画城市管理的动态特性。

A Method for Integrating Network Voronoi and Spatial Clustering

The planar space assumption of spatial cluster detection is invalid in the real world. The network space has opened a new gate to finer-scale spatial analysis, and provides a perspective for human dynamics. The urban street network is shaped by social and economic forces over time and also reflects the influences of governmental policies and cultural values. In the real world, any phenomenon whose location is represented through a street address system is inherently constrained by the street network. Hence, both events and their movements are constrained by the street network in the urban area. For example, the street-side business will cause more traffic, which lead to both management and commuting costs. Since the spatial point process is inherently probabilistic, it’s hard to set a fixed set of criteria, which would otherwise be dealt with as a spatial optimization problem. The weight of a given street segment will vary across space and over time when the activities of street-side business on this focal segment and nearby ones are considered. It is crucial to incorporate this information into urban management and urban studies, because equally-weighted street segments do not exist in the real world. The extension of Voronoi diagrams to the network space provides a useful tool in estimating service area in cities. Weighted Voronoi diagrams have been widely adopted to describe the capacity constraints. This proposed method develops a network Voronoi diagram with weighted links based on spatial cluster analysis. It borrows the strength from two large and growing literatures: Voronoi diagram and spatial cluster analysis. The weight is a central component in the construction of weighted Voronoi in urban street network. The weights are generated using local Moran’s I statistic. The weights, either additive or multiplicative, are normalized and transformed into the link length for constructing network Voronoi diagrams. The Monte Carlo simulation process is adopted to illustrate the statistical significance of detected clusters, and only links with significant p values are chosen to be weighted. The normalization interval determines how the clustering level influences the weights. The additive weight reflects the added constraint of the link attribute, while the multiplicative weight demonstrates the degree of influences imposed on the links. The conceptual foundations and technical details of this approach are elaborated in the case study of Wuhan City, China. The results show that the method is effective in incorporating the clustering criteria into the Voronoi construction, and provides an alternative tool for service area division. The constructed network Voronoi diagrams explicitly take into account the characteristics of underlying event distribution instead of a fixed set of criteria. This method sheds new light on micro-level spatial analysis, providing a perspective for observing how socio-economic urban activities, represented as network-constrained point distribution, shape the spatial structure and patterns in the metropolitan area. These constraints are often modelled on the generator points according to a set of predefined criteria. However, human activities are highly dynamic and constantly evolving. The outcomes of these activities are often represented as spatial point processes, which are also constrained by the network space. By means of clustering, the influences of the event points can be modelled as weights posed on the network links. Such weights can represent the congestions caused or times consumed in handling the events. Accounting these weighted links in the network Voronoi diagram will effectively capture the probabilistic nature of underlying point processes, and therefore more faithfully approximate the partition of street space. In a follow-up study, sensitivity analysis needs to be carried to test how the results might vary based on the street segment size to select.