在大型异构路网中,不同区域的交通运行特征存在显著差异,因此需要针对各个区域的具体特征制定相应的交通管理和控制策略.合理划分路网以获得交通特征均质的子区,对于有效的交通管控和分析至关重要.首先提出了一种改进的密度峰值聚类方法(Enhanced Density Peak Clustering,En-DPC),用于路网子区的初始划分.该方法基于质量概率相似性并考虑路网连接性约束,提升了算法对异常数据的鲁棒性,避免子区内路段不连续的问题.接着,利用En-DPC方法对初始划分的子区进一步合并,形成大小适中的新子区.最后,通过边界调整提高子区边界的平滑度,获得最终的划分结果.该方法能够根据路网交通状态自动确定子区数量,确保划分的合理性.此外,考虑到路网拥堵状态的时空演变,在静态划分基础上设计了一种动态划分方法,根据车辆密度的变化动态调整边界,以提升其在实时交通管控中的适用性,并利用瑞士苏黎世的线圈检测器数据验证了所提出方法的有效性.结果表明,本文提出的方法能够有效地将大型异构路网划分成均质子区,且每个子区都可获得一个清晰的宏观基本图.与现有文献中的路网划分方法如归一分割和"蛇"方法相比,本文方法不仅在归一化总方差、平均NcutSil-houette和模块度等性能评价指标上表现更优,而且子区划分时间明显低于其他两种方法.
Dynamic partitioning method for urban road networks based on enhanced density peak clustering
In large heterogeneous road networks,traffic characteristics vary significantly across re-gions,necessitating tailored traffic management and control strategies specific to each region.Effec-tive traffic management and analysis require reasonable partitioning of road networks to obtain subre-gions with homogeneous traffic characteristics.Therefore,this paper proposes an enhanced density peak clustering(En-DPC)method for the initial partitioning of the road network.Based on mass-based probabilistic similarity and network connectivity,this method enhances robustness to anoma-lous data and avoids discontinuities within road segments.Subsequently,the En-DPC method merges these initial subregions to form new subregions of appropriate size.Finally,boundary adjustment is performed to improve the smoothness of subregion boundaries,to obtain the final result.This meth-od can automatically determine the number of subregions based on network traffic conditions.More-over,considering the spatiotemporal evolution of congestion,a dynamic method is developed based on static partitioning,whereby boundaries are dynamically adjusted according to changes in traffic density,to enhance the applicability of the method to real-time traffic management.The proposed method was validated using loop detector data from Zurich,Switzerland,thereby demonstrating the effective partitioning of large heterogeneous road networks into homogeneous subregions,each with a clear macroscopic fundamental diagram.Compared to existing network partitioning methods such as the normalized cut(Ncut)and the"snake"method,our approach not only performed better in terms of evaluation metrics such as normalized total variance,average NcutSilhouette,and modulari-ty,but also significantly reduced the subregion partitioning time.
traffic engineeringnetwork partitioningenhanced density peak clusteringmacroscopic fundamental diagramdynamic partitioning
国家科技期刊平台
NSTL
万方数据
