Modeling the heterogeneous traffic correlations in urban road systems using traffic-enhanced community detection approach
Introduction
Modeling the spatial interaction between components of urban road systems is essential for understanding traffic patterns. Urban road traffic has inherent spatial co-variations. Traffic conditions detected at different locations within a certain distance are often correlated [[1], [2], [3], [4], [5]]. A good characterization of this spatial co-variation can be used for better understanding urban traffic rhythm from a spatial perspective [6], or for improving traffic-related applications, such as traffic interpolation and forecasting [[7], [8], [9], [10], [11], [12]]. The fundamental issue of spatial interaction analysis is the determination of spatial neighbors: those elements surrounding a given element that could be considered to influence the element [13].
In previous studies, spatial neighborhoods are usually defined by distance. For instance, Kamarianakis and Prastacos [7] used Euclidean distance in geographical space to model the spatial correlation of traffic flow. Min et al. [9] used spatial neighbors within a defined topological distance in a city road network to improve traffic forecasting. Zou et al. [11] also used topological distance to determine neighboring roads, hence to improve the interpolation of traffic data, while Ding et al. [10], Min and Wynter [3] determined spatial neighbors by time-dependent route distance. However, distance based models usually cannot depict the urban traffic well due to the existence of spatial heterogeneity, which is resulted from the uneven urban structures and travel demands.
In urban road systems, some roads carry more traffic than others, and thus have higher impacts on their neighbors [14]. This leads to two types of spatial heterogeneity in city traffic. First, from a global perspective, the influence of road traffic varies at different locations. This may lead to difficult definition of traffic-relevant neighbors. Second, from a local perspective, the influence of road traffic is anisotropic where traffic flow on a road segment will not spread evenly to all downstream road segments, but instead concentrate in certain directions [15]. Cheng et al. [4] explored dynamic space–time autocorrelation on journey time data in London and revealed that the spatial heterogeneity are due to the variation of the level of correlation between individual road links. Furthermore, the size of the spatial neighborhood is time dependent [12].
Community detection is an algorithm to partition all the nodes in a network into different clusters. Nodes are densely connected within the same clusters and sparsely connected between different clusters [[16], [17], [18], [19], [20]]. In the case of road network with road segment as node and the connection between road segments as edge, nodes within the same clusters are densely connected, which may also be interacted strongly in traffic.
Therefore, in this paper, we propose a new approach called “traffic-enhanced community detection” to extract the traffic-correlated road segment clusters (i.e., spatial neighborhood) by considering both the topological connections/distances and the traffic correlations among road segments. First, the urban road network is modeled as a dual graph where the road segments represent nodes, the connections among road segments represent edges, and the weights on the edges represent the traffic correlation of the road segments. Second, the traffic-correlated road segments are gathered together through the Infomap community detection algorithm. The Mutual information index is used to evaluate the robustness of the community detection results, while the Moran’s I and Calinski–Harabaz Index are used to verify the approach’s capacity to uncover the heterogeneous traffic relevance on the road system.
The remainder of the paper is organized as follows: Section 2 introduces traffic-enhanced community detection method in detail. Section 3 conducts a series of experiments using mass traffic data to evaluate the robustness and effectiveness of the identified traffic-related road segment clusters. Section 4 is devoted to discussions. The conclusion is drawn in Section 5.
Section snippets
Modeling urban road system
Traditionally, an urban road system is represented as a network where the interactions are mapped as nodes and road segments are mapped as edges [21]. Such a plain representation is called “primal graph”, which is often used to describe the spatial configuration of the road segments and interactions. However, a “dual-graph” with road segments represented as nodes and the connections among road segments represented as edges is more suitable to model the relationships among road segments, which
Data
The road system of downtown Beijing, China (as shown in Fig. 3) is used in this case study. It contains 4616 road segments, including expressways, arterial streets, and collector streets, which are rendered by red, orange and green respectively.
Vehicular speeds on road segments are obtained from more than 50,000 GPS-equipped taxis at 5-min time intervals, from April 1, 2012 to June 30, 2012. Since the traffic conditions are markedly different on workdays and weekends, two datasets are built
Discussion
In this paper, we propose a novel approach to extract the traffic-related road segment clusters by conducting the Infomap community detection algorithm on the traffic-enhanced dual graph. The highlights of this paper can be summarized as follows.
(1) A traffic-enhanced topological dual-graph is constructed to represent the road system, which can well describe the topological connections and traffic correlations among road segments at the same time.
(2) The Infomap community detection algorithm is
Conclusions
Identifying and quantifying the traffic correlations within road systems is essential for both traffic management and effective vehicle navigation. In this paper, we have proposed a novel approach to measure the traffic correlations within city road systems. It uses a traffic-enhanced topological dual-graph to represent the road network, and then divides a road network into closely connected and strongly traffic correlated road segment clusters. It is argued the data-driven method can reveal
Acknowledgments
This research is supported by the National Natural Science Foundation of China (Grant No. 41631177), Key Project of the Chinese Academy of Sciences, China (Grant No. ZDRW-ZS-2016-6-3), and the National Key Research and Development Program, China (Grant No. 2016YFB0502104). We also thank the anonymous referees for their helpful comments and suggestions.
References (33)
- et al.
Tracking and predicting a network traffic process
Int. J. Forecast.
(1997) - et al.
A multivariate state space approach for urban traffic flow modeling and prediction
Transp. Res. C
(2003) - et al.
Real-time road traffic prediction with spatio-temporal correlations
Transp. Res. C
(2011) - et al.
Space–time modeling of traffic flow
Comput. Geosci.
(2005) - et al.
Optimized and meta-optimized neural networks for short-term traffic flow prediction: A genetic approach
Transp. Res. C
(2005) Community detection in graphs
Phys. Rep.
(2010)- et al.
Robustness of city road networks at different granularities
Physica A
(2014) - et al.
Community structure in traffic zones based on travel demand
Physica A
(2016) - et al.
A network centrality measure framework for analyzing urban traffic flow: A case study of Wuhan, China
Physica A
(2017) - et al.
Spatio-temporal autocorrelation of road network data
J. Geogr. Syst.
(2012)
Urban traffic state explained by road networks and spatial variance: approach using floating car data
Transp. Res. Rec.
Encapsulating urban traffic rhythms into road networks
Sci. Rep.
Forecasting traffic volume with space–time ARIMA model
Adv. Mater. Res.
An improved distance metric for the interpolation of link-based traffic data using kriging: a case study of a large-scale urban road network
Int. J. Geogr. Inf. Sci.
Dynamic spatial weight matrix and localized space-time autoregressive integrated moving average for network modeling
Geogr. Anal.
Cited by (31)
Structure and robustness of China’s railway transport network
2023, Transportation LettersComplex-network-based traffic network analysis and dynamics: A comprehensive review
2022, Physica A: Statistical Mechanics and its ApplicationsCitation Excerpt :However, the study disregarded network dynamics. Lu et al. [68] used the Infomap algorithm to detect community structures based on topological structures and dynamic traffic flows. They built a dual graph and treated the correlation coefficients of traffic conditions between two endpoints as link weights.
Unfolding community homophily in U.S. metropolitans via human mobility
2022, CitiesCitation Excerpt :Infomap is based on the map equation framework (Rosvall et al., 2009), can be applied to the directed weighted graphs, and presents superiority over modularity methods (Bae et al., 2013; Farage et al., 2021). Notable efforts include Hong and Yao (2019), who employed Infomap to detect the hierarchical community structure of urban roads, and Lu et al. (2018), who used the Infomap algorithm to detect traffic-correlated segment clusters for revealing the road traffic correlation. Given the strong capability and high efficiency of Infomap in handling network structures with massive vertices and edges, we adopted the Infomap algorithm to partition our network into communities with strong spatial interactions.
Revealing spatiotemporal correlation of urban roads via traffic perturbation simulation
2022, Sustainable Cities and SocietyCitation Excerpt :In addition, several types of correlation measurement methods have been proposed for different road network scenarios. Specifically, the Pearson correlation coefficient and its improved form are widely utilized to capture the spatiotemporal correlations between road segments, which demonstrate that traffic flow and travel speed distinctly enhance the accuracy of road network correlation estimation (Su et al., 2017; Barroso et al., 2020; Lu et al., 2018; Sun et al., 2005). Natural language processing techniques, including word embedding and topic models, can also be used to detect correlations between road segments by comparing them with words in corpora (Liu et al., 2017; Liu et al., 2019).
Discovering latent structures with integrated propagation algorithms in geographical information networks
2021, Physica A: Statistical Mechanics and its ApplicationsCitation Excerpt :And the structure information such as community numbers and overlap is usually unknown. For example, the road network [7] is affected by the road positions and distances, and has no clear classifications. In the geographical information networks, the distances between different positions should be taken into consideration, which can be represented as different edge weights.
Visualizing and exploring POI configurations of urban regions on POI-type semantic space
2020, CitiesCitation Excerpt :Therefore, in this section, we evaluate the effectiveness of our proposed approach by making spatial autocorrelation analysis toward the POI configurations of urban regions on semantic space (i.e., “thematic maps”) using the Global Moran's Index (Moran, 1950). The Global Moran's I measures spatial autocorrelation based on both element locations and values (Lu, Liu, Duan, Cheng, & Du, 2018). It evaluates whether the pattern expressed is clustered, dispersed, or random.