Transport Network Graph Representation

Transport Network Graph Representation#

There are several ways to represent transportation networks, mappings from the real world to another representation, e.g., a visual representation as a hiking map or a graph representation as a network. From application to application, useful representations can differ. We will use a directed multigraph \(G = (V, E, l)\) with edges \(e \in E\) and vertices \(v \in V\). The edges \(e\) are weighted with length \(l\), but can have more attributes, like a type or a name. Also, the vertices \(v\) can have attributes, e.g., geographical latitude and longitude. Edges represent streets, and vertices represent intersections, junctions, or dead ends. Streets are specifically not the semantic entity of a road, but a part of a road between exactly two intersections. Another way of dealing with a road network is grouping edges to ways, inspired by the semantics of named roads [elgouj2022], or a dual construction is defining road sections drivable without turns as nodes and streets connected by a turn as edges [lagesse2015]. For the street graph, \(G\) we require a few more properties:

  • Directed: The edges have a direction, e.g., from intersection \(a\) to intersection \(b\). In the case of two-way streets are represented by two edges, one from \(a\) to \(b\) and one from \(b\) to \(a\).

  • Strongly connected: There is a path from every vertex to every other vertex. In a street graph, this means that every intersection is reachable from every other intersection.

  • Loops: An edge can start and end at the same vertex.

As the transportation network can have bridges and tunnels, the graph is not necessarily planar. The Python package osmnx [boeing2017a] implements such functionality to standardizedly retrieve OSM data and simplify the network into a graph representation of the transportation network that the above requirements after some filtering. It is based on the NetworkX [SciPyProceedings11] package, which implements graph algorithms and data structures.

Partition Requirements#

The street graph \(G\) will be split into partitions, one for each Superblock and one for the sparse network. This can be described by a partitioning \(\mathcal{P} : G \mapsto \left(G_\mathrm{sp} \cup G_1 \cup \dots \cup G_k\right)\) returning subgraphs \(G_i\subseteq G\), one sparse \(G_\mathrm{sp}\), and \(k\) \(G_i\). Such a partitioning function \(\mathcal{P}\) must satisfy a union property

\[\bigcup_{i=1}^k G_i \cup G_\mathrm{sp} = G\]

and an edge-wise disjoint property

\[\forall i, j \in \{\mathrm{sp}, 1, \dots, k\} : i \neq j \Rightarrow E_i \cap E_j = \emptyset,\]

where \(E_i\) is the set of edges of \(G_i\). The union property states that the partitioning function \(\mathcal{P}\) must return a partitioning of the whole graph \(G\), in other words, no street should be left out. The disjoint property states that the partitions must be edge-wise disjoint, i.e., no street should be part of more than one partition. This also means that from the set of edges \(E_i\) we can exactly reconstruct the set of vertices \(V_i\). Our goal is to compare performance of automatized \(\mathcal{P}\), before any restrictions are applied, to restricting paths to only use edges of the start and, end and sparse network. Such for all paths \(p = (e_s, \dots, e_t)\), where \(e_s \in E_\mathrm{s}\) and \(e_t \in E_\mathrm{t}\), the path is a subset \(p \subseteq E_\mathrm{s} \cup E_\mathrm{sp} \cup E_\mathrm{t}\), including paths starting or ending in the sparse network. To satisfy connectivity for \(\mathcal{P}\), a sufficient condition is that the sparse network is strongly connected and that the Superblocks are connected to the sparse network. From anywhere in a neighborhood, it must be possible to reach anywhere else in a city, without passing a foreign Superblock. However, it is possible that with a start and end inside the same Superblock one must, by car, use the sparse network.

[elgouj2022]

El Gouj, H., Rincón-Acosta, C. & Lagesse, C. Urban morphogenesis analysis based on geohistorical road data. Appl Netw Sci 7, 1–26 (2022). DOI 10.1007/s41109-021-00440-0

[lagesse2015]

Lagesse, C. Lire les Lignes de la Ville. (Universite Paris Diderot-Paris VII, 2015).

[boeing2017a]

Boeing, G. OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks. Computers, Environment and Urban Systems 65, 126–139 (2017). DOI 10.1007/978-3-030-12381-9_12

[SciPyProceedings11]

Hagberg, A. A., Schult, D. A. & Swart, P. J. Exploring network structure, dynamics, and function using NetworkX. in Proceedings of the 7th python in science conference (eds. Varoquaux, G., Vaught, T. & Millman, J.) 11–15 (2008).