Estimating Peak-Hour Traffic Congestion Patterns For Interacting Agents On Urban Networks

TL;DR

This paper models urban traffic congestion emergence by simulating vehicle interactions and network loading, introducing a new measure for cumulative road usage and analyzing congestion patterns and their impact on network connectivity.

Contribution

It presents a novel dynamical measure for cumulative road usage and provides insights into how congestion develops and disrupts urban network connectivity.

Findings

Congestion begins with scattered dysfunctional edges.

A new measure better proxies cumulative road usage.

Network transitions abruptly to widespread congestion.

Abstract

We study the emergence of congestion patterns in urban networks by modeling vehicular interaction by means of a simple traffic rule and by using a set of measures inspired by the standard Betweenness Centrality (BC). We consider a topologically heterogeneous group of cities and simulate the network loading during the morning peak-hour by increasing the number of circulating vehicles. At departure, vehicles are aware of the network state and choose paths with optimal traversal time. Each added path modifies the vehicular density and travel times for the following vehicles. Starting from an empty network and adding traffic until transportation collapses, provides a framework to study network's transition to congestion and how connectivity is progressively disrupted as the fraction of impossible paths becomes abruptly dominant. We use standard BC to probe into the instantaneous out-of-equilibrium network state for a range of traffic levels and show how this measure may be improved to build a better proxy for cumulative road usage during peak-hours. We define a novel dynamical measure to estimate cumulative road usage and the associated total time spent over the edges by the population of drivers. We also study how congestion starts with dysfunctional edges scattered over the network, then organizes itself into relatively small, but disruptive clusters.