A two-dimensional data-driven model for traffic flow on highways

Michaël Herty, Giuseppe ViscontiNetworks 2018 · DOI

TL;DR

This paper introduces a novel two-dimensional macroscopic model for traffic flow on highways, capturing dynamics both along and across lanes. By fitting closure relations to experimental data, the model offers a more realistic representation of traffic than traditional one-dimensional approaches, demonstrating improved predictive accuracy.

What Problem Does This Paper Address?

Traffic flow modeling has traditionally relied on one-dimensional (1D) macroscopic models, which treat traffic as a single entity moving along a road. While these models are mathematically tractable, they often oversimplify real-world traffic dynamics, particularly on multi-lane highways where lane changes and lateral movements are crucial. Existing multi-lane models often resort to empirical interaction terms for lane changes, which can be difficult to estimate from data. This paper addresses the need for a more comprehensive model that inherently accounts for the two-dimensional nature of traffic flow, using real road geometry and experimental data to derive its behavior, rather than relying on heuristic assumptions.

Methods

The core of this work is the development of a two-dimensional first-order macroscopic traffic flow model based on the continuity equation:

$$ \partial_t \rho + \partial_x q^x + \partial_y q^y = 0 $$

Here, $\rho$ is the traffic density, $q^x$ and $q^y$ are the fluxes in the x (along-road) and y (across-lanes) directions, respectively. The key innovation lies in how the "closure relations" (i.e., the functional relationships between density and flux, $q^x = q^x(\rho)$ and $q^y = q^y(\rho)$) are determined. Instead of using pre-defined theoretical forms, the authors derive these relations by fitting smooth, parameterized functions to experimental fundamental diagrams (plots of flux vs. density and speed vs. density) obtained from real traffic data. For the along-road dynamics ($q^x$), a three-parameter family of strictly concave functions is used, while for the across-lanes dynamics ($q^y$), a simpler two-parameter family is employed due to less available data at higher densities. The numerical solution of the model is achieved using a dimensional splitting method with a second-order finite volume scheme.

Datasets

The primary dataset used in this study is experimental traffic trajectory data collected from an 80-meter stretch of the westbound A3 highway near Aschaffenburg, Germany. This data, captured by laser scanners with a temporal resolution of 0.2 seconds, provides two-dimensional positions of vehicles. The authors also refer to US highway data (US101 from the NGSIM dataset) for comparison, noting that this data exhibits a "naive" behavior across lanes, where lane changes are less pronounced.

Key Insights and Ideas

  • The importance of lateral dynamics: The study highlights that lane changes, though often a small percentage of total vehicle movements, significantly impact traffic dynamics and potential conflicts. The inclusion of a y-direction flux is crucial for a more accurate representation.
  • Data-driven closure relations: Deriving flux functions directly from experimental data, rather than relying on theoretical assumptions, allows the model to capture the specific behaviors of traffic on real roads and under particular traffic rules.
  • Two-dimensional models outperform 1D: The paper provides strong evidence that the 2D model offers superior predictive accuracy compared to its 1D counterpart, especially when dealing with complex traffic scenarios.
  • German highway traffic rules enable 2D modeling: The authors note that the specific traffic rules on European highways, which allow for more fluid lateral movement, make the German data particularly suitable for developing and validating a 2D model. US highways, with stricter lane discipline, show less pronounced lateral dynamics.

Key Results

The paper demonstrates that the proposed two-dimensional model achieves better predictive accuracy than a comparable one-dimensional model. This improvement is quantified through error plots (1-norm error on density and traveling time) which consistently show lower errors for the 2D model across various initial conditions and simulation times. For instance, the relative errors in density prediction are shown to be significantly lower for the 2D model. The data-fitting procedure for the closure relations is shown to be robust, with small variations in fitting parameters leading to only minor changes in the model's predictive error.

Limitations and Future Work

  • Data limitations for y-direction: The authors acknowledge that the available data provides less information on the behavior of the y-direction flux at higher densities, leading to a simpler functional form for $q^y$.
  • Idealized road geometry: While the model accounts for the road's curvature in data cleaning, it primarily models a continuous domain rather than discrete lanes, which could be a limitation for very specific lane-based analyses.
  • Future work: The authors suggest extending the model to higher-order schemes and investigating the impact of traffic regulations, such as lane reductions, using the 2D framework. They also mention ongoing work on a second-order 2D model and a hybrid kinetic model.

Summary updated

Abstract

Based on experimental traffic data obtained from German and US highways, we propose a novel two-dimensional first-order macroscopic traffic flow model. The goal is to reproduce a detailed description of traffic dynamics for the real road geometry. In our approach both the dynamics along the road and across the lanes is continuous. The closure relations, being necessary to complete the hydrodynamics equation, are obtained by regression on fundamental diagram data. Comparison with prediction of one-dimensional models shows the improvement in performance of the novel model.
BibTeX
@inproceedings{herty2018two,
  author = {Michaël Herty and Giuseppe Visconti and Giuseppe Visconti},
  title = {{A two-dimensional data-driven model for traffic flow on highways}},
  booktitle = {Networks},
  year = {2018},
  doi = {10.3934/nhm.2018010},
}

Loading related content...

Figure