Effect of the choice of stagnation density in data-fitted first- and second-order traffic models

TL;DR

This paper investigates how the choice of stagnation density affects the accuracy of data-fitted first- and second-order macroscopic traffic models, showing that lower stagnation densities improve predictive performance.

Contribution

It demonstrates that reducing stagnation density from common values improves the accuracy of traffic models fitted to real data.

Findings

Lower stagnation densities (90-100 vehicles/km/lane) enhance model accuracy.

Higher stagnation densities (120 vehicles/km/lane and above) cause slower information propagation.

Adjusting stagnation density improves predictive capabilities of LWR and ARZ models.

Abstract

For a class of data-fitted macroscopic traffic models, the influence of the choice of the stagnation density on the model accuracy is investigated. This work builds on an established framework of data-fitted first-order Lighthill-Whitham-Richards (LWR) models and their second-order Aw-Rascle-Zhang (ARZ) generalizations. These models are systematically fitted to historic fundamental diagram data, and then their predictive accuracy is quantified via a version of the three-detector problem test, considering vehicle trajectory data and single-loop sensor data. The key outcome of this study is that with commonly suggested stagnation densities of 120 vehicles/km/lane and above, information travels backwards too slowly. It is then demonstrated that the reduction of the stagnation density to 90-100 vehicles/km/lane addresses this problem and results in a significant improvement of the predictive accuracy of the considered models.