Abstract
This paper is concerned with highway traffic estimation using traffic sensing data, in a Lagrangian-based modeling framework. We consider the Lighthill-Whitham-Richards (LWR) model (Lighthill and Whitham, 1955; Richards, 1956) in Lagrangian-coordinates, and provide rigorous mathematical results regarding the equivalence of viscosity solutions to the Hamilton-Jacobi equations in Eulerian and Lagrangian coordinates. We derive closed-form solutions to the Lagrangian-based Hamilton-Jacobi equation using the Lax-Hopf formula (Daganzo, 2005; Aubin et al., 2008), and discuss issues of fusing traffic data of various types into the Lagrangian-based H-J equation. A numerical study of the Mobile Century field experiment (Herrera et al., 2009) demonstrates the unique modeling features and insights provided by the Lagrangian-based approach.
Introduction
Traditional highway traffic estimation has historically relied on the Eulerian perspective, which views traffic from fixed locations, such as loop detectors. While effective for counting vehicles at specific points, this "bird's-eye view" is increasingly supplemented by Lagrangian data—the "driver's-seat view" provided by GPS-enabled smartphones and connected vehicles. This mobile data tracks the position and speed of individual vehicles over time.
The classic Lighthill-Whitham-Richards (LWR) traffic model is typically formulated in Eulerian coordinates. However, integrating mobile GPS data into Eulerian models is often mathematically cumbersome and computationally intensive. This research bridges the gap by providing a rigorous mathematical foundation for moving between these two coordinate systems, ensuring that data from fixed sensors and mobile probes can be fused seamlessly to estimate traffic states.
Methodology
The core of this framework lies in reformulating the LWR partial differential equation (PDE) into Lagrangian coordinates, where the independent variables are time ($t$) and vehicle label ($n$).
As shown in Figure 1, the transformation establishes a direct link between the spatial position $x$ and the vehicle index $n$. A key contribution of this work is the rigorous proof that the viscosity solution to the Hamilton-Jacobi (H-J) equation in the Eulerian frame is equivalent to the one in the Lagrangian frame. This ensures stability even when traffic conditions are discontinuous, such as during the formation of shock waves (traffic jams).
To handle real-world data that may be noisy or conflicting, the authors utilize viability theory and the Lax-Hopf formula. This approach treats data as a set of constraints rather than exact points, providing a semi-analytical, grid-free solution. Unlike traditional simulations that require a 2D grid, the Lax-Hopf approach allows for calculating the traffic state at any specific point $(t, n)$ independently, making it highly parallelizable and efficient for real-time applications.
Experimental Validation
The model was validated using data from the Mobile Century field experiment conducted on a 3.45-mile stretch of I-880 Northbound in California. The dataset included:
- Eulerian Data: Loop detector counts recorded every 30 seconds.
- Lagrangian Data: GPS and velocity data from 97 probe vehicles recorded every 3 to 4 seconds.
The study used 44 vehicle trajectories as training data to reconstruct the traffic state and evaluated the model's accuracy against the remaining 53 vehicles.
Key Results and Insights
One significant advantage of the Lagrangian frame is that the relationship between vehicle spacing and speed (the fundamental diagram) is monotonically increasing, which simplifies the mathematical resolution compared to the concave relationship found in Eulerian coordinates.
The experimental results demonstrated high accuracy in several areas:
- Trajectory Reconstruction: The model successfully reconstructed individual vehicle paths with high correlation to actual sensing data.
- Velocity Field Visualization: The framework generated detailed velocity fields that clearly captured the transitions between free-flow and congested traffic.
- Travel Time Accuracy: Travel time estimation errors were remarkably low, with most errors falling within a ±20-second window for the multi-mile trip, even when probe data was sampled at lower frequencies (every 30 to 90 seconds).
Conclusion and Future Work
While this study provides a robust framework for linear freeway segments, several avenues for future research remain. Extending the Lagrangian PDE framework to complex network settings—including intersections, merges, and diverges—is a critical next step. Additionally, while the viability approach handles noisy data well, further work on data reconciliation is needed to address extreme outliers and ensure the model's reliability across diverse cyber-physical infrastructures.
Summary updated