Estimation of Traffic Stream Density Using Connected Vehicle Data: Linear and Nonlinear Filtering Approaches

Area of Science:

• Transportation Engineering
• Traffic Flow Theory
• Data Science

Background:

• Estimating traffic stream density on signalized approaches is crucial for traffic management.
• Connected vehicle (CV) data offers a novel data source for traffic state estimation.
• Nonlinear filtering methods, like particle filters (PF), are often employed for complex state estimation problems.

Purpose of the Study:

• To develop and evaluate a nonlinear filtering approach, specifically a particle filter (PF), for estimating traffic stream density using connected vehicle (CV) data.
• To compare the performance of the PF with linear estimation techniques, namely the Kalman filter (KF) and adaptive Kalman filter (AKF).
• To analyze the sensitivity of these estimation techniques to factors like CV penetration rate and initial conditions.

Main Methods:

• A particle filter (PF) was developed using CV travel-time measurements.
• State and measurement equations were derived from traffic flow continuity and hydrodynamic traffic flow relationships, respectively.
• The PF was compared against a Kalman filter (KF) and an adaptive Kalman filter (AKF) using simulated data under oversaturated conditions.

Main Results:

• All three techniques (PF, KF, AKF) produced accurate traffic density estimates.
• The Kalman filter (KF) surprisingly yielded the most accurate estimates among the tested methods.
• Estimation accuracy improved with increased CV market penetration and, for the PF, with a higher number of particles.
• The KF demonstrated less sensitivity to initial conditions compared to the PF.

Conclusions:

• A simple linear estimation approach, specifically the Kalman filter (KF), is highly effective and suitable for estimating traffic density using connected vehicle data on signalized approaches.
• Despite the complexity of nonlinear methods, linear filters can provide superior or comparable performance in this specific application.
• The findings highlight the potential of leveraging CV data for real-time traffic monitoring and management, emphasizing the robustness of linear filtering techniques.