Abstract
The emergence of connected and automated vehicles (CAV) indicates improved traffic mobility in future traffic transportation systems. This study addresses the research gap in macroscopic traffic modeling of mixed traffic networks where CAV and human-driven vehicles coexist. CAV behavior is explicitly included in the proposed traffic network model, and the vehicle number non-conservation problem is overcome by describing the approaching and departure vehicle number in discrete time. The proposed model is verified in typical CAV cooperation scenarios. The performance of CAV coordination is analyzed in road, intersection and network scenario. Total travel time of the vehicles in the network is proved to be reduced when coordination is applied. Simulation results validate the accuracy of the proposed model and the effectiveness of the proposed algorithm.
Introduction
As we move toward an autonomous future, we face a decades-long "mixed traffic" phase where Connected and Automated Vehicles (CAVs) and Human-Driven Vehicles (HDVs) share the road. Current macroscopic traffic models—which treat traffic like a fluid—often struggle with two primary issues:
- The Non-Conservation Problem: In discrete-time models, mathematical errors can cause vehicles to "leak," essentially appearing or disappearing during state updates.
- CAV Blindness: Traditional models often fail to differentiate between a human driver’s reaction time and a CAV’s ability to follow closely or receive global routing commands.
This paper introduces a novel macroscopic traffic model designed for this transition era, explicitly accounting for CAV-specific behaviors and fixing the "disappearing vehicle" problem.
Methodology
The researchers built their framework on a modified "S model" (a simplified version of the Berg-Lin-Xi model). Key technical innovations include:
- Discrete-Time Conservation: By redefining how vehicles approach and depart queuing zones in discrete time steps ($T_s$), the model ensures every vehicle entering a road segment is accounted for upon exit.
- MPR-Aware Density Modeling: The model treats the Market Penetration Rate (MPR) as a dynamic state variable, adjusting the maximum traffic flow density based on the shorter headways (following distances) that CAVs can safely maintain.
Three-Tier Coordination
The model implements coordination at three distinct levels:
- Road Level: Simulating Cooperative Adaptive Cruise Control (CACC) and platooning.
- Intersection Level: Synchronizing CAV starts to eliminate the "slingshot" delay seen when traffic lights turn green.
- Network Level: A global route allocation algorithm using a Genetic Algorithm to minimize the Total Travel Time (TTS) across the entire grid.
Simulation and Parameters
The study relies on high-fidelity simulations using SUMO (Simulation of Urban MObility) for microscopic verification. The simulated environment consists of a 12-node urban grid with 200m road segments and a 60 km/h speed limit. Five distinct flow indices were tested, ranging from light traffic (720 veh/h) to heavy congestion (3600 veh/h), with CAV penetration rates varying from 20% to 100%.
Key Insights and Results
- The 20% Tipping Point: The benefits of CAV coordination on average vehicle speed become obvious once the MPR exceeds 20%. Below this, HDVs create too much disruption for CAV platooning to stabilize the flow.
- Uni-modal Optimization: Road coordination provides the most significant speed boosts at medium densities. At very low density, there is no traffic to optimize; at extreme gridlock, even CAVs lack the space to maneuver.
- Speed and Travel Time: In high-density scenarios, coordination increased average speeds from ~42 km/h to the 60 km/h limit at 100% MPR. Under heavy flow, the coordination algorithm reduced Total Travel Time (TTS) from approximately 122.4s to 119.6s as MPR increased.
- Conservation Accuracy: The model successfully resolved the vehicle number non-conservation issue, providing a more reliable foundation for real-time traffic management.
Conclusion and Future Work
While the model is a significant step forward, future research could expand the objective function to include multi-objective optimization for fuel economy and environmental impact. Additionally, integrating more stochastic human driving behaviors and utilizing Reinforcement Learning could help the model handle the massive state space of larger, more complex city networks in real-time.
Summary updated