Generalized Multi-hop Traffic Pressure for Heterogeneous Traffic Perimeter Control

Xiaocan Li, Xiaoyu Wang, Ilia Smirnov, Scott Sanner, Baher AbdulhaiarXiv 2024 · DOI · arXiv

Introduction

Urban traffic management increasingly relies on Perimeter Control (PC), a "gating" strategy designed to limit vehicle inflow into a protected city center to prevent gridlock. Traditional PC systems often assume a "homogeneous" city center where congestion is uniformly distributed. However, real-world urban environments are inherently heterogeneous; one sector may be severely congested while another remains clear.

Existing heterogeneous controllers typically utilize 1-hop pressure—the difference in traffic density between an entry point and the immediate downstream link. This approach is fundamentally myopic. A gate might permit entry because the immediate street is clear, unaware of a massive bottleneck just a few blocks further. This research introduces a generalized multi-hop downstream pressure metric to bridge the gap between local intersection-level views and city-level aggregate management.

Methodology: A Two-Stage Hierarchical Controller

The researchers propose a two-stage control scheme that balances macro-level stability with micro-level distribution efficiency.

  1. Stage 1 (Macro View): A pre-trained Deep Reinforcement Learning (DRL) agent monitors the density of the entire protected region. It determines the total permitted inflow required to maintain optimal network performance.
  2. Stage 2 (Micro Distribution): Instead of distributing the total inflow uniformly, a Multi-hop Pressure Calculator evaluates each entry gate based on downstream conditions.
The proposed hierarchical control framework, integrating a DRL-based macro controller with a multi-hop pressure-based micro distributor.

The Multi-hop Metric and Markov Chain Theory

The core innovation lies in the multi-hop metric, grounded in Markov chain theory. The urban network is modeled as a directed graph where vehicle movements are treated as state transitions. The metric calculates pressure by looking $h$ hops deep into the network, incorporating a "decaying influence" where distant congestion is weighted less than immediate traffic.

Graph representation of the traffic network, illustrating transition probabilities between links and the inclusion of a 'supersink' node (Ω) to represent trip completion.

To mathematically account for vehicles reaching their destinations, the authors introduce a "supersink" node ($\Omega$). The final allocation of inflow uses a Softmax function to prioritize feeder links that lead to less congested multi-hop paths.

Experimental Results

The system was tested using the AIMSUN traffic simulator on a grid network of 36 signalized intersections. The researchers evaluated the model against several benchmarks, including homogeneous PC and standard myopic (1-hop) heterogeneous PC.

Key Findings

  • Significant Efficiency Gains: In scenarios with highly imbalanced traffic, the 8-hop controller reduced the Total Time Spent (TTS) within the protected region by approximately 31% compared to the homogeneous approach (1979 hours vs. 2883 hours).
  • The 10-Hop Horizon: Performance improves as the "hop count" increases, but plateaus around 10 hops. Beyond this horizon, the influence of distant traffic decays to near zero, providing minimal additional benefit.
  • Robustness: Sensitivity analysis showed that the controller remains effective even when turning ratios are not perfectly known, outperforming baseline methods despite estimation errors.

Conclusion and Future Work

By replacing myopic local metrics with a "farsighted" multi-hop pressure metric, this approach significantly improves the resilience of urban networks to imbalanced demand. Future research will focus on incorporating dynamic turning ratio estimation to account for real-time incidents and testing the model on the irregular, "organic" topologies characteristic of older cities. Additionally, the authors suggest that integrating upstream potential—the volume of vehicles waiting to enter—could further refine the pressure balancing mechanism.

Summary updated

Abstract

Perimeter control (PC) prevents loss of traffic network capacity due to congestion in urban areas. Homogeneous PC allows all access points to a protected region to have identical permitted inflow. However, homogeneous PC performs poorly when the congestion in the protected region is heterogeneous (e.g., imbalanced demand) since the homogeneous PC does not consider specific traffic conditions around each perimeter intersection. When the protected region has spatially heterogeneous congestion, one needs to modulate the perimeter inflow rate to be higher near low-density regions and vice versa for high-density regions. A naïve approach is to leverage 1-hop traffic pressure to measure traffic condition around perimeter intersections, but such metric is too spatially myopic for PC. To address this issue, we formulate multi-hop downstream pressure grounded on Markov chain theory, which ``looks deeper'' into the protected region beyond perimeter intersections. In addition, we formulate a two-stage hierarchical control scheme that can leverage this novel multi-hop pressure to redistribute the total permitted inflow provided by a pre-trained deep reinforcement learning homogeneous control policy. Experimental results show that our heterogeneous PC approaches leveraging multi-hop pressure significantly outperform homogeneous PC in scenarios where the origin-destination flows are highly imbalanced with high spatial heterogeneity. Moveover, our approach is shown to be robust against turning ratio uncertainties by a sensitivity analysis.
BibTeX
@misc{li2024generalized,
  author = {Xiaocan Li and Xiaoyu Wang and Ilia Smirnov and Scott Sanner and Baher Abdulhai},
  title = {{Generalized Multi-hop Traffic Pressure for Heterogeneous Traffic Perimeter Control}},
  year = {2024},
  eprint = {2409.00753},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2409.00753},
}
Figure