A Distributed Architecture for Real-time Hybrid Traffic Light Control in Urban Transportation Networks

TL;DR

This paper introduces a distributed, real-time hybrid traffic light control system for urban networks, combining macroscopic traffic modeling, machine learning, and hybrid control strategies to improve traffic flow and convergence.

Contribution

It presents a novel hybrid control architecture integrating machine learning-based congestion detection with distributed traffic light management.

Findings

Convergence achieved under cyclic flow profiles

Effective congestion level identification via machine learning

Hybrid control strategy improves traffic flow management

Abstract

A macroscopic model is proposed to depict the traffic dynamics involved in urban traffic systems. The link dynamics are described based on the cell-transmission model and bounded by the link capacities, while the flow dynamics are proposed based on the discharge headways and saturation flow at intersections. To fulfill the requirement of a closed-loop traffic light control strategy, an approach to estimate the branching ratios at intersections is proposed and simulations show that the convergence would be achieved under constant cyclic flow profiles. Furthermore, a system partitioning approach is proposed based congestion level identification, which is achieved via a machine learning method and a hybrid traffic network control strategy is proposed to integrate different traffic light control schemes together.