Data-Driven Intersection Management Solutions for Mixed Traffic of Human-Driven and Connected and Automated Vehicles

TL;DR

This paper introduces two data-driven solutions for urban intersection management involving connected and automated vehicles, improving traffic flow and reducing delays without additional sensors.

Contribution

It presents a novel centralized platoon-based controller and an adaptive signal control system that leverages data-driven methods and V2I communication, reducing costs and enhancing efficiency.

Findings

The neural network-based signal timing achieves up to 25% delay reduction.

Estimation error decreases with higher market penetration of connected vehicles.

The proposed system outperforms traditional traffic control methods.

Abstract

This dissertation proposes two solutions for urban traffic control in the presence of connected and automated vehicles. First a centralized platoon-based controller is proposed for the cooperative intersection management problem that takes advantage of the platooning systems and V2I communication to generate fast and smooth traffic flow at a single intersection. Second, a data-driven approach is proposed for adaptive signal control in the presence of connected vehicles. The proposed system relies on a data-driven method for optimal signal timing and a data-driven heuristic method for estimating routing decisions. It requires no additional sensors to be installed at the intersection, reducing the installation costs compared to typical settings of state-of-the-practice adaptive signal controllers. The proposed traffic controller contains an optimal signal timing module and a traffic state estimator. The signal timing module is a neural network model trained on microscopic simulation data to achieve optimal results according to a given performance metric such as vehicular delay or average queue length. The traffic state estimator relies on connected vehicles' information to estimate the traffic's routing decisions. A heuristic method is proposed to minimize the estimation error. With sufficient parameter tuning, the estimation error decreases as the market penetration rate (MPR) of connected vehicles grows. Estimation error is below 30% for an MPR of 10% and it shrinks below 20% when MPR grows larger than 30%. Simulations showed that the proposed traffic controller outperforms Highway Capacity Manual's methodology and given proper offline parameter tuning, it can decrease average vehicular delay by up to 25%.