Open Collections

Ghoul, Tarek

University of British Columbia

The proliferation of Connected Vehicles and their ability to collect a large amount of data present an opportunity for real-time safety optimization of traffic networks. At intersections, Adaptive Traffic Signal Control (ATSC) systems and dynamic speed advisories are among the proactive real-time safety interventions that can assist in preventing rear-end collisions. This thesis proposes two systems that utilize connected vehicle data to optimize traffic safety in real-time using reinforcement learning approaches. The first system utilizes a Deep Deterministic Policy Gradient (DDPG) reinforcement learning agent in conjunction with a dynamic programming approach to optimize vehicle trajectories and issue speed advisories. The second proposed system is a Signal-Vehicle Coupled Control (SVCC) system incorporating ATSC and speed advisories to optimize safety in real-time. By applying a rule-based approach in conjunction with a Soft-Actor Critic Reinforcement Learning framework, the system assigns speed advisories to platoons of vehicles on each approach and extends the current signal time accordingly. Dynamic traffic parameters are collected in real time and used to estimate the current conflict rate at the intersection, which is then processed and input into the respective models. The systems were tested on two different intersections modelled using real-world data through the simulation platform VISSIM. Significant reductions in traffic conflicts and delay were observed, with the simple speed advisory system yielding a 9-23% reduction in traffic conflicts and the SVCC system yielding a 41-55% reduction. Similarly, delay reductions of about 24% were observed. Both systems function at lower levels of market penetration, with diminishing returns beyond 50% Market Penetration Ratio (MPR). The thesis thus proposes and demonstrates the effectiveness of two unique CV-based systems that are low in computational intensity and applicable in the near future.

Text

2021-09-01

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/79571

Civil Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/