| Author: | Hu, Zijian |
| Title: | Relieving network congestion : an AI-empowered traffic management framework for large-scale urban road networks |
| Advisors: | Ma, Wei (CEE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Subject: | Traffic congestion -- Prevention Traffic engineering -- Data processing Artificial intelligence Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Civil and Environmental Engineering |
| Pages: | xvi, 193 pages : color illustrations |
| Language: | English |
| Abstract: | Traffic congestion is one of the most severe urban issues in many metropolises. To relieve the traffic congestion, in the thesis, we formulate a traffic management framework with three elementary components, sensing, modeling and managing. Traffic sensing is the method that collects traffic state information (i.e., speed, density, and flow) in real-world traffic networks from specialized sensors such as loop detectors and cameras. Traffic modeling aims to reproduce traffic dynamics in the real world through analytical formulation and agent-based simulation. Traffic managing consists of a series of control policies to improve traffic efficiency and safety. These components are tightly connected and function synergistically in the system. However, the urban road network is extensive and heterogeneous. Practically, it is challenging to extend the sensing, modeling and managing techniques to entire road networks due to the limited financial budgets and exponentially increasing difficulty in solutions. Therefore, the major research question in this thesis is, how we can relieve the traffic congestion and increase the traffic efficiency on the scope of the entire urban road network. To answer this question, this thesis presents a holistic Artificial Intelligence (AI)-empowered framework for traffic management. Moreover, we decompose the aforementioned research question into three sub-questions: how can we fulfill efficient and effective sensing/modeling/managing components on the scope of the entire urban road network, and answer them with AI-empowered methods separately. For the sensing component, we complete the traffic-related information in sensor-free zones in two ways. Firstly, we attempt to estimate traffic-related information from other types of sensors. Specifically, we develop a holistic framework to estimate the traffic density from surveillance cameras since cameras are widely spread in most metropolises. We decouple the traffic density estimation into two parts: camera calibration and vehicle detection, and solve them through non-linear programming and deep learning methods. Secondly, we aim to infer traffic-related information from existing sensors. In this way, we propose an attention-based graph neural network to improve the accuracy of the network-wide traffic flow estimation by incorporating Global-Open Multi-Source (GOMS) data. Both methods have been validated in multiple cities with decent accuracy. For the modeling component, we propose a hybrid "meso-macro" traffic modeling tool to solve the gridlock issue and to enhance the modeling accuracy and efficiency under traffic congestion. The model resolution in local areas can be adjusted according to road characteristics and user settings, and multiple resolutions in different areas can be simultaneously achieved in a hybrid traffic model in a single simulation. We validate the modeling efficiency in large-scale networks, and we further develop a calibrated scenario with real-world data to reflect the traffic dynamics in real-world scenes, which provides a foundation for subsequent traffic managing. For the managing component, we categorize the road network into freeways and urban roads, and we propose a demonstration-guided Deep Reinforcement Learning (DRL) method for coordinated ramp metering and perimeter control on freeways and urban roads in the entire road network. The demonstration guidance from traditional controllers is leveraged to improve the convergence of the vanilla DRL method in large-scale road networks. Numerical experiments have shown that the proposed method can achieve better performance in multiple scenarios. We also validate the proposed method in a real-world large-scale network in Hong Kong. The thesis outcome is to achieve a systematic framework that senses traffic conditions, models the traffic operations and provides adaptive management policies in the realm of urban networks, which improves the efficiency, safety and sustainability of urban mobility. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13680