Intelligent urban traffic signal control using multiagent reinforcement learning
Abstract/Contents
- Abstract
- Urban traffic management plays an important role in building smart and sustainable cities, from social, economical, and environmental points of view. However, modern traffic signal control systems have no been significantly improved over the last four decades despite extensive studies conducted in urban transportation research. One major reason is that most traditional signal control models rely heavily on statistical assumptions and heuristic criteria, which cannot support adaptive microscopic signal controls, while most proposed intelligent signal control methods are not robust and scalable enough for managing real-world large-scale urban traffic flows. This thesis proposes a systematic and realistic approach for implementing reinforcement learning (RL) to design adaptive, optimal, and scalable intelligent urban signal control systems. The direct adoption of existing RL models is impractical for large-scale signal control because of the extremely high computational complexity of learning and searching of control policies. To overcome this difficulty, this thesis proposes a practical multiagent RL framework that is scalable for large-scale traffic signal control, by applying various approximation methods. These methods are developed from three major perspectives: (1) integrating heuristic rules in traditional urban transportation studies to reduce the complexity of control policy searching; (2) applying state-of-the-art supervised learning techniques to achieve efficient feature extraction and Q-function approximation; and (3) implementing distributed control techniques to achieve effective communication and coordination in large-scale traffic networks. To evaluate the performance of this MARL framework, various numerical experiments are conducted in microscopic traffic simulations. The results demonstrate that proposed MARL methods outperform the major traditional and intelligent signal control methods, with a linear computational time complexity which is acceptable for real-world applications.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2016 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Chu, Tianshu |
|---|---|
| Associated with | Stanford University, Department of Civil and Environmental Engineering. |
| Thesis advisor | Leckie, Jim, 1939- |
| Thesis advisor | Lepech, Michael |
| Thesis advisor | Wang, Jie |
| Advisor | Lepech, Michael |
| Advisor | Wang, Jie |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Tianshu Chu. |
|---|---|
| Note | Submitted to the Department of Civil and Environmental Engineering. |
| Thesis | Thesis (Ph.D.)--Stanford University, 2016. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2016 by Tianshu Chu
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...