An Exploration of Traffic Signal Control using Multi-agent Market-based Mechanisms

Raphael, J
(2018) An Exploration of Traffic Signal Control using Multi-agent Market-based Mechanisms. PhD thesis, University of Liverpool.

[img] Text
200983747_May2018.pdf - Unspecified

Download (20MB)


Traffic congestion is a major issue on many urban road networks around the world. The distributed and stochastic nature of traffic has attracted the multi-agent and market mechanism community to the traffic domain which has resulted in many novel approaches to both traffic control and traffic assignment. However, the real-world application of many market-based traffic control systems remains in question because they require technology that has not yet been developed, e.g., autonomous cars. This thesis focuses on the use of market mechanisms for traffic control, more specifically, the application of market principles set forth in market-based multi-robot systems to the traffic domain. Thus, the primary goal of this thesis is the design, implementation and evaluation of a multi-agent market-based traffic control system which does not rely on vehicle agents and other major changes to vehicles or transportation infrastructure. Evaluation of the traffic control system is conducted on two grid-based maps using six different traffic scenarios. The traffic scenarios represent various traffic patterns which include changes in traffic intensity and direction. The traffic scenarios are simulated in SUMO, an open source, macro traffic simulator. Additionally, performance is measured using three metrics: travel time, traffic density, and number of stops. This thesis makes five contributions: (i) demonstration of the efficacy of a novel multi-agent market-based traffic control methodology; (ii) demonstration of the efficacy of a market-based technique for dynamic coalition formation; (iii) analysis of three key traffic control parameters used by SCOOT, a popular urban adaptive traffic control mechanism used in over a dozen countries; (iv ) development of a Python implementation of SCOOT for use on SUMO and (v) a thorough evaluation of the novel market-based mechanisms introduced here, along with SCOOT and a reinforcement-learning traffic controller, over a variety of road traffic conditions. This thesis provides a unique insight into the behaviour of three key traffic control parameters and results show that the novel market-based mechanism has the potential to improve traffic performance in traffic conditions that are less than ideal for SCOOT.

Item Type: Thesis (PhD)
Divisions: Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User: Symplectic Admin
Date Deposited: 23 Aug 2018 12:38
Last Modified: 15 Apr 2021 07:15
DOI: 10.17638/03021623
  • Sklar, Elizabeth
  • Maskell, Simon