Traffic signal phasing optimisation using enhanced q-network (EDQN)
Loading...
Supplementary material
Other Title
Authors
Cui, Jiping
Author ORCID Profiles (clickable)
Degree
Master of Applied Technologies (Computing)
Grantor
Unitec, Te Pūkenga – New Zealand Institute of Skills and Technology
Date
2024
Supervisors
Keivanmarz, Ali
Sharifzadeh, Hamid
Sharifzadeh, Hamid
Type
Masters Thesis
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
traffic signs and signals
traffic flows
traffic management
algorithms
traffic flows
traffic management
algorithms
ANZSRC Field of Research Code (2020)
Citation
Cui, J. (2024). Traffic signal phasing optimisation using enhanced q-network (EDQN) (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Applied Technologies (Computing)). Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology
https://hdl.handle.net/10652/6843
Abstract
Traffic signal control is key in managing urban traffic flows and volumes. How ever, traditional traffic signal control methods typically struggle to face real-time traffic fluctuations, leading to frequent congestion. In contrast, reinforcement learn ing (RL), a machine learning approach that learns and optimises behaviour from its environment, offers an adaptive way to adjust signal control strategies. It can maximise traffic flow and minimise vehicle delay compared to conventional methods.
Against this backdrop, this research aims to optimise signal phasing by reducing the queue length of vehicles waiting at the intersection and improving travelling efficiency by using reinforcement learning (RL) used in traffic signal phasing control, summarising key theoretical foundations related to both fields. It then discusses commonly used microscopic traffic simulation software, such as Vissim, Simulation of Urban MObility (SUMO), and Paramics.
Focusing on optimising traffic signal phasing, this research proposes an enhanced Deep Q-networks (DQN) called EDQN. The control problem is modelled with a Markov Decision Process (MDP), having the speed and position of every vehicle within 140 meters of the intersection serving as inputs. To demonstrate the model’s optimisation, we define the reward function with the weighted sum of standardised metrics, including the average queue length of all vehicles waiting on the lane before entering the intersection.
The simulation results indicate that, when we compare with other models such as fixed-time control, actuated control, and the standard DQN algorithm, the improved EDQN algorithm consistently delivers the best performance that the queue length of vehicles has been decreasing obviously and convergence speed across light, moderate, and heavy traffic conditions that are faster than other control methods. The results conclude that the proposed EDQN model effectively enhances vehi cle throughput and improves overall traffic efficiency. Reducing the queue length significantly enhances traffic signal control at single intersections, achieving over 70% optimisation across three types of traffic conditions and significantly improving overall traffic signal phasing in the queue length metric.
Publisher
Permanent link
Link to ePress publication
DOI
Copyright holder
Author
Copyright notice
All rights reserved