Date of Award

Summer 2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Civil/Environmental Engineering

Program/Concentration

Civil Engineering

Committee Director

Mecit Cetin

Committee Member

Sherif S. Ishak

Committee Member

Hong Yang

Abstract

Connected vehicles (CVs) will enable new applications to improve traffic flow. The focus of this dissertation is to investigate how reinforcement learning (RL) control for the variable speed limit (VSL) through CVs can be generalized to improve traffic flow at different freeway bottlenecks. Three different bottlenecks are investigated: A sag curve, where the gradient changes from negative to positive values causes a reduction in the roadway capacity and congestion; a lane reduction, where three lanes merge to two lanes and cause congestion, and finally, an on-ramp, where increase in demand on a multilane freeway causes capacity drop. An RL algorithm is developed and implemented in a simulation environment for controlling a VSL in the upstream to manipulate the inflow of vehicles to the bottleneck on a freeway to minimize delays and increase the throughput. CVs are assumed to receive VSL messages through Infrastructure-to-Vehicle (I2V) communications technologies. Asynchronous Advantage Actor-Critic (A3C) algorithms are developed for each bottleneck to determine optimal VSL policies. Through these RL control algorithms, the speed of CVs are manipulated in the upstream of the bottleneck to avoid or minimize congestion. Various market penetration rates for CVs are considered in the simulations. It is demonstrated that the RL algorithm is able to adapt to stochastic arrivals of CVs and achieve significant improvements even at low market penetration rates of CVs, and the RL algorithm is able to find solution for all three bottlenecks. The results also show that the RL-based solutions outperform feedback-control-based solutions.

DOI

10.25777/dc8z-c232

ORCID

0000-0001-5251-0748

Share

COinS