Coordinated Intersection Control Through Reinforcement Learning with Special Consideration of Freight Traffic
-
2021-12-14
-
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report
-
Corporate Publisher:
-
Abstract:This project specially studies signal timing with special consideration of freight traffic in urban areas. The rationale is that freight logistics are critical to quality of life and economies. However, freight mobility, especially along major freight corridors in urban areas, rarely gets special consideration in signal timing. The advent of the Internet of Things (IoT) makes vast information collection a reality. The rich data environment, combined with the boost in computational power, has brought unprecedented opportunities closer to reality than ever before for real-time, information-driven intersection traffic control under variants of traffic scenarios. The research advances the conventional traffic signal control through introduction of control theories and reinforcement learning methods to design highly efficient network control algorithms. This research focuses on developing a new traffic responsive network signal control in general, and specially with freight traffic considered. When dealing with network signal control, unlike the traditional formulations that either face challenges to quantify promptly such as total delay or using simple linear combinations of observations as reinforcement learning’s reward that lack a theoretical basis (e.g., sum of weighted waiting time and queue length). Hence, this study first utilizes Lyapunov optimization to minimize the long-term average queue across the network and proposes backpressure as the network performance measure. Then the study builds a network signal control algorithm with reinforcement learning (RL) that utilizes backpressure as reward and uses double Deep Q-Network (Double-DQN) in the training process. The proposed algorithm is compared with traditional transportation methods and other RL-based methods.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: