Date of Award

2023

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Civil Engineering

Committee Chair

Rui Ma

Committee Member

Michael D. Anderson

Committee Member

Virginia Sisiopiku

Committee Member

Abdullahi Salman

Committee Member

Avimanyu Sahoo

Subject(s)

Traffice engineering, Machine learning, Neural networks (Computer science)

Abstract

In the quest to optimize traffic management through machine learning (ML), this dissertation delves into three foundational areas: Traffic state estimation (TSE), traffic pattern classification, and traffic crash detection. The first facet of this dissertation focuses on TSE. Traditionally, TSE methods are categorized into two approaches: model and data-driven methodologies, each with its limitations. Data-driven ML models can falter with limited training data or misleading samples. Moreover, their “black-box” nature makes them hard to interpret. This research introduces a physics-based neural network (PINN) framework that combines the strengths of both TSE methods. The presented approach uses limited observational speed data to produce high-quality TSEs across traffic networks. The Lighthill–Whitham–Richards (LWR) physical traffic flow model, coupled with a PINN, highlights the potential of this approach in improving highway network TSE accuracy. The second part of this research focuses on categorizing traffic speed patterns using unsupervised learning techniques. The study combines convolution neural networks (CNN) and K-means clustering to analyze reconstructed spatiotemporal images of speed data obtained from major corridors in the Huntsville metropolitan area. The data spans a consequential seventeen-month period, encompassing pre-, intra- and post-COVID-19 phases. The obtained patterns offer a detailed understanding of how the pandemic has influenced traffic behavior. The potential outcomes and uses of this analysis are vast. It sets the stage for determining the ideal data amount needed to comprehensively train models for accurate traffic forecasting. By assessing how traffic patterns adapt during major disruptions, city planners and decision-makers can craft stronger infrastructure and transport plans. Lastly, the research delves into the crucial field of traffic crash detection, an area with immense implications for road safety. The study employs graph neural networks (GNN) to construct a framework that can forecast traffic crashes by utilizing speed and traffic crash report data. The model detects the crash, its duration, and the impacted traffic links. In conclusion, employing ML algorithms to estimate traffic conditions, recognize patterns, and detect crashes holds great promise for improving traffic management. This dissertation has demonstrated the efficiency and precision of ML algorithms in carrying out these tasks.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.