Dissertation
Big spatiotemporal data analytics of traffic congestion propagation patterns
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Spring 2023
DOI: 10.25820/etd.007122
Abstract
New approaches are needed to find geographic information embedded in big data and enhance large scale surveillance and prediction. In the context of vehicle movement, congestion can propagate to upstream segments as vehicles continue to queue up, leading to large-scale traffic congestion. If the spatiotemporal patterns of such dynamic process can be identified, city managers and traffic engineers will be better able to mitigate traffic congestion. Although recent research has led to new methods for the detection of spatiotemporal propagation patterns of traffic congestion, there is a lack of adequate evaluation as well as investigation into the use of data of high temporal granularity. The heterogeneous spatiotemporal patterns of traffic congestion propagation also make predicting future propagation difficult. Challenges, therefore, remain if such tools are to become of practical use. To address these challenges, three data driven methods are proposed. The first research task aims to detect congestion propagation patterns and comprehensively evaluate the proposed method using simulated data. The results show the proposed method is better than a state-of-the-art approach. Real-world cases are also presented to show the effectiveness of the proposed method given currently available data. The second research task addresses the problem of predicting congestion propagation patterns via a Markov based approach. A sparse matrix query algorithm is proposed to reduce the computational time of the method, and the Area under the ROC Curve (AUC) of predicted results ranges from 0.535 to 0.616 for most future time steps. The third research task provides an alternative method based on an existing deep-learning technique. The AUC of the alternative method is generally 0.25 higher than the Markov based approach.
Details
- Title: Subtitle
- Big spatiotemporal data analytics of traffic congestion propagation patterns
- Creators
- Haoyi Xiong
- Contributors
- David A Bennett (Advisor)Xun Zhou (Advisor)Caglar Koylu (Committee Member)Kang Zhao (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Geography
- Date degree season
- Spring 2023
- DOI
- 10.25820/etd.007122
- Publisher
- University of Iowa
- Number of pages
- ix, 79 pages
- Copyright
- Copyright 2023 Haoyi Xiong
- Language
- English
- Date approved
- 06/30/2023
- Description illustrations
- illustrations, tables, graphs
- Description bibliographic
- Includes bibliographical references (pages 75-79).
- Public Abstract (ETD)
- This project develops methods to detect the dynamic process of how large-scale traffic congestion evolves from a small road segment and predict such process in the near-future. Economic cost of traffic congestion is expected to rise from $166 billion in 2017 to $200 billion in 2025 in the U.S. Understanding the dynamic process of traffic congestion evolution will help city managers and traffic engineers mitigate future traffic congestion. The first method is designed to detect the evolution process of traffic congestion from vehicle trajectory data by making better usages of recently available fine-grained data than existing studies. The results show the proposed method always outperform the most recent existing study with simulated data. Real-world cases are also detected to show the effectiveness of the proposed method. Two methods are presented to predict the near-future evolution process of traffic congestion, which can also be used to estimate the potential impact of current traffic congestion on the future traffic. One is a traditional Markov-based method which simply assumes congestion is affected by nearby locations and time. The other one is a deep-learning method which trades computational time for better ability to capture the complex evolution process of traffic congestion. The results show the deep-learning method outperform the Markov-based method.
- Academic Unit
- Geographical and Sustainability Sciences
- Record Identifier
- 9984425393202771
Metrics
5 File views/ downloads
48 Record Views