東華大學圖書館 |

Low-Rank Models for Spatiotemporal Traffic Data Analysis /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Low-Rank Models for Spatiotemporal Traffic Data Analysis // Xudong Wang.
Author:	Wang, Xudong,
Description:	1 electronic resource (175 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 86-04, Section: B.
Contained By:	Dissertations Abstracts International86-04B.
Subject:	Sparsity. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31527353
ISBN:	9798342121507

Low-Rank Models for Spatiotemporal Traffic Data Analysis /
Wang, Xudong,

Low-Rank Models for Spatiotemporal Traffic Data Analysis /Xudong Wang. - 1 electronic resource (175 pages)

Source: Dissertations Abstracts International, Volume: 86-04, Section: B.

Smart cities have gained substantial attention due to rapid urbanization and the need to transform congested urban areas into sustainable, efficient, and safe environments. Transportation is a critical element in ensuring smooth and efficient mobility within the smart city framework. With the widespread deployment of ubiquitous sensing technology and advanced intelligent transportation systems (ITS), extensive amounts of urban traffic data, characterized by high spatial and temporal resolution, are routinely collected.Although machine learning and big data techniques have rapidly advanced in the realm of transportation, many general-purpose learning-based methods may not be suitable for traffic data. This is primarily attributed to intricate spatiotemporal correlations, dynamic characteristics, scalability challenges, and the presence of missing values and anomalies in the data. Therefore, there is a compelling need to develop customized machine learning approaches tailored to the unique characteristics of traffic-related tasks. To address these challenges, this thesis focuses on the development of data-driven methodologies, specifically utilizing low-rank models, to improve the understanding of urban transportation systems.This thesis comprises four scientific papers, each with a distinct focus on research objectives: addressing missing data, identifying mobility patterns, and detecting anomalies. To achieve these goals, low-rank models are employed. These models are designed to represent data with a reduced-dimensional structure, often characterized by a more concise set of features or latent patterns.In the first paper, a Hankel-structured tensor low-rank model is introduced to estimate complete traffic data from sparse mobile sensor observations. Unlike traditional methods relying on predefined traffic models or extensive simulations, this method formulates the problem as spatiotemporal Hankel tensor completion. Assuming a low-rank structure for the Hankel tensor enables data-driven approximations of both global and local traffic patterns.Subsequently, building upon the complete traffic speed dataset, a low-rank model called the anti-circulant dynamic mode decomposition (circDMD) is proposed to explore the mobility patterns of highway systems. It can extract spatial-temporal coherent structures with explicit physical interpretations through a data-driven approach. This framework aids in identifying traffic bottlenecks and enables long-term predictions.In the third part of this thesis, a dynamic framework is introduced for the detection of anomalies in highway systems. This framework utilizes a low-rank tensor structure to represent the temporal variations in system dynamics, providing a robust tool for uncovering abnormal generative mechanisms and unexpected temporal patterns in traffic networks.Finally, the primary emphasis is on the detection of anomalies when working with partial data observations. A robust matrix completion method with an anti-circulant structure is introduced. This method capitalizes on the periodicity evident in traffic data, revealed through a short-time Fourier transform (STFT), enabling a more profound understanding of localized changes in the frequency content over time.In summary, this thesis represents an exploration of low-rank models as a means to address several critical challenges in traffic data. The proposed methods contribute to the field of transportation research by providing tailored solutions for addressing missing data imputation, analyzing mobility patterns, and enhancing anomaly detection.

English

ISBN: 9798342121507Subjects--Topical Terms:

3680690
Sparsity.

Low-Rank Models for Spatiotemporal Traffic Data Analysis /
LDR:08470nmm a22003733i 4500 001 2434472
005 20250912102024.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798342121507
035 $a (MiAaPQD)AAI31527353
035 $a (MiAaPQD)McGillj098zh50q
035 $a AAI31527353
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Wang, Xudong, $e author. $3 3824465
245 1 0 $a Low-Rank Models for Spatiotemporal Traffic Data Analysis / $c Xudong Wang.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (175 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-04, Section: B.
500 $a Advisors: Sun, Lijun.
502 $b Ph.D. $c McGill University (Canada) $d 2024.
520 $a Smart cities have gained substantial attention due to rapid urbanization and the need to transform congested urban areas into sustainable, efficient, and safe environments. Transportation is a critical element in ensuring smooth and efficient mobility within the smart city framework. With the widespread deployment of ubiquitous sensing technology and advanced intelligent transportation systems (ITS), extensive amounts of urban traffic data, characterized by high spatial and temporal resolution, are routinely collected.Although machine learning and big data techniques have rapidly advanced in the realm of transportation, many general-purpose learning-based methods may not be suitable for traffic data. This is primarily attributed to intricate spatiotemporal correlations, dynamic characteristics, scalability challenges, and the presence of missing values and anomalies in the data. Therefore, there is a compelling need to develop customized machine learning approaches tailored to the unique characteristics of traffic-related tasks. To address these challenges, this thesis focuses on the development of data-driven methodologies, specifically utilizing low-rank models, to improve the understanding of urban transportation systems.This thesis comprises four scientific papers, each with a distinct focus on research objectives: addressing missing data, identifying mobility patterns, and detecting anomalies. To achieve these goals, low-rank models are employed. These models are designed to represent data with a reduced-dimensional structure, often characterized by a more concise set of features or latent patterns.In the first paper, a Hankel-structured tensor low-rank model is introduced to estimate complete traffic data from sparse mobile sensor observations. Unlike traditional methods relying on predefined traffic models or extensive simulations, this method formulates the problem as spatiotemporal Hankel tensor completion. Assuming a low-rank structure for the Hankel tensor enables data-driven approximations of both global and local traffic patterns.Subsequently, building upon the complete traffic speed dataset, a low-rank model called the anti-circulant dynamic mode decomposition (circDMD) is proposed to explore the mobility patterns of highway systems. It can extract spatial-temporal coherent structures with explicit physical interpretations through a data-driven approach. This framework aids in identifying traffic bottlenecks and enables long-term predictions.In the third part of this thesis, a dynamic framework is introduced for the detection of anomalies in highway systems. This framework utilizes a low-rank tensor structure to represent the temporal variations in system dynamics, providing a robust tool for uncovering abnormal generative mechanisms and unexpected temporal patterns in traffic networks.Finally, the primary emphasis is on the detection of anomalies when working with partial data observations. A robust matrix completion method with an anti-circulant structure is introduced. This method capitalizes on the periodicity evident in traffic data, revealed through a short-time Fourier transform (STFT), enabling a more profound understanding of localized changes in the frequency content over time.In summary, this thesis represents an exploration of low-rank models as a means to address several critical challenges in traffic data. The proposed methods contribute to the field of transportation research by providing tailored solutions for addressing missing data imputation, analyzing mobility patterns, and enhancing anomaly detection.
520 $a Les villes intelligentes ont suscite une attention considerable en raison de l'urbanisation rapide et de la necessite de transformer les zones urbaines congestionnees en environnements durables, efficaces et securises. Les transports sont un element essentiel pour garantir une mobilite fluide et efficace dans le cadre des villes intelligentes. Avec le deploiement generalise de la technologie de detection omnipresente et des systemes avances de transports intelligents (ITS), d'importantes quantites de donnees urbaines sur la circulation, caracterisees par une resolution spatiale et temporelle elevee, sont regulierement collectees.Alors que les techniques d'apprentissage automatique et de gestion des megadonnees ont rapidement progresse dans le domaine des transports, de nombreuses methodes d'apprentissage generalistes peuvent ne pas etre adaptees aux donnees de circulation. Cela est principalement du aux correlations spatiotemporelles complexes, aux caracteristiques dynamiques, aux defis d'evolutivite, ainsi qu'a la presence de valeurs manquantes et d'anomalies dans les donnees. Par consequent, il est imperatif de developper des approches d'apprentissage automatique personnalisees adaptees aux caracteristiques uniques des taches liees a la circulation. Pour relever ces defis, cette these se concentre sur le developpement de methodologies axees sur les donnees, en particulier en utilisant des modeles a faible rang, pour ameliorer la comprehension des systemes de transports urbains.Cette these se compose de quatre articles scientifiques, chacun ayant un objectif de recherche distinct : traiter les donnees manquantes, identifier les schemas de mobilite et detecter les anomalies. Pour atteindre ces objectifs, des modeles a faible rang sont utilises. Ces modeles sont concus pour representer les donnees avec une structure de dimension reduite, souvent caracterisee par un ensemble plus concis de caracteristiques ou de schemas latents.Dans le premier article, un modele a faible rang de tenseur structure de Hankel est introduit pour estimer des donnees de circulation completes a partir d'observations spoiii radiques de capteurs mobiles. Contrairement aux methodes traditionnelles qui reposent sur des modeles de circulation predefinis ou des simulations etendues, cette methode formule le probleme comme une completion de tenseur Hankel spatiotemporel. En supposant une structure a faible rang pour le tenseur Hankel, elle permet des approximations basees sur les donnees des schemas de circulation a la fois globaux et locaux.Ensuite, en s'appuyant sur l'ensemble de donnees complet de vitesse de circulation, un modele a faible rang appele decomposition dynamique en mode anticyclique (circDMD) est propose pour explorer les schemas de mobilite des reseaux routiers. Il peut extraire des structures coherentes spatiales-temporelles avec des interpretations physiques explicites grace a une approche basee sur les donnees. Ce cadre aide a identifier les points de congestion routiere et permet des previsions a long terme.Dans la troisieme partie de cette these, un cadre dynamique est introduit pour la detection d'anomalies dans les reseaux routiers. Ce cadre utilise une structure de tenseur a faible rang pour representer les variations temporelles de la dynamique du systeme, offrant un outil robuste pour decouvrir les mecanismes generateurs anormaux et les schemas temporels inattendus dans les reseaux de circulation.Enfin, l'accent principal est mis sur la detection des anomalies lors de l'utilisation d'observations de donnees partielles. Une methode de completion de matrice robuste avec une structure anticyclique est introduite.
546 $a English
590 $a School code: 0781
650 4 $a Sparsity. $3 3680690
650 4 $a Missing data. $3 3697670
650 4 $a Speed limits. $3 3705129
650 4 $a Fourier transforms. $3 3545926
650 4 $a Dynamical systems. $3 3219613
650 4 $a Mathematics. $3 515831
690 $a 0405
710 2 $a McGill University (Canada). $e degree granting institution. $3 3759190
720 1 $a Sun, Lijun $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-04B.
790 $a 0781
791 $a Ph.D.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31527353