서브메뉴
검색
Spatial-Temporal Traffic Flow Recovery and Prediction in Large-Scale Urban Transportation Networks
Spatial-Temporal Traffic Flow Recovery and Prediction in Large-Scale Urban Transportation Networks
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153121
- ISBN
- 9798346872405
- DDC
- 385
- 저자명
- Fu, Sicheng.
- 서명/저자
- Spatial-Temporal Traffic Flow Recovery and Prediction in Large-Scale Urban Transportation Networks
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 162 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-06, Section: B.
- 주기사항
- Advisor: Ran, Bin.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2024.
- 초록/해제
- 요약In urban transportation systems, comprehensive and accurate traffic flow information is essential for effective planning, infrastructure development, route optimization, and enhancing Intelligent Transportation Systems (ITS) to improve safety, mobility, and sustainability. However, significant spatial and temporal challenges hinder the efficient use of this data. Spatially, urban traffic detectors are often sparsely placed due to cost and physical constraints, leading to data gaps that limit network coverage and monitoring accuracy. Temporally, urban traffic flow is highly variable, influenced by factors such as road types, lane configurations, intersections, and recurring congestion patterns, making it difficult to rely solely on historical and real-time data for proactive decision-making. To address these challenges, this thesis introduces a comprehensive approach to urban traffic flow recovery and prediction, utilizing a vehicle-road-cloud architecture within the ITS to support obtaining traffic flow information across the city network in both spatial and temporal dimensions.To address spatial data gaps, this dissertation proposes two urban traffic flow recovery methods based on a sparse optimization framework and an analytical optimization framework. These methods utilize GPS speed data from connected automated vehicles (CAVs) alongside sparse flow data to achieve comprehensive, network-wide traffic flow recovery. Both methods employ a dynamic traffic assignment matrix to connect large-scale network speed data with individual link-level flows. The Sparse Recovery method utilizes the LASSO framework to induce sparsity, which is particularly effective when link flow data is missing due to limited sensor coverage. The Analytical Recovery method formulates the optimization problem using a quadratic objective function, which offers intuitive insights into traffic flow dynamics by using Stochastic Gradient Descent (SGD) and Lagrange Relaxation (LR) for parameter fine-tuning. These methods were validated through both a real-world case study and a hypothetical case study in Futian District, Shenzhen. Utilizing a city-level simulation platform built with SUMO to replicate urban traffic, both methods demonstrated close alignment with actual traffic flows and consistently maintained low estimation errors. This underscores their potential for effective large-scale traffic flow recovery using GPS data, even when flow observations are limited.To extend the temporal dimension of traffic flow data, this thesis further introduces a Dynamic Urban Spatial Temporal Graph Convolutional Network (DUST-GCN) that leverages historical traffic data and road information integrated to predict future urban traffic flow. Accounting for the dynamic and heterogeneous nature of urban traffic, DUST-GCN incorporates a temporal-spatial attention mechanism to capture dynamic dependencies, an adaptive graph structure to reflect changing inter-road relationships, and road-specific embeddings that capture the unique characteristics of various road types. To improve long-term prediction accuracy, a hybrid periodic input design integrates daily and weekly patterns, supported by GRU layers for long-term temporal modeling. Using newly developed real-world urban traffic datasets with corresponding road information for experiments, DUST-GCN consistently outperformed baseline models across different urban traffic scenarios, proving its reliability and adaptability.Overall, the traffic flow recovery and prediction approaches developed in this work provide comprehensive spatial-temporal traffic information across city networks. These advancements support urban traffic departments in traffic management, scheduling, control, and work zone deployment, offering robust tools for proactive and efficient urban transportation planning.
- 일반주제명
- Transportation
- 기타저자
- The University of Wisconsin - Madison Civil & Environmental Engr
- 기본자료저록
- Dissertations Abstracts International. 86-06B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017165079
■00520250211153121
■006m o d
■007cr#unu||||||||
■020 ▼a9798346872405
■035 ▼a(MiAaPQ)AAI31762613
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a385
■1001 ▼aFu, Sicheng.
■24510▼aSpatial-Temporal Traffic Flow Recovery and Prediction in Large-Scale Urban Transportation Networks
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a162 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-06, Section: B.
■500 ▼aAdvisor: Ran, Bin.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2024.
■520 ▼aIn urban transportation systems, comprehensive and accurate traffic flow information is essential for effective planning, infrastructure development, route optimization, and enhancing Intelligent Transportation Systems (ITS) to improve safety, mobility, and sustainability. However, significant spatial and temporal challenges hinder the efficient use of this data. Spatially, urban traffic detectors are often sparsely placed due to cost and physical constraints, leading to data gaps that limit network coverage and monitoring accuracy. Temporally, urban traffic flow is highly variable, influenced by factors such as road types, lane configurations, intersections, and recurring congestion patterns, making it difficult to rely solely on historical and real-time data for proactive decision-making. To address these challenges, this thesis introduces a comprehensive approach to urban traffic flow recovery and prediction, utilizing a vehicle-road-cloud architecture within the ITS to support obtaining traffic flow information across the city network in both spatial and temporal dimensions.To address spatial data gaps, this dissertation proposes two urban traffic flow recovery methods based on a sparse optimization framework and an analytical optimization framework. These methods utilize GPS speed data from connected automated vehicles (CAVs) alongside sparse flow data to achieve comprehensive, network-wide traffic flow recovery. Both methods employ a dynamic traffic assignment matrix to connect large-scale network speed data with individual link-level flows. The Sparse Recovery method utilizes the LASSO framework to induce sparsity, which is particularly effective when link flow data is missing due to limited sensor coverage. The Analytical Recovery method formulates the optimization problem using a quadratic objective function, which offers intuitive insights into traffic flow dynamics by using Stochastic Gradient Descent (SGD) and Lagrange Relaxation (LR) for parameter fine-tuning. These methods were validated through both a real-world case study and a hypothetical case study in Futian District, Shenzhen. Utilizing a city-level simulation platform built with SUMO to replicate urban traffic, both methods demonstrated close alignment with actual traffic flows and consistently maintained low estimation errors. This underscores their potential for effective large-scale traffic flow recovery using GPS data, even when flow observations are limited.To extend the temporal dimension of traffic flow data, this thesis further introduces a Dynamic Urban Spatial Temporal Graph Convolutional Network (DUST-GCN) that leverages historical traffic data and road information integrated to predict future urban traffic flow. Accounting for the dynamic and heterogeneous nature of urban traffic, DUST-GCN incorporates a temporal-spatial attention mechanism to capture dynamic dependencies, an adaptive graph structure to reflect changing inter-road relationships, and road-specific embeddings that capture the unique characteristics of various road types. To improve long-term prediction accuracy, a hybrid periodic input design integrates daily and weekly patterns, supported by GRU layers for long-term temporal modeling. Using newly developed real-world urban traffic datasets with corresponding road information for experiments, DUST-GCN consistently outperformed baseline models across different urban traffic scenarios, proving its reliability and adaptability.Overall, the traffic flow recovery and prediction approaches developed in this work provide comprehensive spatial-temporal traffic information across city networks. These advancements support urban traffic departments in traffic management, scheduling, control, and work zone deployment, offering robust tools for proactive and efficient urban transportation planning.
■590 ▼aSchool code: 0262.
■650 4▼aTransportation
■650 4▼aEnvironmental engineering
■653 ▼aSpatial-temporal data
■653 ▼aTraffic flow data
■653 ▼aTraffic flow prediction
■653 ▼aTraffic flow recovery
■653 ▼aTraffic simulation
■653 ▼aUrban transportation
■690 ▼a0709
■690 ▼a0543
■690 ▼a0775
■71020▼aThe University of Wisconsin - Madison▼bCivil & Environmental Engr.
■7730 ▼tDissertations Abstracts International▼g86-06B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17165079▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


