東華大學圖書館 |

Statistical Inference of Spatio-Temporal Transportation Networks through Large-Scale Multi-Source Data.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Statistical Inference of Spatio-Temporal Transportation Networks through Large-Scale Multi-Source Data./
作者:	Ma, Wei.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2019,
面頁冊數:	213 p.
附註:	Source: Dissertations Abstracts International, Volume: 81-03, Section: B.
Contained By:	Dissertations Abstracts International81-03B.
標題:	Transportation. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=22617832
ISBN:	9781085673235

Statistical Inference of Spatio-Temporal Transportation Networks through Large-Scale Multi-Source Data.
Ma, Wei.

Statistical Inference of Spatio-Temporal Transportation Networks through Large-Scale Multi-Source Data. - Ann Arbor : ProQuest Dissertations & Theses, 2019 - 213 p.

Source: Dissertations Abstracts International, Volume: 81-03, Section: B.

Thesis (Ph.D.)--Carnegie Mellon University, 2019.

This item must not be sold to any third party vendors.

The increasing complexity and interconnectivity of transportation networks call for a better understanding of the stochasticity and uncertainty of the spatio-temporal network variables such as demand, flow, and speed. On the one hand, classical network models overlook the statistical features of network dynamics, a critical feature of recurrent traffic. The statistical features of spatio-temporal networks are also critical inputs for off-line and on-line transportation management for large-scale networks. On the other hand, various data resources, from traditional traffic sensors (loops, cameras, etc.) as well as emerging sensors (Bluetooth, GPS probe, parking spot occupancy, etc.), are available and have been archived for decades in many mega-cities. With the network models and sensing technologies being developed for decades, there is a lack of study on the understanding of inter-relations of spatio-temporal vehicles/passengers in the network, their causes from demand characteristics, and how big data can help estimate, predict and ultimately intervene any component of network flow aiming for system optimum. This dissertation presents a comprehensive study on the statistical traffic assignment models and probabilistic demand estimation models in the constant-time networks as well as the time-varying networks. The following three major questions are answered throughout the dissertation: 1) How to characterize the spatio-temporal relationship of flow dynamics (traffic flow rate, speed, choices of time and routes) given uncertain traffic demand and system performance? 2) How to infer the flow dynamics using both archived and real-time data from multiple resources? 3) How to use the new methodology to improve the reliability and sustainability of large-scale networks through improving both recurrent and non-recurrent traffic conditions? As the preliminary research, we first present a complete study on the generalized statistical traffic assignment model (GESTA) and probabilistic demand estimation model in the constant-time networks. The GESTA builds the relationship among probability distributions of link/path flow and their travel cost where the variance stems from three sources, demand, route choice and unknown errors. A novel theoretical framework for estimating demand distribution using multi-source traffic data is proposed. Both models can be solved efficiently in a large-scale network to provide insights for decision making and demonstrate computational efficiency.To understand the flow dynamics in time-varying networks, we propose a novel concept of dynamic assignment ratio (DAR) matrix. The DAR matrix enables the realistic representation and efficient computation of dynamic traffic flow. With DAR matrix, we propose a theoretical formulation for estimating dynamic demand using computational graph. The computational graph can be evaluated on multi-core CPUs or Graphics Processing Units (GPU) efficiently, and hence the proposed method can be efficiently applied to the large-scale networks. Using the concept of DAR matrix, we present a solution framework for multi-class dynamic OD demand estimation (MCDODE) in large-scale networks. The proposed framework is built on a computational graph with tensor representations of spatio-temporal flow and all intermediate features involved in the MCDODE formulation. A forward-backward algorithm is proposed to efficiently solve the MCDODE formulation on computational graphs. Additional, we rigorously formulate the spatio-temporal structure of probabilistic traffic demand and propose a statistical inference framework to infer the structure of dynamic demand using multi-source data. Lastly, we present two real-world applications that are built on top of the transportation network models. The first application analyzes the impact of the Greenfield Bridge closure using the dynamic network analysis for Pittsburgh Metropolitan Area. The second application builds a real-time traffic management application based on the dynamic traffic assignment. The two applications demonstrate the potentials of our proposed models in improving the mobility, safety and reliability of the transportation systems.

ISBN: 9781085673235Subjects--Topical Terms:

555912
Transportation.
Subjects--Index Terms:

Data-driven approach

Statistical Inference of Spatio-Temporal Transportation Networks through Large-Scale Multi-Source Data.
LDR:05483nmm a2200385 4500 001 2272756
005 20201105110215.5
008 220629s2019 ||||||||||||||||| ||eng d
020 $a 9781085673235
035 $a (MiAaPQ)AAI22617832
035 $a AAI22617832
040 $a MiAaPQ $c MiAaPQ
100 1 $a Ma, Wei. $3 1675305
245 1 0 $a Statistical Inference of Spatio-Temporal Transportation Networks through Large-Scale Multi-Source Data.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2019
300 $a 213 p.
500 $a Source: Dissertations Abstracts International, Volume: 81-03, Section: B.
500 $a Advisor: Qian, Zhen (Sean).
502 $a Thesis (Ph.D.)--Carnegie Mellon University, 2019.
506 $a This item must not be sold to any third party vendors.
520 $a The increasing complexity and interconnectivity of transportation networks call for a better understanding of the stochasticity and uncertainty of the spatio-temporal network variables such as demand, flow, and speed. On the one hand, classical network models overlook the statistical features of network dynamics, a critical feature of recurrent traffic. The statistical features of spatio-temporal networks are also critical inputs for off-line and on-line transportation management for large-scale networks. On the other hand, various data resources, from traditional traffic sensors (loops, cameras, etc.) as well as emerging sensors (Bluetooth, GPS probe, parking spot occupancy, etc.), are available and have been archived for decades in many mega-cities. With the network models and sensing technologies being developed for decades, there is a lack of study on the understanding of inter-relations of spatio-temporal vehicles/passengers in the network, their causes from demand characteristics, and how big data can help estimate, predict and ultimately intervene any component of network flow aiming for system optimum. This dissertation presents a comprehensive study on the statistical traffic assignment models and probabilistic demand estimation models in the constant-time networks as well as the time-varying networks. The following three major questions are answered throughout the dissertation: 1) How to characterize the spatio-temporal relationship of flow dynamics (traffic flow rate, speed, choices of time and routes) given uncertain traffic demand and system performance? 2) How to infer the flow dynamics using both archived and real-time data from multiple resources? 3) How to use the new methodology to improve the reliability and sustainability of large-scale networks through improving both recurrent and non-recurrent traffic conditions? As the preliminary research, we first present a complete study on the generalized statistical traffic assignment model (GESTA) and probabilistic demand estimation model in the constant-time networks. The GESTA builds the relationship among probability distributions of link/path flow and their travel cost where the variance stems from three sources, demand, route choice and unknown errors. A novel theoretical framework for estimating demand distribution using multi-source traffic data is proposed. Both models can be solved efficiently in a large-scale network to provide insights for decision making and demonstrate computational efficiency.To understand the flow dynamics in time-varying networks, we propose a novel concept of dynamic assignment ratio (DAR) matrix. The DAR matrix enables the realistic representation and efficient computation of dynamic traffic flow. With DAR matrix, we propose a theoretical formulation for estimating dynamic demand using computational graph. The computational graph can be evaluated on multi-core CPUs or Graphics Processing Units (GPU) efficiently, and hence the proposed method can be efficiently applied to the large-scale networks. Using the concept of DAR matrix, we present a solution framework for multi-class dynamic OD demand estimation (MCDODE) in large-scale networks. The proposed framework is built on a computational graph with tensor representations of spatio-temporal flow and all intermediate features involved in the MCDODE formulation. A forward-backward algorithm is proposed to efficiently solve the MCDODE formulation on computational graphs. Additional, we rigorously formulate the spatio-temporal structure of probabilistic traffic demand and propose a statistical inference framework to infer the structure of dynamic demand using multi-source data. Lastly, we present two real-world applications that are built on top of the transportation network models. The first application analyzes the impact of the Greenfield Bridge closure using the dynamic network analysis for Pittsburgh Metropolitan Area. The second application builds a real-time traffic management application based on the dynamic traffic assignment. The two applications demonstrate the potentials of our proposed models in improving the mobility, safety and reliability of the transportation systems.
590 $a School code: 0041.
650 4 $a Transportation. $3 555912
650 4 $a Systems science. $3 3168411
650 4 $a Civil engineering. $3 860360
653 $a Data-driven approach
653 $a Machine learning
653 $a Origin-destination demand estimation
653 $a Traffic assignment problem
653 $a Traffic operation
653 $a Transportation network modeling
690 $a 0709
690 $a 0790
690 $a 0543
710 2 $a Carnegie Mellon University. $b Civil and Environmental Engineering. $3 2097787
773 0 $t Dissertations Abstracts International $g 81-03B.
790 $a 0041
791 $a Ph.D.
792 $a 2019
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=22617832