Files in this item



application/pdfWANG-DISSERTATION-2015.pdf (27MB)
(no description provided)PDF


Title:Hybrid state estimation applications for joint traffic monitoring and incident detection
Author(s):Wang, Ren
Director of Research:Work, Daniel
Doctoral Committee Chair(s):Work, Daniel
Doctoral Committee Member(s):Ouyang, Yanfeng; Peschel, Joshua; Sowers, Richard
Department / Program:Civil & Environmental Engineering
Discipline:Civil Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Traffic state estimation
Traffic incident detection
Particle filter
Kalman filter
Hybrid state estimation
Abstract:This dissertation is motivated by the practical problems of highway traffic estimation and incident detection using measurements from various sensor types. It proposes a framework to jointly estimate the traffic state and incidents in a hybrid state estimation problem where a continuous variable models the traffic state and a discrete model variable identifies the location and severity of an incident. Clearly, knowledge of an incident can improve post-incident traffic state estimates. Moreover, knowledge of the traffic state can be used to improve detection of incidents, by observing when the predicted traffic state differs significantly from the observed measurements. Two macroscopic traffic flow models are deployed to describe the evolution of traffic. Both the first order model and the second order model are extended to hybrid models by embedding a model parameter to denote the number of lanes open along the highway. The resulting traffic incident models are capable of describing traffic dynamics under both non-incident and incident scenarios that result in lane blockages. Next, several nonlinear filters are proposed to solve the joint traffic state estimation and incident detection problem. First, a multiple model particle filter and an interactive multiple model ensemble Kalman filter are proposed, where the particle filter or the ensemble Kalman filter are used to accommodate the nonlinearity of the traffic model, and multiple model methods are deployed to address the switching dynamics of traffic when incidents occur. Next, the multiple model particle filter is extended to a multiple model particle smoother to improve the estimation accuracy when data is limited. Finally, a variant of the multiple model particle filter, called the efficient multiple model particle filter, is developed for field implementations, which requires significantly less computation time compared to the other filters considered in this thesis. To validate the framework, the proposed nonlinear filters are implemented on the first order and second order traffic flow models, and tested in the microscopic traffic simulation software CORSIM and on field data collected on I-880 in California, which includes density measurements from inductive loops and speed measurements from GPS equipped vehicles. The results show that with either traffic flow model, the proposed traffic estimation algorithms are capable of jointly estimating the traffic state and detecting incidents when the traffic flow is high (i.e., when an incident results in congestion). The proposed algorithms are also compared with existing algorithms that independently estimate the traffic state or incidents. The results show that jointly estimating the state and incidents in one algorithm may perform better than two dedicated algorithms working independently, especially when loop detectors are sparse and the penetration rate of GPS equipped vehicles is high.
Issue Date:2015-11-19
Rights Information:Copyright 2015 Ren Wang
Date Available in IDEALS:2016-03-02
Date Deposited:2015-12

This item appears in the following Collection(s)

Item Statistics