Dissertation/Thesis Abstract

Optimization Methods for Calibration and Analysis of Congested Freeway Facilities
by Trask, Joseph Lake, Ph.D., North Carolina State University, 2017, 160; 10610841
Abstract (Summary)

Congestion and delays on freeways result in countless hours of frustration for drivers and cost the nation billions of dollars each year. Consequently, projects are constantly being undertaken to both improve the physical capacities of roadways and develop new strategies for demand management and congestion reduction. A variety of planning and analysis tools are used to aid in project planning and execution, and the Highway CapacityManual is one such tool whose goal is to provide methodologies and algorithms to assess the mobility effects of highway projects. In particular, the manual includes a macroscopic simulation approach for analyzing congested freeway facilities methodology rooted in the hydrodynamic theory of traffic flow. The methodology has been shown to be an effective approach for planning and operational analyses, but despite being widely used, little has been done to utilize optimization techniques in conjunction with the method.

This work begins by utilizing classical linear programming to provide a novel optimization framework for conducting congestion analysis using the methodology. As it is currently constructed in the manual, the computational procedure is inherently limited to computing outputs from a specified set of inputs. By reformulating these relationships and embedding them in an optimization framework, the methodology can be used to characterize vehicle flows based on any number of optimality criteria. The resulting approach bears many similarities to system optimal dynamic traffic assignment techniques, and its effectiveness is explored with respect to peak demand spreading and ramp metering applications.

Next, the work takes a step back and focuses on a key challenge of the model creation process. Before any planning or congestion management analysis can be conducted, an analyst must first create a calibrated model that faithfully represents the conditions of the real-world facility. This work develops a new calibration framework using a genetic algorithm metaheuristic to automate the demand estimation and adjustment process. Two encoding approaches for the problem are proposed based on the quality of available data. Additionally, the genetic algorithm approach is expanded to include three key capacity parameters, allowing the metaheuristic to consider pre-breakdown capacity adjustments, queue discharge flow rate, and the facility-wide jam density alongside demand volumes. At each stage, the effectiveness of the metaheuristic approach is demonstrated for examples representing both ideal conditions and real-world case studies. As an example, a real world case study is presented for the proposed calibration approach that focuses on a section of I-540 outside of Raleigh, NC. The calibration process is shown to remove 70% of the error for lightly congested conditions, and as much as 90% of the error when speeds indicate flow is highly congested. Further, the balance of the approach in matching both facility travel times and individual segment speeds provides an improvement over current methods focusing on just the former.

Finally, a principle benefit of both the linear programming formulation and the GA calibration framework is their inherent flexibility. This flexibility provides an ideal basis for further innovation in terms of both optimization applications and improved model calibration. For example, numerous additional single and multi-objective functions can be studied for each approach, only a small number of which are presented in this work. Further, the LP model provides a strong optimization base that can be greatly expanded through the incorporation of integer variables, or even nonlinear constraints and objective functions. Additionally, the calibration framework can be improved and tailored to specific instances, utilizing more complex genetic operators or by implementing modelspecific intergenerational learning heuristics.

Indexing (document details)
Advisor: Baugh, John, Rouphail, Nagui
Commitee:
School: North Carolina State University
Department: Operations Research
School Location: United States -- North Carolina
Source: DAI-B 78/10(E), Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Transportation planning, Operations research
Keywords: Freeway facilities, Genetic algorithms, Linear programming
Publication Number: 10610841
ISBN: 9781369857245
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest