Event Type: PhD Defense

The objective of this dissertation is to develop and apply an analytic procedure that estimates the amount of traffic congestion (vehicle hours of delay) that is caused by different types of accidents that occur on urban freeways, as well as to develop a model for prediction of real-time accident information such as how long an accident will affect traffic congestion and to the extent of the traffic congestion. Although it has been speculated that non-recurrent congestion caused by accidents, disabled vehicles, spills, weather events, and visual distractions accounts for one-half to three-fourths of the total congestion on metropolitan freeways, there are insufficient data to either confirm or deny this conjecture. 
 
The first part of this dissertation develops a method to separate the non-recurrent delay from any recurrent delay that is present on the road at the time and place of a reported accident, in order to estimate the contribution of non-recurrent delay caused by the specific accident. The procedure provides a foundation for a forecasting model that will assist transportation agencies such as Caltrans to allocate resources in the most effective way to mitigate the effects of those accidents that are likely to result in the greatest amount of delay. Additionally, since freeway travelers may be able to alter their driving routes based on the real-time accident information, the forecasting model may reduce traffic congestion and the incidence of secondary accidents.
 
Since a number of estimated delay results were censored by time and/or space boundary conditions, general statistical approaches were not available. An approach based on survival analysis was applied to analyze estimated delay and to predict traffic congestion impact in terms of time and space. Specifically, a statistical model based on the Cox type proportional hazard analysis is estimated that describes non-recurrent delay as a function of day of week, time of day, weather, and observable (e.g., from emergency calls and/or aerial or on-scene observation) characteristics of the accident. These accident characteristics, which are available to Freeway Traffic Management Systems, include time of day, number of involved vehicles, whether a truck is involved, and collision location (by lane or side of road). This statistical model can be used to inform a manager as to the expected delay associated with an accident as soon as the accident is reported and its characteristics are observed. This can in turn be used in improving resource allocation.
 
Additionally, this dissertation develops three prediction models regarding the spatio-temporal impact caused by a traffic accident as well as an accident duration model based on AFT metric model. Information provided by such predictions can play an important role in public sector transportation agencies providing freeway travelers with real-time traffic information under incident conditions.

The so-called activity-based approach to analysis of human interaction within social and physical environments dates back to the original time-space geography works of Hägerstrand and his colleagues at the Lund School in 1970, with a unique kernel problem termed “household activity scheduling”. The problem attempts to derive estimates of activity decisions taking into account the time, duration, mode, location and route of the given activity sets performed by individuals. 
This dissertation research studies the activity scheduling/rescheduling problem under an uncertain environment. Theories and models for predicting activity-travel behavior are developed within the context of an activity-based approach built on the general consensus that the demand for travel is derived from a need or desire to participate in activities. Computationally-tractable systems are developed that inherently incorporate factors of uncertainty that can potentially increase the ability to address the household activity scheduling problem and the related dynamics of human movement required for social interaction and household sustenance. A stochastic mixed integer linear program is formulated to model travel behavior in which each activity of the prescribed household agenda has a known probability of being completed (or cancelled). Further, a chance-constrained program is proposed to determine the optimal activity/travel pattern when travel time and activity duration are assumed to be stochastically distributed, while the remaining inputs are precisely known. Finally, under the assumption that the activity/travel pattern involves a dynamic decision-making process of rescheduling/adaptations to initial plans subject to unexpected events, a predictive model of activity rescheduling behavior is developed in the form of a mixed integer linear program. 
The dissertation presents solution methodologies to the proposed models. Data drawn from a comprehensive on-line survey are utilized to verify the proposed activity schedule/reschedule models. Numerical results are presented to demonstrate the performance of the proposed models. Finally, conclusions and directions for future research are summarized. 

The aviation industry has sought to address the negative externality of aircraft noise using a variety of approaches, but there has been little theoretical work to date encompassing both the market implications and the social optimality of air transportation noise policy. This dissertation develops simple theoretical models to analyze the effects of noise regulation on an airline’s scheduling, aircraft ‘quietness’, and airfare choices. Monopolistic and duopolistic airline competition are modelled, and two types of noise limits are considered: maximum cumulative noise from aircraft operations and noise per aircraft operation. As expected, tighter noise limits, which reduce community exposure to noise, also cause airlines to reduce service frequency and raise fares, which hurts consumers. Welfare analysis investigates the social optimality of noise regulation, taking into account the social cost of exposing airport communities to noise damage, as well as consumer surplus and airline profit. Numerical simulations show that the type of noise limit has a significant effect on the magnitude of the first-best and second-best optimal solutions for service frequency, cumulative noise, and aircraft size and level of quietness. Furthermore, the numerical analyses suggest that under the more realistic second-best case, the cumulative noise limit might be a preferable policy instrument over the per-aircraft noise limit. In the monopoly’s parameter space exploration, welfare is found to be slightly higher, cumulative noise is lower, and the fare is slightly lower when the planner controls cumulative noise rather than per-aircraft noise. In the duopoly case, the per-aircraft limit offers only modest welfare gains above the levels achieved with the cumulative limit, whereas the cumulative limit substantially increases welfare above that achieved with the per-aircraft limit in some regions of the parameter space explored. The effects of noise taxation and the optimal level of noise taxes are also investigated with the duopoly model; the analysis shows equivalence between noise taxation and the cumulative noise limit. 

Within the existing body of activity scheduling behavior models, the Household Activity Pattern Problem (HAPP) model is an activity-based model characterized by a rigorous mathematical programming formulation. The HAPP model can deal with detailed activity patterns including spatial, temporal, personal and modal information with complex constraints. The HAPP model is in the form of a Mixed Integer Programming model (MIP) which includes both continuous variables and discrete variables. Such temporal attributes of an activity pattern as starting time, duration and ending time are continuous variables, and those spatial attributes associated with the sequencing of activities, travel modes, participation persons and vehicles are discrete variables.
As formulated, the HAPP model is a constrained utility maximizing model. Empirical application of the model to a demand context involves estimation of the components of the objective function, based on data from observed patterns. However, due to computational difficulties in HAPP model, genetic algorithms (GA) have been proposed to estimate the set of factors influencing the objective function that “best” reproduces the observed spatial and temporal interrelationships. The fitness score in the GA approach used to evaluate the quality of the representation is the difference between the observed activity pattern scheduling (OAPS) and predicted activity pattern scheduling (PAPS), or the similarity between the two.
In this dissertation, we propose a new similarity metric for the GA estimation procedure. The metric considers the problem based on the continuous representation of discrete activity variables along the temporal dimension. Three similarity judging rules work together to form the similarity definition of similarity metric. They are: the temporal overlap among activities of different type, correspondence between participant person and vehicle used for each activity; permutations in the temporal sequence of activities and activity duration length similarity. The estimation procedure is tested on data drawn from a well-know activity/travel survey.

This dissertation defines and studies a class of dynamic problems called the “Mass Transport Vehicle Routing Problem” (MTVRP) which is to efficiently route n vehicles in real time in a fast varying environment to pickup and deliver m passengers, where both n and m are large. The problem is very relevant to future transportation options involving large scale real-time routing of shared-ride fleet transit vehicles. Traditionally, dynamic routing solutions were found using static approximations for smaller-scale problems or using local heuristics for the larger-scale ones. Generally heuristics used for these types of problems do not consider global optimality.
 
The main contribution of this research is the development of a hierarchical methodology to solve MTVRP in three stages which seeks global optimality. The first stage simplifies the network through an aggregated representation, which retains the main characteristics of the actual network and represents the transportation network realistically. The second stage solves a simplified static problem, called “Mass Transport Network Design Problem” (MTNDP). The output of stage 2 is a set of frequencies and paths used as an initial solution to the last stage of the process, called Local Mass Transport Vehicle Routing Problem (LMTVRP), where a local routing is performed.
 
The thesis presents the proposed methodology, gives insights on each of the proposed stages, develops a general framework to use the proposed methodology to solve any VRP and presents an application through microsimulation for the city of Barcelona in Spain.

Auction based market clearing mechanisms are widely accepted for
conducting business-to-business transactions. This dissertation focuses on
the development of auction mechanism decision tools for freight
transportation contract procurement in spot markets and long-term markets.
Spot markets have found their niche because of the Internet and standard
classic auctions are widely employed. For long-term markets, large
shippers (typically manufacturing companies or retailers) have begun to
use combinatorial auctions to procure services from trucking companies and
logistics services providers. Combinatorial auctions involve very
difficult optimization problems both for shippers and carriers. In the US
truckload market very few carriers have the technical know-how to bid in
combinatorial auctions. To reduce these problems we look at a different
auction scheme termed a unit auction, where the shipper can exploit the
economies of scope in the network and give the carriers the chance to bid
on pre-defined packages similar to ‘lotting’ in supply chain procurement.
Shippers have non-price business constraints, which must be included in
the winner determination problems to closely match shipper business
objectives. We develop allocation formulations incorporating the non-price
business constraints and Lagrangian based heuristics for solving them in
both unit auctions and combinatorial auctions. We provide carrier bidding
framework for classic auctions in spot markets using concepts from
economic auction theory. For bidding in combinatorial auctions, we study
the effects of demand uncertainty, carrier network synergies and strategic
pricing, and shipper’s winner determination problems on carrier bidding
using optimization-based simulation analysis. We also provide a framework
for volume-based contracts using insights from classical transportation
problem.
Further, we also present a mechanism for cross shipper auctions for
shipper collaboration and alleviate logistical inefficiencies like
deadheading and dwell times for carriers. Finally we develop pareto
efficient profit sharing mechanisms among shippers using co-operative game
theory.

Since the early 1990’s, public policies for transportation planning have evolved towards modally balanced transportation systems, requiring planning agencies to more precisely evaluate the capacity of their transportation systems, considering all feasible modes as well as low-cost capacity improvements. However, existing methods for capacity analysis are limited to either an individual facility or a single mode network, and thus appear insufficient for multimodal systems capacity analysis. This dissertation presents an advanced method for capacity assessment that can serve as an analytical tool in the freight transportation planning process, particularly from a multimodal perspective. The multimodal network capacity model formulated in this research takes a mathematical form of a nonlinear bi-level optimization problem with an embedded user equilibrium network assignment problem at its lower level. The bi-level formulation, referred to as the MNCP model in this dissertation, is comprehensive in the sense that many crucial factors are incorporated including multiple modes and commodities, behavioral aspects of network users, external factors, as well as the physical and operational conditions of a network. The numerical tests designed to illustrate the application of the MNCP model indicate that the algorithm developed for solving the bi-level problem has been successfully implemented. These results show the capability of the model not only to estimate the capacity of a multimodal network, but also to identify the capacity gaps over all individual facilities in the network, including intermodal facilities. By incorporating more precise capacity measures into the planning process, planning agencies would benefit from the proposed MNCP model in articulating investment priorities across all transportation modes, thus achieving their goal of developing sustainable transportation systems in a cost-effective manner. 

Advances in traffic surveillance technology can provide more complete and intelligent data from detectors. This dissertation describes an improved method of freeway performance measurement that integrates multi-sensor data fusion with a vehicle-monitoring algorithm capable of identifying the same vehicle/s at different locations. To obtain a more robust and effective data set for vehicle monitoring, data fusion from two state–of–the-art traffic detectors — loop detectors and video detectors — was introduced. Investigations and development of a new algorithm for data fusion and real-time vehicle monitoring – TRASURF (TRAffic SURveillance and perFormance) were also described. The algorithm’s development was based on an examination of feature vector extraction from each advanced traffic sensor, data fusion across multiple technologies and analysis of sensor performance. A real-world data set from one section of the I-405 freeway was applied to develop and evaluate the algorithm for a single freeway section. Based on extensive analysis of these field data, the PARAMICS (PARAllel MICroscopic Simulation) model was used to generate simulated fused data. This simulation served as the means to test and evaluate the performance of TRASURF as a multi-section vehicle-monitoring algorithm. The algorithm’s ability to reconstruct individual vehicle trajectories will enable more efficient and effective traffic surveillance, and will enhance the collection and analysis of network-wide traffic information including path travel time and origin-destination matrices. Furthermore, investigations and descriptions of various applications of advanced detectors for traffic analysis, especially in the context of the single-loop configuration widely used within California and many other locations were made. Traffic data extraction based on advanced loop detectors will make a vital contribution to many aspects of traffic operations and management, as these data are not available from conventional detectors.

A fundamental requirement for the successful implementation of advanced transportation management and information systems (ATMIS) is the development of a real-time traffic surveillance system that can produce accurate and reliable traffic performance measures. My dissertation presents a new framework for anonymous vehicle tracking that is capable of tracking individual vehicles using vehicle features. The core of the proposed method is a vehicle reidentification algorithm for signalized intersections based on inductive vehicle signatures. This has two major components: search space reduction and probabilistic pattern recognition. Both real-time intersection performance and intersection origin-destination (OD) information can be obtained as basic outputs of the algorithm. I developed an evaluation framework for vehicle tracking performance using a microscopic traffic simulation. I conducted a systematic simulation investigation of the performance and feasibility of anonymous vehicle tracking along multiple detector stations using the proposed simulation evaluation framework. The proposed system produces a rich data source for OD estimation. This application is the focus of my dissertation. I also present other useful applications of inductive vehicle signatures. These include the development of a methodology for evaluating traffic safety based on individual vehicle information and the prediction of section travel times. The proposed methodology can also be a valuable tool for operating agencies to support a number of intelligent transportation systems (ITS) strategies including congestion monitoring, adaptive traffic control, system evaluation and provision of real-time traveler information.

The real-life traffic flow process of signalized surface street networks
can be naturally decomposed into the traffic flow and the control strategy
components. The first challenge is faced when modeling the traffic flow
process on the surface streets. Specifically, the mathematical
description of traffic dynamics can be obtained by a number of models that
have a direct effect on the complexity of the corresponding optimal
control problem. Moreover, in order for the mathematical model to be
considered as part of a constraint optimization problem the constraints
must be expressed by inequalities (linear or not). However, the
consistent mathematical description of the traffic flow process inevitably
includes conditional piece-wise functions. For example, the traffic flow
at the approach of a signalized intersection is a piece-wise function
whose range depends (is conditional) on the prevailing traffic conditions
and the signal indication. Expressing this function (or others of similar
form) as a set of constraints that are additionally linear with respect to
the corresponding variables is a non-trivial task. The practices
typically followed include ignoring this function by averaging the outflow
during green over the cycle length, or approximating it with inexact
representations, or manipulating it during the solution process. These
approaches result to modeling inconsistencies and solutions of
questionable quality due to the involved heuristics. On the other hand,
there are cases of such functions that are equivalently represented by a
Mixed Integer Model (MIM) i.e., a set of inequality constraints in both
continuous and discrete variables.
The next challenge appears in designing the control strategy model.
Surprisingly, while someone would expect the modeling of the control
strategies to be driven by the currently followed state-of-the-practice,
one discovers that even the most recent modeling approaches follow the
outmoded concept of single-ring controllers with cycles of fixed duration
for a single pair of conflicting movements. Moreover, despite the fact
that the aforementioned control strategy (or its variations) is the most
widely adopted and optimized over surface street networks of comparable
dimensions, the corresponding solution time that offers a qualitative
measure of the strategy performance is not reported. Similarly, for those
control strategy models that are solved as Mixed Integer Programming
Problems no relative information on the solution time is provided.
 
The general objective of this dissertation is to address the issues
associated with developing a model for signalized surface street networks
and solving the corresponding optimal control problem. In order to
accomplish this goal the following specific aims are fulfilled:
1.   Based on analogies from the theory of mathematical logic we develop two
methodologies for transforming conditional piece-wise functions into an
equivalent MIM representation.
2.   We demonstrate the potential of both methodologies by their application
to a number of conditional piece-wise functions that are found during the
process of developing a mathematical representation for signalized surface
street networks that is based either on the dispersion-and-storage or on
the cell transmission traffic flow models. For example, we have developed
MIM representations that describe the cases of the outflow at the approach
of a signalized intersection when assuming a 2-band (Green-Red) signal, or
when assuming a 3-band (Green-Yellow-Red) signal, etc.
3.   We demonstrate the capability of both methodologies in analyzing the
structure of existing MIMs, which subsequently enables us to provide
improved (in terms of the variables and the constraints) representations.
4.   We develop a control strategy model that describes a dual-ring,
8-phase, variable cycle controller, and we further propose an alternative
formulation that is based again on logic-based methods that could
potentially be useful (in terms of the solution time of the corresponding
problem) within the context of a customized branch-and-bound solution
algorithm.
5.   We examine the performance of our control strategy under various
hypotheses both quantitatively (solution time) and qualitatively using the
CPLEX solver that is based on a branch-and-cut solution algorithm.
 
This dissertation demonstrates the potential of the optimal control
approach as a powerfull tool in solving the complex problem of optimizing
traffic signals for surface street networks.