Event Type: PhD Defense

Crowdsourced delivery, or crowd shipping, is a delivery service in which logistics service providers contract delivery services from the public (i.e., non-employees), instead of providing delivery services exclusively with an in-house logistics workforce. Studies in the literature formulate the crowdsourced delivery problem as a Vehicle Routing Problem (VRP) and propose a variety of solution approaches for small problem instances. Conversely, this dissertation focuses on large-scale crowdsourced systems and problem instances, which have significant promise and importance, respectively, for effective planning and efficient operation crowdsourced systems in the real-world.

This dissertation provides a taxonomy of urban last-mile crowdsourced delivery, defines the crowdsourced shared-trip delivery problem, presents two separate models for the crowdsourced shared-trip delivery problem, and develops a novel decomposition heuristic tailored to solve large-scale crowdsourced delivery problems. The taxonomy identifies three types of urban last-mile crowdsourced delivery—crowdsourced trip-based delivery, crowdsourced time-based delivery, and crowdsourced shared-trip delivery. This dissertation focuses on the crowdsourced shared-trip delivery problem in which crowdsourced drivers communicate their origin and destination of an upcoming trip to the logistics provider and, if the logistics provider can identify packages for the driver to pickup and deliver without significant detours, the logistics provider assigns packages to and compensates the driver. The dissertation models the crowdsourced delivery problem by adapting VRP formulations from the literature as well as using a set partitioning formulation. The set partitioning formulation leads to an alternative solution method for large-scale instances that involves decomposing the problem into several subproblems. The novel decomposition heuristic contains a series of subproblems that are solved in this dissertation using various solution algorithms, including, budgeted k-shortest path, large scale bi-partite matching, package switching and vehicle routing. The decomposition heuristic achieves high quality results and significantly shorter computational time in comparison to an exact solution method.

In the numerical case study, the dissertation analyzes various factors that may impact the effectiveness and efficiency of urban crowdsourced delivery. The results indicate that crowdsourced shared-trip delivery service can reduce total delivery cost by between 20% to 50%, compared to a delivery service that exclusively uses its own dedicated vehicles and drivers. Vehicle miles travelled (VMT) savings depend on the origin location (i.e., home locations or the store/depot) of crowdsourced drivers that participate in the service. In addition, the results show that dedicated vehicles are still required since even a considerably large set of candidates shared crowdsourced vehicles cannot usually serve all packages.

There have been laws passed in California (SB32) that would require the State to cut its Greenhouse Gas Emissions (GHG) to 40% of 1990 levels by 2030 in order to combat climate change. With cars contributing to 41% of GHG emissions in California it is clear that to reach that goal there will need to be a significant reduction in Vehicle Miles Travelled (VMT). A way to quickly reduce VMT is to invest in existing rail systems specifically commuter rails. An investigation was conducted to model the potential effects of improving commuter rail services on
a state vs. national level, station-by-station level, and a regional level. To conduct the research data was gathered from the National Transit Database, Longitudinal Employer-Household Dynamics site, and the Environmental Protection Agencies Smart Location Database (EPA-SLD) for the year 2014.
The California Model unlinked passenger trips are more sensitive to the hours of service than the National Model. Also, the California Model is more sensitive to log peak vehicles operated which would imply that the more vehicles or frequency of the vehicles servicing people can have a large impact on passenger trips.
The Station boarding and egress models were the best when there were exogenous latent variables in the regression model. The latent variables Mixed-Use Density and Work Opportunity play a significant role in transit boardings and egress by stating that if the mixed-use density increases the employment, employment entropy, and ratio of jobs accessible in 45 minutes increases.
Model 2 is superior of the SEM models created. The ridership factors that the passenger rates to all the observed variables and the measure of their satisfaction with the variables can be a tool to use for improving service quality and for planning for future services. In the long run, this could have cost savings because if there is information about the riders’ preferences there can be improvements made specific to what is valued as important. This model can be easily modified to fit other transit services in many different regions or countries because of the framework structure which can be used for analyzing any type of service from survey responses.

Understanding the spatiotemporal distribution of commercial vehicle is essential for facilitating strategic pavement design, freight planning, and policy making. Hence, analysts and researchers have been increasingly interested in collecting more diverse high granularity truck data across different truck characterization schemes to meet these variegated needs across the roadway network to better understand their distinct operational characteristics and dissimilar impacts on infrastructure and the environment. Existing truck monitoring infrastructure are limited in spatial coverage across the roadway network due to their high installation and maintenance costs. The recently developed Truck Activity Monitoring System (TAMS) by the University of California Irvine Institute of Transportation Studies provides a cost-effective solution for monitoring truck movements statewide across California along major freeways networks through existing inductive loop infrastructure enhanced with inductive signature technology. Nonetheless, it possesses three major limitations: model bias against underrepresented truck classes, spatial coverage limitation on rural highways, and system obsolescence over time.

This dissertation explored a resilient spatiotemporal truck monitoring framework using inductive loop signature and multi-array Light Detection and Ranging (LiDAR) sensor technologies to address the aforementioned issues and to improve truck monitoring capabilities across the roadway network. The designed framework comprises three major parts: Inductive loop sensors for major highway truck monitoring, multi-array LiDAR sensors for rural highway truck monitoring, and a self-learning truck classification through a sensor integration framework.

The first part of the framework was built upon the existing Truck Activity Monitoring System (TAMS) developed by ITS Irvine and addresses prediction model biasness caused by inherently imbalanced truck datasets to provide reliable truck speed estimation and truck classification data.

The second part explored non-intrusive LiDAR-based sensing technologies to fill the surveillance gap along rural highway corridors. This section developed a truck classification method using a LiDAR sensor oriented to provide a wide field-of-view of roadways.

Finally, a self-learning framework for truck classification systems was designed to address system obsolescence through the integration of inductive loop sensors and LiDAR sensors, the latter of which has been proved to have the ability to recognized the truck axle configuration in this dissertation. This framework enhances the resilience of signature-based FHWA classification model with an intelligent system update to accommodate the change of the truck designations over time and significantly reduces the overall burden of periodic model calibration by utilizing the information stored in the legacy model.

Traffic advisories to travelers are based upon traffic state information at the link level. This is due to existing infrastructure which sometimes can only provide link-level information. However, the primary justification for providing link-level data is the reluctance of Traffic Management Agencies to consider more detailed traffic state data for operational and safety reasons. However, with the advances in automotive technology, sensing equipment, and the Internet of Things (IoT), we can do better. Research shows that faster and more accurate travel paths can be obtained by using lane data rather than link data. Our contention is that for vehicles to be able to change lanes to improve their travel times, operationally, they would need to enter into Peer-to-Peer negotiations with surrounding vehicles, where they can trade their position in time and space in exchange for monetary benefits. Our work is an exploration of this idea.
We begin with a simple in-vehicle advisory control policy, partially inspired by the Kinetic theory of traffic. We then move towards an individual-level Peer-to-Peer negotiated lane change framework by first investigating its efficacy by means of microsimulation studies. We then propose an agent-based optimization framework for this system, which minimizes both travel time and the ”envy” induced among drivers when they are assigned paths that are inferior to their peers. Numerical results from running our optimization on an illustrative off-ramp network show that the proposed model converges to both envy-free and system optimum traffic states, even at a net zero budget, meaning this system can be used by transportation agencies without exacting tolls or giving subsidies.
Our proposed framework of routing vehicles on a lane to lane basis can only be realized in the field if the mediating agency (TMC, or a mobility service) has accurate information about traffic conditions. We propose multiple algorithms, including a LSTM (Long Short Term Memory) neural network architecture-based framework to estimate traffic states solely using information collected from sensor-equipped probe vehicles, without the need for any other data such as those obtained from traditional embedded loop detectors.

Measuring the geographic extent of travel-activity patterns is very important to develop our knowledge on potential and actual activity spaces around individual travel routes and activity locations which will enrich our understanding of human activities. Previous activity space studies, primarily from the field of geography, demonstrate analysis techniques to characterize and assess spatial dimensions of areas that individuals come into contact within daily life. Although a handful of studies have begun to integrate activity space within the travel behavior analysis in Europe and U.S. context, few studies have measured the size, structure, and implications of human activity spaces in the context of developing countries. To address these concerns, this dissertation examines the impact of land-use characteristics, socio-demographics, individual trip characteristics, and personal attitudes on travel-activity based spatial behavior in Dhaka City, capital city of Bangladesh.
This dissertation focuses on two separate subareas: Mirpur from the Dhaka North City Corporation and Dhanmondi from the Dhaka South City Corporation based on their distinctive socio-economic and transportation characteristics. The first stage of this dissertation (presented in Chapter 2) is comprised of a household-based travel diary pilot survey that was conducted in 2017. Regular activity locations were geocoded using Geographic Information System (GIS). Network analyst based Shortest Path Network (SPN) with Road Network Buffer (RNB) was used to calculate activity space of the participants. Daily activity areas for individual respondents range from 0.38 to 6.18 square miles. Land-use mix is found to be a significant predictor of activity space size. Larger activity space is recorded for the residents of one subarea over another due to less land-use diversity. Pilot study results identify specific socio-economic and travel differences across the two study subareas (by car ownership, income, modal share, distance traveled, trip duration).
The second stage of this dissertation (presented in Chapter 3) builds on lessons learned from the pilot study and comprises of a weeklong household-based travel diary survey collected in 2018. Using Artificial Neural Network (ANN) and Regression Analysis, results show that weekly (weekdays/weekend days) activity areas for individual respondents range from 0.08 to 10.13 square miles. Dhanmondi respondents are found to have larger weekday activity space while Mirpur respondents have larger weekend activity space. Trip characteristics (distance, duration, and cost) are found to be significant predictors of individual activity space size. In case of household activity space, Density variables are found to play the most significant role. Higher density of retail shops and employment locations within a household’s activity space decrease the weekday activity space. Also, households without car have limited activity area during weekday. Unlike weekday finding, smaller households are found to have smaller activity space for weekend.
Exploratory and Confirmatory Factor Analysis shows that people’s perceptions (Perceived neighborhood amenities, Car attachment, Monetary concerns, Perceived daily travel area and environmental concern) mainly shape weekend spatial behavior. RNB activity space measure indicates that 44.4% of respondents do not have access to a recreational facility within their weekly activity space. Positive correlations are found between activity area and number of different opportunities except open space. The association is strong for hospital, retail shop and restaurant facility. Weekend activity spaces are found to be more compact than those for weekdays. Individual day-to-day variability is less during weekdays than for weekends. Also, weekday to weekend variability is found to be much larger in Dhanmondi compared to Mirpur. Female respondents and high-income people are found to have smaller activity spaces. While examining heterogeneity in activity spaces, results indicate that at an aggregate level activity spaces vary from day to day. To further analyze the impact of different indicator characteristics on this variability, Fixed Effects Panel Regression using Least Squares Dummy Variable approach, General Linear Model and Random Effects Panel Regression: Mixed Models is used. Model estimation results show that several time-varying predictors: trip characteristics (duration, distance, and cost), Density of schools and retail shops, intersection Design within the activity space, and few time-invariant predictor variables are found to significantly affect day to day activity space variability. Three attitudinal factors (Perceived neighborhood amenities, Environmental awareness, and Monetary concerns) also show significance in predicting activity space variability.
This dissertation study contributes to travel-activity space literature and planning practice in several ways. To my knowledge, this is the first study of activity space calculation in any South East Asian city and therefore contributes significantly to transportation science literature of the region. Also, only a few previous studies have assessed the influence of individual perceptions and values on activity spaces. Finally, understanding the day-to-day variability of activity spaces and examining accessibility to potential urban opportunities provide planners and policy makers specific guidance for future planning implications on spatial behavior in the study area.

According to the California Air Resources Board (CARB, 2020), light-duty vehicles are responsible for 13 percent of statewide NOx emissions and 28 percent of statewide greenhouse gas emissions. Scientists, policymakers, and car manufacturers have been striving to reduce the air pollution and greenhouse gas emissions from the transportation sector using various measures, ranging from cleaner engines to alternatives to driving to reduce VMT. In this dissertation, I focus on a subset of these measures: carsharing programs and Battery Electric Vehicles (BEVs).

In the first part of this dissertation, I explore the profile of households engaging in carsharing by estimating zero-inflated negative binomial (ZINB) models on data from the 2017 National Household Travel Survey (NHTS). My results show that households who are more likely to carshare are those who participate in other forms of sharing, have more Silent generation members, are less educated (the highest educational achievement is a high school degree), and have fewer vehicles than drivers. Conversely, households with more young adults (18 – 20 years old), with 2 or more adults and no children, take part in carsharing program less often. Moreover, households who took more part in ridesharing and have fewer vehicles than drivers are less likely to never carshare. Furthermore, households whose annual income between $75,000 and $150,000 are more likely to never carshare.

In the second part of this dissertation, I concentrate on the adoption of BEVs. More specifically, I focus on two questions: 1) What are the characteristics of households who own battery electric vehicles (BEVs)?; and 2) Does the travel behavior of these households differ from the travel of households who have motor vehicles but not BEVs? To answer those questions, I characterize three groups of households based on their vehicle holdings: BEV-only, BEV+ (i.e., households with both one or more BEV and at least one conventional vehicle), and non-BEV households. I analyze data from the 2017 NHTS using mixed methods. Results show that BEV households are more likely to be Asian, well-educated, with a higher income and to live in higher population and employment density areas. Furthermore, BEV-only households are more likely to be composed of one adult (not retired) with fewer Baby Boomers. Yet, BEV+ households are more likely to be larger households with 2 or more adults. Also, BEV+ households are more likely to have more Generation X (37-52 years old in 2017) and Z members (20 years old or younger in 2017). They are also more likely to own their home. My analysis on gender (at the individual level) concluded that BEV owners are more likely to be men. Furthermore, I find that BEV households travel as much as non-BEV households.

Although carsharing and BEVs could substantially decrease the environmental footprint of transportation, they are currently far from mainstream. To promote carsharing programs, their reach could be extended, they could be made more affordable, while increasing the cost of owning and operating private vehicles. Similarly, state and federal governments could continue to provide financial incentives to lower the purchase price difference between conventional and BE vehicles, manufacturers could provide extended warranties on batteries, and the charging infrastructure needs to be developed in order to attract more customers.

The Covid-19 crisis is giving governments around the world an opportunity to invest in clean technologies to jumpstart the economy. It is critical to take advantage of this crisis to reduce air pollution and greenhouse gas emissions from transportation for the good of current and future generations.

The incorporation of alternative fuel vehicles has been essential in reducing emissions in the transportation sector. Particularly to heavy-duty trucks, zero-emission technologies are becoming more attractive. However, batteries and fuel cells still face a long way until they became became a viable solution in terms of price, autonomy, weight, and infrastructure. An interim solution is the use an overhead catenarysystem, also known as eHighway. The pilot project demonstrated the feasibility of the eHighway system; however, the literature exploring this type of technology is lacking. This dissertation aims to cover this literature gap and propose a new framework to comprehensively explore the aspects of an eHighway implementation in terms of optimal placement, effects on the well-to-wheel (WTW) emissions, and impacts on the power grid. This methodology was applied to a California model using data from the California Statewide Freight Forecasting Model.

First, we defined the optimal eHighway placement to maximize vehicle miles traveled in the system or minimize emissions around disadvantage communities in four different scenarios for the years of 2020 and 2040. This process shows that most eHighways would be located along the I-5 or close to ports to maximize vehicle miles traveled or in Central Valley to maximize the benefit for disadvantage communities.

Second, we estimated the WTW emissions for heavy-duty truck according to the truck fuel type for each of the scenarios with adoption rates from 25% to 100%. The total emissions in terms of CO2 and NOx were compared to a scenario without eHighway. All the eHighway scenarios for 2020 and 2040 reduced the total WTW heavy-duty truck emissions. The best-case scenario for 2020, with 500 miles of total eHighway length and adoption rate of 100%, reached a reduction of almost 8% in CO2 emission and over 20% of NOx. The same scenario showed a reduction of 16% in CO2 and 20% of NOx for the year 2040. 

Finally, we analyzed the impacts of the eHighway energy demand on the state’s power grid. We showed that some of the systems would require up to 1 MWh of daily energy from some power substation. However, due to the unavailability of public data on California’s power grid, we could not draw conclusions in terms of the ability of these substation to handle such demand.

These results show the applicability of the proposed methodology for the deployment and impacts of the eHighway system. Furthermore, although there are other aspects to be considered before large-scale implementation of the eHighway system (e.g., costs), the results presented in this study support the deployment of an eHighway system in California to support the urgent need for making road freight transport more sustainable. 

Complex travel behavior places travel in a broader context than in the conventional single-trip based approach. The activity-based approach provides an analysis framework that positions travel decisions as dependent on a collection of activities that form an agenda for participation and, therefore, cannot be properly analyzed on individual trip basis. The basic units of analysis for activity-based approaches are tours, which can be defined as sequences of trips and activities that begin and end at the same location. In this dissertation, I apply a tour-based approach to analyze complex travel behavior of individuals from three broad perspectives: environmental, technological, and economical.

 First, I examine the complex travel behavior of workers, who utilize a sustainable transport option, namely public transit. I identify dominant patterns of work tours — tours that contain at least one work activity and have at least one link made by public transit — and analyze the factors that determine tour choice using Structural Equation Modeling (SEM). The results obtained by using the 2017 National Household Travel Survey (NHTS) dataset suggest that 80 percent of work tours consist of seven dominant tour patterns and that tour choice is influenced by a set of socio-demographic, built environment, and activity-travel characteristics. Second, the complex travel behavior of people who use technology-enabled ride-hailing services, such as Uber/Lyft, is explored. In particular, I identify heterogeneous groups of ride-hailing users by using Latent Class Analysis, analyze the activity-travel patterns of each of these groups, and discuss the ramifications of that behavior to policy directives.

Lastly, I explore the travel behavior of workers, again in terms of tours, when they are exposed to an economic downturn, the 2007-2009 great recession. I apply multi-group SEM to analyze changes in tour choice during the recession (2009) compared to pre- (2006) and post-recession (2012) years. Using American Time Use Survey data, this study shows that activity-travel relationships and their role in tour choices differed significantly in the recession year. The results of this study provide insights into potential changes in worker’s travel demand during a recession, which would contribute to building better pattern choice sets in tour-based models.

The common thread throughout this dissertation is the development of a framework for analyzing complex travel behavior under disruptive changes due to environment, technology, and economics forces.

High-occupancy vehicle (HOV) lanes are those reserved for cars with a minimum of two or three occupants and other qualified vehicles. However, some HOV lanes could be
underutilized, even when the corresponding general purpose (GP) lanes on the same roads are congested. A type of relatively recent congestion pricing strategy is realized with the high-occupancy toll (HOT) lane, where single-occupancy vehicles (SOVs) can pay a price to use HOV lanes during peak periods. The purpose of this study is to propose control theoretic approaches to congestion pricing for HOT lanes. This dissertation focuses on three operation objectives: (1) maintaining a certain level of service on the HOT lanes; (2) improving the overall system performance; and (3) maximizing revenue for the operators. Inspired by the work of (Yin and Lou, 2009), this study first proposes a simultaneous estimation and control method for a freeway with HOT and GP lanes. An integral controller is applied to estimate the average value of time (VOT) of SOVs, and the dynamic prices are calculated based on the logit model. The closed-loop system is proved to be stable and guaranteed to converge to the optimal state both analytically and numerically. Two convergence patterns, Gaussian or exponential, are revealed. The effect of the scale parameter in the logit model is also examined.

The heterogeneity of SOVs is essential for determining pricing schemes for HOT lanes. A vehicle-based user equilibrium (UE) principle is proposed to incorporate heterogenous SOVs. A general lane choice model is derived based on the characteristics of the logit and the vehicle-based UE principle. We obtain an insight regarding
the pricing schemes by analytically solving the optimal dynamic prices with constant demands of HOVs and SOVs. Then, we design a feedback controller to determine the dynamic prices without knowing SOVs’ lane choice models, but to satisfy the two control objectives: maximizing the flow-rate but not forming a queue on the HOT lanes.
Agencies have stated that revenue maximization should generally coincide with the optimization of freeway performances. In order to verify this statement, an optimal control problem is formulated to find the pricing scheme that maximizes the revenue for HOT lane operators. Numerical results show that there is conflict between those two
objectives. Operators need to make different strategies based on the traffic demand patterns. It has long been known that drivers’ departure time choice behavior is one fundamental cause of congestion. Further in this dissertation, pricing schemes are proposed to consider both lane choice equilibrium and departure time equilibrium. It turns out that flat pricing schemes are able to meet the following three constraints: (1) the nontransfer disutility (NTD) is minimized; (2) the cost for each HOV is not worse-off; and (3) the costs for each nonswitching and switching SOV are the same. We show that different revenue requirements lead to different pricing schemes.

Smart Urban Mobility in the future demands a paradigm shift. Transportation supply needs to be designed to incorporate individual-level preferences in an era of readily-
available information about other users and network performance. It is, therefore, reasonable to expect that an individual would have information to compare her transportation allocation with other users. For individuals having the same goal (e.g., the shortest path to the destination from the same departure location and time), the peer to peer comparison may induce ‘envy’ if the user perceives his/her assigned travel option to be worse than that of his/her peers.

In turn, a user adjusts his/her travel options until he/she does not feel envy. This concept is an extension of the well-known travel behavior assumption called “User
Equilibrium”. Existing behavior models, however, do not allow users to compare their allocations with others. Furthermore, it is assumed that users have perfect information about their own alternative and all users are homogeneous.

This dissertation is dedicated to modeling a smart shared-mobility system, which accounts for individual
level of allocation. More specifically, we focus on the optimization of the allocation problem to achieve both system-wide efficiency and minimum envy among individuals. We consider envy to be an important allocation aspect in the transportation system. Maximizing the efficiency of a system necessarily brings about some level of unfairness where some users are allocated to inferior alternatives. When users having superior alternatives can compensate the envy of groups having inferior alternatives, an envy-free state can be achieved—which can be shown to be Pareto efficient state. Using a combination of pricing and incentives, we propose an optimization model to arrive at this new equilibrium.

This research has significant contributions in that the proposed model provides a framework to combine system-wide objectives with individual users’ utility objectives. Furthermore, we consider user heterogeneity, which has not been researched in the general area of transportation assignment. The proposed optimization model can be applied to pricing strategies both for commercial and public agencies, who have real-time information about customer characteristics and system performance.

Numerical results obtained from running our optimization model on both toy and real networks show that the proposed model converges to both envy-free and system optimum states with appropriate allocation and pricing schemes. Our findings show that the proposed smart mobility system technically works efficiently without governmental subsidy since the budget-balance mechanism trades off credits among users. In addition, the level of user heterogeneity affects the amount of credits charged or disbursed.