Download File

Vehicle-to-Vehicle Inductive Charge Transfer Feasibility and Public Health Implications

Promiti Dutta

Submitted in partial fulfillment of the requirements for the degree of Doctor of Public Health under the Executive Committee of the Mailman School of Public Health

COLUMBIA UNIVERSITY

2021

Promiti Dutta

Abstract

Vehicle-to-Vehicle Inductive Charge Transfer Feasibility and Public Health Implications

Promiti Dutta

There has been an increased push away from the traditional combustion-engine powered vehicle due to environmental, health, and political concerns. As a result, alternative methods of transportation such as electric vehicles (EVs) have gaining popularity in the market. However, the

EVs are not penetrating the market as quickly as expected, due in part to a combination of range, charge anxiety, and their financial costs. EVs cannot travel far due to limited driving range and require longer charge times than combustion-engine powered vehicles to recharge. Coupled with a lacking infrastructure for charging, the feasibility of an all-electric transportation market is still not possible.

We propose a novel system in which we study and characterize the feasibility of increasing the effective driving range of a battery electric vehicle by utilizing inductive charge transfer to create an ad-hoc charging network where vehicles can “share” charge with one another. The application of wireless charge transfer from vehicle-to-vehicle (V2V) is the first of its kind and does not require any changes to current metropolitan infrastructures. Through the use of computer networking and communications algorithms, we analyze real-world commuter and taxi data to

determine the potential effectiveness of such a system. We propose a participation and incentive mechanism to encourage participation in this network that enables the system to be functional.

To illustrate proof of principle for V2V charging at traffic lights, we simulate a simplified model in which vehicles only exchange charge at traffic lights without coordination with other vehicles. Using a greedy heuristic, vehicles only exchange charge if they happen to meet another vehicle that has charge to share. The heuristic is greedy since decisions are made at each iteration with longer optimality not being considered. We are able to demonstrate an increase in effective driving range of EVs using these simplistic assumptions.

In this thesis, we develop and quantify a complete simulation framework, which allows

EVs to operate using charge sharing. We analyze data from the United States Department of

Transportation, New York City Taxi and Limousine Commission, and Regional New York City data sources to understand the cumulative driving distance distributions for passenger/commuter vehicles and taxicabs in large metropolitan areas such as New York City. We show that the driving distributions can best be represented as heavy-tail distribution functions with most commuter vehicles not requiring additional charge during a typical day’s usage of their vehicle as compared to taxicabs, which regularly travel more than 100 miles during a 12-hour shift.

We develop and parameterize several variables for input into our simulation framework including driving distance, charge exchange heuristics, models for determining pricing of charge units, traffic density, and geographic location. The inclusion of these parameters helps to build a framework that can be utilized for any metropolitan area to determine the feasibility of EVs.

We have performed extensive evaluation of our model using real data. Our current simulations indicate that we can increase the effective distance that an electric vehicle travels by a factor of at least 2.5. This increased driving range makes EVs a more feasible mode of

transportation for fleet vehicles such as taxicabs that rely heavily on commuting long cumulative distances. We have identified areas for future improvement to add further parameters to make the model even more sensitive.

Finally, we focus on the application of our charge sharing framework in a real-world application for utilizing this methodology for the New York City bus system. In partnership with the New York City MTA, we launched a feasibility study of converting the currently majority hybrid bus fleet into a complete electric bus fleet with charging available at bus stops during scheduled bus stops. Unlike the earlier charge sharing framework, this simulation focuses on discrete distances that are traveled by the bus before having an opportunity to charge at the next bus stop. In this scenario, a large source of variability is the amount of time that the bus is able to stop at a bus stop for charging since this is determined by the amount of time needed to successfully embark and disembark the passengers at the given bus stop. This particular variability impacts how much charge the bus is able to gain during any given stop.

We conclude with a list of opportunities for future work in expanding the model with additional parameters and conclusions of our work. Further, we identify areas of further research that outline the potential positive and negative outcomes from a charge sharing system that can be extended to various other applications including micro-mobility applications such as electric scooters and bicycles.

Table of Contents

List of Abbreviations ...... vi

List of Figures ...... viii

List of Tables ...... xiii

Acknowledgments ...... xv

Dedication ...... xvi

Specific Aims ...... 1

Chapter 1: Introduction ...... 4

1.1 Motivation ...... 4

1.2 Disadvantages of combustion-engine powered based vehicles ...... 5

Environmental Concerns ...... 6

Health Concerns ...... 7

Political issues ...... 18

1.3. Contributions ...... 18

1.4. Outline of Thesis ...... 21

Chapter 2: Current and New Charging Technologies ...... 24

2.1 Types of Electric Vehicles ...... 24

Categorizing battery electric vehicles ...... 26

Advantages of BEVs ...... 27

Disadvantages of BEVs ...... 27

2.2 Current battery technology ...... 28

Types of Batteries ...... 29

Battery life ...... 30

2.3 Current battery charging technology ...... 31

Inductive Charge Transfer (ICT) ...... 31 Alternative methods for wireless charging ...... 35

2.4 Related Work: EV Charging Applications ...... 36

2.5 Vehicle-to-vehicle wireless charging ...... 38

2.6 Conclusions ...... 39

Chapter 3: Vehicle to Vehicle Charge Sharing Framework ...... 41

3.1 Development of simulation framework ...... 41

Proof of principle model ...... 42

Framework and Parameters ...... 42

Advanced Framework: Scheduling Rendezvous Points using Fisheye State Routing ...... 47

3.2 Input data ...... 53

Driving distance ...... 54

Heuristics for exchanging charge ...... 54

Modeling incentives using game theory ...... 54

Nash bargaining solution ...... 61

3.3 Testing and convergence ...... 63

3.4 Conclusions ...... 63

Chapter 4: Driving Data ...... 64

4.1 Defining Fleet Vehicles ...... 64

Structured fleet vehicles ...... 64

Unstructured fleet vehicles ...... 65

4.2 Unstructured Fleet Vehicle Data Sets ...... 66

Commuter Vehicles ...... 66 Omnibus Household Survey ...... 66

Regional Household Transportation Survey ...... 68

New York City Taxicabs ...... 69

4.3 Conclusions ...... 79

Chapter 5: Simulation Framework and Findings ...... 80

5.1 Proof of Principle: General Charge Sharing Network ...... 80

Driving distance ...... 80

Stopping Distance ...... 81 Unit Charge Transfer ...... 81

5.2 Case study: can all taxicabs be electric? ...... 85

Probability of refueling vehicle ...... 86

Percent loss of energy in system ...... 89

Effectiveness of System ...... 90

Simulation Conclusions ...... 92

5.3 Offering incentives for participation ...... 94

5.4 Conclusions ...... 97

Chapter 6: New York City Fleet Bus Study ...... 98

6.1 Background ...... 100

6.2 Data ...... 102

iii

6.3 Data Preprocessing ...... 104

6.4 Probabilistic Model ...... 105

6.5 Optimization ...... 109

6.6 Conclusion ...... 113

Chapter 7: Future Work and Considerations ...... 114

7.1 Extension to structured fleet vehicles ...... 114

7.2 Extension to other types of unstructured fleet vehicles ...... 115

7.3 Determine charge transfer heuristics that optimize transfer based on priorities ... 115

7.4 Application to other geographic areas ...... 116

7.5 Add time as a factor to stochastic model ...... 117

7.6 Vary population density of model ...... 118

7.7 Model Charge Entering System ...... 118

7.8 Potential Positive and Negative Public Health Consequences ...... 119

7.9 Introducing Charge Sharing as a Reality ...... 121

7.10 Conclusions ...... 121

Chapter 8: Conclusions ...... 123

References ...... 126

List of Abbreviations

σDD standard deviation driving distance

μDD mean driving distance

AIC Akaike Information Criterion

BC black carbon

BEV battery electric vehicle

CATI Computer Assisted Telephone Interviewing

COPD chronic obstructive pulmonary disease csv comma separated value d driving distance

DD driving distance distribution

DOT Department of Transportation

EC elemental carbon

EF efficiency transfer

EPA Environmental Protection Agency

EV electric vehicle

FCFS first come first serve

FedEx Federal Express

FSR fisheye state routing gps global positioning system

ICE internal combustion engine

ICT inductive charge transfer

MTA Metropolitan Transit Authority

NAAQS National Ambient Air Quality Standards

NBS Nash bargaining solution

NHTS National Household Transportation Survey

NJTPA New Jersey Transportation Planning Authority

NYC New York City

NYCAAS New York City Community Air Survey

NYMTC New York Metropolitan Transportation Council

RHTS Regional Household Transportation Survey

SD stopping distance

TLC Taxi and Limousines

TRAP traffic related air pollution

UPS United Parcel Service

UTRC University Transportation Research Center

UT unit charge transfer

V2V vehicle-to-vehicle

VMT vehicle miles traveled

WAVE Wireless Advanced Vehicle Electrification

WTA willingness to accept

WTP willingness to pay

vii

List of Figures

Chapter 1 Figure 1.1: Figure adapted from USAfacts.org to show how different types of electric vehicles have sold in the US market since 2000.

Figure 1.2: Figure adapted from National Geographic Energy Blog [Chameides 2014]. The graph shows an increasing trend for VMT and VMT per capita until 2004 when a downward trajectory ensues.

Figure 1.3: Process flow diagram used to identify papers for literature search.

Chapter 2 Figure 2.1: Charge rates for varying types of batteries. Figure adopted from [Botsford and Szczepanek 2009].

Chapter 3 Figure 3.1: Stochastic simulation framework showing the inputs and parameters required.

Figure 3.2: Simplified proof of principle simulation framework showing inputs for the model.

Figure 3.3: Fisheye scopes for vehicles. The smallest radii occurs when cars are least desperate for charge and expands until the radii, r, approaches the number of units of charge remaining for the vehicle at critical point.

Figure 3.4: The desperation factor model is represented by an inverted parabola. The maxima of the parabola is the largest radius that a vehicle will travel and is set

viii

at charge = 50 which is also the mean of the driving distribution of all vehicles. X- axis represents charge remaining (in units) and y-axis represents distance (in units).

Figure 3.5: Process schematic for each vehicle in the simulation at each iteration.

Figure 3.6: Screenshot from simulation showing movement of cars. Each square represents an active car in the simulation with the color indicating its appropriate state.

Figure 3.7: Example of WTA and WTP pricing using logistic functions. The formulation of these functions is described in Section II.

Figure 3.8: Graphical interpretation for NBS.

Chapter 4 Figure 4.1: Figure adopted from [United States Census Bureau 2013] showing main modes of transportation for commuting to work for the United States in 2011.

Figure 4.2: Histogram showing commuter driving distribution for typical day commute.

Figure 4.3: Cumulative driving distance histogram for 12,815 vehicular cases in the RHTS survey.

Figure 4.4: Number of active taxicabs on the streets of New York based on hour of day for the 365 days in 2012. Each color represents a different day.

Figure 4.5: Scatterplot showing variation of paid driving distances by date. Notably, the beginning of the year has larger cumulative driving distances than the

rest of the year. The larger distances are likely due to increased taxicab rides to the area airports and inclement weather usage.

Figure 4.6: Total number of paid taxicab rides per month in 2012. March has the most number of rides, while November has the least. The average number of rides per month is more than 14 million.

Figure 4.7: Minimum distance histogram of distances traveled by NYC taxicabs during 12-hour shifts can be modeled by a Rician distribution.

Figure 4.8: Maximum distance histogram of distances traveled by NYC taxicabs during 12-hour shifts can be modeled by a Rician distribution.

Chapter 5

Figure 5.1: p(refueling) for varying μDD and σDD = 45.5 units. We notice the increase in p(refueling) for driving distances greater than 150 units. Compared to the average driving distance of 100 units, we show a 50% increase in the distance a car can travel without needing to refuel.

Figure 5.2: Using a DD with μDD = 92.5 units and σDD = 45.5 units, the effect of SD is shown to increase following a logarithmic curve.

Figure 5.3: Using a DD with μDD = 92.5 units and σDD = 45.5 units, the effect of an increased SD still follows a logarithmic trend.

Figure 5.4: Effect of UT on p(refueling) with varying SD with constant DD (μDD =

92.5 units and σDD = 45.5 units).

Figure 5.5: Graph showing the cumulative arrival of vehicles into the simulation.

Figure 5.6: The probability of refueling decreases as the ratio of taxis to commuter vehicles increases.

Figure 5.7: The percentage energy loss in system increases for both the charge leaving system and the charge lost as the ratio of taxicabs to commuter vehicles increase.

Figure 5.8: The percentage of total energy needed in the system decreases as the ratio of taxicabs to commuter vehicles increases.

Figure 5.9: shows the effect of varying the cost of a unit charge, cost of failure and distance between charge transfers when a charge/unit = $0.25.

Figure 5.10: demonstrates the effect of varying the cost of a unit charge, cost of failure and distance between charge transfers when a charge/unit = $0.50.

Figure 5.11: shows the effect of varying the cost of a unit charge, cost of failure and distance between charge transfers when a charge/unit = $0.75.

Figure 5.12: demonstrates the effect of a charge unit on the probability of refueling (i.e. not reaching destination without failing) for varying costs of charge unit. We notice that the as distance between charge transfer increases, the probability of refueling also increases.

Chapter 6

Figure 6.1: Map of the B63 bus route. Figure adapted from MTA.

Figure 6.2: Six different trips from the archive showing a variety of frequencies of recorded data-points for different geographic locations.

Figure 6.3: Depiction of a single trip, using different granularity parameters. As granularity is increased, the data-points are more dispersed throughout the map.

Figure 6.4: The graphical model, which describes the conditional dependencies of the observed variables, namely histograms and hidden parameters which are multinomial distributions.

Figure 6.5: An MM distribution example. Each bin number represents a stop and the height correspond to amount of time spent at a particular stop and each color represents a trip pattern.

Figure 6.6: The result of clustering the B63 bus route data set into K=10 categories. Every cluster then is averaged on the time that the vehicles have spent at stops only, obtaining 10 different histograms. These histograms are then used to optimize the electric charge down time across all different trips that might happen.

Figure 6.7: The graph shows the best achievable utility decreasing with respect to number of chargers that can be installed throughout the B63 route.

xii

List of Tables

Chapter 1 Table 1.1: Illustrative example of how inclusion and exclusion criteria for search on Pubmed filtered number of results for literature search.

Table 1.2: Excluded papers from literature search upon further investigation.

Table 1.3: Included papers from literature search and generic study descriptors.

Table 1.4: Review of papers found showing direct health outcomes for exposure to

NO2 and PM2.5.

Table 1.5: Hospital Admissions associations to PM2.5 and NO2 for identified studies. Unit change for measurements vary among studies. Measurements are taken in microgram per cubic meter.

Table 1.6: Example pricing of full-size 2014 model BEVs currently available in the United States vehicle market.

Chapter 2 Table 2.1: Main types of electric vehicles and components available for refueling.

Table 2.2: Examples of currently available BEVs (since 1998) for each category of electric vehicle.

Table 2.3: Examples demonstrating range and charging limitations of currently available BEVs.

xiii

Table 2.4: Technical specifications and time needed for charging at different levels (normal and fast).

Table 2.5: Examples of currently available IPT devices on the market along with their published specifications.

Chapter 3 Table 3.1: Description of tests performed to determine sensitivity of our model.

Chapter 4 Table 4.1: Differences between Structured and Unstructured Fleet Vehicles.

Table 4.2: Description of relevant data fields in NYC TLC FOIL 2012 data set.

Chapter 5

Table 5.1: Comparison of refueling probabilities based on different methods

assuming a 1:1 ratio of taxicabs to commuter vehicles.

Chapter 6

Table 6.1: Metadata for data provided by MTA on bus route B63.

xiv

Acknowledgments

I would like to thank my dissertation committee for their time and their suggestions and guidance throughout my doctoral studies. Professors Ana Navas-Acien and Greg Freyer have provided unending support, encouragement, and flexibility in the pursuit of my doctoral thesis. Words cannot express the amount of gratitude I have for their passion to help me succeed and for teaching me how to become a better researcher. I am also thankful for the other members of my thesis committee who have taken time to help guide me on this journey by providing helpful feedback at critical points throughout the process. Thanks

Drs. Markus Hilpert, Darby Jack, and Marianthi-Anna Kioumourtzoglou.

This journey would have not been feasible without the support of family. Thanks to my parents (Probir and Indrani) and brother (Preetam), who have given me love and support from a young age to follow an academic path. I would also like to express my deepest gratitude to my husband, Alexander, for his patience and support throughout my academic tenure. He has spent endless hours discussing, participating, and helping.

Most importantly, thanks to my children, Maya and Neil, who have been an inspiration for the completion of this thesis. Whether attending meetings and conferences with me or attempting to sit idly while I give a remote presentation for a conference, my little children have demonstrated patience in helping their mother complete her work.

Dedication

To My Maya and Neil.

xvi

Specific Aims

In an effort to shift from traditional combustion-engine powered vehicles given concerns around health and environmental effects from traffic-related air pollution (TRAP), electric vehicles (EVs) have gained popularity as an alternative solution. Numerous studies have reported an association between TRAP and several adverse health outcomes. Therefore, reducing TRAP through the utilization of EVs can be a method to reduce such health outcomes.

However, the driving range of a typical EVs leaves much to be desired for those wanting to travel long distances, especially in areas where charging a vehicle is difficult. Our goal is to demonstrate that vehicle-to-vehicle (V2V) charging can increase the driving range of an EV and reduce health and environmental exposures to harmful pollutants. We illustrate our proof of principle for V2V charging at traffic lights.

SPECIFIC AIM 1:

Develop a quantifiable simulation framework allowing EVs to operate using charge sharing

We hypothesize that charge sharing can effectively extend the driving distance of an EV. To prove this, we create a simulation with a simplified model where vehicles only exchange charge at traffic lights without coordination with other vehicles for charge transfers. As a result, vehicle only exchange charge if they happen to meet another vehicle that has charge to share. With these simplistic assumptions, we will perform extensive evaluation of this hypothesis using real data to determine the validity of our assumption.

SPECIFIC AIM 2:

Design and parametrize variables for input into our simulation framework including driving distance, charge exchange heuristics, pricing of charge units, traffic density, and location

We analyze data from the United States Department of Transportation, New York City Taxi and

Limousine Commission, and Regional New York City data sources to understand the cumulative driving distance distributions for passenger/commuter vehicles and taxicabs in large metropolitan areas such as New York City. Our hypothesis is that driving distributions can be represented as heavy-tail distribution functions with most commuter vehicles not requiring additional charge during a typical day’s usage of their vehicle as compared to taxicabs, which regularly travel more than 100 miles during a 12-hour shift.

We create a more complex simulation to be more realistic of how charge-sharing interactions may occur through coordinated efforts. The addition of these parameters will help to build a framework that can be utilized for any metropolitan case scenario to determine the feasibility of switching to EVs. Our main objective is to retain excess charge in the system to increase the opportunities for vehicles needing charge to acquire charge in order to reach their destinations while minimizing the delay for vehicle to reach its destination.

SPECIFIC AIM 3:

Examine the feasibility of converting New York City transit buses to all-electric buses

We partner with the New York City Metropolitan Transit Authority (NYC MTA), to determine whether the conversion of their current hybrid bus fleet is possible to become all electric. Unlike the charging paradigm, we establish in Specific Aims 1 and 2, we recognize that the simplicity of this model relies on guaranteed stops at pre-designated bus stops during which the bus would be

2 able to gain charge wirelessly through a charge pad at the bus stop. Our main consideration is to determine given existing data the likelihood of a bus unable to complete its route due to lack of charge. We focus on a proof of concept using one bus route and using historical data for that bus route. We then model the bus driving distribution and determine whether an optimal solution is available for installing charge pads that would allow the bus to run continuously on its route without failing due to lack of charge on its route. We create the generalizable framework that can be used to model each of the bus routes to determine the optimality of the number of chargers needed for each route for the MTA to be able to convert to an all-electric bus fleet.

Chapter 1: Introduction

1.1 Motivation

Vehicles, both personal and commercial, have become a ubiquitous form of transportation in the developed world. The auto industry is amidst a technological transformation in identifying alternative sources of energy to power vehicles due to two driving forces: environmental pollution prevention and depletion of fuel resources. This drive for developing “smarter” solutions to create a “smarter planet” is crucial to advancing the technological science of electric vehicles (EVs). As alternative methods for creating energy are being sought, we see an increased interest in electric vehicles as one potential solution for lessening our dependence on fossil fuels.

In 2011, President Obama announced in his State of the Union address that his administration would push to have 1-million electric vehicles one the road by 2015. In 2019, about

727,000 electric-drive vehicles were sold and just under half of those were plug-in electric cars capable of operating on electricity alone. For comparison, a total of 17 million new light-duty vehicles were sold in 2019 according to the Bureau of Transportation Statistics.

Currently, the main hurdle with adoption of EVs is due to their limited driving range and ease of recharging, in addition to their financial cost. The main aim of this thesis is to propose a model framework in which we can increase the feasibility of EVs in the current market with already available battery and charging technology.

Figure 1.1: Figure adapted from USAfacts.org to show how different types of electric vehicles have sold in the US market since 2000.

1.2 Disadvantages of combustion-engine powered based vehicles

The internal combustion engine (ICE) has been the predominant energy choice for vehicles for more than a hundred years [Tauber 1995]. Especially, in the United States, driving is typically a primary mode of transportation with each American traveling on an average of less than 10,000 miles a year. The total country averages near 3 trillion total vehicle miles traveled per year (VMT).

There has been a paradigm shift in the recent years, with a drop in both miles traveled and vehicle miles traveled per capita [Chameides 2014]. According to a graph produced by the National

Geographic using data from Federal Highway Administration’s Traffic Volume Report, there has been a steady increase in vehicle miles traveled and vehicle miles travelled per capita since 1985 until 2004 (Figure 1.1). However, in 2004, there was a steady downhill trajectory to this figure.

Some speculate that a combination of reasons may have led to this downward trend including a failing economy, increasing gasoline prices and a general increase in environmental awareness.

While the internal combustion engine has seen minor improvements and changes since its original creation over a hundred years ago, there has been a growing concern in the various effects caused by combustion-engine powered vehicles. Environmental, health, and political reasons have all spurred the advancement of technology to search for alternative methods to power vehicles.

Environmental Concerns

The Environmental Protection Agency (EPA) states that 28% of all greenhouse gas emissions in

2011 for the United States were from transportation related sources [Greenhouse 2014]. These are the second largest sources of greenhouse gases in the United States after electricity. The United

States is also the second highest carbon dioxide emitter after China [Houghton 2008]. These emissions are primarily from burning fossil fuels for use in cars, truck, ships, trains and planes.

Interestingly, while vehicles have become more and more environmentally friendly with lower emissions, there has still been a steady rise in greenhouse gases from these modes of transportation.

The EPA estimates that there has been an increase of approximately 18% since 1990 for transportation related greenhouse gases, which is likely due to the fact that there are more vehicles on the road (34% increase from 1990 to 2011 for total vehicles miles traveled)!

Air pollution from internal combustion engines occur long before the vehicle emissions as simple fuel storage transfers at gas stations release from the gas pump [Adria Mora and Hilpert

2017]. Common markers used to measure such pollution include but are not limited to pollutants as ultrafine particles, carbon monoxide, nitrogen dioxide, black carbon, polycyclic aromatic hydrocarbons, and some metals [WHO 2013].

Figure 1.2: Figure adapted from National Geographic Energy Blog [Chameides 2014]. The graph shows an increasing trend for VMT and VMT per capita until 2004 when a downward trajectory ensues.

Health Concerns

Traffic-related air pollution (TRAP) has gained increased interest and more studies are trying to quantify and qualify air pollution caused from traffic [Hilpert et al. 2019]. Studies have shown that traffic-related air pollution from traditional vehicles leads to several adverse health issues such as pregnancy loss [Kioumourtzoglou et. al. 2019], asthma and adverse cardiorespiratory effects

[McConnell et. al. 2010], as well as an increased premature mortality related to increased risk of cardiovascular disease and other chronic diseases [Finkelstein, Jerrett, Sears 2004]. For other health outcomes, there are inconsistent findings. For example, studies related to TRAP’s association with childhood autism are controversial. Even though there are hundreds of studies 7 available on TRAP and health outcomes, not all studies are of equal quality. Definitions of exposures, markers, criteria for inclusion/exclusion, and impact are not standardized making it difficult to compare outcomes from different studies. Even with these issues in comparing studies, it is widely recognized by the scientific communities that exposures to primary traffic-generated pollutants are of public health concern and thus require more attention.

Nitrogen oxide (NO2) is considered a commonly known traffic-related pollutant, since it is emitted from vehicles with internal combustion engines [Richmont-Bryant et. al. 2017]

[Hesterberg et. al. 2009]. It is a known irritant of the respiratory system since it can penetrate deep into the lungs creating respiratory conditions such as coughing, wheezing, dyspnea, bronchospasm, and even pulmonary edema when exposure is at high. High concentrations were even shown to affect T-lymphocytes, particularly cells that are known to produce immune response [Chen,

Gokhale, Shofer, Kuschner 2007]. Longer term exposure to NO2 can lead to chronic lung disease and even impair the sense of smell [Chen, Gokhale, Shofer, Kuschner 2007]. [Chen, Gokhale,

Shofer, Kuschner 2007] Other symptoms aside from respiratory symptoms can be seen from exposure to NO2 such as symptoms of eye irritation.

PM2.5 is another pollutant commonly associated as a surrogate for TRAP. PM2.5 are tiny airborne solid and liquid particles less than 2.5 microns in diameter. In NYC, PM2.5 comes from inside and outside the city from all kinds of combustion activity [Kheirbek et. al. 2016]. These can include the fuel burning in vehicles, buildings, power plants, and construction equipment, as well as commercial cooking and industrial activities [Kheirbek et. al. 2016]. PM2.5 can either come directly from these sources or be formed in the atmosphere from other pollutants and be transported via air to other geographic areas. For NYC, approximately 20% of PM2.5 is from traffic related air pollution and varies by season and location [Kheirbek et. al. 2016]. Therefore, while PM2.5 is

8 generally accepted as a surrogate for TRAP, only a portion of the concentrations recorded can be directly associated to direct vehicle emissions.

There have been many epidemiological studies on the health effects of particulate matter.

These have shown a positive relation between long term and short-term exposure of PM2.5 and acute nasopharyngitis [Zhang et. al. 2019]. In addition, long-term exposure to PM over the course of years is found to related to cardiovascular diseases and infant mortality [Zhang et. al. 2019].

Long-term chronic effects showed respiratory diseases and issues with the immune system [New

Hampshire 2019]. People with pre-existing conditions such as asthma, pneumonia, diabetes, respiratory and cardiovascular diseases are more susceptible and vulnerable to the effects of particulate matter [Kappos et. al. 2004]. PM2.5 is strongly associated with respiratory system diseases [Kappos et. al. 2004] since the size of the particulate matter permits for it to more easily piece into interior spaces [Boschi 2012].

Literature Review of Health Effects

We performed a literature search for peer-reviewed published scientific papers with a focus on studies from the New York City area (Figure 1.2) to understand the known health effects of

TRAP to health outcomes, This is since the simulation and data analysis that we conduct in this is done for the New York City area (including the 5 boroughs). We utilized the following digital databases: Google Scholar, Pubmed, Embase. Search terms, including free text and indexing vocabulary (e.g. Mesh terms) included the following words and phrases: traffic, air pollution, nitrogen dioxide, health effects, cardiovascular, respiratory, environmental effects.

We limit our papers to those where the research is conducted post the year 2000. The air pollution created by vehicles in recent days is different from cars in earlier decades. Years of

9 technological advances have helped for cars to become lower in emissions and thus, we wanted to understand the current day effects of traffic related air pollution on health.

There are many papers that address air pollution and health outcomes in various geographic areas. However, the search narrowed with the inclusion of only TRAP related outcomes. We narrowed the results focusing only on papers that did their studies in the NYC area since the data that we use for the purposes of our simulation and modeling framework to show the feasibility of all electric vehicles is from NYC. The idea was to understand the potential health benefits that could be expected with the feasibility of vehicle-to-vehicle charge utilizing electric vehicles in

NYC. Table 1.1 provides an illustrative example of how search results were narrowed down for our literature search on Pubmed.

The result of this cursory inclusion/exclusion principle leads to 22 papers (Table 1.2 and

Table 1.3). However, upon further investigation of the papers by reading through the full manuscripts, many of the papers did not directly study TRAP emissions in direct correlation to health outcomes even though keyword searches indicated presence of such words in the paper.

Conference abstracts that were not full manuscripts were also excluded. A total of 17 manuscripts were included for final review.

The specific association of explicit-stated TRAP to health outcomes in the NYC area is limited from the search criteria we utilized. After review of the 17 papers, we noticed that many of the papers did not fit the criteria for the following reasons: 9 papers in which the study was not specific to TRAP, 1 was a grant submission and had no actual outcomes yet, 2 papers were simulation studies, 4 papers were there was no association to specific health outcomes (Table 1.2).

This resulted in 6 papers for actual review. These papers are from two main research teams who have published additional information in each subsequent publication from their original study.

We also note that the 5 of the 6 studies are completed as prospective cohort studies, while one was completed as a case control study comparing asthmatics to non-asthmatics (Table 1.3). For many of the studies, the cohorts followed children that were known be asthmatics and thus associated the severity of the condition as it related to TRAP. Given that the subjects were already asthmatic, the study could not conclude the causality of asthma as a result of TRAP. Additionally, the time duration of each of these studies is limited, with most studies lasting between 3 weeks and 5 months, and only one study running for 5+ years. Thus, the long-term effects through follow-up are lost to understand the longer-term effects of the TRAP to the human (Table 1.3).

The selection of these studies limits us to younger children since no papers were identified using this search criteria to associate outcomes for adults and senior populations regarding TRAP and health outcomes using the search criteria for this review (Table 1.3). This is a limitation of the search criteria used to identify papers for the literature search. Opening the search to include known key surrogates of traffic pollution as search terms would have likely resulted in a more comprehensive outcome of the effects. For instance, [Ito et. al. 2011] showed that cardiovascular disease hospitalizations and cardiovascular disease mortality are not correlated but each are independently associated with PM2.5 exposure. [Ito et. al. 2011] also demonstrated that elemental carbon and NO2 showed the most consistent associations with cardiovascular disease outcomes throughout the year and are important in adverse health effects in NYC.

The outcomes for the three prospective cohort studies [Thurston et. al. 2004] [Spira-Cohen et. al. 2006] [Spira-Cohen et. al. 2011] where the subjects in the study were known to be asthmatics indicate that while continuous PM2.5 exposure did not have consistent results for its effect to asthma wheezing, coughing, lung function, and shortness of breath, we notice that elemental carbon (EC) was correlated to these health conditions in these individuals consistently in these

11 studies. This implies that presence of EC from TRAP can be associated with asthma related health outcomes and severity of the individual’s asthma condition. [Patel, Quinn, et. al. 2011], [Patel,

Chillrud, et. al. 2013], and [Patel et. al. 2010] show that there is an increased correlation with age of asthma wheezing and shortness of breath as it relates to TRAP related PM2.5, EC, and NO2.

PM2.5 was not consistently shown in these three studies to have an effect, but black carbon (BC),

NO2, EC, had strong correlations that held consistently among the 3 different studies [Patel, Quinn, et. al. 2011] [Patel, Chillrud, et. al. 2013] [Patel et. al. 2010]. Since all these studies were conducted with minors (under the age of 18), we need to recast our literature search to be able to be more inclusive of other studies in order to capture the correlation of TRAP with adult health outcomes.

Identify records searching databases Screen records to (PubMed, EmBase, Remove duplicate ascertain that they Google Scholar, etc.) records include quantitative using keywords synthesis of findings identified

Review remaining records to finalize Exclude records that records: TRAP in are not in the NYC NYC area to health region and were not outcomes published post-2000

Figure 1.3: Process flow diagram used to identify papers for literature search.

Table 1.1: Illustrative example of how inclusion and exclusion criteria for search on Pubmed filtered number of results for literature search. Search Term Results air pollution 58,674 air pollution + health effects 22,477 traffic related air pollution + health effects 1,386 traffic related air pollution + health effects + New York City 22 Final number of full text papers after review 17

Table 1.2: Excluded papers from literature search upon further investigation. Study Exclusion Reason Maciejczyk et. al. 2004 Claudio et. al. 2006 Matte et. al. 2013 Kheirbek et. al. 2014 Kheirbek, Ito, et. al. 2014 not specific to TRAP Humphrey et. al. 2019 Choe et. al. 2019 Connerton et. al. 2020 Shearston et. al. 2020 Spira-Cohen 2006 grant submission Jensen et. al. 2009 simulation based study Kheirbek et. al. 2016 Spira-Cohen et. al. 2009 Patel et. al. 2009 no association to specific health outcomes Ross et. al. 2011 Clougherty et. al. 2013

Table 1.3: Included papers from literature search and generic study descriptors. Sample Age Study Study Environmental Outcome Study Location Size (years) Duration Design Ethnicity Markers Measure PM2.5 and Asthma Thurston et. prospective not elemental (wheezing), al. 2004 South Bronx 10 10-12 3 weeks cohort reported carbon cough, lung function, shortness of Spira-Cohen ~5 prospective not breath et. al. 2006 South Bronx 40 10-12 months cohort reported African American/ Hispanic of Spira-Cohen prospective Caribbean et. al. 2011 South Bronx 40 10 - 12 1 month cohort descent 249 (57 PM2.5, EC, BC, Asthma

14 NYC High asthmatics/ NO2 wheezing, lung

Schools 192 Hispanic/A function, Patel et. al. (location nonasmathi 1 - 1.5 prospective frican shortness of 2010 confidential) cs) 13-20 months cohort American breath Northern Patel, Manhattan Dominican Quinn, et. and South 593* (727 prenatal prospective /African al. 2011 Bronx recruited) - 5 5+ years cohort American Patel, NYC (school 18 African Chillrud, et. locations asthmatics/ case American al. 2013 confidential) 18 healthy 14-19 1 month control /Hispanic

Since the literature review resulted in very few findings, we expand the literature search to use different keywords, specifically “hospital admission and air pollution.” Unlike our prior search, this time we focused on identifying papers where the outcome resulted in a hospital admission for humans and all papers were considered as long as they were written in English. We also did not limit the search by geography to NYC but rather focused globally on any study.

Further, we only included papers where exposures for NO2 and PM2.5 were investigated as these are known surrogates for traffic-related air pollution. We excluded studies that did not have analysis of outcomes by such exposure. Table 1.4 shows the studies identified by location, study design, sample size, study duration and hospital admissions reason.

Table 1.5 shows the hospital admissions reason associations to PM2.5 and NO2 for the studies identified in Table 1.4. The review shows that the effect of air pollution like NO2 and PM2.5 have an association to increased risk for hospital admission for certain cardiovascular and respiratory illnesses. The papers we identified also showed correlations to other air pollutants such as ozone, sulfur dioxide, and PM10. PM2.5 is shown to have a higher influence on hospital admissions for cardiovascular and respiratory conditions than NO2 and other air pollutants. The biological reasoning for this is since particulate matter can penetrate deeper into the lungs and heart and has been shown to alter the autonomic control of the heart [Brook et. al. 2002]. However, this maybe an over-simplification as most of the particulate matter is secondary and more homogenous in space. As a result, there is smaller exposure measurement error than other more local pollutants such as NO2 and elemental carbon.

Since we have seen the association of PM2.5 to adverse cardiovascular and respiratory health effects, we can utilize this finding to understand the potential correlation to PM2.5 measurements in New York City. The New York City Community Air Survey (NYCCAS)

Table 1.4: Review of papers found showing direct health outcomes for exposure to NO2 and PM2.5. Study Location Study Sample Study Hospital Admission Design Size Duration Chen et. al. Australia Time-series 36,204 2003 - Asthma 2016 (Adelaide) & case- 2013 crossover Ferreira et. al. Brazil Time-series 2,727 2010 – Cardiovascular and 2016 2011 respiratory Xie et. al. 2014 China Time-series 422,716 2010 - Ischemic Heart Disease 2012 Zhang, Wang, China Time-series 1,461 2008 – Respiratory and stoke et. al. 2014 2011 Iskandar et. al. Denmark Case- 8,226 2001 - Asthma 2011 crossover 2008 Andersen et. al. Denmark Cohort 57,053 1993 – Asthma 2012 2007 Milojevic et. England Case- ~3 2003 – Cardiovascular, atrial al. 2014 crossover Million 2009 fibrillation, arrhythmia, heart failure Atkinson et. al. England Cohort 812,063 2003 - Chronic Obstructive 2014 (London) 2007 Pulmonary Disease (COPD) Tonne et. al. England Cohort 18,138 2003 - Readmission of 2016 (London) 2007 myocardial infarction

Kollanus et. al. Finland Time-series 165,259 2001 – Cardiovascular and 2016 2010 respiratory Wong et. al. Hong Cohort 66,820 1998 – Peptic, gastric, and 2016 Kong 2001 duodenal ulcers Ghozikali et. Iran Case- n/a 2008 – Chronic Obstructive al. 2015 crossover 2009 Pulmonary Disease (COPD) Alimohammadi Iran Retrospectiv 379 2012 – Ischemic stroke et. al. 2016 e cross 2013 sectional Vidale et. al. Italy Time-series 939 2000 – Ischemic stroke 2010 2003 Jevtić et. al. Serbia Time-series 10,469 2007 – Cardiovascular 2014 2009 admission Cheng et. al. Taiwan Case- 81,836 2006 – Asthma, pneumonia, 2015 crossover 2010 Chronic Obstructive Pulmonary Disease (COPD) Phung et. al. Vietnam Time-series 76,640 2004 – Cardiovascular and 2016 2007 respiratory

Table 1.5: Hospital Admissions associations to PM2.5 and NO2 for identified studies. Unit change for measurements vary among studies. Measurements are taken in microgram per cubic meter. PM2.5 NO2 Hospital Hospital Study Admission RR/OR/HR Admission RR/OR/HR Chen et. al. 2016 Asthma RR 30.2% Asthma RR 12.5% Iskander et. al. 2011 Asthma OR 1.09 Asthma OR 1.10 Andersen et. al. Asthma HR 1.12 2012 COPD HR 1.08 Asthma OR 1.10 Cheng et. al. 2015 COPD OR 1.11 pneumonia OR 1.12 Zhang, Wang, et. al. 2014 Respiratory RR 1.94 Atkinson et. al. 2014 COPD HR 1.05 COPD HR 1.06 Ghozikali et. al. 2015 COPD OR 1.0038 Respiratory RR 8.5% Ferreira et. al. 2016 cardiovascular RR 19.6% Respiratory RR 1.08 Phung et. al. 2016 Cadiovascular RR 1.04 Kollanus et. al. Respiratory RR 10.5% 2016 cardiovascular RR 1.5% Jevtic et. al. 2014 Cardiovascular RR 1.049 Milojevic et. al. Cardiovascu 2014 lar OR 1.7% myocardial Myocardial Tonne et. al. 2016 infarction HR 1.02 infarction HR 1.05 Ischemic heart Xie et. al. 2014 disease RR 0.27% Alimohammadi et. al. 2016 ischemic stroke RR 1.09 ischemic stroke RR 1.07 Vidale et. al. 2010 ischemic stroke RR 1.039 Peptic ulcer HR 1.18 Wong et. al. 2016 gastric ulcer HR 1.29 duodenal ulcer HR 0.98

measures PM2.5 in nearly 60 locations across NYC. In 2018, seasonal PM2.5 average concentrations ranged from 3.4 to 16.5 ug/m3 [NYCAAS 2018]. In 2012, the EPA lowered the National Ambient

3 Air Quality Standards (NAAQS) of annual PM2.5 to an average of 12 μg/m [Hinsdale 2020]. The

World Health Organization’s guideline is even lower at 10 ug/m3 [Hinsdale 2020]. Twelve NYC neighborhoods do not even meet the WHO standards, indicating the need for us to find ways to better control PM2.5 concentrations across NYC.

Political issues

Dependence on vast amounts of oil has political issues and can be of particular concern with the general availability of this resource. The Middle East is a large supplier of oil for the world given large deposits in that region. Other countries such as Iran and Venezuela also have large deposits of oil available in their region. Therefore, these countries have an inherent stronghold on other countries because of their precious commodity. Being able to find other methods so that a country is not dependent on other particular countries for resources, helps maintain independence.

1.3. Contributions

Our main objective is to find a method for increasing the driving range of an electric vehicle for usage in metropolitan areas without the costly need to change existing infrastructure (whether by adding charging stations or altering roadways). We propose a cost sensitive solution that does not require increasing the price of an electric vehicle. Since the battery is the most expensive component of the EV, our method makes it possible to decrease current battery size and still obtain the same driving range, helping defray the total cost of an EV.

Table 1.6: Example pricing of full-size 2014 model BEVs currently available in the United States vehicle market. Vehicle Average Miles 2014 MSRP Price BMW i3 80 – 100 miles $41,350 Chevrolet Spark EV 82 miles $26,685 Ford Focus EV 76 miles $35,170 Honda Fit 76 miles $36,625 Nissan Leaf 84 miles $28,980 Tesla Model S >200 miles $69,900 Toyota RAV 4 112 miles $49,800

Currently, batteries are one of the most expensive components of a battery electric vehicle.

Therefore, solutions that involve increasing battery capacity through improved battery technology are oftentimes increasing the overall cost of an electric vehicle significantly. Table 1.6 shows the current purchase prices of various battery EVs that are currently available on the market.

Our proposed solution of using inductive charge transfer (ICT) to have vehicles share excess energy vehicle-to vehicle (V2V) is a novel method for solving the traditional range and charge anxiety problems that are currently the reasons most often cited for the lack of interest in electric vehicles. We are the first to propose the concept of V2V wireless charging via ICT. Other research groups have investigated the usage of wireless charge transfer but in a more traditional sense that employs using a power source as the charge provider. Chapter 2 describes the background for these studies, which of course, require infrastructure changes for implementation.

However, in busy metropolitan areas, such as New York City, Boston, and San Francisco, these changes are extremely difficult to implement given governmental restrictions and overall inconvenience of closing roadways in an already congested city. Our V2V solution creates mobile

“charging” stations for electric vehicles to engage in short recharging periods at stoplights and/or predetermined rendezvous points. Each vehicle in the charging network has the option to “buy” or

“sell” charge from/to another vehicle depending on charge transfer heuristics. We apply methods

from computer communications, networking, and game theory to create a “social network” amongst participants in our electric vehicle charge sharing schema:

(1) to encourage participation through pricing incentives

(2) to lessen the burden of finding an appropriate partner for charge exchange that

minimizes the burden on both sides.

This “ad-hoc” network that is formed is constantly changing with participants entering and leaving the system as well as with potential constantly moving charge stations.

Feasibility studies were performed using Omnibus Household Survey by the United States

Department of Transportation (DOT). The survey collected average commuting distance in miles for a one-way commute between home and work for a person. We extrapolate round-trip commute distances from this data and model the distribution as a heavy-tailed lognormal with a mean, μ =

32 miles, and standard deviation σ = 26 miles. Using this data, we find that 100% of all commuter vehicles would be able to reach their destinations using our V2V charge sharing framework as opposed to the 4.9% of cars that would otherwise need to recharge in order to reach their destinations.

To determine the efficacy of our V2V framework, we test our model using data from the

Regional Household Transportation Survey in 2010 - 2011 conducted by the New York

Metropolitan Transportation Council (NYMTC) and the North New Jersey Transportation

Planning Authority (NJTPA). This large data set contains responses to self-reported survey questions to 28 regional counties in the New York, New Jersey and Connecticut area.

We perform a feasibility analysis for a system as such for vehicles that do not have a predetermined set path and are random in their usage for destinations. We use taxicab usage data for New York City in 2012 for our analysis of this problem. Most NYC taxicabs operate for 12

hours each shift and thus a standard electric vehicle would not be able to perform without require recharging during the shift. This recharge time is extremely costly to taxicab drivers as it is a loss of customers during the time and as a result a loss of potential revenue. Using our proposed network model, we are able to show that approximately 80% to 99% of all NYC taxicabs would be able to operate during their 12-hour shift without having to stop for recharging depending on the ratio of taxicabs to cars available in the system. This is in stark contrast to 18.8% - 20.7% of taxis that would be able to operate without recharging through an entire shift.

We also applied this framework to test the hypothesis of whether electric buses are feasible in New York City in partnership with the Metropolitan Transit Authority (MTA). Instead of exchanging charge with another vehicle at a rendezvous point, the bus would acquire charge during a scheduled stop at a bus stop during which time passengers were embarking and disembarking the vehicle. These stops unlike scheduling cars for rendezvous points are guaranteed since the bus has regularly scheduled stops with distance traveled between the two spots also discretely defined.

1.4. Outline of Thesis

In Chapter 2, we start by describing briefly the types of electric cars that currently exist. From there we have a brief description about the current battery and charging standards for these vehicles. We then focus our attention to a literature search that describes other methods for battery charging. We conclude the chapter with our proposal of car-to-car wireless charging.

Chapter 3 focuses on our charge sharing model and framework. In this chapter, we describe in detail the development of our stochastic simulation framework as well as technical information such as programming and running environment. We conclude the chapter with a discussion on the various inputs for the model and the testing conducted to ascertain model sensitivity and accuracy.

Chapter 4 describes our analysis of large data sets for the purposes of modeling and simulating in a real-world situation environment. We analyze data from the National Household

Transportation Survey of 2009 by the Federal Highway Association to identify driving distributions for vehicles. We also obtain and analyze data from the New York City Taxi and

Limousine Commission for all paid medallion rides in 2012 to understand best how taxicabs are operated in a large metropolitan area.

In Chapter 5, we turn our attention to the results obtained from our simulations. In the first two sections, we focus on a proof of principle simulation to ascertain that our method for vehicle- to-vehicle charge sharing will in fact increase the driving range of an EV. We extend our experimentation to a case study involving New York City taxicabs to determine whether an all- electric taxi fleet would be possible given our framework. Finally, we explore methods for pricing incentives to encourage participation in our charge-sharing network.

Chapter 6 focuses on the application of our charge sharing framework in a real-world application for utilizing this methodology for the New York City bus system. In partnership with the New York City MTA, we launched a feasibility study of converting the currently majority hybrid bus fleet into a complete electric bus fleet with charging available at bus stops during scheduled bus stops. Unlike the earlier framework of having to schedule rendezvous for charge sharing, this simulation focuses on discrete distances that are traveled by the bus before having an opportunity to charge at the next bus stop. The variance is the amount of time that the bus is able to stop at a bus stop for charging since this is determined by the amount of time needed to successfully embark and disembark the passengers at the given bus stop. This particular variability impacts how much charge the bus is able to gain during any given stop.

Chapter 7 discusses future work including variables that can be added to our charge sharing framework to make it more robust and adaptable to other types of environments beyond extremely urban settings like New York City. There is an entire body of work that can be done to evaluate the air pollution and health impacts of having all EVs that participate in a charge sharing network as opposed to traditional combustion-engine vehicles. Considering the broarder framework of the infrastructure needed to support each type of vehicle and the potential outcomes from having EVs requires more analysis to determine if the overall effect of EVs in a charge sharing network is truly a benefit to public health. Chapter 8 concludes this thesis with thoughts on the adaptability of the charge sharing framework to other types of transportation devices such as electric bicycles and scooters.

Chapter 2: Current and New Charging Technologies

The internal combustion engine has been the predominant energy choice for vehicles for more than

100 years [Tauber 1995]. However, there has been a push in recent years to find alternative fuel sources to the traditional combustion-engine powered vehicle due to environmental, health and political concerns. Between environmental pollution and fossil fuel availability concerns, the vehicle market has been evolving relatively quickly [Harris 2009]. In this thesis, we focus on the feasibility of fully battery powered electric vehicles (BEVs). However, prior to BEVs there were several other types of electric vehicles, which were available on the market.

This chapter focuses on an introduction to the types of electric vehicles (EVs) and the classification used to describe the models. We then provide a description of the current battery and charging technology that exists for EVs. We proceed to describe the current state of art in wireless charging technology and existing research for wireless charging implementations for EVs. To conclude this chapter, we discuss the novel concept of vehicle-to-vehicle wireless inductive charging to create a new method and application for wireless charging that may serve to make EVs more feasible in their current technological state.

2.1 Types of Electric Vehicles

Electric vehicles originated in the mid-19th century when electricity was one of the preferred methods for motor vehicle propulsion. An EV uses one or more electric motors for propulsion.

There are many types of electric vehicles such as electric cars, electric trains, electric boats, scooters, etc. There are three main types of electric vehicles as described in Table 2.1. The difference between a hybrid and a plug-in hybrid is how the battery pack within the hybrid is

Table 2.1: Main types of electric vehicles and components available for refueling.

Car Type Gasoline Battery Electric Charge Hybrid X X Plug-in Hybrid X X X Battery Electric (BEV) X X

recharged. In a hybrid, this is done through regenerative breaking as opposed a plug-in hybrid whose battery pack can be recharged similar to a BEV.

We focus on BEVs, which function solely on energy stored in rechargeable battery packs.

The other two types of EVs do not have the same operating range constraints due to batteries as a

BEV since they can still operate on fuel. In this thesis, we do not concern ourselves with a vehicle’s ability to utilize regenerative braking. Regenerative braking occurs in vehicles when the brake is pushed and there is an energy recovery mechanism that slows down a moving vehicle or object by converting its kinetic energy into a form that can be either used immediately or stored until needed.

It is very commonly to recharge vehicle batteries.

Currently only a few BEVs exist on the market. Similar to other electric vehicles, BEVs use electric motors and motor controllers instead of internal combustion engines for propulsion.

BEVs are different from plug-in hybrid vehicles which function partly on batteries and partly on gasoline similar to typical cars. A major difference aside from the use of gasoline that separates

BEVs from hybrids is that BEVs plug into charging stations whereas hybrid vehicles cannot be externally charged.

Table 2.2: Examples of currently available BEVs for each category of EV. Vehicle Name Top Speed Charging Driving Market Release Time Range Date (Level 2) Low Speed Vehicles Dynasty iT Sedan 25 mph 6 hours 30 miles April 2001 GEM Car (Chrysler) 25 mph 6 - 8 hours 30 - 40 miles April 1998 ZEV Smiley 31 mph 10 hours 75 miles Spring 2007

City Speed Vehicles Citroën C1 ev'ie 60 mph 6 - 7 hours 60 - 70 miles April 2009 NICE Mega City 40 mph 8 hours 60 miles Oct. 2006 Stevens Zecar 56 mph 6 - 8 hours 50 miles March 2008

Highway Vehicles BMW i3 93 mph 4 hours 81 – 99 miles 2013 Nissan Leaf 93 mph 8 hours 73 miles Dec. 2010 Tesla Model S – 85 kWh 125 mph 3.5 hours 265 miles 2013

Categorizing battery electric vehicles

Current battery electric vehicles on the market are grouped into categories based on the maximum speed they are capable of reaching:

• low speed: cars not capable of reaching 37 mph. Vehicles are economy sized cars.

• city speed: cars capable of reaching at least 37 but less than 62 mph. Vehicles are intended

for city-use.

• highway: cars capable of reaching at least 65 mph and intended for highway use.

Another category available is raceway cars, which have top speeds close to 120 mph and long driving ranges.

Table 2.2 summarizes the charging times and driving ranges for a some currently available models for each of the three types of BEVs. While most of the vehicles in the current market generally require between 6 to 8 hours to completely charge using Level II charging through a

North American standard 240-volt outlet power outlet, the driving range varies greatly between

the three types of vehicles. This variation in range is caused by multiple factors, such as curbside weight of vehicle, type of battery (chemical composition), etc.

Advantages of BEVs

There are several advantages of having EVs over typical internal combustion engine (ICE) based vehicles [Hori 2004]. These include:

• Energy Efficiency: Electric motors convert 75% of the chemical energy from batteries to

power the wheels versus ICEs, which only convert 20% of the energy stored in gasoline.

• Environmentally Friendly: EVs do not emit any tailpipe pollutants. While the power plan

producing the electricity may however emit some, sources of renewable energy (i.e. water,

solar, wind) cause no pollution at all.

• Performance Benefit: EVs are quieter and provide smoother operation and stronger

acceleration. They also require less maintenance than ICEs.

Disadvantages of BEVs

Although there are advantages, EVs also have significant battery-related challenges [Eberle and von Helmolt 2010]:

• Driving Range: EVs can only drive approximately 80 - 100 miles before recharging.

Combustion-engine powered vehicles can travel over 300 miles before needing refueling.

• Recharge Time: Fully recharging batteries can take from 4 to 8 hours. “Quick charging” a

battery to 80% capacity can take 30 minutes. (Table 2.3).

• Bulk and Weight: Battery packs are heavy and consume considerable vehicle space.

Table 2.3: Examples demonstrating range and charging limitations of currently available BEVs. 2014 Vehicle Level 2 EPA Driving Lithium Ion MSRP Model Charging Time Range Battery Capacity BMW i3 3 hours 80 – 100 miles 35 kWh $41,350 Chevrolet 7 hours 82 miles 21.3 kWh $26,685 Spark EV Nissan Leaf 8 hours 84 miles 24 kWh $28,980 Tesla S 3.5 hours 265 miles 85 kWh $69,900 Ford Focus 4 hours 76 miles 23 kWh $35,170 EV Honda Fit 3 hours 82 miles 20 kWh $36,625

Range anxiety is the term used to describe the fear that a vehicle has insufficient charge to reach its destination and therefore would cause the vehicle’s occupants to be stranded on the side of the road. This is considered as one of the major barriers along with availability of charge stations causing the limited adoption of BEVs. However, in a recent poll conducted by Deloitte, 80% of its participants in the United States stated wanting a range greater than 100 miles for an electric vehicle [Deloitte 2011]. A smaller percentage, approximately 60%, wanted a range greater than

200 miles, and 37% expected a range greater than 300 miles in order to compare to internal combustion engine vehicles [Deloitte 2011].

2.2 Current battery technology

A battery is any device that stores energy for later use. A "battery" is limited to an electrochemical device that converts chemical energy into electricity with a galvanic cell. Batteries are the main component of a BEV as there is no other source of energy for the vehicle available. Currently, the battery is the most expensive component to a BEV and many advancements are being made to battery technology to make them more efficient and usable in EVs.

Most importantly, BEVs boast improvements in driving range through advances in battery technology (i.e. using larger cell batteries) [Dhameja 2001]. However, there are major downsides to these hardware-based solutions [Dhameja 2001]:

• Larger batteries are more expensive and as a result increase the cost of a vehicle.

• Larger batteries may require additional charging time.

• Larger batteries may require the need for larger sized vehicles, which adds additional

weight and in turn drains the additional battery charge.

Types of Batteries

BEV batteries are deep cycle, since they have to provide power over longer periods of time

[Rosenkranz 2003]. Deep cycle batteries are designed to be regularly discharged to most of its capacity, usually between 50% - 80% depending on the manufacturer and construction of the battery [Rosenkranz 2003]. They are characterized by their relatively high power to weight ratio, energy to weight ratio, and energy density [Rosenkranz 2003]. Smaller and lighter batteries reduce the weight of the vehicle and in turn improve the performance. The maximum range of BEVs is due to current limitations in battery technologies such as the lower amount of specific energy

(energy per unit mass) compared to fossil fuels.

Deep cycle batteries that are cycled down to only 20% charge, have a lower life span than those that maintain average cycles at 50% discharge [Serrao, et. al. 2011]. There is a direct correlation between depth of discharge on the battery and the number of discharge cycles.

BEV batteries are commonly charged from the power grid. While most power for this is from domestic resources such as coal, hydroelectricity, nuclear, etc., renewable sources of energy may also be used such as solar, hydro, and wind. The capacity of the grid connection limits

charging time. The normal household outlet provides between 1.5 kilowatts in countries with 110- volt supply to 3 kilowatts in countries with 240-volt supply. Higher power levels allow for faster charging, but there are constraints within the battery, as most batteries do not accept charge greater than their charge rate. High charge rates have adverse effects on battery discharge capabilities.

The charge rate, which is often denoted as C or C-rate, signifies a charge or discharge rate equal to the capacity of a battery in one hour. For example, a 1.6Ah battery has a C-rate = 1.5A. If the charge rate were C/2, it would require two hours to charge/discharge completely versus having a C-rate of 2C, in which it would need only 30 minutes. Charge rate (C) for batteries depend on the battery type. There are several batteries that exist currently on the market that allow for higher

C’s that would be beneficial (Figure 2.1). This implies that charging at faster rates would be feasible and that batteries would not have adverse effects (i.e. explode) given charging conditions!

Battery life

The lifespan of batteries vary with how it is used, maintained, and charged as well as temperature and other factors. Overcharging has negative consequences on batteries as does undercharging.

Over usage has the same type of negative consequence as under usage. Thus, optimizing proper battery usage and charging is very important and necessary in maintaining long-term battery life.

In Germany, the Centre for Solar Energy and Hydrogen Research Baden-Württemberg,

(ZSW) has developed lithium-ion batteries retain more than 85% of initial capacity after 10,000 complete charging and discharging cycles (with a complete charge and discharge cycle per hour)

[Yirka 2013]. This indicates that batteries would be able to last upwards of 27.4 years. The power density for these batteries is also relatively high (1,100 watts/kg) which indicates that EVs would have shorter charging times and better acceleration capabilities [Yirka 2013].

2.3 Current battery charging technology

The most common solution to range anxiety is to install battery chargers similar to gas stations.

This requires changing infrastructure to both build and install these charge stations and to ascertain that the grid in the desired area will be able to handle the additional load of multiple vehicles recharging at the same time. The main drawback aside from the need for infrastructure changes is the amount of time required for a vehicle to charge, which are currently orders of magnitude greater than the amount of time needed to currently refill a gas engine tank with fuel. Electric vehicle charging equipment is typically categorized into three types as summarized in Table 2.4.

Another option for recharging batteries is through induction charging. Induction charging uses an electromagnetic field to transfer energy between two objects. Several small household appliances are now available that utilize this technology such as rechargeable electric toothbrushes, cell phones, and remotes for gaming consoles.

Inductive Charge Transfer (ICT)

Several major companies as well as start-ups are also working on ICT for BEVs (Table

2.5). Of particular interest is the recent merger of HaloIPT, a New Zealand based company, with

Qualcomm to form Qualcomm Halo. HaloIPT was the first company to make inductive charge

31 metal hydride batteries. Nickel metal hydride, Lithium titanate batteries allow charging on the however, in addition to relatively low energy order of ten minutes and have been shown to have density, also contains nickel, an expensive extremely long cycle life—on the order of 5000 material. Three of the more popular lithium based full depth of discharge cycles. Lithium titanate has electrochemical systems vying for EV applications high inherent safety because the graphite anode of are lithium cobalt or lithium manganese oxides standard format and iron phosphate batteries is (stand ard format), lithium iron phosphate, and replaced with a titanium oxide. lithium titanate. Figure 5 shows the pack level specific energy of BattTableeries s u2.4:ch as Technical those for la ptospecificationsp computers and and time neededEV batteries for achargings a function at of differentcharge rate, levels called cell(normal phones, w andhich fast)use st anind 2014ard form. at lithium “C”. A “one C” charge rate is the time it takes to cells would seem to be the obvious solution charge a battery in one hour. A C/2 charge rate becaTypeuse they have a relaChargingtively low co st andPower very Supplrequiresy twChargero hours, w hile a 6C cChargingharge rate requ Timeires high specific energy. In factLevel, a conc ept developed only ten minutPoweres. The values in Figure 5 are by NormalAC Propu lsion usiLevelng “bricks” 1 of 18650120V, single approxim1.4a kWte, as shown by th16e error hours bars, and computer cells configured into large packsphase, has 12A depend on many factors such as State of Charge been adopted by bothLevel BMW 2(Min i Cooper240V,) and single and ba3.3ttery kW che mistry. 8 hours Tesla Motors. Computer cells, which are phase, 15A manufactured by the billions in Asia, have 4.2 Range vs. Pack size disadvantages. Due to safety and performance240V, single 6.6 kW 4 hours phase, 30A Batteries, even lead acid batteries, are expensive. reasons they are limited to one- to two-hour The larger the battery pack, the more expensive it chargeFast rat es at the packLevel leve l3 so as not to i480mpact VDC, 3- is. Wha50kWt is the right size for 30an EVminutes battery pack? cycle life. phase Figure 6 shows EV range per charge versus pack Lithium iron phosphate batteries are manufactured size for a variety of EVs. One measure of EV by many companies worldwide and have gained efficiency is Wh/mi on a plug-to-wheels basis. credibility through their use in power tools. The lower the Wh/mi, the more efficient the EV Lithium iron phosphate cells have a much lower drive train. A smaller, less expensive battery pack energy density than standard format cells, but can may one day allow the full functionality of today’s be charge d much faster—on the order of twenty to conventional vehicles if fast charging is part of the thirty minutes. equation.

Figure 2.1: FigureCharge 5: Pack rates level for spe varyingcific energy types versus ofcha batteries.rge rates for differentFigure battery adopted chem istriefroms [Botsford and Szczepanek 2009].

EVS24 International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium 6

transfer (ICT) available for electric vehicles [Halo Web]. Their current technology requires that the objects be separated by at most 400 mm (~15.75 in) and boasts transfer efficiencies greater than 85%. Start-ups such as, Plugless Power by Evatran, a company located in Virginia, boasts efficiencies greater than 90% for their ICT system while charging at the same rates as direct plugins

[Carlson 2014]. Fleet testing is currently underway by Google on Evatran’s technology.

Efficiency refers to the loss of energy during a charge transfer. Currently available technologies for ICT boast efficiencies well over 90%. The efficiency of the system is overall important to make sure that not too much energy is lost during a transfer. Since the system will have a limited amount of excess energy available; we need to ascertain that we can conserve as much energy as possible. If transfers are highly inefficient, then there may not be enough energy available in the system to support the total amount of charge needed by the participating vehicles.

Power levels for transfer are important to determine the amount of charge that can be transferred between vehicles in a 30 second stop at an intersection. We perform a quick back of the envelope calculation to determine the feasible rates of transfer for a Nissan Leaf with a standard total capacity 24 kWh battery. In order to transfer 5 units of charge in 30 seconds, we need to transfer 1.2 kWh in 30 seconds. Using a conservative efficiency of 90%, we assume that 10% is lost to heat, which indicates that we need to transfer 1.33 kWh in 30 seconds. Using a simple power calculation (eq. 2.1), we note that we would need a transfer rate of 159.6 kW.

While this is a relatively high transfer rate, we note that according to Table 2.5, there is already technology available that can sustain these rates.

()$%*+ 1.33 23ℎ (2.1) !"#$% = = = 159.6 23 ,-.$ 30 6$7

Table 2.5: Examples of currently available ICT devices on the market along with their published specifications (in 2014). Name Technology Efficiency Distance Transfer Rates (in sec) Bombardier PRIMOVE Up to 200 kW

Qualcomm Qualcomm Up to 7 kW Halo

Companies Siemens >90% 15 cm 3.6 kW

ORNL 90 – 94% 25 cm 7 kW KAIST OLEV 85% 20 cm 180 kW

Research Research Groups Stanford >97% 6.5 feet 10 kW Delphi 20 cm 3.3 kW

EvaTran Plugless >90% 10 cm 3.3 kW ups - Power

Start WiTricity >90% 15 – 20 cm 3.3 kW

Safety is a concern for humans from the magnetic waves that are present during a transfer and for determining battery safety since the temperature will increase for the battery to accept charge. IPT has been shown to be safe to humans even at high transfer rates [Wu, Gilchrist, et. al. 2011]. We perform a quick calculation to determine whether the temperature increase during a charge transfer would affect the integrity of the battery and cause adverse events (i.e. battery explosion).

Using the earlier example of transfer to a Nissan Leaf, we estimate a lower bound of 10% of the transferred Joules becoming heat. Using equation 2.2, we obtain that 478,800 Joules of heat are produced. To calculate the amount that increases the battery temperature with the heat being spread equally, we apply equation 2.3. We assume that the specific heat capacity of Lithium Ion battery is 0.83 J/g C at the operating temperature of an electric vehicle. We note that the temperature of the battery simply increases 1.96 degrees Celsius during a charge transfer. This small increase of temperature is negligible and as a result does not pose any safety hazards for the technology.

((") = $)$%*+ ∗ ,-.$ = 159.6 23 ∗ 30 sec = 4,788,000 C"DE$6 (2.2)

F = .7∆H (2.3)

478,800 = (293.93 ∗ 1000)* ∗ 0.83 C/*M ∗ ∆H

∆H = 1.96 C

Alternative methods for wireless charging

Several alternative methods have been proposed for wireless charging. While each method has its pros and cons, these methods currently have not gained much traction for vehicle charging. Among notable alternatives are: inductive coupling, laser, and strong electromagnetic resonance.

• Inductive Coupling: The inductive coupling works under the resonant coupling effect

between coils of two LC circuits. The maximum efﬁciency is only achieved when

transmitter and receiver are placed very close from each other. The cons of this method

include the short range and the loss of efficiency as fast as the coils are separated.

• Laser: LaserMotive, a company that focuses on using laser technology to recharge

batteries, flew a quadrocopter for 12.5 hours [Boen 2019]. This proof of principle

experiment demonstrated an extension of the battery life by 150 times from its traditional

5 miles. Prior to this, in 2009, LaserMotive demonstrated the feasibility of their technology

with transmission ranges over 1 kilometer.

• Strong Electromagnetic Resonance: In [Karalis et. al. 2009] and [Kurs 2007] introduced

the method of wireless energy transfer via "Strong" electromagnetic resonance. This

method uses the "strong" electromagnetic resonance phenomenon, achieving energy

transfer efﬁciently at several dozens of centimeters.

2.4 Related Work: EV Charging Applications

There has been significant work done on identifying battery-charging mechanisms for EV batteries. Among applications being considered are swapping solutions in which the battery is physically swapped out with another battery as well as wireless charging opportunities.

Swapping Solutions

Battery swapping stations are another proposed method for recharging electric vehicles [Chen, et. al. 2013] [Ellis, et. al. 2013] [Worley and Klabjan 2011]. One noted leader in the industry, Better

Place, announced bankruptcy and that it would no longer be working in this domain [Motavalli

2013]. The company had raised over $850 million from investors, with the idea to create a network of swapping stations that allow EV drivers to avoid long recharging times by simply swapping batteries through an automated process. This concept was being pursued in Israel, Denmark,

Australia, China, and United States. American operations for Better Place were seen in localities such as Northern California and Hawaii. In Northern California (San Francisco and San Jose),

Better Place had been commissioned to create all electric taxi fleets with swappable batteries.

Tesla is still investigating battery swapping as a solution to the range anxiety issue for their vehicles. Currently, the Tesla is designed such that its battery can be swapped in less than half the time it takes to refill a gas tank. However, the first Tesla battery swapping stations have not yet been implemented for public use.

The issue with swapping batteries is the lack of uniformity in batteries. Finding an automatic method that could easily replace batteries for all types of EVs was not feasible until standardization occurs in EVs. Coupled with large infrastructure changes and heavy machinery that were needed to sustain the concept, the idea to swap batteries has not seen much popularity.

Another idea to increase effective driving range of EVs was to have vehicles available on demand and to share EVs [King et. al. A] [King et. al. B]. This car-sharing program was much like that of hourly rental cars, where unused vehicles would be parked in a central location charging when not in use. This idea while useful for the adaptation of EVs is not sustainable as people own their vehicles and the concept of on demand and car sharing only works for rental vehicles and not personally owned vehicles.

Wireless Charging Solutions

There are many proposed methods for using wireless charging for electric vehicles. Wireless charging typically occurs through inductive charge transfer (ICT) for the purposes of electric vehicles. ICT is a method in which electrical power is efficiently transferred between two objects without wires through an electromagnetic field. Several small household appliances are now available that utilize this technology such as rechargeable electric toothbrushes, cell phones, and remotes for gaming consoles. It is unaffected by dust, dirt, water, etc. and as a result has less wear and tear on electrical connections associated to the vehicle.

Currently, the most common application is to install wireless charge pads in homes. This application of wireless charge transfer offers the owner many benefits, from being able to utilize a program to offer guidance on smart charging to avoiding the hassle of physically plugging in the

EV to the charging system. Still the disadvantage of this method is that it does not offer any additional driving range when the EV is being used on a day-to-day basis.

A current proposed method for using ICT for recharging EVs involves placing inductive charging strips on roadways to provide charge for BEVs driving on the street. Several research groups have suggested this method as a method to increase driving range of EVs, however have

done little work to quantify the results from such a system [Lakshari, Shladover, Lechner 1986]

Yilmaz, Buyukdegirmenci, Krein 2012]. The focus has been on the hardware and infrastructure changes needed to sustain this concept [Zhang, Wong, Chen 2011] [Zhang et. al. 2014].

In South Korea, researchers at the Korean Advanced Institute of Science and Technology

(KAIST), have developed an Online Electric Vehicle System (OLEV) designed to allow EVs travel longer distances with smaller batteries [Lee et. al. 2010] [Ko, Jang, Jeong 2012]. KAIST has been able to successfully demonstrate their wireless roadway charging system with high transfer efficiencies (>70%) on a bus and SUV. A major disadvantage with this concept is still the need to alter existing infrastructure through roadway installations of energy transfer pads.

Another proposal involved installing wireless charging devices at traffic lights such that

EVs can be charged while waiting at a traffic light [Mohrehkesh and Nadeem 2011]. The proposal further stated that intelligent transportation methods would need to be investigated to determine the best method for coordinating and extending stops at traffic lights to enable EVs to acquire charge. The method suggests EVs would be routed such that their frequency and duration at traffic lights are increased. This method not only requires changes to infrastructure but also adds a level of complexity by having all vehicles affected with the new traffic light schema. Additionally, there is no mention in the literature about the number of vehicles that can be serviced at one traffic light.

2.5 Vehicle-to-vehicle wireless charging

While others have explored V2V applications, these were always proposed for stationary periods at parking locations using wired connections [Lui, Wu, Gao 2013]. We are the first to propose

V2V wireless charging [Dutta 2013]. The basic premise for our charge-sharing network is to have two vehicles wirelessly transfer charge between each other at coordinated rendezvous spots and/or

traffic light intersections. Through the use of computer networking and communications algorithms we show the feasibility of such a scheme to increase the effective driving distance of an electric vehicle. The main advantage of our proposed solution is the lack of needing to change current roadway and grid infrastructures in metropolitan cities. Unlike installing charging or battery swapping stations, our method relies on vehicles serving as ad hoc moving charge stations that can offer charge to each other.

To make our concept feasible, the EVs will need to have specific hardware installed including inductive power transfer (IPT) devices as well as mechanisms on the vehicles to minimize the distance between the two vehicles during the energy transfer. The feasibility of our network-based approach will be determined by modeling realistic average distances between opportunities for charge transfers by using real-world data. The frequency in which a BEV is able to receive charge units currently determines the success of the network given battery size and the amounts of energy that can be transferred in short intervals. Therefore, the average distance between charge transfers is a parameter that needs to decrease. This can be accomplished by increasing the number of vehicles participating and/or by communicating between vehicles to arrange rendezvous points.

2.6 Conclusions

This chapter provided a summary of the types of EVs and the classification used to describe the various types. It also provided an overview pertaining to the battery and charging technology that exists for EVs. Our focus for this thesis on the BEV since it offers the most opportunity for reducing dependence on any version of the internal combustion engine. We reviewed the various efforts that have occurred in wireless charging technology space and the existing research for

wireless charging implementations for EVs. To conclude this chapter, we introduced the novel concept of vehicle-to-vehicle wireless inductive charging to create a new method and application for wireless charging that may serve to make EVs more feasible in their current technological state.

Chapter 3: Vehicle to Vehicle Charge Sharing Framework

In this chapter, we describe the framework and development of our charge-sharing network. We discuss the considerations and assumptions made to develop the various parameters for our discrete time stochastic model.

3.1 Development of simulation framework

Our vehicle-to-vehicle (V2V) charge-sharing network was developed in two main iterations. The first iteration was a simplified version developed in Matlab to demonstrate proof of principle, while the second version was developed in Java to serve as a more robust simulation model with more inputs and parameters. Figure 3.1 represents the second version of our model framework.

• Charge and travel distance are normalized such that 1-unit charge results in a travel distance

driven of 1 distance unit. (Many BEV’s can travel 100 miles on a fully charged battery, so

that the unit of distance is 1 mile.)

• A car does not need its excess remaining charge units upon reaching its destination.

• External factors that can cause a car to use more than 1 unit of charge to travel 1 unit of

distance are not included in the model (i.e. cars are never stuck in traffic nor do they lose

additional charge for using headlights, heating/air conditioning, radio, etc.).

Routing •Fisheye Routing •All Taxi EV Fleet Scheduling Rendezvous Incentives for Participation •First Come First Serve Algorithms •Nash Bargaining for 2 Players

Driving Distance Distribution Data Stochastic Testing and •NYC Taxicab Simulation Convergence Analysis •NHTS Commputer Vehicles

Figure 3.1: Stochastic simulation framework showing the inputs and parameters required.

Proof of principle model

We model a probabilistic charge-sharing network, where there is no use of communications to predetermine rendezvous points for vehicles. Cars randomly meet at traffic lights.

Framework and Parameters

The simulation framework is described in Figure 3.2 and contains several parameters that are adjustable to model different scenarios:

• Driving Distance Distribution (DD) is the distribution by which cars are assigned to the

various travel distances, d. It is a modified Gaussian distribution with mean, μDD, and

standard deviation, σDD. We use a modified Gaussian distribution since a standardized

Gaussian distribution would contain negative values for distance needing to be travelled.

We clip all values under 0.

• Stopping Distance (SD) is the distance that a car travels before having the opportunity to

exchange charge with another car. SD is modeled similar to DD as a modified Gaussian

distribution with mean, μSD, and standard deviation, σSD.

• Unit Charge Transfer (UT) is the maximum amount of charge that can be transferred

between two cars during a charge transfer. We are modeling quick charge transfer

opportunities (e.g. at a stoplight which offers a 30 second transfer opportunity).

• Efficiency factor (EF) is the efficiency of the charge transfer between vehicles. We vary

this factor for the various simulations to accommodate the understanding that there is some

amount of charge that is lost since no transfer is loss-less. For most experiments, we hold

this value constant at 0.9 given literature references (Chapter 2) on the efficiencies of

wireless charge transfer. Since the value is accommodated in our modeling, we can vary

this variable for further understanding.

The basic simulation is structured for discrete time analysis, which we run to represent the passing of 30 days (one month). In each iteration the simulation uses the following algorithm (Figure 3.3):

• Step 1: add cars – Cars enter the simulation with a Poisson distribution and fully charged

battery. Each car is randomly assigned a driving distance from the DD.

• Step 2: remove cars – We remove cars, which have completed their life cycle, either by

reaching their destination or fully depleting their charge.

• Step 3: calibrate cars – We simulate expected stops for a subset of cars via the SD, e.g. red

light stops. Stopped cars are randomly paired and charge exchange heuristics applied.

• Step 4: drive cars – Cars drive m units of time and expend SD units of charge for SD units

of distance travelled.

Charge Transfer Heuristics

A car has two properties: battery charge remaining, c, and a distance to travel, d. Cars are never permitted to have more than a fully charged battery of 100 units. When two cars meet one of the following three scenarios occur: (Note that for each scenario, we ascertain that any UT that occurs does not exceed the total battery capacity of 100 units. If the added UT will exceed the total capacity of the battery, we limit the UT to the amount that will yield a fully charged battery.)

• Scenario 1: Both cars have excess charge, which means that c > d such that (cx > dx)

and (cy > dy). We transfer charge to the car that has the longer distance to travel before

reaching its destination as it will be in the system longer and have more opportunities

for transferring charge to another car that may be in need. We calculate the excess

charge for the car that will be travelling the shorter distance using equation 3.1. We

determine whether $$ is greater than the anticipated unit charge transfer, UT, such that

$$ ≥ OH. If this is true, then we transfer UT from one car to another and calculate the

% resulting new charge amounts for the vehicles with equations 3.2 and 3.3 where 7$

denotes the new amount of battery charge the car will have after exchanging charge.

The car receiving the UT is the car, which will remain in the system longer (P$ > P&)

and thus has a longer distance to travel before reaching its destination.

$$ = 7$ − P$ (3.1)

% 7$ = 7$ − OH (3.2)

% 7& = 7& + OH (3.3)

Stopping Distance (SD)

Driving Distance Unit Charge Distribution (DD) Transfer (UT)

Proof of Principle Simulation

Figure 3.2: Simplified proof of principle simulation framework showing inputs for the model.

Step 1: Add Cars

Step 4: Step 2: Drive Remove cars Cars

Step 3: Calibrate Cars

Figure 3.3: Schematic diagram demonstrating basic algorithm used for each iteration of the stochastic simulation.

However, if $$ < OH, then we transfer ex since transferring any amount greater than ex

would violate the greedy behavior for the rules. Thus, the charge transferred between the

two cars is the excess, ex, is calculated by equations 3.4 and 3.5:

% 7$ = 7$ − $$ = P$ (3.4)

% 7& = 7& + $$ (3.5)

• Scenario 2: One car has excess charge, ex > 0, and the other has a deficit, ey < 0. For

example if we have two cars where cx > dx and cy < dy. We calculate the excess and deficit

with equations 3.6 and 3.7.

$$ = 7$ − P$; 6D7ℎ ,ℎV, $$ > 0 (3.6)

$& = 7& − P&; 6D7ℎ ,ℎV, $& < 0 (3.7)

Two sub-scenarios can occur:

o If the excess is less than UT ($$ < OH), all the excess energy is transferred from car x

to car y. The new amount of charge in each of the vehicles after the transfer is calculated

by equations 3.8 and 3.9.

% 7$ = P$ (3.8)

% 7& = 7& + $$ (3.9)

While this does not ensure that the car y will have enough energy to reach its destination,

that car at least gains a few more opportunities to find another vehicle from which it can

gain a bit more energy. If the excess is greater than or equal to UT ($$ ≥ OH), we

transfer UT from car x to car y. The new amounts of charge in each of the vehicles after

the transfer is calculated by equations 3.10 and 3.11.

% 7$ = 7$ − OH (3.10)

% 7& = 7& + OH (3.11)

• Scenario 3: Both cars have a charge deficiency, which means that c < d such that (cx< dx)

and (cy< dy). The greedy heuristic maintains that no charge is transferred to either vehicle.

Advanced Framework: Scheduling Rendezvous Points using Fisheye State Routing

Common examples of scheduling rendezvous points involve public transit systems such as para- transit [Baker, Franz, Sweigart 1993] and for private transportation systems such as package delivery [Rivers 2002]. In both these cases, one objective is to minimize the time that any passenger or package has to wait in order to reach their destination. The other objective is to reduce the number of delivery trucks and/or transit vehicles needed to accomplish the delivery and transportation of people and packages.

Our main objective is to retain excess charge in the system to increase the opportunities for vehicles needing charge to acquire charge in order to reach their destinations. Using GPS, we know the start, node i, and ending, node j, points for all vehicles entering and exiting the system.

We propose a novel application of fisheye state routing (FSR) to coordinate the meeting of vehicles at rendezvous points to exchange charge [Jaap, Bechler, Wolf, 2005]. The FSR approach translates to maintaining accurate distance and path quality information about the immediate neighborhood of a node, with progressively less detail as the distance increases [Pei, Gerla, Chen

2000 and Johansson, et. al. 2004].

For our case, FSR is useful since we have the most information about vehicles that are number of levels and the radius of each scope will depend on the size of the network. Entries corresponding closest to each other at a given time iteration. It is harder to predict with reasonable accuracy the to nodes within the smaller scope are propagated to the neighbors with the highest frequency and the arrival time for a vehicle to a destination that is further away given driving conditions. Therefore, exchanges in smaller scopes are more frequent than in larger. That makes the topology information about to minimize both charge consumed and time for the vehicles that are set to rendezvous, we want near nodes more precise than the information about far away nodes. FSR minimized the consumed bandwidth as the link stateto find update the optimal packets pairing and that routing are that exchangedrequires the least amountonly ofamong deviation fromneighboring original nodes and it manages to reduce the message sizerouting of for the both vehicles.topology information due to removal of topology information concerned far-away nodes. Even if a nodeWe definedoesn’t our fisheye have scope accurate based on a “desperation information factor” with about a first- comefar- firstaway-serve nodes, the packets will be routed correctly because (FCFS)the route methodology info forrmation the vehicle becomesneeding charge. more This “desperation and more factor” accurate determines the as the packet gets closer to the destination. This meansradius of thatthe scope FSR that the scales vehicle will well communicate to large and travel mobile within to find ad a feasiblehoc carnetworks for as the overhead is controlled and supports hichargegh ratestransfer (Figureof mobility. 3.3).

Fig. 2 illustrates how the fisheye technique is48 applied to a MANET. When the size of a network increases, sending update messages may potentially consume the bandwidth. FSR uses the fisheye technique to reduce the size of the update message without affecting routing. In the figure, three fisheye scopes are defined with respect to the focal point, node 11. 3.3 Zone Routing Protocol

Zone Routing Protocol or ZRP was the first hybrid routing protocol with both a proactive and a reactive routing component. ZRP was first introduced by Haas in 1997. ZRP is proposed to reduce the control overhead of proactive routing protocols and decrease the latency caused by routing discover in reactive routing protocols. ZRP defines a zone around each node consisting of its k-neighborhood (e. g. k=3). In ZRP, the distance and a node, all nodes within hop distance from node belong to the routing zone of node. ZRP is formed by two sub-protocols, a proactive routing protocol: Intra-zone Routing Protocol (IARP), is used inside routing zones and a reactive routing protocol: Inter-zone Routing Protocol (IERP), is used between routing zones, respectively. A route to a destination within the local zone can be established from the proactively cached routing table of the source by IARP therefore, if the source and destination is in the same zone, the packet can be delivered immediately. Most of the existing proactive routing algorithms can be used as the IARP for ZRP. For routes beyond the local zone, route discovery happens reactively. The source node sends a route requests to its border nodes, containing its own address, the destination address and a unique sequence number. Border nodes are nodes which are exactly the maximum number of hops to the defined local zone away from the source. The border nodes check their local zone for the destination. If the requested node is not a member of this local zone, the node adds its own address to the route request packet and forwards the packet to its border nodes. If the destination is a member of the local zone of the node, it sends a route reply on the reverse path back to the source. The source node uses the path saved in the route reply packet to send data packets to the destination.

Consider the network in Fig. 3 The node S has a packet to send to node X. The zone radius is r=2. The node uses the routing table provided by IARP to check whether the destination is within its zone. Since it is 240

Basic Optimization Problem

Our main objective is to minimize the energy lost in the total system and reduce the probability of failure of the electric vehicles. To do this, we need to minimize the distance that a vehicle needing charge has to travel out of its desired path to reach another vehicle in order to exchange charge.

Our objective function is defined such that we minimize the total distance cost of traveling from node i to node j for each car in our system, cij (eq. 3.12). Our decision variable, Xij (eq. 3.13), determines whether a car travels from node i to node j.

(3.12) min Z = [ [ 7'"\'" ' "

1 -^ 7V% ,%V_$E6 ^%". `"-), - ," a (3.13) \ = ] '" 0 -^ 7V% P"$6 bcH ,%V_$E ^%". `"-), - ," a

Our main constraint is the amount of charge available in a car at a certain time since the distance traveled by a car to reach a rendezvous point has to be less than or equal to the amount of charge that the vehicle has available (eq. 3.14).

P'"% ≤ 7ℎV%*$()('*(+*, (3.14)

Framework Set-up and Assumptions

We model a discrete-time simulation in which each iteration represents one time unit. During one time unit, a car can travel one unit distance, which costs the car one unit charge. Charge and travel distance are normalized such that 1-unit charge results in a travel distance driven of 1 distance unit. We model our space to represent the New York City grid structure (250 units by 20 units);

however, unlike Manhattan streets, all streets in the system accommodate two-way traffic flowing in both directions. When a car enters the system, it has the following properties:

1. Type of car: Cars are classified as commuter vehicles or taxicabs.

2. Charge, c: Cars enter the system with 100 charge units representing a full tank (fully

charged battery).

3. Distance to destination, d: The distance a car needs to drive in order to reach its

destination is randomly assigned using the real-world data described in Chapter 4.

4. Start point: The starting location of a car when it enters the system.

5. End point: The destination for a car when it participates in the system.

Charge Transfer Heuristics

We use a simple heuristic for charge exchange. Cars only exchange charge if they have enough to reach their own destination and are willing to give away all excess charge. No priorities are assigned to distinguish between commuter vehicles and taxicabs.

Our simulation assumes that only the car needing charge will travel out of its predetermined shortest path to obtain more charge. We create a desperation factor model to show how far out of their predetermined route a car needing charge will travel. Those cars with the excess charge will continue on their original path. Once a rendezvous point has been set by two cars, we will not recalculate either cars path to look for new rendezvous partners since this can lead to having the car just travel in circles. We apply a first-come-first serve (FCFS) principle for all car interactions since both vehicles are considered “busy” until the charge transfer occurs.

Figure 3.4: The desperation factor model is represented by an inverted parabola. The maxima of the parabola is the largest radius that a vehicle will travel and is set at charge = 50 which is also the mean of the driving distribution of all vehicles. X-axis represents charge remaining (in units) and y-axis represents distance (in units).

Desperation Factor Model

Desperation Factor Model: The desperation factor determines the radii of our fisheye and is modeled as an inverted parabola (Figure 3.4) where the cars radius for traveling to a rendezvous point is dependent upon the amount of charge remaining. We model the parabola to be tangent to the y = x line that indicates the upper bound on how far a car can travel with the amount of charge remaining (eq 3.15). We also model the parabola such that a car does not expend all its energy to reach a rendezvous point unless it is very desperate for additional charge. This limit is seen as charge approaches 0 in our model.

1 100 (3.15) ^-6ℎ$+$ %VP-D6 = − 7ℎV%*$- + 7ℎV%*$ 99 99

As cars have less charge remaining, they are more likely to travel the maximum distance possible with their current charge allocation to obtain energy. However, a car that has nearly a full tank of

charge will likely not be as open to traveling out of its way to find additional charge since it can only add marginal amounts of charge to fill up its charge tank completely.

Simulation

We simulate 480 iterations to represent a 24-hour time period with each unit time iteration equivalent to three minutes of the day. Each time unit represents the amount of time needed for a car to drive 1 unit distance, which is representative of 1 mile at a rate of 20 miles per hour. The general heuristic for the simulation is as follows and is summarized in Figure 3.5:

1. Cars enter system. Each car is randomly placed on the grid (node i) and is randomly

assigned a destination to travel (node j) on the grid.

2. Determine shortest path from node i to node j for each car using Dijekestra’s algorithm

[Skiena 1990]. We apply Dijkestra’s algorithm to find the shortest distance from one point

to another for a vehicle.

3. Determine whether car has enough charge to reach destination.

a. If 7 − P ≥ 0: Follow route determined in step 2. We only exchange charge if a

vehicle is available on the current path. This only done to increase the amount of

time that excess energy remains in the system.

b. If 7 − P < 0: Determine fisheye scope based on “desperation factor” (eq. 3.15) and

look for potential car helper. If there are multiple helpers available, the car that has

the greatest value from the calculation of distance required to travel to rendezvous

point and charge available for transfer. This is to ascertain that the best candidate

for receiving the maximum amount of charge is chosen.

Step 1. Cars enter system.

Step 2. Determine shortest path using Dijkestra's algorithm.

Step 3. Determine if car has enough charge to reach destination.

Yes: follow shortest path to destination in Step 2.

No: Find rendesvous partner using desperation factor.

Figure 3.5: Process schematic for each vehicle in the simulation at each iteration.

We are able to visualize the movement of cars during each iteration (Figure 3.6). Cars are given colors to represent the current state in which passenger cars are cyan, taxicabs are orange. If a car has successfully reached its destination, it is green whereas a car that has failed to reach its destination is indicated in red.

3.2 Input data

Our model has various input parameters. Some of these parameters are characterized using actual data sources such as the driving distances. Others are varied for experimental purposed to understand optimal methods for the charge transfer mechanism.

Driving distance

The models for driving distance are obtained from real-world data sources as shown in Chapter 4.

Heuristics for exchanging charge

The rules for exchanging charge maximize the amount of charge that remains in the system. We implement a simple set of heuristics, which follow two underlying principles:

1. Cars only transfer charge if at least one car has excess charge available.

2. Cars accept charge if it a has charge deficit or if it keeps the charge in the system longer.

Modeling incentives using game theory

As a method to increase participation in the charge-sharing network, we model a method of offering incentives to participants using Nash bargaining. In this section, we describe the formulation of this part of the model.

Cars that require charge (c < d) in order to successfully reach their goal are referred to as buyers while cars that have excess charge (c > d) are referred to as sellers. Buyers want to purchase units of charge from sellers at the lowest price possible while sellers want to maximize their revenue. Incentives for participation are offered to encourage participation in the system. The buyer’s incentive is reaching its destination without needing to stop for recharging, while the seller’s incentive is to earn money on route to a destination.

Figure 3.6: Screenshot from simulation showing cars. Each square represents an active car in the simulation with the color indicating its appropriate state.

In each transaction, two cars, one buyer and one seller, attempt to determine a price for the potential transaction. They both have a price that they are willing to pay/receive for each unit charge, referred to as buyer’s Willingness to Pay (WTP) and seller’s Willingness to Accept (WTA).

We model WTP and WTA as logistic functions (Figure 3.7). Since logistic functions have two horizontal asymptotes, we are able to set the lower asymptote at the minimum price that a unit charge is worth (the market value) and the maximum price that one is willing to pay or can receive for a unit charge (i.e. cost of failure for buyer and governmental regulations for seller).

Nash Bargaining theory assumes there is one buyer and seller. The seller values the object less than the buyer such that WTP ≥ WTA. However, if WTP < WTA, no trade is possible. We determine at what price, p, the buyer and seller agree upon for the transaction assuming that WTA

≠ WTP using the Nash Bargaining Solution (NBS) from cooperative game theory.

* * The NBS finds {u1 ,u2 }, the point in S (the set of all possible agreements) that maximizes

the Nash product: u1 *u2 u.*u-. This point lies on the boundary of S , and it has a geometrical interpretation: it is the point that maximizes the area of an inscribed rectangle (Figure 3.8).

Parameter formulation

We model the buyer’s willingness to purchase (WTP) and seller’s willingness to accept (WTA) as logistic functions (Figure 3.7).

Buyer’s Willingness to Purchase (WTP) Formulation: Our buyer is defined as an EV who needs

to purchase units of charge in order to successfully reach its destination. It has a charge, cb , and a

distance to goal, db , such that cb < db . The amount of charge needed by the buyer to reach its

destination is denoted by cneed = db − cb c/001 = d2-c2. To determine the maximum total number

of opportunities remaining for charge transfer, we divide the distance to goal ( db ) divided by the

average distance between stops ( dstop ) (eq. 3.16). We also calculate the number of charge transfer

opportunities needed to accumulate enough charge to reach its destination, needb ,where the total charge needed, c , is divided by maximum amount of charge that can be transferred during a need charge transfer transaction, cmax (eq. 3.17).

Figure 3.7: Example of WTA and WTP pricing using logistic functions. The formulation of these functions is described in Section II.

Figure 3.8: Graphical interpretation for NBS.

P+ (3.16) "``+ = P3456

77,,8 (3.17) )$$P+ = 79($

WTP is modeled as a logistic function as shown in equation 3.18, such that the upper horizontal asymptote, a , determined by :;<=;>>?@ABC0. This represents the price per charge unit that the WTP (.DD-:!) will never exceed. Similarly, d , is the lower horizontal asymptote which represents the price of a unit charge. The price for the seller’s willing to accept (WTA) will never lower than this amount.

We have two other parameters, b , which represents the shift of the function or in our case defined

by needb and the rate of charge, c , where c < 0 .

? WTP = (#-!) + d for 0 ≤ l ≤ ,5663,+ (3.18) .-0 &

Sellers Willingness to Accept (WTA) Formulation: The seller is defined as car who has excess

charge available such that it has a charge, cs , and a distance to goal, P3, where 73 > P3. The excess charge available for selling, 73,**, is denoted by equation 3.19.

73,** = P3 − 73 (3.19)

Similar to the buyer’s calculation for total charge opportunities available for charge

transfer, the seller’s opportunities, are denoted by "``3 as calculated by equation 3.20 where dstop

is the average distance between stops. We also calculate the number of charge transfer opportunities needed to transfer all excess charge prior to reaching its destination, )$$P3, where the total charge needed, 73,**, is divided by maximum amount of charge that can be transferred during a charge transfer transaction, 79($ (eq. 3.21).

P3 (3.20) "``3 = P3456

73,** (3.21) )$$P3 = 79($

WTA is modeled as a logistic function as shown in equation 3.22, such that the upper horizontal upper asymptote, h, will be determined by maximum price of unit charge minus price of unit charge. This symbolically represents, the price, WTA, will never exceed this amount. The shift of function, f, is denoted by our needs while the rate of change, g, is constrained to g > 0. The lower horizontal asymptote, d, is the price of unit charge which symbolically, the price, WTA, will never be lower than this amount.

G (3.22) 3Hm = '((')) + P for 0 ≤ x ≤ topps,s (.H, * )

Nash Bargaining Formulation

We assume that at each intersection, only two cars will be involved in a charge transfer: one buyer, and one seller. The seller is willing to sell with cost, WTA and the buyer places a value on the unit of charge, WTP. We assume that as long as WTP > WTA, the transaction will occur. Our problem

is to determine at what price, p, will the unit of charge be sold at. We define the buyers and sellers utility as shown in equations 3.23 and 3.24, respectively. If no transaction occurs, the both the buyer and seller utility are 0.

Set S is the set of all feasible outcomes where the outcomes for the players are denoted by

(u1,u2 ) . The buyer’s utility is u1 (eq. 3.23) and the seller’s utility is u2 (eq. 3.24). If no price agreement is reached, then both seller and buyer are at a disagreement point (0, 0).

D. = OI(3H! − !%-7$) = (3H! − !%-7$) (3.23)

D- = OJ(!%-7$ − 3Hm) = (!%-7$ − 3Hm) (3.24)

* * Bargaining solution will be given by single point (u1 ,u2 ) that satisfies the following assumptions:

* * • Individual rationality: (u1 ,u2 ) ≥ (0, 0)

Simply stated, the bargaining solution must be at least as good for each player as

what they would get from no agreement.

* * • Feasibility: (u1 ,u2 ) must be in the set S, which basically states that we cannot

allocate something that is not available to be given out.

• Pareto optimality: This means that there can’t exist any other proposed bargaining

* * solution that provides a better solution than (u1 ,u2 ) .

• Independence of irrelevant alternatives

• Independence of linear transformations

• Symmetry

The value added by a trade, M, is determined when the buyer and seller agree to a trade and is determined by equation 3.25. Consequently, we can calculate the value added to the buyer, UB, and seller, US, using equations 3.26 and 3.27, respectively.

n = 3H! − 3Hm (3.25)

OI = 3H! − !%-7$ (3.26)

OK = !%-7$ − 3Hm (3.27)

When a disagreement occurs, we assign a value referred to as the best alternative to negotiated agreement (BATNA). If both players in a 2-player bargaining game disagree on how to

divide M, then each receives their disagreement value. Let Db represent the disagreement player

for the buyer and Ds represent the disagreement value for seller. We set our disagreement to 0, since if no transaction occurs then both the buyer and seller receive no utility

Nash bargaining solution

The bargaining problem is a pair, B = (U, d) , such that U is the set of utility pairs that can be

obtained as p varies between c and v. The element of U is pair u = (u1,u2 ) . If u ∈ U, then buyer gets utility D. and seller gets utility D-. As noted before, the disagreement point is (0,0). Thus, if no agreement is reached then buyer and seller get utility 0. The Nash Solution function will tell us what utility the buyer and the seller receive, and hence the price at which the object is traded.

Table 3.1: Description of tests performed to determine sensitivity of our model. Test Hypothesis Vary ratio of cabs to passenger As ratio decreases, amount of excess energy decreases and cars should result in higher failure rates for cabs. Vary total driving distance As the total driving distribution mean and standard deviation distributions increases, higher failure rates should ensue due to lack of excess energy in the system. Start with lower initial charge Given a constant driving distance distribution, we should notice higher failure rates as we decrease the initial charge available in cars when they enter the system. This is due to the lack of excess energy in the system. Vary transfer efficiency As transfer efficiency decreases, higher failure rates will occur since we will be reducing the amount of excess energy in system. Size of city (grid size) Increasing size of city will result in higher failure rates, assuming number of cars in the system are kept constant, since there will be lower vehicle density available in the system which will require cars to travel further to find rendezvous partners for charge sharing.

To find the solution, we determine the Nash Product (eq. 3.28) and identify the price, p, which maximizes the equation. We find feasible utilities in the pair [(!%-7$ − 3Hm), (3H! −

!%-7$)] as p varies in interval [WTA, WTP]. We maximize by choice of p in constraint: 3Hm ≤

!%-7$ ≤ 3H!, we find the NBS is determined by equation 3.29, since p satisfies the constraint

3Hm ≤ !%-7$ ≤ 3H!.

Nash Product = (!%-7$ − 3Hm)L(3H! − !%-7$)M (3.28)

3Hm + 3H! (3.29) `%-7$ = 2

3.3 Testing and convergence

It is important to perform tests on our model to determine whether the model has sensitivity to various parameters that we have set. For example, if we change the driving distances of the cars in the system, we should notice that the success rate of vehicles reaching the system starts diminishing at a certain threshold given that there is not enough excess energy within the system to support such long driving distances. Table 3.1 summarizes the tests and hypothesis that we have for our model. We provide the analysis of our results from these tests in Chapter 5.

3.4 Conclusions

In this chapter, we describe the basic heuristic-based simulation framework that was initially built to demonstrate the feasibility of a V2V wireless charging paradigm. We subsequently develop a more complex model that allows for the tuning of more parameters and coordinates the vehicles that are going to engage in charge transfers as opposed to relying on mere probability of finding another vehicle for charge transfer. Finally, we develop a framework to model the incentive structure needed to entice vehicles to participate in such an environment.

Chapter 4: Driving Data

Driving behavior varies for a vehicle based on the type of vehicle. Commuters use their vehicles very differently from service vehicles. Vehicles can be classified based on method of usage. One common classification is referred to as fleet vehicles. In this dissertation, we focus our analysis on unstructured fleet vehicles. This chapter defines our use of the term unstructured fleet vehicle and discusses the real-world data sets acquired and analysis performed.

4.1 Defining Fleet Vehicles

Fleet vehicles are groups of vehicles that are owned or leased by a business as opposed to an individual or family. Types of fleet vehicles range from cars to vans to trucks, depending on the need of the company. Multiple drivers, multiple paths, or any combination of the two can use any single vehicle in a fleet system. Fleet vehicles can be represented as moving micro-grids or networks with nodes and parameters. Fleets of vehicles can be clustered into groups: structured and unstructured fleet vehicles and each serve very different functions (Table 4.1).

Structured fleet vehicles

Structured fleet vehicles are those vehicles that follow a set path and schedule with very little variance in daily activity. These include vehicles such as Federal Express (FedEx), United Parcel

Service (UPS), Metropolitan Transit Authority (MTA) buses, etc. In each of the example cases, the buses and trucks have a defined schedule. This allows for very little variability except for times when there is major traffic congestion or other conditions that prohibit from keeping its course.

Given this, we know the hours of operation for the vehicles and the number of miles traveled for each vehicle. More importantly, we know the location of a vehicle at a given time.

Table 4.1: Differences between Structured and Unstructured Fleet Vehicles Structured Unstructured Schedule Yes No Network path Yes No Example FedEx, UPS, MTA Taxicabs, Rental Cars, Commuter Vehicles

Unstructured fleet vehicles

Unstructured fleet vehicles are those, which belong to a business or government agencies, but have no set path or schedule. These vehicles are seemingly spontaneous in their daily usage and routes.

Examples include taxicabs, limos, car rental vehicles, utilities vehicles, etc. Unstructured vehicles are the hardest to model as there is very little information that is definite, leaving most scenarios as assumptions. Taxicabs drive around busy streets to find passengers. Once the cab has a passenger, the cab tries to find the shortest and most time efficient path to dropping off the passenger at their desired location. The iterative cycle continues throughout a cab driver shift. The only constants in this scenario are the times during which the cab driver is working and as a result the time the vehicle is running as well as the general radius that cabs travel. Similarly, car rental agencies rent out their vehicles with very little information about the travel pattern for the vehicle.

They have a date of rental pickup and a date and time for rental drop off. Some car rental companies place stipulations on where the vehicles can be taken (i.e., only within tri-state area), while others place mileage limits on their vehicles.

4.2 Unstructured Fleet Vehicle Data Sets

We focus our analyses on the less deterministic paths by using unstructured fleet vehicles as our main point of study. The use of these data sets requires us to use real world data and apply best-fit techniques to acquire mathematical models that can be used for the analysis of our methods.

Commuter Vehicles

According to the United States Census Bureau, approximately 600,000 people in the United States travel 90 minutes and 50 miles to work. There are approximately 10.8 million who travel an hour each way, which is representative of 8.1% of works in the United States. These results are from the United States Census Bureau’s annual American Community Survey in 2013, which provides local statistics on a variety of topics [United States Census Bureau 2013]. Interestingly, driving is the most common method of commuting amongst all commuters and even amongst those who commute for more than 60 minutes each way to get to work (Figure 4.1) since mass transit options are common alternatives for populations near metropolitan areas.

Omnibus Household Survey

For our preliminary data set, we utilize the Omnibus Household Survey results conducted by the

United States Department of Transportation to understand the driving behaviors for commuters.

The survey collects the average commuting distance in miles for a one-way commute between home and work for people in the United States.

How We Get to Work: All Commutes and Commutes of 60 Minutes or More

All Workers Who Did Workers With Travel Times Not Work At Home of 60 Minutes or Longer

79.9%

Drove Alone 61.1%

10.1% Carpooled 12.9%

5.3%

Public Transportation 23.0%

4.8% Bicycle, Walk and Other 2.9%

Source: U.S. Census Bureau, 1-year American Community Survey, 2011.

Figure 4.1: Figure adopted from [United States Census Bureau 2013] showing main modes of transportation for commuting to work for the United States in 2011.

Frequency

Driving Distance (miles)

Figure 4.2: Histogram showing commuter driving distribution for typical day commute.

To estimate a round-trip commute from this data, we double the one-way commute distances and add a “miscellaneous trip factor” (eq. 4.1). The miscellaneous trip factor is to account for all work unrelated tasks or emergencies for which the vehicle is driven during the day. For the purposes of this study, we assume that the miscellaneous trip factor is 10 miles.

s"D)P ,%-` = 2(c)$ 3V+) + n-67$EEV)$"D6 H%-` tV7,"% (4.1)

The resulting driving distribution can be represented as a lognormal distribution with mean,

μ = 32, and standard deviation, σ = 26 (Figure 4.2). Thus, given this distribution approximately

4.9% of the commuting vehicles would be unable to reach their destination without recharging.

Regional Household Transportation Survey

To obtain more detailed information from a larger dataset about commuter vehicle usage, we turn our attention to the 2010 - 2011 Regional Household Transportation Survey (RHTS) which is conducted by the New York Metropolitan Transportation Council (NYMTC) and the North New

Jersey Transportation Planning Authority (NJTPA). It is similar to the National Household

Transportation Survey (NHTS), which is conducted by the United States Department of

Transportation Federal Highway Administration. The NHTS survey is conducted every 5 – 8 years, with the most recent survey in 2009 for data collected between March 2008 and May 2009. This survey includes data for 150,147 households that were contacted based on randomly generated telephone numbers using the Computer Assisted Telephone Interviewing (CATI) system.

The RHTS contains information about the demographic and travel behavior for 28 counties of New York, New Jersey and Connecticut. The purpose is to obtain household travel data to 68

understand traffic and congestion in the metropolitan area. In total, there were 143,925 linked trips reported from 18,965 households with 43,558 participants. A sub-sample of 1,930 households participated by wearing global positioning system (GPS) devices. The GPS method was used to validate the biases in the data set from reporting biases.

After removing duplicates from the data set and preprocessing the data to contain trips involving personal cars as the mode of transportation, we arrive at 12,815 aggregate cases to compute the cumulative travel distance of a vehicle each day. The data follows and exponential distribution with a mean = 38.108 miles and standard deviation = 42.664 miles (Figure 4.3).

New York City Taxicabs

The New York City Department of Taxis and Limousines (NYC TLC) captures data about its medallion owned taxicabs. This data serves multiple purposes for the organization and is available to the public via Article 6 of the New York State Public Officer’s Freedom of Information Law

(FOIL). The Department of Transportation (DOT) does not make the data readily available via any mass dissemination methods (i.e. websites), but allows people to acquire access to copies by bringing brand-new unopened hard drives to their offices. The files are available as comma separated value (csv) files split by month with each file containing approximately 2.5 gigabytes of data for the trip data and 1.75 gigabytes of data for the corresponding supplementary fare data.

The total data set inclusive of both trip and fare data is approximately 50 gigabytes. We use this driving data set to understand and model how taxicabs are used in New York City.

Figure 4.3: Cumulative driving distance histogram for 12,815 vehicular cases in the RHTS survey.

The data set contains over 192 million data entries for each paid ride in a medallion owned taxicab from January 1, 2012 to December 31, 2012. The relevant information contained in the primary data set is summarized in Table 4.2. A supplementary data set that links to this original data contains payment information for the taxicabs. It is de-identified by medallion number and driver’s license ID but keeps track of the fare paid and the type of payment.

The data contains information about the amount in tolls and tips paid to the driver. This information is seemingly accurate for credit card purchases as the system registers the different between the fare and total amount paid by the passenger, however, in cases of cash payments, there

is reporting bias present in the data set since taxicab drivers oftentimes do not key in the total cash amount tendered into the system. Additionally, if a cash tip was made on top of a credit card payment, the system does not have a method to record this information.

While the data provides a detailed and in-depth glimpse of how taxicabs are utilized in the city, the data quality is not completely accurate. Much of the information available in the data set is retrieved automatically based on the turning on and off of the meter during a paid passenger ride adding an inherent reporting bias to the data. The time at which a cab driver turns on the meter varies for each cab driver. For example, certain cab drivers, will turn on the taxi meter prior to stepping out of their vehicles to help load bags into the trunk, while others will start driving the vehicle after the passenger is seated and will turn on the meter as the drive is ongoing. In certain instances, the meter takes time to reset from the previous ride or has to turn on when a shift has just begun for the taxicab driver and causes a longer delay as to when the data is acquired.

There are several types of analysis that can be performed on a large data set. The NYC TLC FOIL data has endless possibilities in characterizing usage in a large metropolitan area ranging from simple analysis of the time and travel distance of a typical taxicab ride to more involved calculations that seek the hourly/daily/monthly usage pattern of taxicabs. Analysis has also been done by the TLC to understand the usage patterns with relation to the origin and destination of paid rides. To that end, more than 95% of taxi trips begin in Manhattan or at the local airports.

Taxis in New York City operate in shifts with two daily shift changes occurring in the early morning hours and late afternoon hours. Coincidentally, there is a shift change on during the typical weekday evening commute. However, there are negligible differences in the driving distances between taxicabs during morning and afternoon shifts.

Table 4.2: Description of relevant data fields in NYC TLC FOIL 2012 data set. Data Field Data Type Definition Medallion number String De-identified medallion number for taxicab. Drivers license ID String De-identified drivers license ID since a medallion can have multiple drivers. Pick up date and Date and time Describes the date and time at which a passenger’s time (DD/MM/YYYY meter started measuring the fare. This does not HH:MM:SS) necessarily the exact moment at which the passenger entered the vehicle as cab drivers often press the button after the cab has already started driving. Drop off date and Date and time Date and time at which a passenger’s meter has time (DD/MM/YYYY stopped measuring the fare. Does not indicate the HH:MM:SS) exact moment the passenger has reached their destination as cab drivers can deactivate the meter before arriving at the destination to offer the passenger time to provide payment earlier for quicker drop off. Passenger count Number Describes the number of passengers that were in the vehicle. This number is often inaccurate since fares in NYC are not dependent on the number of passengers. As a result, many cab drivers do not modify this number during a passenger trip.

Trip time in seconds Number Calculated trip duration from pickup date and time to drop off date and time. For the reasons mentioned above, this trip duration number is not fully indicative of the total trip duration. Trip distance in Number Calculated based on the number of miles registered miles by the taxicab from the time pickup was initiated to drop off time. This number may be slightly skewed due to the reasons mentioned above regarding accuracy of measurement.

Pick up longitude Number Longitude coordinates where the passenger was picked up. More specifically coordinates at which point the meter was turned on. Pick up latitude Number Latitude coordinates where the passenger was picked up. More specifically coordinates at which point the meter was turned on. Drop off longitude Number Longitude coordinates where the passenger was dropped off. More specifically coordinates at which point the meter was turned off. Drop off latitude Number Latitude coordinates where the passenger was dropped off. More specifically coordinates at which point the meter was turned off.

We determine the number of taxicabs active on the streets during the day, with 0 indicating midnight through 23 representing 11 PM (Figure 4.4). There are varying margins of availability based on day of year, but certain times regardless of day of week or month there are very small margins such as rush hour traffic around 4PM. Most shift changes occur between 4PM and 5PM and again between 4AM and 5AM daily. Shift changes require the driver to return the taxicab to its assigned garage. Failure to arrive by a certain time can result in a fine for the taxicab driver.

The data shows that the typical number of taxicabs available on the streets of New York between

4PM and 5PM diminished by approximately 20% each day. Coincidentally, this shift change occurs during one of the highest demand times for taxicabs for the evening commute.

There is also a gradual decline in the number of taxicabs available each day between midnight and 5AM. While this corresponds to the typical shift change problem between 4AM and

5AM, the demand for taxicabs during these early hours is also limited. Hence, many taxicabs end their shifts early. The usage of taxicabs during these hours are increased during certain holidays

(i.e. New Year Eve) as well as weekend when a nightlife is more prominent in the city.

From preliminary analysis of the data set, we note an interesting pattern in taxicab usage throughout the year, in which the largest driving distances for taxicabs is at the beginning of the year from January – February and then tapers to a constant range for the rest of the year (Figure

4.5). This could be due to the colder temperatures during that time in the city coupled with increased visitors in the area.

Figure 4.4: Number of active taxicabs on the streets of New York based on hour of day for the 365 days in 2012. Each color represents a different day.

Figure 4.5: Scatterplot showing variation of paid driving distances by date. Notably, the beginning of the year has larger cumulative driving distances than the rest of the year. The larger distances are likely due to increased taxicab rides to the area airports and inclement weather usage.

16500000 16000000 15500000 15000000 14500000 14000000 13500000 Number of Paid Trips 13000000 12500000

April May June July January March August February October September NovemberDecember

Figure 4.6: Total number of paid taxicab rides per month in 2012. March has the most number of rides, while November has the least. The average number of rides per month is more than 14 million.

However, analysis of taxicab usage by number of paid taxicab rides as opposed to distance traveled shows that the in the 2012 year, there were a 178,544,324 rides with an average of

14,878,693 rides per month and standard deviation of 638,965 taxicab rides.

Figure 4.6 summarizes the number of rides per month. In this case, we see that March has the most number of paid taxicab rides and November has the least. Using this data, we extrapolate the minimum and maximum cumulative distance a taxicab may have driven during a 12-hour shift.

Cumulative distance is the total distance that a taxicab drives both with a paid passenger as well as while in search of the next passenger.

We estimate a minimum cumulative distance and a maximum cumulative distance using two calculations. For the minimum distance, we use the GPS coordinates between the drop off and pick up points to calculate a “Manhattan” street distance between the two points. This calculation is slightly skewed, as all streets in Manhattan do not run in both directions.

To calculate the maximum distance, we compute the time elapsed between the drop off and pick up of passengers. We extrapolate a distance driven during this time with an assumption that the taxicab traveled at the 30 miles per hour speed limit during this time interval. Figures 4.7 and

4.8 show the minimum and maximum cumulative distance driven by a taxi that can be fitted to a heavy-tailed Rician distribution.

Given these driving distributions for taxicabs in New York City and the known possible ranges of driving distances in common EVs (approximately 75 miles according to Environmental

Protection Agency), only 18.8% - 20.7% of NYC taxis would be able to complete an entire taxicab shift without needing to recharge. If we increase the driving range of electric vehicles to 150 miles, approximately 50.7% - 69.4% of taxis would be able to complete a taxi shift. However, if the driving range of electric vehicles increases to 300 miles, then 99% of all taxis would be able to

Figure 4.7: Minimum distance histogram of distances traveled by NYC taxicabs during 12-hour shifts can be modeled by a Rician distribution.

Figure 4.8: Maximum distance histogram of distances traveled by NYC taxicabs during 12-hour shifts can be modeled by a Rician distribution.

complete a 12-hour shift without needing to refuel. Of course, the shorter the driving range of EVs, the less money that a taxicab driver can make during a standard 12 hour shift. To ensure desirability of EVs in the taxicab fleet, there is a need to ascertain longer driving distances for EVs without the need to frequently recharge. Current quick charging rates for EVs currently require approximately 30 minutes to regain 80% of the battery life, thus taxicabs may not be willing to have so much idle time.

4.3 Conclusions

This chapter focuses on describing the types of driving behavior. Driving behaviors vary based on whether a vehicle is a fleet vehicle or commuter. Even within fleet vehicles, there is deviation on how driving patterns occur. For the purpose of the thesis we focus on unstructured fleet vehicles.

These are vehicles, which while belonging to a fleet do not have a predefined route and as a result are a bit more difficult to predict the usage patterns. We acquired a dataset through the NYC FOIL for 2012 NYC registered taxicab rides. We analyzed this dataset to understand the usage of these unstructured fleet vehicles and modeled their behavior as a Rician distribution that will be used for our simulation purposes.

Chapter 5: Simulation Framework and Findings

In this chapter, we discuss the results from our simulation. We start by determining the feasibility of our charge-sharing network using our proof of principle model. We then use our more developed framework described in Chapter 3 to determine the effect of routing and coordination in our model.

We focus our attention to a case study to determine if converting all taxicab fleets into EVs is a possibility given our charging schema. Finally, we quantify two possible incentive structures for participation in the charge-sharing network.

5.1 Proof of Principle: General Charge Sharing Network

A study conducted in 2005 on the average commute distance found that most Americans drove between 47 to 138 miles with an average, μDD, of 92.5 miles and standard deviation, σDD, of 45.5 miles [Langer 2005]. For our experiments, we use these statistics as the basis for the mean and standard deviation for the driving distribution. We also assume that transfers are loss-less such that a car giving n units of charge implies that the car receiving the charge will also receive n units.

Experiments are run to simulate a 30-day time period.

Driving distance

The average BEV travels approximately 100 units without recharging. We test the effect that various driving distributions have on the need to recharge given ICT (Figure 5.1). With a driving distribution of μDD = 60 units and σDD = 45.5 units, none of the cars fail since most cars travel less than 100 units. For a driving distribution as such, cars can succeed even without ICT. Longer driving distances for the driving distribution is where our proposed method would be useful.

For μDD = 92.5 units and σDD = 45.5 units [Langer 2005], we note that we can extend the driving distance of a BEV by 50 units without adding failures by using ICT. Of course, as the driving distance increases, we notice that the p(refueling) also increases gradually. This is a significant improvement over the standard 100 units that a BEV can travel.

Stopping Distance

To test the basic proof of concept, we first determine the effect that stopping distance, SD, has on p(refueling) by varying stopping distance between 0.25 unit to 1 unit in 0.25 unit steps for a driving distribution where μDD = 92.5 and σDD = 45.5 units. With UT = 1 unit, we note that as the SD increases for a given driving distribution, the overall p(refueling) also increases (Figure 5.2).

Increasing SD implies that cars must travel further before having the opportunity to transfer charge, which reduces the number of opportunities a car is given to obtain charge while trying to drive to its destination. Similarly, we note that the effect remains the same when the SD and UT are scaled by a factor of 20 using the same driving distributions (Figure 5.3).

Unit Charge Transfer

We investigate the effect of increasing the UT from 0.5 unit to 5 units (Figure 5.4). We note that the greater the UT, the lower the p(refueling). This is due to the ability of providing cars with significantly more charge than what was expended between two opportunities for charging.

For stopping distances that are greater than or equal to UT, the p(refueling) is larger. When cars expend more energy between opportunities to transfer charge, they need to have the ability to acquire more charge than the amount expended. Otherwise, the limited addition of charge is

p(refueling) for Driving Distributions with Standard

Deviation σDD = 45.5 1 0.9 0.8 0.7 0.6 0.5 0.4

p(refueling) 0.3 0.2 0.1 0 0 50 100 150 200 250 300 350 400 Driving Distance (units)

μ = 60 units μ = 92.5 units μ = 125 units μ = 157.5 units μ = 190 units

Figure 5.1: p(refueling) for varying μDD and σDD = 45.5 units. We notice the increase in p(refueling) for driving distances greater than 150 units. We show a 50% increase in the distance a car can travel without needing to refuel.

Stopping Distance Effect: Energy Transfer = 1 unit 100.00% 90.00% 80.00% 70.00% 60.00% 50.00% 40.00%

p(refueling) 30.00% 20.00% 10.00% 0.00% 75 100 125 150 175 200 Driving Distance (units)

SD = 0.25 SD = 0.5 SD = 0.75 SD = 1.0

Figure 5.2: Using a DD with μDD = 92.5 units and σDD = 45.5 units, the effect of SD is shown to increase following a logarithmic curve.

Stopping Distance Effect: Energy Transfer = 20 units 100.00% 90.00% 80.00% 70.00% 60.00% 50.00% 40.00% p(refueling) 30.00% 20.00% 10.00% 0.00% 75 100 125 150 175 200 Driving Distance

SD = 5 SD = 10 SD = 15 SD = 20

Figure 5.3: Using a DD with μDD = 92.5 units and σDD = 45.5 units, the effect of an increased SD still follows a logarithmic trend.

Effect of Amount of Charge Transfer on p(refueling) 45.00% 40.00% 35.00% 30.00% 25.00% 20.00%

p(refueling) 15.00% 10.00% 5.00% 0.00% 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Stopping Distance (units)

UT = 0.25 UT = 0.50 UT= 1 UT = 5

Figure 5.4: Effect of UT on p(refueling) with varying SD with constant DD (μDD = 92.5 units and σDD = 45.5 units).

insignificant in helping them reach their destination since the next opportunity to acquire charge would require more energy than what they had acquired.

Thus, it is crucial for stopping distance to be less than the unit charge transfer for our ICT- based approach to be feasible. One method of solving this problem is to have the availability of frequent rendezvous points for charge exchange. Communication methods will allow for cars to find other potential vehicles interested in participating for a charge transfer and will help make stopping distance more negligible in the future.

Our simulation demonstrates that wireless charge sharing increases the potential maximum distance that an EV can travel. Using our model, we show that cars can travel with reasonable probabilities of success beyond the 100-mile distance that typical EVs average. We are able to extend the driving distance of a typical BEV by 50% through charge sharing for a standard driving distribution. The probability for refueling is affected by the unit charge transfer, UT, as well as stopping distance, SD. As UT increases, p(refueling) decreases. Similarly, as SD increases, p(refueling) increases as well.

The probability of meeting is based on the number of cars participating in the system and plays a significant role in the success of the system. With fewer participants in the system, cars would have a reduced opportunity for potential transfers. Because more opportunities for charge transfer indicate a better chance of success, frequent opportunities for charge transfer and an increased unit charge transfer are needed to ascertain an increased probability of success.

There are many remaining open challenges to making this charge transfer network feasible.

Aside from the need of having many participants in the system, which requires people to purchase such vehicles, and for these vehicles to be equipped with ICT devices, the state-of-the art for the transfer technology will need to be sufficient. This involves ascertaining that small periods of time

to rendezvous is sufficient for charge transfer, i.e. enough charge can be transferred in one interaction that does not require either vehicle to stop for long periods of time or to rendezvous frequently with other vehicles. The idea that we suggest in this paper is that cars can interact with one another at a stoplight, which would mean that the technology would have to be efficient to transfer energy between two vehicles that are safely stopped at an intersection.

5.2 Case study: can all taxicabs be electric?

We perform a case study to determine whether it is possible for a fleet of New York City taxicabs to become all electric using the data from the RHTS and NYC TLC. Given the longer driving distances that taxicabs have to endure during a 12-hour shift, the current driving ranges of standard electric vehicles does not suffice without necessitating that the car stop to refuel. While quick charge methods would only require a 30-minute charging time, this duration causes a taxicab driver to lose potential business during that time. Unlike many large metropolitan cities in the United

States and around the world, New York City taxicabs are hailed on the street as well as dispatched from taxi stands. Most NYC cabs pick up passengers on the sides of the roads as opposed to traditional methods of calling for specific passenger pick-ups.

To determine the efficacy of our charge sharing system, we determine the probability of refueling vehicles in the system as a percentage and the percent loss of energy of the total system.

We simulate 480 iterations to represent a 24-hour time period with each unit time iteration equivalent to three minutes of the day. Each time unit represents the amount of time it would take a vehicle to drive 1 unit distance (equivalent to 1 mile) at 20 miles per hour.

Probability of refueling vehicle

The probability of refueling a vehicle is highly dependent on the assumed driving distributions. As discussed earlier, only 4.9% of commuter vehicles would need refueling without our coordination schema compared to the 80% of taxis that would need refueling when assuming that a single charge on a battery will last an average of 75 miles of driving.

The arrival of cars is simulated to model actual data between 5am and 5pm on an average day in NYC. The number of cabs kept constant for each simulation run to end at 9000 cabs (Figure

5.5) to represent typical number of taxicabs in the system as represented by the NYC TLC data.

We achieve this by increasing our lambda for the Poisson distribution to account for a larger entry of cars at the beginning of the simulation. Once we have 9000 cabs in the system, we decrease the lambda value for the Poisson distribution such that we do not add any more cabs to the simulation.

The number of passenger car varies as determined by ratio of cabs to passenger vehicles.

We model different ratios of taxicabs to commuter vehicles (Figure 5.6). As the ratio increases, there is a direct correlation to the decrease in probability of refueling for taxicabs. Given the driving distribution of commuter vehicles, there is no need to refuel any of the commuter vehicles with our coordination schema. This means that the original 4.9% of commuter vehicles that would need to be refueled no longer face that issue.

Given such a high density of vehicles, we notice that most vehicles are able to coordinate charge exchange as needed. There is more than enough energy available in the system for the vehicles to successfully contain the charge in the system. However, if cabs could just exchange charge with each other, we notice that there is not enough charge in the system since there would always be approximately a ~24% – 26% charge deficit. As a result, participation by passenger vehicles is crucial.

30000

25000

20000

15000

10000

5000

0 1 16 31 46 61 76 91 106 121 136 151 166 181 196 211 226 241 256 271 286 301 316 331 346 361 376 391 406 421 436 451 466

Cabs Passenger (0.5) Passenger (.75) Passenger (1) Passenger (2) Passenger (3)

Figure 5.5: Graph showing the cumulative arrival of vehicles into the simulation.

Similarly, when the ratio of taxicabs to commuter vehicles is 1:1, approximately 0% of taxicabs need to be refueled. This is in stark comparison to the 80% of taxicabs that would need to be refueled without our coordination schema.

We consider refueling a failure as this would indicate that the taxi was not able to stay on the roads to serve its customer and is costing the taxicab driver time recharging and potential revenue. We also notice an exponential trend in the probability of refueling based on the ratio of taxicabs to commuter vehicles. Table 5.1 summarizes our findings.

p(Refueling) for Taxicabs during 12 Hour Shift 50%

40%

30%

20%

p(Refueling) 10%

0% 1 to .5 1 to .75 1 to 1 1 to 2 1 to 3 -10% Ratios of Taxis to Commuter Vehicles

Figure 5.6: The probability of refueling decreases as the ratio of taxis to commuter vehicles increases.

Table 5.1: Comparison of refueling probabilities based on different methods assuming a 1:1 ratio of taxicabs to commuter vehicles. Taxicab Fleet Our Method 0% Lower Bound >80% Upper Bound 0% Probabilistic Method 34.2%

Percent loss of energy in system

To determine the total charge lost in the system as a percentage, we determine the total charge in system (eq. 5.1) and then calculate the total charge leaving the system (eq. 5.2).

Total charge in system = # of cars * 100 (5.1)

Total charge leaving system = ∑(100 − 7ℎV%*$ D6$P) (5.2)

The total charge leaving the system is determined from vehicles that successfully reached their destination and as a result left the system before having the opportunity to transfer their excess charge. The total charge wasted is the distance that cars needing charge travel in order to reach their rendezvous points (eq. 5.3). Total charge lost is the summation of the total charge leaving system, charge that was used by vehicles that still resulted in the vehicle needing to be refueled and total charge wasted (eq. 5.4). The total charge leaving system and total charge lost (in percentage) can both be modeled as exponential curves with r2 values greater than 0.95 (Figure

5.7) whereas the charge wasted (in percentage) remains approximately the same for all ratios of taxicabs to commuter vehicles. The total charge increases as the ratio of taxicabs to commuter vehicle increases since the total leaving system also increases. The total energy needed in the system is calculated by the sum of total distance that the vehicles need to travel (eq. 5.5) in order to reach their destinations. As long as the total energy needed in the system is less than the total energy available in the system, we should be able to ensure success for vehicles in the coordination schema. The amount of excess charge available will determine how effective we are in distributing the excess charge to other vehicles since vehicles have to expend energy to rendezvous. We note

that the % of charge needed in our system can be represented as an exponential curve with an r2 value of 0.97 (Figure 5.8).

MℎV%*$ 3V6,$P = [ distance traveled to reach rendezvous point (5.3)

H",VE 7ℎV%*$ E"6, (5.4) = [(charge wasted + total charge leaving system + charge lost from cars that needed refueling)

Total energy needed = ∑ driving distance for each vehicle (5.5)

Effectiveness of System

To evaluate the effectiveness of our coordination scheme, we determine the upper and lower bounds of our system. Our upper found is determined assuming that the system had perfect communication and that no excess energy was lost by vehicles when traveling to a rendezvous point. Essentially this assumes, that anytime a car was in need of charge, there was a car conveniently located right next to it. This upper bound assumes 0% failure since all cars would be able to obtain enough energy and 0% energy loss since no car has to travel to a rendezvous point to obtain the extra energy needed to reach its destination and that cars with excess energy are kept in the system until they transfer their charge to a car in need.

Similarly, our lower bound assumes that none of the vehicles communicate with each other and that no vehicle participates in the charge-sharing network. With these assumptions, we can expect a failure rate of approximately 4.9% for commuter vehicles and over 80% for taxicabs given the driving distributions discussed earlier in this thesis (Chapter 4).

Energy Loss in System (%)

Total Charge Leaving System Total Charge Lost 35.0% 30.0% 25.0% 20.0% 15.0%

% energy loss 10.0% 5.0% 0.0% 1 to .5 1 to .75 1 to 1 1 to 2 1 to 3 Ratio of taxicabs to commuter vehicles

Figure 5.7: The percentage energy loss in system increases for both the charge leaving system and the charge lost as the ratio of taxicabs to commuter vehicles increase.

% of Total Energy Neded in System 120%

100%

80%

60%

40%

20% % of total energy needed

0% 1 to .5 1 to .75 1 to 1 1 to 2 1 to 3 Ratio of taxicabs to commuter vehicles

Figure 5.8: The percentage of total energy needed in the system decreases as the ratio of taxicabs to commuter vehicles increases.

We compare our results for a coordinated method to a probabilistic approach. In our probabilistic method, there is no communication between vehicles. Charge transfers only occur if two vehicles, regardless of vehicle type, meet at an intersection. Charge is transferred between vehicles during the duration of a stoplight which is assumed to be 30 seconds and a maximum of

5 units of charge can be transferred during this time. We use a very basic rule for probabilistic charge transfer, which entails keeping the charge in the system as long as possible. Therefore, only vehicles that need charge are given charge, and when two vehicles meet where neither vehicle is in need of charge, we proportionally redistribute the 5 charge units to make sure that it stays in the system as long as possible. With this model, we show that approximately 34.2% - 37% of taxicabs would be able to last a shift without needing to recharge.

Given these upper and lower bounds, our coordination schema is able to obtain results close to the lower bounds, indicating almost perfect coordination. While there will always be energy loss due to distances that will need to be traveled to reach the rendezvous points, we will never be able to attain perfect coordination at 0% p(refueling) for both vehicle types.

Simulation Conclusions

We investigate the effect of communications for coordinating rendezvous points for electric vehicles to share charge wirelessly. We define a novel method of applying fisheye state routing

(FSR) with first-come-first-serve pairing to solve the coordination problem. We show through simulation that this method is one possible solution to minimizing distance traveled by a vehicle for coordination. Our approach reduces the probability of refueling to 0% for commuter cars and significantly lowers the probability of refueling for taxicabs traveling more than 75 miles.

We determine the upper and lower bounds of our system in order to compare our results and system effectiveness. The theoretical upper bound of a system as such would assume perfect communication amongst vehicles which would mean that we are not hindered by the fisheye scope but rather know about the entire system. This would mean that vehicles are communicate rendezvous points more effectively and as a result would not have any failures in the system, since the system always contains more than enough energy for all vehicles to reach their destination. We would expect to see 0% refueling for both vehicle types.

Similarly, our lower bound assumes that there is no communication, coordination or charge exchange. Under these conditions, 4.9% of the commuter vehicles would not reach their destination without refueling while approximately over 80% of the taxicabs would not last through a 12-hour shift without refueling. This is based solely on the modeled driving distributions using the data described in this thesis.

However, one limitation of FSR is the short-term view of the system which means that cars pick the best possible option at a given time point in order to exchange charge. Additionally, no priorities were assigned to vehicle types in this study so both car types were given the same consideration for pairing. Given that taxicabs have the larger driving distance distribution, we should explore the implications of modifying the heuristics such that taxicabs are always given priorities for establishing rendezvous points in addition to having commuter vehicles actually approach taxicabs for rendezvous’ even though the car providing charge currently does not alter its path to help another vehicle.

5.3 Offering incentives for participation

The basic simulation is modeled as a discrete time analysis, in which experiments represent the passing of 30 days (1 month). For our proof of experimentation, we vary the cost of a unit charge and the cost of total failure. Total failure is considered when a car cannot reach its destination due to a lack of charge.

We vary the cost of a unit charge from $0.25/unit to $0.75/unit. Our simulation results provide us with the average Nash bargaining price per unit charge based upon varying costs for failure. We also vary our distance between charge transfer to determine whether there is any effect from increased or decreased opportunities. Figures 5.9 – 5.11 demonstrate how the agreed upon

Nash Bargaining Price approaches stability as the distance between charge transfer opportunities increase regardless of the cost of a charge unit or cost of failure. We notice that the Nash Bargaining price for each of these figures follow the same trend and reach a horizontal asymptote for pricing.

Figure 5.12 demonstrates the effect of a price of charge unit on the probability of refueling, when holding the cost of failure at a constant. We show that the probability of refueling (failing to reach destination) increases as the distance between charge transfer opportunities increase as well as the price for a charge unit increases. In general, this is because the number of opportunities for charge transfers during a given trip decreases and as a result the chances of reaching a Nash

Bargaining solution for a given rendezvous between a buyer and seller also decreases.

Our simulation shows that utilizing Nash Bargaining Solution to offer incentives for participation in a charge-sharing network to increase the distance an electric vehicle can travel without needing to recharge is possible. We show that the driving distance of a vehicle can be increased with low probabilities of failure. We are able to guarantee that more cars will be able to

Charge/Unit = $0.25 2.5

1.5

1 Nash Price 0.5

0 0 2 4 6 8 10 12 Distance between Charge Transfer (units)

fail = 0 fail = 50 fail = 100 fail = 500

Figure 5.9: shows the effect of varying the cost of a unit charge, cost of failure and distance between charge transfers when a charge/unit = $0.25.

Charge/Unit = $0.50

3 2.5 2 1.5 1 Nash Price 0.5 0 0 2 4 6 8 10 12 Distance between Charge Transfer

fail = 0 fail = 50 fail = 100 fail = 500

Figure 5.10: demonstrates the effect of varying the cost of a unit charge, cost of failure and distance between charge transfers when a charge/unit = $0.5.

Charge/Unit = $0.75 3.5 3 2.5 2 1.5

Nash Price 1 0.5 0 0 2 4 6 8 10 12 Distance between Charge Transfer

fail = 0 fail = 50 fail = 100 fail = 500

Figure 5.11: shows the effect of varying the cost of a unit charge, cost of failure and distance between charge transfers when a charge/unit = $0.75.

Effect of Charge/Unit on p(refueling), Failure = $100 0.5 0.4 0.3 0.2

p(refueling) 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Distance Between Charge Transfer (units)

Charge/Unit = 0.75 Charge/Unit = 0.50 Charge/Unit = 0.25

Figure 5.12 demonstrates the effect of a charge unit on the probability of refueling (i.e. not reaching destination without failing) for varying costs of charge unit. We notice that as the distance between charge transfer increases, the probability of refueling also increases.

reach their destination without needing to stop for a longer recharge or needing to be towed in the middle of the road because they ran out of the charge.

A key benefit to our model is the lack of needing to change currently existing infrastructure, but rather relying on a crowdsourcing method to encourage vehicles to assist each other. In future work, we will explore the effect of auctions and multiple cars for bargaining.

5.4 Conclusions

In this chapter, we discussed the results from our simulation. We showed the feasibility of our charge-sharing network using our proof of principle model. We then apply our more advanced simulation framework as described in Chapter to 3 to determine the effect of coordinating routing and rendezvous points. We focus on the possibility of converting all NYC taxicab fleets into EVs and determine that given the parameters defined in our model. The model results show promise given that there would not be a high probability of failure of the vehicles participating in the system. As a result, we quantify two possible incentive structures to encourage participation in the charge-sharing network.

Chapter 6: New York City Fleet Bus Study

We conduct a feasibility study to determine primarily whether wireless charging at specifically designated bus stops throughout New York City can help to increase the feasibility of electric buses for city use both from an operational and a financial standpoint. We partnered with the Office of Strategic Innovation and Technology at the Metropolitan Transit Authority (MTA) in order to obtain data about bus operations. With a letter of support from Thomas Lamb, the Chief of

Innovation and Technology for the New York City Transit division, we applied for a grant to sponsor the project through the University Transportation Research Center (UTRC) Region 2. The

Region 2 UTRC is located at City College and is one of ten original Centers established by

Congress in 1987 with the recognition that transportation plays a key role in the nation's economy and the quality of life of its citizens [UTRC Web]. UTRC Region 2 represents the USDOT Region

II that includes New York, New Jersey, Puerto Rico and the U.S. Virgin Islands. City College functions as the lead institution of a consortium of twelve universities. The UTRC makes available yearly project seed grants for its consortium universities to gain sponsorship for preliminary work with the hopes of these projects to gain further funding from larger funding sources.

The final outcome was a statistical model that can be utilized with bus data to determine the efficacy of wireless charging at bus stops. Our model can be adjusted to see to how the placement of varying numbers of charging stations changes the outcome of buses being able to continuously and successfully complete their routes.

Using probabilistic modeling we cluster bus trips to discover patterns of travel for buses which include time spans that are spent at different stops. Using model selection, we choose the best model parameter i.e. number of clusters equivalent to number of travel patterns. This probabilistic model allows us to simulate data since it is a generative model and can be easily

applied to other bus lines. Also, this model allows us to simplify the optimization problem of finding good spots to install wireless charging pads that cater to as many bus trip types as possible with minimum disruption of the current schedule.

In this study, we examined the MTA since it is comprised of over 5,900 buses in a fixed- route service in New York City as well as 2,000 vans and cabs for ADA para-transit service. Buses operate in the city on a continuous cycle with increased coverage during peak transit times. Mass transportation systems are critical in sustaining large metropolitan cities. Currently, bus transit is the second leading carbon emitter/passenger mile after private commuting vehicles.

The MTA has converted much of its fleet to hybrid buses that are designed with electric- drive systems that consist of a battery pack and electric motor. There is regenerative braking that supplies additional power to accelerate and for inclines. This technology combined with the use of a diesel particulate filter and ultra-low-sulfur fuel has reduced emissions of particulate matter by

90%, nitrogen oxides by 40%, and greenhouse gases by 30%. Additionally, fuel consumption of hybrid buses is approximately 25% to 35% less than a standard diesel bus.

After an interview with the Chief of Innovation and Technology at the New York City

Transit, we learned that the MTA committed to adopting greener environmentally friendly options for operating the large transit system in NYC. This sentiment is shared by several major U.S. cities.

Still opportunities exist to further reduce the emissions and increase “greenness” of buses.

Fleets of buses for transit (e.g., Greyhound) and school buses have not adopted these technologies.

For example, there are school bus emission reduction programs to encourage school buses to convert to better vehicle types. Similarly, there are several different trials ongoing worldwide for adoption of electric vehicles such as a 5-year trial ongoing in the United Kingdom and a three- month zero-emissions bus trial in Bangalore, India [BYD K9].

The goal of our study is to demonstrate the feasibility of electrical buses that use wireless charging technology throughout their route without disrupting the established pattern of operation for the fleet. We introduce a methodology by which patterns of operation of the transportation route is recognized and using this information we formulate a combinatorial optimization problem whose solution is the location of wireless charging pads that maintains the operation of the fleet.

This methodology provides the necessary framework to investigate any bus route and wireless charging technology and find the appropriate numbers and locations to install the charger pads.

6.1 Background

There have been several electric bus trials around the globe. While there is not a lot of academic literature available about these trials, there are several articles published to indicate that this is key trend for the future of electric buses. A novelty of this study is the quantification of the feasibility of electric buses in more research-based terms for future modeling and simulation as the technologies improve.

Ongoing trials in the United State include trials through a Utah-based company, Wireless

Advanced Vehicle Electrification (WAVE) [WAVE Web]. WAVE started operating buses on college campuses. While the idea of electric buses is not novel, WAVEs approach to reducing battery size and placing constraints on bus travel speeds to minimize over-usage of the battery is a new method for utilizing wireless charge transfer at designated areas [WAVE Web]. This downsizing of the battery size helps with cost reduction for these vehicles. WAVE currently has a trial in California and looks to expand its trial to 10-20 cities in the upcoming years [WAVE Web].

A long 5-year wireless electric bus pilot is ongoing in the United Kingdom since the beginning of 2014. This trial has bus operating recharging the buses to two-thirds of their battery capacities while parked over charging places when the bus drivers have their scheduled breaks.

100

The novelty is that the trial will try to utilize only two charging plates to support a fleet of 8 electric buses.

There was even an electric bus trial in New York City that was run in conjunction with

BYD (electric bus manufacturer) and the NYC MTA in 2012. The pilot tested several routes throughout Manhattan and covered a total distance of 1,481 miles. While the results from the trial were favorable, the main issue for the MTA’s adoption of the technology is the large initial costs required to convert to the electric buses. Additionally, this pilot test did not address the potential of wireless charging but rather relied on having the bus return to a depot. The main issue with this method is the long amounts of time needed to recharge the vehicle while it remains out of service, adding to the number of buses needed to maintain a working fleet.

A 4-bus fleet trial was run on a 6.1-kilometer (3.8 mile) pilot project route and will utilize

Bombardier’s inductive charging system throughout the route to keep recharging in Berlin in 2015.

At the proposed transfer rate output of 200kW, the bus needed only need a few minutes of charging points at end points of the line in order to recharge and continue on trek. The trial was a success and showed the feasibility of this application. Since then many such feasibility studies have been completed to show how buses can operate using wireless charging.

With several electric bus companies on the current market, as well as several competing wireless charge transfer devices also available, the current technical specifications available for these devices and buses are typically from the manufacturer. The United States Department of

Transportation (DOT) retained a company to produce a report which contains in detail a survey up to 2014 about the technologies that exist for Electric Buses. Current bus chargers due to advanced superconductors and better batteries can now charge as quickly as full charges in 3 minutes. There are some that even boast 10 seconds, although not much literature is available to validate the claim.

101

Our goal in this study, however, was not to pick a particular technology but to show the feasibility of having wireless charging at bus stops.

6.2 Data

We focus our analysis and feasibility study on the B63 bus route in Brooklyn, where it serves the communities of Bay Ridge, Sunset Park, Park Slope, Gowanus, Boerum Hill, and Cobble Hill

(Figure 6.1). The entire route is 5.8 miles long and is serviced 24 hours by the B63. It saw an annual patronage of 3,807,497 paid passengers in 2019.

We were provided data for the B63 route from April 3, 2011 through May 3, 2011. Each record in this data set contains, for a single bus, the time of observation, bus location, bus route, next stop, distance from that stop, and other variables (Table 6.1).

The numbers of individual trips (travels between starting points and final destinations) that are recorded in this data set exceed tens of thousands but are not perfectly clean. This volume of data can make any optimization problem extremely difficult. In the following sections we discuss how to overcome this challenge while gaining further insight into the dataset by using a generative probabilistic model and preprocessing the data to make it appropriate for our model.

We noticed a few more caveats on the data. First, there is no formal integration with the schedule, so trip ID's in this dataset cannot be used at any point to infer whether a bus was early or late. The trip ID’s only indicating a particular stopping pattern. Secondly, this data does not indicate the particular time that a given bus served (or passed) a particular stop. It simply relates, for every observation the server received from a bus, the ID of the next stop on that trip and the distance to that stop. To infer when a bus served a given stop, one can look at consecutive observations where the ID of the next stop changed.

102

Figure 6.1: Map of the B63 bus route in Brooklyn. Figure adapted from MTA.

103

Table 6.1: Metadata for data provided by MTA on bus route B63. Data Label Definition vehicle_id the 4-digit ID of the bus timestamp the date and time of the observation Latitude Latitude of the bus Longitude Longitude of the bus Phase the phase of the bus in its duty cycle; current extract includes only observations when the bus is inferred to be IN_PROGRESS (i.e. driving on the route) or LAYOVER_DURING (i.e. waiting at a terminal for a trip to begin) Trip_id GTFS trip_id representing the stopping pattern inferred for the given bus at the given time direction_id GTFS direction_id for the direction the bus is traveling trip_headsign GTFS destination sign value for the inferred representative trip shape_dist_traveled distance the bus has traveled (in meters) along the precise geographic route of the inferred representative trip Stop_id GTFS stop_id of the next stop the bus will serve Stop_sequence GTFS stop_sequence of the next stop the bus will serve dist_from_stop – the distance of the bus (in meters) from that next stop

6.3 Data Preprocessing

We examine the dataset from a perspective that would help with defining our probabilistic model.

The reasoning behind choices that are made is explained in the next section, which describes the details of the model. Previously, we saw that the records in the dataset belong to buses from different times of the day and locations. We group these records into individual trips defined by the process in which a particular bus travels from the starting point to the final destination. Note that the route is the same across all trips. Therefore, we break this route in equidistant segments.

Hereby, each record will fall into one of these geographical segments. This allows a single trip to be represented by a histogram of the number of records versus the geographical location. Bus routes generally do not follow a straight-line pattern. Thus, we project the trajectory of the route on two-dimensional surfaces onto straight lines.

104

Let \., \-, … . , \N denote the partition of all records into N individual observed trips.

Consider the smallest latitude and longitude intervals that confine all the records of the dataset which are defined by minimum and maximum values of each record. For simplicity, assume this geographical area be a square. We divide the latitude and longitude information into m equal size intervals which creates an area that can be thought of as a grid. Therefore, the grid has * = . ∗ . bins. Following the formation of the grid, each individual record can be assigned to a single bin in the grid. All the data from a single partition \' can form a two-dimensional histogram which is

∗ N denoted by Ñ'. The superposition of all the two-dimensional histograms, namely Ñ = ∑' Ñ', is a sparse matrix since much of this geographical region is not on the route of the vehicles. Therefore,

9∗9 Q Ñ' for all - can be encoded into a vector histogram using a map sP∗: Ν → Ν where M is the number of non-zero elements of H*. Let sP∗(Ñ') = ℎ', where the ℎ'" is the j-th non-zero component of Ñ∗ ordered first by column numbers followed by row numbers.

Figure 6.2 shows the visual representation of six different trips from the data archive. It demonstrates a variety of frequencies of recorded data-points for different geographic locations.

Figure 6.3 is a depiction of a single trip, using different granularity parameters. As granularity is increased, the data-points are more dispersed throughout the map.

6.4 Probabilistic Model

We are interested in the way a bus trip unfolds from the starting point till the final destination. The most relevant information is the amount of time that a bus might spend at certain locations.

Theoretically, if we know the exact amount of time information across all observed bus trips and possible distinct locations along the route, we can formulate an optimization problem to find the locations for which there should exist a wireless charging pad such that no bus in previously

105

observed trips would have run out of fuel (electricity). There are two major problems with this framing. The exact information is not available, and new trip types might emerge in the future.

In the next step, we use a multinomial mixture model as a probabilistic model in order to capture the nature of the observations (Figure 6.4). We imply the way that bus trips unfold from their origins to their destinations (which is fixed in our setting) can be categorized into histograms that are generated from a mixture of multinomial distributions (Figure 6.5). This is incentivized and can be characterized by the observation that during different hours of day and based on various weather patterns, i.e. the extent of road traffic, the time-location trajectory of trips can change. In similar circumstances it is expected for these trajectories to be similar as well. The individual multinomial distribution might represent a circumstance, which creates a certain type of bus trip.

The set Ñ = {ℎ., ℎ-, … , ℎN} which consists of all the observed trip histograms. Assuming there are ä mixtures, the model is described in equation 6.1. In equation 6.1, ãR is a normalized vector of dimension M and å is the categorical distribution of each cluster, which is depicted by

Figure 6.2.

N Y (6.1)

`çℎ.:N|U+:-V+:.,Wé = è `(ê'|å) è `(ℎ'|ãR, ê') 'X. RX.

The generative model is described as follows:

ê' ∼ MV,$*"%-7VE(å) (6.2)

ℎ' ∼ nDE,-.")-VE(ãê') (6.3)

106

Figure 6.2: Six different trips from the archive showing a variety of frequencies of recorded data-points for different geographic locations.

Figure 6.3: Depiction of a single trip, using different granularity parameters. As granularity is increased, the data-points are more dispersed throughout the map.

107

For inference, we use the Expectation-Maximization (EM) Algorithm which consists of iteratively updating the hidden parameters ê.:N and ã.:Y that is the conditional probability

`(ê.:N|å, ã.:Y, Ñ.:N). In the maximization step, the cluster parameters are updated using the membership assignments ê.:N.

Expectation Step: For all - = 1, … , b and 2∗ = 1, … , ä:

Q G/0 (6.4) ∏"X. ΘR∗" ê'R∗ = Y Q G/0 ∑RX.( ∏"X. ΘR" )

Maximization Step: For all . = 1, … , n and 2∗ = 1, … , ä

N ∑'X. ê'R ∗ ℎ'9 (6.5) ãR∗9 = Q N ∑"X. ∑'X. ê'R ∗ ℎ'"

N ∑'X. ê' (6.6) å = Q N ∑"X. ∑'X. ê'"

The EM algorithm requires initialization of parameters that is achieved by randomly selecting K normalized histograms as the original ã.:Y parameters and setting å to be a uniform categorical distribution. In order to find the number of clusters K, we perform model selection by using the

Akaike Information Criterion (AIC). We use AIC since it estimates the quality of each model relative to each of the other models given a collection of data models.

The result of running an instance of multinomial mixture model with 10 clusters on our data can be seen in Figure 6.6. Each histogram shows the distribution of time spent at each bus stop. Each bin number represents a stop and the height correspond to amount of time spent at a particular stop and each color represents a trip pattern. Note that the stops appear on the histogram

108

with the order that they appear on their physical route. The unit of time in the graphs is 30 sec.

6.5 Optimization

We formalize the problem of locating suitable places for installing wireless charging pads for a particular bus route. Let Ñ = {ℎ'} denote the available histogram of distribution of time over stops for each individual trip. These histograms indicate the amount of time that is spent at each individual bus stoop across all the observed trips. Without loss of generality, we assume all the stops to be equidistant. This allows us to establish a single unique depletion orate of the battery from any stop to the one right after it. We denote this universal stop-to-stop depletion rate by ï

(where ï is the percentage of the battery’s capacity, which is discharged if the vehicle moves from one stop to the one immediately after it). Note that, it is trivial to generalize the problem in order to account for various distances between any two immediate stops by introducing ï., ï-, … , ïQ where M is the number of stops. Batteries are charged at a rate of ñ(,) percent for , unit of time spent at each stop. We assume that ñ is a linear function in other words ñ(,) = % ∗ ,.

The goal is to ensure that the solution to the optimization problem, which consists of a set of charging pad locations, ensures no bus trip type is at risk of depletion in entire electric charge and thus fails to arrive at the final destination. Subsequently, we would like to decrease the amount of energy that the buses will require by other means (fossil fuels). In the formulation, a solution is an ordered tuple of locations ó = (ó., … , óN) for a fixed b where 1 ≤ ó. < ó- < ⋯ < óN ≤ n.

The following set of equations describes the utility function. Note that, ^'"(ó) indicates the remaining battery charge in trip - where the bus reaches stop ó" before recharging at that location.

For simplicity, we assume that óD = 1 for all possible solutions.

109

Figure 6.4: The graphical model, which describes the conditional dependencies of the observed variables, namely histograms and hidden parameters which are multinomial distributions.

Figure 6.5: An MM distribution example where each bin number represents a stop and the height correspond to amount of time spent at a particular stop and each color represents a trip pattern.

110

Q N (6.7) O(ó) = [ [ ôç^'"(ó) < 0é ∗ ^'"(ó) 'X. "X. "H. (6.8) ( ) ( ) ^'" ó = 100 + [(− óR − óRH. ï + ñℎ'Z1) − (ó" − ó"H.)ï RX.

Given öN, the set of all feasible solutions ó = (ó., … , óN) for a fixed b where 1 ≤ ó. < ó- <

⋯ < óN ≤ n which our problem can be formalized as .-)Z[K- O(ó).

Since it is computationally intractable to find the optimal solution to this combinatorial problem, we propose a local search algorithm. It is crucial to notice this solution would be computationally feasible only if, instead of all the possible trip variations, we consider the categories of trips that we have derived using the probabilistic model described in the previous

111

Figure 6.7: The graph shows the best achievable utility decreasing with respect to number of chargers that can be installed throughout the B63 route.

section. We initialize the solution to a randomly selected feasible solution. Then one of the stops in the solution is selected according to a uniform distribution. Then a value (direction) P is drawn from a beta distribution õ(1, ú) for an alpha parameter. Then the selected stop of the solution is updated to either ó = ó + ùP Qû "% ó = ó − ùP Qû at random. We make sure to choose the ' ' - ' ' - feasible one if one of them is not feasible. It is important to make sure there are no duplicate stops in the solution and that the new solution is an ordered tuple. If the new solution improves the utility, we update the solution. This is iteratively repeated for a fixed number of iterations.

This algorithm can be used to find the optimal utility for different available technologies

(i.e. different charging and depletion rates). The graph in Figure 6.7 shows the best achievable utility decreasing with respect to number of chargers that can be installed throughout the route.

112

6.6 Conclusion

With partnership from the MTA and through project funding through the UTRC, we were able to build a robust framework that is able to take the current bus performance and route data to model and show the feasibility of using wireless charging at bus stops to be able to have an all-electric bus fleet. We showed that the B63 bus route can perform optimally using this framework. We would need to study each of the bus lines in the NYC area to ascertain that all routes could be serviced by electric buses using this framework in order to be able to convert all bus routes to all electric buses.

113

Chapter 7: Future Work and Considerations

Future work can be useful in extending our charge transfer framework to become more robust, sensitive to various parameters, and addition of parameters to better model a situation. We identify a number of areas for future work, which include the development of better heuristics to determine charge transfer. Some specific ideas for possible future work are discussed in this chapter along with potential framework ideas for pursuing the development of these areas.

7.1 Extension to structured fleet vehicles

In the current framework that we have described in this thesis, we focus our analysis to unstructured fleet vehicles that include commuter vehicles and taxicabs. These unstructured vehicles are especially difficult to quantify and characterize in behavior since they have no defined path and schedule. However, as discussed in Chapter 4, structured fleet vehicles are those vehicles that have a defined path and schedule. These vehicles, such as buses and Federal Express trucks, have more regularity and precise timing information available to a social network. While we find that unstructured vehicles themselves provide convincing results to show that a V2V wireless charging network is a feasible solution to the charging availability problem for adaptation of EVs in the current market, it would be interesting to find out whether the addition of structured fleet vehicles would enhance the availability of charge in the system. Our hypothesis is that having structured fleet vehicles in the network would make available more charge to the participants in the network, since most structured fleet vehicles do not need most of their charge and have ample time to recharge at a depot.

Federal Express and UPS delivery trucks are an interesting addition to the model as they also spend time idly parked on the side of a road especially in large metropolitan areas where it is

114

simpler for the delivery person to walk packages within a certain radius after finding appropriate parking. As a result, these vehicles could serve as mobile charge stations that are constantly moving around in the metropolitan area to which other vehicles can simply preschedule rendezvous for charge exchange. This concept of a roaming mobile charge station would be an interesting addition to the current conceptual framework.

7.2 Extension to other types of unstructured fleet vehicles

Our model focuses on two specific types of unstructured fleet vehicles: the traditional household commuter vehicle and a taxicab fleet. There are several other types of unstructured fleets, which are equally as important and have large populations of vehicles, such as car rental companies.

Without infrastructure currently readily available, rental EVs may not come by charging stations easily. Specifically in NYC, anyone who rents an EV would be obligated to park at a garage overnight in order to charge their EV. The obvious downside to this is not only the lacking number of parking garages that currently have charging capabilities but also the burden of the high costs for parking on the renter.

To make the model more robust, it would be interesting to see what the effect of adding rental vehicles that are unable to recharge during a rental period to the system. Our hypothesis is that we would not see a significant change in the system since there is ample charge available for vehicles to share.

7.3 Determine charge transfer heuristics that optimize transfer based on priorities

We apply a very simple heuristic that only allows vehicles to exchange charge if at least one vehicle has enough charge available to reach its primary destination. This is an extremely

115

conservative method for charge sharing, as vehicles do not interact unless they can satisfy their own energy requirements first. A more aggressive approach and one that should potentially be modeled to determine impact on the system is for people to exchange charge even if they are giving away charge that would be needed for their own journey. The expectation being that they can then acquire charge at a later point in the journey to complete their trip. One reason for this may be due to increased incentives that can be gained during these transactions. For example, if a car can sell charge for a higher amount in the beginning of the journey and know that they can purchase charge for a lower amount later, then there is an implicit net profit, which may make the decision to exchange more units than necessarily available lucrative.

Another method to exchange charge especially given varying types of fleets on the roadway is by assigning priorities. For example, taxicabs have a higher priority than another type of vehicle as ICT is their primary means for the charge transfer. One way to enhance their priority is by weighing their incentives for participation to be greater than other types of vehicles.

7.4 Application to other geographic areas

The usage of vehicles for transportation varies quite differently in other parts of the world. We focus our simulation to model a large metropolitan area like New York City. We use our data and formulate our model area to represent that of New York City’s as well.

A limitation to this framework is the need for high traffic density and opportunities for charge transfer. Therefore, it is most successful in an urban environment that offers such opportunities like New York City. Even populous cities like Los Angeles would need adaptation of our framework since Los Angeles is a city that is spread out farther apart than how NYC operates in more condensed space. Another example of a populous city that would likely perform well in our current framework is Boston given its traffic similarities to New York City. Similarly, we

116

would need to modify our charging framework a bit to apply to semi-urban and rural areas since opportunities for charge transfer would likely reduce dramatically. Additionally, we would need to consider that vehicles in rural areas need to travel farther to reach their destinations and thus the current framework we propose may lack the energy needed for success. As a result, we would likely need to see more participation in the system for this to be feasible. As an example, imagine the rural areas of Wyoming, the combination of charging opportunities and lacking participating may make EVs infeasible using this framework at the moment. However, given that battery technology is improving, this may be an issue that is easily resolved in the future.

Therefore, it will be interesting to see the results of this model in other large metropolitan, semi-urban, and rural areas. We need to obtain data from these areas and determine how the various parameters of model would need to modify in order for the system to be as successful in these different geographies.

7.5 Add time as a factor to stochastic model

We do not account for time of day, day of week or month in our simulation. We assume that the number of vehicles is consistent no matter what time of day, week or month. This assumption is sufficient for showing the overall feasibility of V2V wireless charging. However, to make our system more robust and to truly represent a 24-hour cycle, we should vary the total number of cars as well as proportionally the types of cars throughout the day. For example, the total number of cars in the late night/early morning hours of the day are significantly lower than during regular business hours. Additionally, there are fewer taxicabs and personal vehicles out during those hours, but more delivery trucks that are cruising the area.

117

7.6 Vary population density of model

Our simulation currently models the vehicle population density to be uniform throughout our geographic area. This is an assumption made to simplify our model, however, a more robust model would account for non-uniform population densities. For example, upper Manhattan does not generally have the same number of vehicles as midtown Manhattan. Additionally, the distribution of types of vehicles also varies by geographic location. For example, midtown Manhattan has a higher ratio of taxicabs to personal vehicles. This variation of population density will certainly lead to the development of a different routing methods for EVs to enable them to acquire enough charge through charge transfers.

7.7 Model Charge Entering System

Our current simulation always assumes that each time a vehicle enters the system, it enters with a fully charged battery containing 100 units of charge. This is an ideal situation since we assume that vehicles are fully charged while they are not in use. However, this assumption is not necessarily true. Especially, if the future smart grid is full of EVs that are constantly charging, there may not be enough supply available to charge all EVs to full amounts whenever not in use.

As a result, testing the model with vehicles entering with varying amounts of charge is something that would be beneficial to determine minimum amounts of energy needed in the system to make the system functional. However, there are challenges in identifying a proper model to determine the distribution of available charge for the EVs entering the system.

118

7.8 Potential Positive and Negative Public Health Consequences

There are many unintended potential positive and negative consequences that are possible given our charge sharing framework. These should be further investigated to truly understand the impact of such a charge sharing framework for EVs.

Among positive outcomes are lowering noise, lowering direct air pollution, and less fuel extraction as an example. Since EVs are known to be quieter than traditional combustion-engine based vehicles, the adoption of EVs will inadvertently lead to lower ambient noise produced on the streets by these vehicles. The reduction of this sound pollution can have positive impacts specifically as health benefits.

EVs cause less direct air pollution but as a consequence, it remains to be seen what type of upstream air pollution concerns might arise from them. EVs require batteries which require production and proper waste management upon utilization. These likely cause additional and different types of pollution considerations especially if we expect mass adoption of EVs. The scale and impact of these pollutants need to be explored and quantified to understand if the negative consequences from this pollution would outweigh the benefits of the EVs. Similar considerations need to be made for fuel extraction as well. As the demand for EVs grow and participation increases in the charge sharing framework, what happens to the amount of energy that is consumed directly from the grid to charge the batteries? Will that cause additional pollution from the production of electricity for the charge. Is the pollution that arises from electricity production worse than TRAP from combustion-engine powered vehicles. This is analogous to the types of pollution considerations that are made for gasoline stations including but not limited to actual environmental remediations at the actual gas station site itself but also the pollution caused by the tankers in transporting gasoline as well as the pollution caused by drilling for the raw materials

119

needed to produce gasoline in addition to the process of producing it. Fundamentally, we would need do a far more in-depth study of all the sources of pollution for a traditional combustion-engine powered vehicle versus an EV to determine the overall impact of having one type of vehicle over the other.

Negative consequences that need to be considered for having a charge sharing network are unintended consequences in traffic congestion from having vehicles pull over to exchange charge.

Places like NYC already have significant traffic congestion issues especially during peak commuting hours and adding an additional component to how cars driving behavior changes as a result of charge sharing, we may actually see more traffic congestion. Additional traffic congestion causes more perturbation to people who have to drive on the streets but also had other negative side effects. In the case of EVs, when an EV is stuck in traffic, it is expending additional energy that should have been used to reach its destination or make available to the charge sharing framework. This would be a waste of charge for our system and would have adverse impact in how the system performs. The added stress to the human stuck in traffic likely also has negative health impact which would then need to be investigated as well. Traffic congestion also causes more accidents. Accidents can be detrimental to humans and can have negative health consequences. They likely also cause additional upstream pollution as damaged vehicles also have to be repaired. Between potential bodily harm to the driver and passengers as well as the need to repair a vehicle, this inevitably adds stress to the person having to deal with all of it. Stress, again, has negative health implications.

Similarly, we would have to worry about the consistency and safety of the wireless charging technology and unintended consequences from failure of the technology such as negative human health outcomes like accidental electrocution.

120

7.9 Introducing Charge Sharing as a Reality

Introducing charge sharing to any given geographic area will require urban planning for a full program implementation. Consider New York City as an example, special traffic lanes similar to bike lanes may need to be scoped and certain areas may need to be carved out as charge transfer opportunity lanes. For this to happen, several public agencies would have to be involved to accommodate this change. Similarly, if there were monetary incentives for exchanging charge, there would need to be regulation and some central authority to oversee to make sure that bad agents were not present in the system. Proper planning is an area of investigation to see what other considerations need to be made for the system to be made feasible in a given setting.

In other settings like highways, safe areas would have to be identified where EVs can safely exchange charge without the potential of causing accidents for other vehicles especially considering that traffic flows at faster rates on highways as opposed to more urban areas like cities.

In the future, self-driving cars that can calibrate speed and communicate with each other may lessen the impact of needing to scheduling physical rendezvous points where vehicles have to stop to exchange charge, but rather allow for EVs to exchange charge while moving at the same speed and interlocking in some fashion.

7.10 Conclusions

While our model can be enhanced to include more parameters that will help make the charge transfer framework to become more robust and provide even more fine-tuning to what can be expected in such a framework. We provide some commentary on the types of variables and parameters that can be incorporated. Of course, the addition of these variables will affect the output of the results we have seen in the dissertation, as it will add more granularity to the framework.

121

Recasting the framework with these additional variables will also help us to have greater understanding for different types of geographic environments such as rural, urban and semi-urban.

We have also identified further research areas to better understand the longer term intended and unintended consequences of having such a charge sharing framework. These impacts can range from impacts to waste management, to health outcomes, to how transportation fundamentally evolves as a result of having such a paradigm available. Of course, in order to make this charge-sharing network an actual solution, there is additional research that needs to be done identify how we can influence change management and participation in the network to encourage more adoption of EVs. Fundamentally, the program needed to roll out such a framework would require not only participation but acceptance from local governments for success.

122

Chapter 8: Conclusions

EVs are gaining popularity in the vehicle market for their environmental friendliness as well as lower costs for operation due to the vehicles being more efficient and needing less service as well as cost of fuel through re-charging being less costly than regular gasoline. We studied one potential method for overcoming the lack of charging infrastructures and method for increasing the effective driving distance of BEVs without needing to alter currently existing roadway infrastructures or current battery technologies. We further performed a case study to determine if large fleets such as taxicabs in a metropolitan area could become all electric without having a great impact on business for the drivers and availability for the consumers. Finally, we offer one method for pricing the charge exchange interactions to encourage participation in our charging exchange network.

In this thesis, we have demonstrated the need to extend the driving range of EVs given current cumulative distances driven by vehicle users through the availability of real-world data sets in the New York region. Our analysis of the Regional Household Travel Survey as well as the

NYC TLC’s data for metered cab rides from 2012 show the necessity of improving cumulative driving distances for electric vehicles in order for more widespread acceptance.

From this data, we propose the idea of extending the current driving range of electric vehicles through the utilization of wireless charge transfer between vehicles. This is a novel concept for vehicle-to-vehicle charge sharing as current utilization of this technology is always between vehicle and grid. We have performed a feasibility study to determine whether the currently available technology (both vehicle batteries and wireless charge devices) would be able to safely handle quick bursts of charging at scheduled rendezvous points and traffic lights. Our calculations show that the minor change in temperature at the node of the battery is not of particular

123

concern and the large transfer rates needed for the method to be feasible is not of concern as well since there is already existing technology on the market.

We have constructed a framework for an agent based discrete time simulation to show as a proof of principle demonstration of the effectiveness of vehicle-to-vehicle wireless charge transfer in an uncoordinated environment. In this case, vehicles exchange charge at traffic lights should there be a potential charging partner who also arrives at the same traffic light at the same time. With this simulation, we show the increased range in cumulative driving distances with a simple probabilistic method.

We also constructed an advanced simulation framework that coordinates rendezvous points between vehicles using advanced routing (fisheye routing) and scheduling (first come first serve) algorithms. We apply this simulation framework to the problem of converting all taxicabs in a large metropolitan area such as New York City. Our study shows that there is a strong possibility of converting all taxicabs to EVs along with participation from other commuter vehicles.

For the wireless charge-sharing network to be successful, we will need to ascertain that there is appropriate vehicle participation. To do so, we present a framework utilizing Nash

Bargaining theory to encourage participation by offering incentives for charge sharing.

We apply the same principles of wireless charge transfer to the NYC MTA buses and find that success of wireless charge transfer as a method of refueling batteries is directly correlated to the ability of a vehicle to have opportunities for charge transfer coupled with the amount of time available for the charging opportunity.

This thesis has demonstrated the ability for a wireless charge transfer to be a feasible method for increasing the adoption of EVs especially given that EVs would help reduce direct emissions and associated health affects as compared to traditional ICE-vehicles. The potential

124

public health benefits of lower TRAP and noise-related air pollution have significant implications for higher traffic-density areas. Since air pollution can also travel due to the movement of particles and gases, the lowering of TRAP benefits not only the immediate community but other neighboring areas as well. However, other potential consequences, including upstream pollution from the production of batteries, as well battery recycling and disposal would need to be assessed in additional research. The generalization of this work, which is tailored to New York City to other urban, semi-urban, and rural areas would also need to be investigated.

Charge sharing can also be used for other applications for transportation such as micro- mobility devices like electric scooters, motorcycles, and electric bikes. The need for these personal transportation vehicles is prominent in areas where there is significant traffic congestion and given other outlying factors like global pandemics involving Coronavirus. In the current pandemic social distancing as a means of reducing virus transmission has yielded a desire for people to find alternative means of transportation where public transportation used to be a primary mean. Our research shows that charge sharing systems provide novel opportunities to advance less polluting transportation systems in the 21st century.

125

References

[Adria Mora and Hilpert 2017] Adrià Mora, B. and Hilpert, M., “Differences in infiltration and evaporation of diesel and gasoline droplets spilled onto concrete pavement.” Sustainability, 9(7),

1271, 2017.

[Alimohammadi et. al. 2016] Alimohammadi, H., Fakhri, S., Derakhshanfar, H., Hosseini-Zijoud,

S.M., Safari, S. and Hatamabadi, H.R. "The effects of air pollution on Ischemic stroke admission rate." Chonnam Medical Journal. 2016; 52(1): 53–58.

[Andersen et. al. 2012] Andersen, Z.J., Bønnelykke, K., Hvidberg M., et. al. "Long-term exposure to air pollution and asthma hospitalisations in older adults: A cohort study." Thorax 2012; 67(1):

6–11. DOI: https://doi.org/10.1136/thoraxjnl-2011-200711

[Atkinson et. al. 2014] Atkinson, R., Carey, I., Kent, A., Van Staa, T., Anderson, H. and Cook, D.

"Long-term exposure to outdoor air pollution and the incidence of chronic obstructive pulmonary disease in a national English cohort." Occupational and Environmental Medicine; 2014. DOI: https://doi.org/10.1136/oemed-2014-102266

[Baker, Franz, Sweigart 1993] Baker, H.M., Franz, L.S., and Sweigart, J.R. “Coordinated transportation systems: An alternative approach to traditional independent systems,” European

Journal of Operational Research, Volume 66, Issue 3, 14 May 1993, Pages 341-352.

126

[Boen 2019] Boen, B., “After the Challenge: LaserMotive.” NASA, 4 Oct. 2019, www.nasa.gov/directorates/spacetech/centennial_challenges/after_challenge/lasermotive.html.

[Boschi 2012] Boschi N. (Ed.). “Defining an educational framework for indoor air sciences education.” In: Education and Training in Indoor Air Sciences. Luxembourg: Springer Science &

Business Media; (2012). 245 p.

[Botsford and Szczepanek 2009] Botsford, C. and Szczepanek, A. "Fast charging vs. slow charging: pros and cons for the new age of electric vehicles." International Battery Hybrid Fuel

Cell Electric Vehicle Symposium. 2009.

[Brook et. al. 2002] Brook, R.D., Brook, J.R., Urch, B., Vincent, R., Rajagopalan, S. and

Silverman, F. "Inhalation of fine particulate air pollution and ozone causes acutearterial vasoconstriction in healthy adults." Circulation. 2002; 105(13): 1534–1536. DOI: https://doi.org/10.1161/01.CIR.0000013838.94747.64

[BYD K9] “BYD K9.” Wikipedia, 18 Feb. 2021, en.wikipedia.org/wiki/BYD_K9.

[Carlson 2014] Carlson, R.W. and Normann, B. Test Results of the PLUGLESS™ Inductive

Charging System from Evatran Group, Inc. No. 2014-01-1824. SAE Technical Paper, 2014.

[Chameides 2014] Chameides, B. "Americans Driving Down Their Driving Miles." The Great

Energy Challenge Blog. National Geographic, 14 Jan. 2014. Web. 02 Sept. 2014.

127

[Chen, et. al. 2013] Chen, Z., Liu N., Xiao, X., Lu X., and Zhang, J., "Energy Exchange Model of

PV-Based Battery Switch Stations Based on Battery Swap Service and Power

Distribution," Vehicle Power and Propulsion Conference (VPPC), 2013 IEEE, vol., no., pp.1,6,

15-18 Oct. 2013.

[Chen et. al. 2016] Chen K, Glonek G, Hansen A, et al. “The effects of air pollution on asthma hospital admissions in Adelaide, South Australia, 2003–2013: Time-series and case–crossover analyses.” Clinical & Experimental Allergy. 2016; 46(11): 1416–1430. DOI: https://doi.org/10.1111/cea.12795

[Chen, Gokhale, Shofer, Kuschner 2007] Chen, T-M, Gokhale, J, Shofer, S, and Kuschner, WG.

“Outdoor air pollution: nitrogen dioxide, sulfur dioxide, and carbon monoxide health effects,”

Am J Med Sci. 333:249–56, 2007.

[Cheng et. al. 2016] Cheng, M.H., Chiu, H.F. and Yang,. C.Y. "Coarse particulate air pollution associated with increased risk of hospital admissions for respiratory diseases in a tropical city,

Kaohsiung, Taiwan." International Journal of Environmental Research and Public Health. 2015;

12(10): 13053–13068. DOI: https://doi.org/10.3390/ijerph121013053

[Choe et. al. 2019] Choe, S.A., et. al. “Ambient Air Pollution during Pregnancy and Risk of

Gestational Diabetes in New York City.” Environmental Research, vol. 175, 2019, pp. 414–20.

Crossref, doi:10.1016/j.envres.2019.04.030.

128

[Claudio et. al. 2006] Claudio, L., et. al. “Prevalence of Childhood Asthma in Urban Communities:

The Impact of Ethnicity and Income.” Annals of Epidemiology, vol. 16, no. 5, 2006, pp. 332–40.

Crossref, doi:10.1016/j.annepidem.2005.06.046.

[Clougherty et. al. 2013] Clougherty, J.E., et. al. “Intra-Urban Spatial Variability in Wintertime

Street-Level Concentrations of Multiple Combustion-Related Air Pollutants: The New York City

Community Air Survey (NYCCAS).” Journal of Exposure Science & Environmental Epidemiology, vol. 23, no. 3, 2013, pp. 232–40. Crossref, doi:10.1038/jes.2012.125.

[Connerton et. al. 2020] Connerton, P., et. al. “Air Quality during COVID-19 in Four Megacities:

Lessons and Challenges for Public Health.” International Journal of Environmental Research and

Public Health, vol. 17, no. 14, 2020, p. 5067. Crossref, doi:10.3390/ijerph17145067.

[Deloitte 2011] Deloitte. (2011). Gaining traction: Will consumers ride the electric vehicle wave?,

(April). Retrieved from http://www.deloitte.com/assets/Dcom-Global/Local

Assets/Documents/Manufacturing/Deloitte EV Survey Summary Findings China US Europe

Japan April 2011 _ Final.pdf.

[Dhameja 2001] Dhameja S., Electric Vehicle Battery Systems. Woburn, MA: Elsevier Science,

2001.

129

[Dutta 2013] Dutta, P., "Use of inductive power transfer sharing to increase the driving range of electric vehicles," Power and Energy Society General Meeting (PES), 2013 IEEE , vol., no., pp.1,5, 21-25 July 2013.

[Eberle and von Helmolt 2010] Eberle U., and von Helmolt, R., "Sustainable transportation based on electric vehicle concepts: a brief overview," Journal of Energy and Environmental Science,

2010, 3, 689 - 699.

[Ellis, et. al. 2013] Ellis, R., Fackrell, C., Gordon, T., Lamb, P., Morris, J.E., Kawasaki, C.,

"Battery recharging and testing swap stations," Technologies for Sustainability (SusTech), 2013

1st IEEE Conference on , vol., no., 208-211, 1-2 Aug. 2013.

[Ferreira et. al. 2016] Ferreira, T.M., Forti, M.C., de Freitas, C.U., Nascimento, F.P., Junger, W.L. and Gouveia, N. "Effects of particulate matter and its chemical constituents on elderly hospital admissions due to circulatory and respiratory diseases." International Journal of Environmental

Research and Public Health. 2016; 13(10): 947. DOI: https://doi.org/10.3390/ijerph13100947.

[Finkelstein, Jerrett, Sears 2004] Finkelstein, M.M., Jerrett, M., and Sears, M.R. “Traffic air pollution and mortality rate advancement periods.” American Journal of Epidemiology, 160(2),

173-177, 2004.

[Greenhouse 2014] Greenhouse Gas Emissions: Transportation Sector Emissions." EPA.

Environmental Protection Agency. Web:

130

www.epa.gov/climatechange/ghgemissions/sources/transportation.html, 04 May 2014.

[Halo Web] http://www.haloipt.com

[Harris 2009] Harris, A., “Charge of the electric car – [power electric vehicles],” Engineering &

Technology, IET Journals, 2009, 52-53.

[Hesterberg et. al. 2009] Hesterberg TW, Bunn WB, McClellan RO, Hamade AK, Long CM,

Valberg PA. “Critical review of the human data on short-term nitrogen dioxide (NO2) exposures: evidence for NO2 no-effect levels,” Crit Rev Toxicol. 39:743–81, 2009.

[Hilpert et. al. 2019] Hilpert, M., Johnson, M., Kioumourtzoglou, M. A., Domingo-Relloso, A.,

Peters, A., Adria-Mora, B., Hernandez, D., Ross, J. & Chillrud, S. N. “A new approach for inferring traffic-related air pollution: Use of radar-calibrated crowd-sourced traffic data,”

Environment International, 127, 142-159, 2019.

[Hinsdale 2020] Hinsdale, J. “By the Numbers: Air Quality and Pollution in New York City.” State of the Planet, 15 Jan. 2020, blogs.ei.columbia.edu/2016/06/06/air-quality-pollution-new-york- city.

[Hori 2004] Hori, Y. "Future vehicle driven by electricity and control-research on four-wheel- motored," UOT Electric March II. Industrial Electronics, IEEE Transactions on 51.5 (2004): 954-

962.

131

[Houghton 2008] Houghton R. A. “Carbon flux to the atmosphere from land-use changes: 1850–

2005,” In TRENDS: A compendium of data on global change. Carbon Dioxide Information

Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn.,

U.S.A. from http://cdiac.ornl.gov/trends/landuse/houghton/houghton.html, 2008.

[Humphrey et. al. 2019] Humphrey, J., et. al. “Putting Co-Exposures on Equal Footing: An

Ecological Analysis of Same-Scale Measures of Air Pollution and Social Factors on

Cardiovascular Disease in New York City.” International Journal of Environmental Research and

Public Health, vol. 16, no. 23, 2019, p. 4621. Crossref, doi:10.3390/ijerph16234621.

[Iskendar et. al. 2011] Iskandar, A., Andersen, Z.J., Bønnelykke, K., Ellermann, T., Andersen,

K.K. and Bisgaard, H. "Coarse and fine particles but not ultrafine particles in urban air trigger hospital admission for asthma in children." Thorax; 2011. DOI:https://doi.org/10.1136/thoraxjnl-

2011-200324.

[Ito et. al. 2011] Ito, K., et al. “Fine Particulate Matter Constituents Associated with

Cardiovascular Hospitalizations and Mortality in New York City.” Environmental Health

Perspectives, vol. 119, no. 4, 2011, pp. 467–73. Crossref, doi:10.1289/ehp.1002667.

[Jaap, Bechler, Wolf, 2005] Jaap, S., Bechler M., and Wolf, L. "Evaluation of routing protocols for vehicular ad hoc networks in typical road traffic scenarios." Proc of the 11th EUNICE Open

European Summer School on Networked Applications (2005): 584-602.

132

[Jensen et. al. 2009] Jensen, S. S., et. al. “Modeling Traffic Air Pollution in Street Canyons in New

York City for Intra-Urban Exposure Assessment in the US Multi-Ethnic Study of Atherosclerosis and Air Pollution.” Atmospheric Environment, vol. 43, no. 30, 2009, pp. 4544–56. Crossref, doi:10.1016/j.atmosenv.2009.06.042.

[Jevtić et. al. 2014] Jevtić M., Dragić N., Bijelović S. and Popović M. "Cardiovascular diseases and air pollution in Novi Sad, Serbia." International Journal of Occupational Medicine and

Environmental Health. 2014; 27(2): 153–164. DOI: https://doi.org/10.2478/s13382-014-0239-y

[Johansson, et. al. 2004] Johansson, E. et al. "An analysis of the fisheye routing technique in highly mobile ad hoc networks." Vehicular Technology Conference, 2004. VTC 2004-Spring. 2004 IEEE

59th. Vol. 4. IEEE, 2004.

[Kappos et. al. 2004] Kappos AD, Bruckmann P, Eikmann T, Englert N, Heinrich U, Höppe P, et al. “Health effects of particles in ambient air.” Int J Hyg Environ Health. (2004) 207:399–407.

[Karalis et. al. 2009] Karalis, A, Joannopoulos, J.D. and Soljačić, M. "Efficient wireless non- radiative mid-range energy transfer." Annals of Physics 323.1 (2008): 34-48.

[Kheirbek, Ito, et. al. 2014] Kheirbek, I., K. Ito, et. al. “Spatial Variation in Environmental Noise and Air Pollution in New York City.” Journal of Urban Health, vol. 91, no. 3, 2014, pp. 415–31.

Crossref, doi:10.1007/s11524-013-9857-0.

133

[Kheirbek et. al. 2016] Kheirbek, I., et. al. “The Contribution of Motor Vehicle Emissions to

Ambient Fine Particulate Matter Public Health Impacts in New York City: A Health Burden

Assessment.” Environmental Health, vol. 15, no. 1, 2016. Crossref, doi:10.1186/s12940-016-

0172-6.

[King et. al. A] King, C., Griggs, W., Wirth, F., and Shorten, R., “Using a car sharing model to alleviate electric vehicle range anxiety,” The 16th Yale Workshop on Adaptive and Learning

Systems, Yale University, New Haven, Connecticut, USA, June 2013.

[King et. al. B] King, C., Griggs, W., Wirth, F., Quinn, K., and Shorten, R. “Alleviating a form of electric vehicle range anxiety through On-Demand vehicle access,” submitted to International

Journal of Control.

[Kioumourtzoglou et. al. 2019] Kioumourtzoglou, M.A., Raz, R., Wilson, A., Fluss, R., Nirel, R.,

Broday, D.M., Yuval, J.B., Hacker M.R., McElrath, T.F., Grotto, I., Koutrakis, P., & Weisskopf,

M.G. “Traffic-Related Air Pollution and Pregnancy Loss,” Epidemiology. 30. 1, 2019.

[Ko, Jang, Jeong 2012] Ko, Y.D., Jang, Y.J., and Jeong, S. "Mathematical modeling and optimization of the automated wireless charging electric transportation system," Automation

Science and Engineering (CASE), 2012 IEEE International Conference on , vol., no., pp.250,255,

20-24 Aug. 2012.

134

[Kollanus et. al. 2016] Kollanus, V., Tiittanen, P., Niemi, J.V. and Lanki, T. "Effects of long-range transported air pollution from vegetation fires on daily mortality and hospital admissions in the

Helsinki metropolitan area, Finland." Environmental Research. 2016;151: 351–358. DOI: https://doi.org/10.1016/j.envres.2016.08.003

[Kurs 2007] Kurs, A. “Power transfer through strongly coupled resonances.” Diss. Massachusetts

Institute of Technology, 2007.

[Lakshari, Shladover, Lechner 1986] Lakshari, K., Shladover, S.E., and Lechner E.H., "Inductive

Power Transfer to an Electric Vehicle," International Electric Vehicle Symposium, 1986.

[Langer 2005] Langer, G. "Poll: Traffic in the United States." ABC News. ABC News Network,

13 Feb. 2005. Web. 05 Sept. 2014.

[Lee, et. al. 2010] Lee, S., Huh, J., Park, C., Choi, N.S., Cho, G.H., Rim, C.T., "On-Line Electric

Vehicle using inductive power transfer system," Energy Conversion Congress and Exposition

(ECCE), 2010 IEEE , vol., no., pp.1598,1601, 12-16 Sept. 2010.

[Lui, Wu, Gao 2013] Liu, C., Wu, D., and Gao, S. "Opportunities and Challenges of Vehicle-to-

Home, Vehicle-to-Vehicle, and Vehicle-to-Grid Technologies," Proceedings of the IEEE , vol.101, no.11, pp.2409,2427, Nov. 2013.

135

[Maciejczyk et. al. 2004] Maciejczyk, P.B., et. al. “Ambient Pollutant Concentrations Measured by a Mobile Laboratory in South Bronx, NY.” Atmospheric Environment, vol. 38, no. 31, 2004, pp. 5283–94. Crossref, doi:10.1016/j.atmosenv.2004.02.062.

[Matte et. al. 2013] Matte, T.D., et. al. “Monitoring Intraurban Spatial Patterns of Multiple

Combustion Air Pollutants in New York City: Design and Implementation.” Journal of Exposure

Science & Environmental Epidemiology, vol. 23, no. 3, 2013, pp. 223–31. Crossref, doi:10.1038/jes.2012.126.

[McConnell et. al. 2010] McConnell, R., Islam, T., Shankardass, K., Jerrett, M., Lurmann, F.,

Gilliland, F., Gauderman, J., Avol, E., Künzli, N., Yao, L., Peters, J., and Berhan, K. “Childhood incident asthma and traffic-related air pollution at home and school,” Environmental health perspectives, 118(7), 1021-1026, 2010.

[Milojevic et. al. 2014] Milojevic, A., Wilkinson, P., Armstrong, B., Bhaskaran, K., Smeeth, L. and Hajat, S. "Short-term effects of air pollution on a range of cardiovascular events in England and Wales: Case-crossover analysis of the MINAP database, hospital admissions and mortality."

Heart. 2014; 100(14): 1093–1098. DOI: https://doi.org/10.1136/heartjnl-2013-304963

[Mohrehkesh and Nadeem 2011] Mohrehkesh, S.; Nadeem, T., "Toward a wireless charging for battery electric vehicles at traffic intersections," Intelligent Transportation Systems (ITSC), 2011

14th International IEEE Conference on , vol., no., pp.113,118, 5-7 Oct. 2011.

136

[Motavalli 2013] Motavalli, J. "Fallout From Failure of Battery Swap Plan." The New York

Times. The New York Times, 01 June 2013. Web. 18 Aug. 2014.

[New Hampshire 2019] New Hampshire Department of Environmental Services Current and

Forecasted Air Quality in New Hampshire. Environmental Fact Sheet (2019). Available online at: https://www.des.nh.gov/organization/commissioner/pip/factsheets/ard/documents/ard-16.pdf

(accessed September 22, 2019).

[NYCAAS 2018] “NYC Environmental Health.” NYC EHS, 2018, nyc- ehs.net/nyccas2020/web/report.

[Patel et. al. 2009] Patel, M.M., et. al. “Spatial and Temporal Variations in Traffic-related

Particulate Matter at New York City High School s.” Atmospheric environment (Oxford, England:

1994) vol. 43,32 (2009): 4975-4981. doi:10.1016/j.atmosenv.2009.07.004

[Patel et. al. 2010] Patel, M. M., et. al. “Traffic-Related Particulate Matter and Acute Respiratory

Symptoms among New York City Area Adolescents.” Environmental Health Perspectives, vol. 118, no. 9, 2010, pp. 1338–43. Crossref, doi:10.1289/ehp.0901499.

[Patel, Quinn, et. al. 2011] Patel, M. M., Quinn, J.W., et al. “Traffic Density and Stationary

Sources of Air Pollution Associated with Wheeze, Asthma, and Immunoglobulin E from Birth to

Age 5 Years among New York City Children.” Environmental Research, vol. 111, no. 8, 2011, pp.

1222–29. Crossref, doi:10.1016/j.envres.2011.08.004.

137

[Patel, Chillrud, et. al. 2013] Patel, M. M., S.N. Chillrud, et al. “Traffic-Related Air Pollutants and

Exhaled Markers of Airway Inflammation and Oxidative Stress in New York City Adolescents.”

Environmental Research, vol. 121, 2013, pp. 71–78. Crossref, doi:10.1016/j.envres.2012.10.012.

[Pei, Gerla, Chen 2000] Pei, G., Gerla, M., and Chen, T.W. "Fisheye state routing: A routing scheme for ad hoc wireless networks." Communications, 2000. ICC 2000. 2000 IEEE International

Conference on. Vol. 1. IEEE, 2000.

[Plugless Web] Evatran. “Meet Plugless | The Wireless EV Charging Station.” Plugless Power, www.pluglesspower.com. Accessed 2 Mar. 2021.

[Phung et. al. 2016] Phung, D., Hien, T.T., Linh, H.N., et. al. "Air pollution and risk of respiratory and cardiovascular hospitalizations in the most populous city in Vietnam." Science of The Total

Environment. 2016; 557: 322–330.

[Richmont-Bryant et. al. 2017] Richmont-Bryant, J., Owen, R.C., Graham, S., Snyder, M.,

McDow, S., Oakes, M., et al. “Estimation of on-road NO2 concentrations, NO2/NOX ratios, and related roadway gradients from near-road monitoring data,” Air Qual Atm Health. 10:611–25,

2017.

[Rivers 2002] Rivers, C. “Coordination in Vehicle Routing.” Ph.D. thesis at Massey University,

New Zealand, 2002.

138

[Rosenkranz 2003] Rosenkranz, C. "Deep cycle batteries for plug-in hybrid application." EVS-20

Plug-in Hybrid Workshop. 2003.

[Ross et. al. 2011] Ross Z., et. al. "Noise, air pollutants and traffic: continuous measurement and correlation at a high-traffic location in New York City." Environ Res. 2011 Nov;111(8):1054-63. doi: 10.1016/j.envres.2011.09.004. Epub 2011 Sep 29. PMID: 21958559.

[Serrao, et. al. 2011] Serrao, L., Onori, S., Sciarretta, A., Guezennec, Y., and Rizzoni, G. "Optimal energy management of hybrid electric vehicles including battery aging," American Control

Conference (ACC), 2011 , vol., no., pp.2125,2130, June 29 2011-July 1 2011.

[Shearston et. al. 2020] Shearston, J.A., et. al. “Opening a Large Delivery Service Warehouse in the South Bronx: Impacts on Traffic, Air Pollution, and Noise.” International Journal of

Environmental Research and Public Health, vol. 17, no. 9, 2020, p. 3208. Crossref, doi:10.3390/ijerph17093208.

[Skiena 1990] Skiena, S. "Dijkstra's Algorithm." Implementing Discrete Mathematics:

Combinatorics and Graph Theory with Mathematica, Reading, MA: Addison-Wesley (1990): 225-

227.

[Spira-Cohen et. al. 2006] Spira-Cohen, A., et. al. “A Health Effects Analysis of Traffic-Related

PM Pollution Among Children With Asthma in the South Bronx, NY.” Epidemiology, vol. 17, no.

Suppl, 2006, p. S229. Crossref, doi:10.1097/00001648-200611001-00588.

139

[Spira-Cohen 2006] Spira-Cohen, A. “The Role of Traffic-Related Pollution in PM-Health Effects

Associations Among Inner-City Children With Asthma.” | Research Project Database | Grantee

Research Project | ORD | US EPA.” EPA, 2006,

cfpub.epa.gov/ncer_abstracts/index.cfm/fuseaction/display.abstractDetail/abstract/8138.

[Spira-Cohen et. al. 2009] Spira-Cohen, A., et al. “Personal Exposures to Traffic-Related Particle

Pollution among Children with Asthma in the South Bronx, NY.” Journal of Exposure Science &

Environmental Epidemiology, vol. 20, no. 5, 2009, pp. 446–56. Crossref, doi:10.1038/jes.2009.34.

[Spira-Cohen et. al. 2011] Spira-Cohen, A., et. al. “Personal Exposures to Traffic-Related Air

Pollution and Acute Respiratory Health among Bronx Schoolchildren with Asthma.”

Environmental Health Perspectives, vol. 119, no. 4, 2011, pp. 559–65. Crossref, doi:10.1289/ehp.1002653.

[Tauber 1995] Tauber, L.A., “Viability and Economics of Building or Purchasing, Driving, and

Maintaining an Electric Car,” Northcon 95. IEEE Technical Applications Conference and

Workshops Northcon 95, 1995, 400-405.

[Tonne et. al. 2016] Tonne, C., Halonen, J.I., Beevers, S.D., et. al. "Long-term traffic air and noise pollution in relation to mortality and hospital readmission among myocardial infarction survivors." International Journal of Hygiene and Environmental Health. 2016; 219(1): 72–78. doi: https://doi.org/10.1016/j.ijheh.2015.09.003.

140

[Thurston et. al. 2004] Thurston, G., et al. “A Study of Traffic-related PM Exposures and Health

Effects among South Bronx Children with Asthma,” Epidemiology, vol. 15, no. 4, 2004, p. S64.

Crossref, doi:10.1097/00001648-200407000-00156.

[United States Census Bureau 2013] "United States Census Bureau." Megacommuters: 600,000 in

U.S. Travel 90 Minutes and 50 Miles to Work, and 10.8 Million Travel an Hour Each Way, Census

Bureau Reports. 5 March 2013. Web. 18 Aug. 2014.

[UTRC Web] “University Transportation Research Center.” The City College of New York, www.ccny.cuny.edu/utrc. Accessed 2 Mar. 2021.

[Vidale et. al. 2010] Vidale, S., Bonanomi, A., Guidotti, M., Arnaboldi, M.and Sterzi, R. "Air pollution positively correlates with daily stroke admission and in hospital mortality: A study in the urban area of Como, Italy." Neurological Sciences. 2010; 31(2): 179–182. DOI: https://doi.org/10.1007/s10072-009-0206-8

[WAVE Web] “Wave | Wireless Advanced Vehicle Electrification | Inductive Charging.” Wave,

30 Jan. 2021, waveipt.com.

[WHO 2013] WHO Regional Office for Europe. Review of evidence on health aspects of air pollution – REVIHAAP Project: Technical Report [Internet]. Copenhagen: WHO Regional Office for Europe, C, Proximity to roads, NO2, other air pollutants and their mixtures. Available from: https://www.ncbi.nlm.nih.gov/books/NBK361807/, 2013.

141

[Wong et. al. 2016] Wong, C.M., Tsang, H., Lai, H.K., et. al. "STROBE long-term exposure to ambient fine particulate air pollution and hospitalization due to peptic ulcers." Medicine. 2016;

95(18). DOI: https://doi.org/10.1097/MD.0000000000003543

[Worley, Klabjan 2011] Worley, O., Klabjan, D., "Optimization of battery charging and purchasing at electric vehicle battery swap stations," Vehicle Power and Propulsion Conference

(VPPC), 2011 IEEE , vol., no., pp.1,4, 6-9 Sept. 2011.

[Wu, Gilchrist, et. al. 2011] Wu, H.H., et al. "A review on inductive charging for electric vehicles."

Electric Machines & Drives Conference (IEMDC), 2011 IEEE International. IEEE, 2011.

[Xie et. al. 2014] Xie, W., Li, G., Zhao, D., et. al. "Relationship between fine particulate air pollution and ischaemic heart disease morbidity and mortality." Heart; 2014.

DOI:https://doi.org/10.1136/heartjnl-2014-306165

[Yilmaz, Buyukdegirmenci, Krein 2012] Yilmaz, M., Buyukdegirmenci, V.T., Krein, P.T.,

"General design requirements and analysis of roadbed inductive power transfer system for dynamic electric vehicle charging," Transportation Electrification Conference and Expo (ITEC),

2012 IEEE , vol., no., pp.1,6, 18-20 June 2012.

[Yirka 2013] Yirka, B. "ZSW Engineers Build Lithium-ion Battery Able to Last for 27 Years."

ZSW Engineers Build Lithium-ion Battery Able to Last for 27 Years. 10 June 2013. Web. 18 Aug.

2014.

142

[Zhang, Wang et. al. 2014] Zhang Z., Wang J., Chen L., et. al. Impact of haze and air pollution- related hazards on hospital admissions in Guangzhou, China. Environmental Science and Pollution

Research. 2014; 21(6): 4236–4244. DOI: https://doi.org/10.1007/s11356-013-2374-6

[Zhang, Wong, Chen 2011] Zhang W., Wong, S.C., Tse, C.K. and Chen Q. "A study of sectional tracks in roadway inductive power transfer system," Energy Conversion Congress and Exposition

(ECCE), 2011 IEEE , vol., no., pp.822,826, 17-22 Sept. 2011.

[Zhang et. al. 2014] Zhang, W.; Wong, S.; Tse, C.; Chen, Q., "An Optimized Track Length in

Roadway Inductive Power Transfer Systems," Emerging and Selected Topics in Power

Electronics, IEEE Journal of , vol.PP, no.99, pp.1,1, 2014.

143