![Heat map of Columbus trip destination frequency in INRIX data set (source [9]).](https://www.researchgate.net/profile/Eleftheria-Kontou/publication/324255956/figure/fig4/AS:644326321295362@1530630774734/Heat-map-of-Columbus-trip-destination-frequency-in-INRIX-data-set-source-9_Q320.jpg)
Content uploaded by Eleftheria Kontou
Author content
All content in this area was uploaded by Eleftheria Kontou on Jul 03, 2018
Content may be subject to copyright.
2018-01-0667 Published 03 Apr 2018
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
Analysis of Fast Charging Station Network for
Electrified Ride-Hailing Services
Eric Wood, Clement Rames, and Eleftheria Kontou National Renewable Energy Laboratory
Yutaka Motoaki and John Smart Idaho National Laboratory
Zhi Zhou Argonne National Laboratory
Citation: Wood, E., Rames, C., Kontou, E., Motoaki, Y. et al., “Analysis of Fast Charging Station Network for Electrified Ride-Hailing
Services,” SAE Technical Paper 2018-01-0667, 2018, doi:10.4271/2018-01-0667.
Abstract
Today’s electric vehicle (EV) owners charge their
vehicles mostly at home and seldom use public direct
current fast charger (DCFCs), reducing the need for
a large deployment of DCFCs for private EV owners.
However, due to the emerging interest among transportation
network companies to operate EVs in their eet, there is great
potential for DCFCs to be highly utilized and become
economically feasible in the future. is paper describes a
heuristic algorithm to emulate operation of EVs within a
hypothetical transportation network company eet using
alarge global positioning system data set from Columbus,
Ohio. DCFC requirements supporting operation of EVs are
estimated using the Electric Vehicle Infrastructure Projection
tool. Operation and installation costs were estimated using
real-world data to assess the economic feasibility of the
recommended fast charging stations. Results suggest that
the hypothetical transportation network company f leet
increases daily vehicle miles traveled per EV with less overall
down time, resulting in increased demand for DCFC. Sites
with overhead service lines are recommended for hosting
DCFC stations to minimize the need for trenching under-
ground service lines. A negative relationship was found
between cost per unit of energy and fast charging utilization,
underscoring the importance of prioritizing utilization over
installation costs when siting DCFC stations. Although this
preliminary analysis of the impacts of new mobility para-
digms on alternative fueling infrastructure requirements has
produced several key results, the complexity of the problem
warrants further investigation.
Introduction
Today’s electric vehicle (EV) owners charge their vehicles
mostly at home and seldom use public direct cur rent fast
charger (DCFCs), reducing the need for a larger deploy-
ment of DCFCs for private EV owners. However, due to the
emerging interest among transportation network companies
(TNCs), whose operation may require quick fueling, there is
potential for DCFCs to be highly utilized and become economi-
cally feasible in the future as EV ride-hailing business evolves.
Despite their ability to charge EVs quickly, the deploy-
ment of DCFCs is currently limited because of the high costs
of both operation and installation that render the deployment
economically infeasible. e potential for high utilization by
ride-hailing EVs is a key to the economics of DCFC deploy-
ment; however, both the operation cost and installation cost
can vary dramatically depending on various factors. Past
studies pointed out great uncertainty in identifying and quan-
tifying signicant cost factors of DCFCs [1, 2]. At the same
time, much uncertainty persists in the operational character-
istics of TNC eets and how they may evolve in the future [3,
4, 5, 6, 7]. erefore, there is much interest in rigorous research
on assessing economic feasibility of DCFCs for TNCs in both
industry and academia.
e U.S. Department of Energy’s SMART (Systems and
Modeling for Accelerated Research In Transportation)
Mobility Advanced Fueling Infrastructure Pillar team
conducted simulations to estimate potential DCFC needs
(location, number of plugs, and electricity demand) by a hypo-
thetica l EV ride-hailing ser vice in Columbus, Ohio. Operation
cost and installation cost were estimated using real-world data
to assess the economic feasibility of DCFCs at the recom-
mended locations. is paper describes the methodology
developed for this study. It also provides key ndings of simu-
lation and analysis conducted by the three participating
national laboratories-Argonne National Laboratory (ANL),
Idaho National Laboratory (INL), and the National Renewable
Energy Laboratory (NREL).
INRIXGlobalPositioning
SystemTravelTrajectories
Understanding vehicle driving and parking patterns is key to
determining EV charging inf rastructure requi rements. Shared
NREL/CP-5400-70438. Posted with permission. Presented at WCX18: SAE World Congress Experience, 10-12 April 2018, Detroit, Michigan.
2 ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
vehicles likely have dierent driving patterns than personal-
use vehicles. erefore, analysis of ride-hailing vehicle use
patterns was necessary as a rst step in estimating DCFC
needs for ride-hailing EVs. Given the lack of available ride-
hailing data sets, this analysis rst develops a procedure for
synthesizing TNC activity patterns from personally owned
vehicle datasets and applies said procedure to an example
dataset from Columbus, Ohio.
Original INRIX GPS Data Set
NREL acquired individual anonymized global positioning
system (GPS) travel trajectories from INRIX [8], which
provided NREL with all GPS travel trajectories (mode imputed
as driving trips by INRIX) that intersected the Columbus
region at any time during 2016. Each trajectory features trip-
level data such as start/end times and GPS coordinates
(including origins, destinations, and intermediate waypoints).
e INRIX data set contains a total of 7.82 million unique
device identiers, 32.9 million trips, 1.04 billion miles of
driving, and 2.58 billion GPS waypoints. e spatial distribu-
tion of trip destinations in the Columbus area is shown in
Figure 1.
Down Sampling and
Processing
e GPS travel trajectories in the INRIX data set are an aggre-
gation of data from several providers and were down-selected
to include only light-duty vehicles. e subset of data from
light-duty consumer vehicles consisted of data sourced from
embedded GPS data (provided primarily by automotive
manufacturers from in-vehicle navigation systems) and
mobile devices (provided primarily by applications installed
in cellular devices). Individual device identiers from the
embedded GPSs were systematically reset aer each trip,
making EV charging simulation impossible. As such,
embedded GPS data were discarded, leaving only light-duty
consumer vehicle data from mobile device sources. is down-
sampling routine leaves approximately 14% of trips from the
total INRIX data set available. is cleansed subset includes
approximately 46.7 thousand unique device identiers, 1.41
million full travel days, 4.48 million trips, and 35.8 million
miles of driving.
Prior to using the INRIX data subset in plug-in electric
vehicle (PEV) driving/charging simulations, several data
processing steps were completed, including:
•Removing the rst and last vehicle-day for each device
identier (in an attempt to remove incomplete
travel days),
•Editing trip origins to match the previous destination in
the trip chain,
•Computing trip driving distance as the sum of haversine
distances between the original trip origin, each
waypoint, and trip destination,
•Estimating home and workplace locations for each
unique device and agging trips to these locations for
use in PEV driving/charging simulations,
•Implementing spatial joins on county, ZIP code, Trac
Analysis Zone, and land use data layers.
Categorizing charging events into home, workplace, and
public charging requires knowledge of the location type of
each trip destination. Unlike a typical travel survey, the INR IX
GPS data set does not report trip purpose. erefore, the desti-
nation type must be inferred from spatial and temporal
heuristics applied at the vehicle level. e INRIX data set
contains multiple travel days for each unique device identier,
which enables the analysis of dwell time patterns at recurring
destinations. e home and workplace location assignment
algorithm proceeds as follows:
•For each unique vehicle identier, destinations with
dwell times greater than a given threshold are selected
and clustered geographically in ~100m x 100m cells.
Nine-hour dwell locations are selected for home location
identication, and 4-hour dwell locations at non-home
locations are selected for workplace
location identication.
•e cumulative dwell time over all travel days is
calculated for each of these cells, and the cell with the
greatest cumulative dwell time is agged as the home or
work location.
•Any trips ending within ¼ mile from the home or work
agged location are considered home or work trips
respectively for this vehicle.
For both home and work/secondary locations, spatial
attributes such as ZIP code, Trac Analysis Zone, and land
use were appended by spatially querying the respective data-
bases and assigned to each vehicle. INRIX travel data were
validated using two travel surveys, the 2012 California
Household Travel Survey and the 2011 Massachusetts
Travel Survey.
FIGURE 1 Heat map of Columbus trip destination
frequency in INRIX data set (source [9]).
© SAE International
ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES 3
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
Ride-Hailing
EmulationAlgorithm
Methodology
City-level analyses on the broad impacts of TNC systems have
been conducted through surveys and eld data collection [10]
due to the lack of detai led data provided by TNCs. e absence
of data sets that may uncover impacts of such mobility systems
is a barrier in quantifying their benets and drawbacks. In
our study, due to the unavailability of data that describe TNC
vehicle movements, a heuristic was deployed that emulated
TNC vehicle data for ride-hailing systems, using as inputs
personal trip data sets. e heuristic process objective is to
enable matching of personal trips to TNC vehicle IDs, by
essentially grouping together trips that can be conducted
consecutively, and by allocating groups to TNC vehicle IDs.
e proposed algorithm, which is portrayed using a sche-
matic representation in Figure 2, rst identies trip candidates
that can be conducted consecutively based on the location and
time of their destinations and origins. In this step, a candidacy
list Ci was created that contains all trips j≠i whose origin is
within a specied space and time distance from a certain trip’s
i destination (repeated for all trips in the set I where i,j∈I) by
imposing two constraints: 1) the down time between trips is
less or equal to an upper bound
ˆ
t
and greater or equal to the
required time td to cover the distance between the trip’s i desti-
nation and the next trip’s j origin with td=dij/
s
(note that dij
is the distance between the trip’s i destination and the next
trip’s j origin and
s
the average speed to cover that distance),
and 2) the deadheading distance dij is less than or equal to an
upper bound
ˆ
d
. ere is no provision that allows prospective
TNC riders to wait for TNC vehicles and depart later than the
desired time (which is the time of departure as dened in the
personal trip data set) since trip origin and destination times
are strictly set and are not exible. is assumption also i mplies
that the trips’ times and distances, as well as routes, have not
changed or been impacted due to the TNC vehicle operation
and are the same as the ones in the personal trip data set.
e second step of the heuristic involves determining
which trip j that is included in the candidacy list of i will be
conducted in sequence-this process constitutes trip-matching.
e trip j that belongs to Ci with the minimum deadheading
distance (MIN(dij), j∈Ci) is selected and conducted aer i,
under the assumption that the driver of the TNC automobile
or the application that assigns that vehicle to the next trip goal
is the minimum of the deadheading distance between the
trips in a sequence.
Note that the heuristic described above does not assign
trips that cannot be grouped with other trips to TNC vehicle
IDs due to the time and location constraints set. e assump-
tion was that those trips were conducted by a personal vehicle.
e heuristic algorithm was implemented in Python 2.7.12
leveraging the processed INRIX data.
FIGURE 2 Schema of heuristic process for pseudo TNC trip data emulation.
© SAE International
4 ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
Results and Discussion
e methodology was applied to 5,000 passenger travel-days
from personal GPS trip data from Columbus. Trips are
assumed to be completed in individual vehicles without
allowing for sharing/pooling. e ride-hailing emulation
attempts to match trips by minimizing deadheading distance
and wait time with constraints of a maximum of 5miles or
20minutes between a given destination and potential next
origins. Ride-hailing vehicles must satisfy the same travel
demand (all trips are served by these vehicles). Trip chains are
thus locally optimized, meeting the objective of minimum
deadheading distance for each TNC vehicle, but not globally
at the system level.
Tab le 1 compares summary statistics for the original
and pseudo-synthetic ride-hailing data sets. Fewer ride-
hailing vehicles complete more trips due to the addition of
deadheading trips, connecting passenger drop-off and
pick-up locations. ese results are highly dependent on the
constraints used for trip matching. With no time or distance
constraints between two consecutive trips, the number of
vehicles is reduced drastically while the total system vehicle
miles traveled (VMT) increases further. In contrast, with
a more stringent deadheading distance constraint of 2miles,
the number of ride-hailing vehicles deployed exceeds the
number of original vehicles due to the inability to match a
large share of trips. One last caveat is that due to computa-
tional limitations, a sample size of only 5,000 vehicle-days
of travel was used. A larger sample size may increase
the probability of matching trips and potentially
increase system-wide eciencies. e small sample set
used here reflects an early-stage ride-hailing market.
Simulating a larger sample set, potentially segregating “ride-
hailing candidates” and “personal vehicles” would provide
a better projection of a more mature ride-hailing
market segmentation.
PEVCharging
InfrastructureSimulation
EVI-Pro Methodology
NRE L developed the Electric Vehicle Infrastr ucture Projection
(EVI-Pro) tool in partnership with the California Energy
Commission to estimate regional requirements for charging
infrastructure to support consumer adoption of PEVs [11, 12].
EVI-Pro uses PEV market projections and real-world travel
data to estimate future requirements for residential, work-
place, and public charging under a variety of scenarios. e
model aims to anticipate spatially and temporally resolved
consumer charging demand while capturing variations with
respect to residents of single-unit dwellings (SUDs) and multi-
unit dwellings (MUDs), weekday/weekend travel behavior,
and regional differences in travel behavior and vehicle
adoption. A graphical representation of the input/output rela-
tionships in EVI-Pro is shown in Figure 3.
EVI-Pro’s charging behavior emulation assumes that
consumers aim to complete all their existing travel electrica lly
while minimizing operating cost. Several charging scenarios
are simulated for each consumer. To identify the optimal
charging scenario, individual travel days from the INRIX
travel data set (originally completed using a conventional
gasoline vehicle) are simulated in the model under dierent
assumptions for charg ing infrastructure availability. e latter
include residential Level 1 (L1) and Level 2 (L2) charging
stations at SUDs, residential L2 charging stations at MUDs,
workplace L2 charging stations, public L2 charging stations,
and public DCFC.
EVI-Pro repeats this charge behavior selection routine
for all travel days in the study and for all vehicle types under
consideration. e modeled PEV eet consists of 20% plug-in
hybrid electric vehicles with a range of 20miles (PHEV20),
20% PHEV50, 20% battery electric vehicles with a range of
100miles (BEV100), 20% BEV250, 10% PHEV20 sport utility
vehicles (SUVs), and 10% BEV250 SUVs for both personal and
ride-hailing vehicles. The default charging behavior is
TABLE 1 Personal and simulated ride-hailing
vehicles comparison.
Metrics Personal vehicles
Ride-hailing
vehicles
Number of vehicles 5,000 individual
vehicles
4,834 total vehicles
3,726 ride-hailing
vehicles chaining
multiple trips
1,108 single-trip
vehicles unable to
chain trips
Number of trips 18,460 individual
trips
25,115 total trips,
including 7,112
additional
“deadheading” trips
Total system VMT 143,139miles 148,149miles
Mean daily VMT 28.6miles 37.0miles *
Trip mean distance 7.8miles 5.9miles
*Note: the mean daily vehicle miles traveled (VMT) reported here are
for the 3,726 ride-hailing vehicles that were able to chain trips and
exclude the additional “single-trip vehicles” that would skew the mean
daily VMT.
© SAE International
FIGURE 3 Graphical representation of inputs/outputs and
data flow in EVI-Pro.
© SAE International
ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES 5
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
“home-dominant,” meaning that consumers prefer to charge
at home, then at their workplace, and then in public locations.
All drivers, whether of personal or ride-hailing vehicles, are
modeled as having access to charging at their residence. In an
automated and driverless future where ride-hailing vehicles
would be owned by eet operators, these vehicles would poten-
tially have reliable access to charging at a depot.
This charging demand simulation generates a set of
charging sessions required to satisfy the travel patterns
displayed in the data in a way that maximizes electric miles
traveled and minimizes operational cost. ese charging
sessions are then post-processed spatially and temporally to
output electric vehicle supply equipment requirements and
use for the Columbus region.
Simulation Results
Estimated PEV charging infrastructure requirements are
shown by mode and plug type in Ta ble 2. Uncertainty in these
estimates is driven by several factors that were not explicitly
modeled in EVI-Pro, including: uncertainty in PHEV demand
for public charging, consumer access to home charging at
MUDs, consumer ability to make shared use of public
charging stations, and consumer tolerance for station/destina-
tion proximity. EVI-Pro provides a range of values in an
attempt to quantif y these uncertainties. e values presented
below are midpoints.
While residential charging requirements remain similar
for personal and ride-hailing vehicles, the demand for non-
residential charging is drastically dierent. Shorter dwell
times at work reduce the demand for workplace charging by
28%, while more frequent dwell events in public locations
combined with higher dai ly VMT increases the need for public
L2 and DCFC by 83% and 82%, respectively.
PEVCharging
InfrastructureCostAnalysis
Siting DCFC where use is expected to be high is important to
increase the economic feasibility of DCFC for station owners.
However, demand for charging is not the only factor that
should be considered when choosing DCFC site locations. e
cost to install and operate DCFC also should be considered.
Not only can costs be high, but they a lso vary widely depending
on how the DCFC is used and where it is located.
Operating and capital costs were estimated for the DCFC
candidate locations output by EVI-Pro to show relationships
between cost, use, and location.
Operation Cost of
ChargingInfrastructure
EV charging station operators must buy electricity from local
utility companies unless the station is owned by a utility
company or generates its own electricity on-site. In this
section, the operation costs of a group of EV charging stations
were assessed. e analysis includes two major assumptions:
1) the station operator must buy all the station’s electricity
from utility companies, and 2) the operator is a standalone
business, that is, it buys electricity exclusively for the
charging station.
Cost Estimation e monthly electricity bill is deter-
mined by the applied rate plan of the utility company, as well
as the electricity consumption and maximum demand of the
charging station.
Rate Schedules. Utility companies usually have multiple
rate plans designed for dierent groups of users with varying
voltage requirements and maximum demands. A rate plan
is composed of base charges including monthly charge,
energy charge, demand charge, and so on, as well as riders,
which may be at rates, may depend on energy or power
consumption; or may be a percentage increase on the base
bill. In addition, some utilities have rates that differ
depending on the season and the time of day. For example,
some utilities charge higher rates during the summer or
daytime when electricity demand tends to be higher [13].
Disincentivizing electricity use during these times reduces
peak demand (peak power usage) and puts less strain on the
utility’s generators [14]. A customer may be eligible for more
than one rate plan. Peak demand usually determines rate
plan eligibility.
Tab le 4 lists types of utility charges and corresponding
symbols for use in bill ca lculation equations. Tabl e 5 lists some
of the rate plans used by Columbus Southern Power Company
[15], one of the two American Electric Power (AEP) companies
that operate in Ohio and dominates the electricity supply in
Columbus, Ohio, the location of the 12 hypothetical charging
stations modeled in this analysis.
To determine which rate plan to use for the simulated
charging station electricity bills, their user ty pe and eligibility
need to be identied. Residential rate types are excluded since
the DCFC stations in this study are targeted for commercial
use. Rates for the primary distribution system, subtransmis-
sion, and transmission at greater than 480-V service are
excluded [16]. Rates with power demand requirements below
24kW are also excluded because even low-power DCFC
stations charge at a minimum of 24 kW [17]. Eligibility is
determined by the charging station maximum demand.
TABLE 2 Infrastructure requirements that would be
necessary to support electrification of the two vehicle groups
(5,000 personal vehicles; 3,726 ride-hailing vehicles plus 1,108
personal vehicles).
Charger type
# plugs
(personal
vehicles)
# plugs
(ride-hailing
vehicles)
Ratio ride hailing
personal
−
Home SUD L1 3,732 3,555 0.95
Home SUD L2 172 212 1.23
Home MUD L2 702 702 1.00
Work L2 222 160 0.72
Public L2 211 387 1.83
Public DCFC 13 24 1.85
© SAE International
6 ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
Tab le 6 shows the basic proles of 12 charging stations
simulated by NREL, including energy usage, maximum
demand, number of charging sessions, and time spent at the
station for one day of various theoretical conditions in a future
Columbus. According to Ta ble 6 , maximum demand ranges
from 20.5kW (Stations 10 and 12) to 82.06 kW (Station 1),
indicating GS-2 Secondary Service is the most appropriate
rate schedule.
Customer Inputs. In addition to the utility company rate
schedule, estimating t he chargi ng station monthly bill requires
knowing its electricity usage. e monthly energy usage is
estimated by multiplying the daily energy usage by 30. Ta ble
6 is the major source of customer data, and Table 7 lists the
customer inputs needed to estimate the monthly bill.
is study uses the Ohio Power Company - Columbus
Southern Power Rate Zone Bill Calculation Spreadsheet to
estimate monthly electricity bills [15]. e spreadsheet receives
a month-long hourly energy usage prole (in kilowatt-hours)
and outputs the approximate monthly bill from that data for
each applicable rate plan. For the GS-2 Secondary Service rate
plan, the simplied equation for the monthly bill is:
Monthly bill MDDR
ccuc
=+ +,
(1)
where Mc is monthly charge, Dc is demand charge, Du is
customer maximum demand, and Rc is total applicable riders.
If more than one utility supplies the target charging
station, the equation for a monthly electricity bill must include
all types of charges in the relevant rate plan of each utility.
e equation is:
Monthly bill MB DAYS,EEDDONON
OFFOFF ONDOND
ccuc
uc
u
cu c
=+
+
(
++
MAX
uuc
uu uu
c
OFFD IF
OFFD OND,OFFD OND, SUR
+
−> −
)
()
+
00
1100 ,
(2)
where.
MBc minimum bill
DAYS number of days this month
Ec energy charge
Eu energy usage
ONc on-peak energy charge
ONu on-peak energy usage
OFFc o-peak energy charge
OFFu o-peak energy usage
ONDc on-peak energy demand charge
ONDu on-peak energy demand
TABLE 4 Types of charges composing utility electric bills.
Type of Charge Symbol Unit Description
Monthly charge Mc$ Customer, metering, and other monthly charges
Energy charge Ec$/kWh Energy, regulatory, and other kWh charges
Demand charge Dc$/kW Charge for highest power demand this month
Non-IDR power charge NIDRc$/NCP kW Power charge for loads <700kW, calculated with max power demand during
this month
IDR power charge IDRc$/4CP kW Power charge for loads >=700kW, calculated with max power during four
critical time periods specified by the utility
Summer seasonal energy
charge
Sc, s$/kWh Energy rate during the summer
Summer on-peak energy
charge
ONc, s$/kWh Charge for energy used during the on-peak during the summer
Summer o-peak energy
charge
OFFc, s$/kWh Charge for energy used during the o-peak during the summer
Summer demand charge Dc, s$/kW Charge for maximum power demanded during the summer
Winter seasonal energy
charge
Sc, w$/kWh Energy rate during the winter
Winter on-peak energy
charge
ONc, w$/kWh Charge for energy used during the on-peak during the winter
Winter o-peak energy
charge
OFFc, w$/kWh Charge for energy used during the o-peak during the winter
Winter demand charge Dc, w$/kW Charge for maximum power demanded during the winter
Riders Rc$ Total rider contribution; fees utilities charge to compensate for various losses,
such as energy loss in electricity transmission
Additional variable costs Ac$ Various other costs (primarily included in subsequent equations to
accommodate complicated energy, demand, and rider pricing schemes)
On-peak energy charge ONc$/kWh Charge for energy used during the on-peak
O-peak energy charge OFFc$/kWh Charge for energy used during the o-peak
On-peak demand charge ONDc$/kW Charge for maximum demand during the on-peak
O-peak demand charge OFFDc$/kW Charge for maximum demand during the o-peak
Minimum bill MBc$/day The minimum amount the customer must pay this month
Surcharge SURc% A percent increase on the total bill
© SAE International
ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES 7
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
OFFDc o-peak energy demand charge
OFFDu o-peak energy demand
SURc surcharge
e function MAX(val_1, val_2) returns the greater of
val_1 and val_2, and the function IF(condition, val_1,
val_2) returns val_1 if the condition is true, and
val_2 otherwise.
Results Tab le 8 shows the estimated monthly electricity
bill for the 12 simulated charging stations in Columbus. e
total electricity cost ranges from $316.5 to $1,397.11. Cost
eciency is calculated by dividing the monthly electricity
TABLE 5 Columbus Southern power rate plans [13].
Schedule Rate Plan Eligibility
R-R Residential service Available for residential electric service through one meter to
individual residential customers
RLM Residential Optional Demand Rate Available for optional residential electric service through one meter
to individual residential customers. Requires the installation of
demand metering facilities.
RS-ES Residential Energy Storage Available to residential customers who use energy storage devices
with time-dierentiated load characteristics approved by the
Company
RS-TOD Residential Time-of-Day Available to individual residential customers. Availability is limited to
the first 500 customers applying for service under this schedule.
RS-TOD 2 Experimental Residential Time-of-Day Available to individual residential customers on a voluntary,
experimental basis. Availability is restricted to customers served by
the circuits designated for the Company's gridSMART pilot program.
GS-1 General Service - Small Available for general service to customers with maximum demands
less than 10 KW.
GS-2 General Service
- Low Load
Factor
Secondary distribution system3Available for general service lo customers with maximum demands
of 10 KW or greater.
Primary distribution system
Subtransmission
Transmission
GS-2-TOD General Service - Time-of-Day Available for general service customers with maximum demands
less than 500 kW, Availability is limited to secondary service and the
first 1,000 customers applying for service under this schedule.
GS-3 General Service
- Medium Load
Factor
Secondary distribution system Available for general service to customers with maximum demands
of 50 KW or greater.
Primary distribution system
Subtratismission
Transmission
GS-4 General Service
-Large
Primary distribution system Available for general service to customers with maximum demands
in excess of 1000 KVA.
Subtransmission
Transmission
Source: Public Utilities Commission of Ohio, “AEP Ohio Standard Tari” AEP Ohio (2017)
© SAE International
TABLE 6 Simulated Charging Stations in Columbus, Ohio.
No. Sessions
No. of
plugs
Daily energy
usage (kwh)
Maximum
demand (kw)
1 6 2 117.81 82.06
2 17 4 297.11 61.5
3 2 2 85.24 61.5
4 2 2 91.51 61.5
5 6 2 94.17 66.85
6 6 2 155.66 61.5
7 48 7 638.35 78.91
8 2 1 17.58 28.93
9 3 1 30.75 28.75
10 2 1 26.63 20.5
11 7 2 52.91 33.55
12 3 1 23.36 20.5
© SAE International
TABLE 7 Customer inputs needed to calculate electricity
bills[14].
Input Symbol Unit Description
Monthly Energy
usage
EukWh kWh used this month
Maximum demand DukW Highest power needed
this month
Maximum demand
from previous year
ADukW Highest power
demand last year
On-peak energy
usage
ONukWh Total on-peak energy
used this month
O-peak energy
usage
OFFukWh Total o-peak energy
used this month
On-peak demand ONDukW Maximum on-peak
demand
O-peak demand OFFDukW Maximum o-peak
demand
Number of days this
month
DAYS days Total number of days
in this month
© SAE International
8 ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
bill by the tota l energy usage. Figure 5 shows a negative corre-
lation between cost efficiency and number of charging
sessions in a month. Cost per unit of energy usage decreases
as number of charging sessions increase at a charging station.
is relationship is caused by the monthly demand charge
averaging out with increased energy use from more
charging sessions.
Installation Cost of Charging
Infrastructure
DCFC operation is important to the DCFC vendor’s long term
economic viability; at the same time, installation costs can be
a signicant burden in materializing a DCFC vendor business.
Installation costs of DCFCs can vary depending on many
dierent technical and environmental factors. Cost data for
EV charge infrastructure are currently limited and can be
found in few peer-reviewed journal articles [1, 2].
DCFC installation costs were collected from the U.S.
Department of Energy’s EV Project and are summarized
as follows:
•Average cost=$23,662
•Median cost=$22,626
•Minimum=$8,500
•Maximum=$50,820
e total cost of the installations cited above includes
only costs paid to the electrical contractors to install Blink
DCFCs. is cost would typically include permit costs, engi-
neering drawings, contrac tor’s installation and administration
labor, subcontracted construction labor or equipment (e.g.,
concrete, asphalt, trenching, and boring), and materials other
than the DCFC itself. To evaluate the cost drivers for DCFC
installations during the EV Project, some of the features of
the installed hardware and site conditions were examined.
e following were found to be signicant DCFC installation
cost drivers observed during the EV Project that are not
specic to the Blink dual-port DCFC. eir impact on instal-
lation costs would be applicable for any installation of a DCFC
unit rated at 20kW or more:
1. Electrical service upgrade
2. Ground surface conditions
3. Materials
DCFC installations oen require new electrical service to be
added to the host’s site. e cost of these installations was
signicantly higher than those that did not require new
service. e magnitude of this cost increase depends on
existing electrical services at the host site and costs from the
electric utility to install a new metered electrical service.
Electrical service extension costs also varied depending on
TABLE 8 Simulated Charging Stations’ Monthly Electricity Bill Estimation.
Location
No. of
Sessions
(monthly)
Daily Total
Energy
Usage (kWh)
Monthly Total
Energy Usage
(kWh)
Maximum
Demand
(kW)
Monthly
Charge ($)
Demand
Charge ($) Riders ($) Total ($)
Cost
Eciency
($/kWh)
1 180 117.81 3,534.40 82.06 9.04 331.11 899.14 1,239.29 0.35
2510 297.11 8,913.42 61.5 9.04 248.03 1,140.04 1, 3 9 7.1 1 0.16
3 60 85.24 2,557.22 61.5 9.04 248.03 669.19 926.26 0.36
4 60 91.51 2,745.31 61.5 9.04 248.03 683.12 940.19 0.34
5 180 94.17 2,825.13 66.85 9.04 269.81 730.34 1,009.19 0.36
6 180 155.66 4,669.68 61.5 9.04 248.03 825.67 1,082.74 0.23
7 1,440 638.35 19,150.64 78.91 9.04 318.20 2,029.16 2,356.40 0.12
8 60 17.58 5 2 7. 3 2 28.93 9.04 116.55 268.79 394.38 0.75
9 90 30.75 922.40 28.75 9.04 115.75 296.71 421.49 0.46
10 60 26.63 922.40 20.5 9.04 82.68 224.78 316.50 0.40
11 210 52.91 1,587.35 33.55 9.04 135.51 383.75 528.30 0.33
12 90 23.36 700.90 20.5 9.04 82.68 217.48 309.19 0.44
Average 260 135.92 4,077.73 50.50 9.04 203.7 69 7. 3 5 910.09 0.36
© SAE International
FIGURE 5 Correlation between cost eciency and number
of sessions
© SAE International
Note that for AEP Southern Power Company’s GS-2 Secondary Service, the nominal energy charge is not included. All the charges related to
energy usage are included in applicable riders. For information about GS-2 Secondary Service, please check “2017-08-28_AEP_Ohio_Standard_
Tari” 6th Revised Sheet No.221-1, available at: https://www.aepohio.com/global/utilities/lib/docs/ratesandtaris/Ohio/2017-08-28_AEP_Ohio_
Standard_Tari.pdf
ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES 9
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
the electric utility’s policies for aboveground or underground
service. Overhead service is typically less expensive and
quicker than trenching for an underground service extension.
e cost of underground service extension varies depending
on the distance (i.e., the length of the underground passage)
from the transformer. erefore, to determine the economical
suitability of the location for DCFC installation, each site
needs to be vetted for available power and proximity to
existing power service. To quantify association between total
installation cost and the above-mentioned cost drivers, several
attributes of the DCFC sites from the EV Project were collected
via invoices and interviews with the contractors as needed.
An ordinary least-squared regression was estimated to
examine the statistical association between total cost and the
identied cost drivers. e coecient estimates and the 95%
condence intervals are shown in Table 10. e mean DCFC
installation cost at a site without electricity service upgrade is
estimated to be $18,290. is installation cost is signicantly
aected by the electricity service upgrade, which adds an addi-
tional cost of between $1,354 and $7,763. e cost of service
upgrade depends on whether the service is overhead or under-
ground. If the electricity service is underground, the cost of
service upgrade is aected by the type of ground service and
the distance of the needed underground power feed. If the
ground surface is gravel rather than concrete or asphalt, instal-
lation cost is estimated to reduce by between $231 and $9,143
and the cost of trenching or boring for DCFC installation, if
required, is estimated at between $38.79 and $174.59 per foot.
Low-cost installations require sucient electrical power
at the site to accommodate the DCFC and a simple installation
with either short underground conduit runs or surface-
mounted conduit. All suggested locations that are within AEP
Ohio’s service territory have adequate facilities to serve a
60-kwh DCFC. Because service upgrade is not necessary, the
relative dierence in installation costs among the proposed
DCFC installation sites primarily would be aected by the
costs of trenching and boring that are required to extend the
service lines if the service lines are underground. Figure 6
shows the map of Location #10 as a location with a potentially
low installation cost. As shown in the map, overhead power
lines, which are shown with blue lines, conveniently extend
around the parking lot of a large retail store, making it conve-
nient for installing DCFCs on parking space without a need
for extensive underground work. As discussed above, instal-
lation cost can be compounded by long underground conduits
and surface conditions that are expensive to restore. On the
other hand, Location #5 (Figure 7) has limited access to the
overhead service line. In Location #5, electrical service is
provided to the nearby amenities mostly via underground
lines. erefore, if DCFC is to be installed in the nearby
parking space, a considerable amount of trenching and boring
would be required, which is estimated to cost from $38.79 to
$174.59 per foot. Moreover, because the ground surface is
either concrete or asphalt, installation costs for this location
can potentially be much more expensive; the above estimates
for the cost model show the cost increase would be between
$231 and $9,143 relative to when the ground surface is gravel.
Total Cost of Charging
Infrastructure
Total capital expenditure was calculated on a monthly basis
and combined with monthly operating expenses to determine
the total cost of charging infrastructure. Capital cost of a
single DCFC unit (i.e., one plug) was assumed to be $40,000.
Each additional DCFC plug per site was assumed to add an
additional $40,000. Because the exact installation location of
DCFCs at each of the recommended sites is unknown and a
slight change in the installation position may signicantly
aect the installation cost, a range of installation costs was
computed based on the results from Table 10. Capital costs
for each site were added to the range of expected installation
costs for each site to provide total capital cost. e total cost
was amortized over 10years at an 8% discount rate to deter-
mine a monthly capital expenditure. Total operating costs for
each site were assumed to be the electricity cost, as shown in
Tab le 8 , plus $100/month for warranty, maintenance, network
service, and other fees. Total monthly cost for each site was
determined by adding the total monthly capital cost and the
total monthly operating cost.
To put total cost in terms that are relatable to revenue,
the total cost of charging infrastructure at each location was
divided by the number of charging sessions. Figure 8 shows
this cost. e horizontal bar represents the total operating
TABLE 9 Attributes data collected.
Variable Name Description
Service Upgrade Binary variable where 1 indicates that new
service was required and 0 indicates new
service was not required
Underground
Service
Binary variable where 1 indicates that new
service was required and 0 indicates new
service was not required
Gravel Binary variable where 1 indicates that
ground surface is gravel and 0 indicates
either asphalt or concrete
Distance Distance of underground power feed in feet
© SAE International
TABLE 10 Coecient estimates.
Coecient Standard. Error P-value 2.5% 97.5%
Intercept 18,290.26 863.02 <0.01 16,574.63 20,005.88
Service Upgrade 4,559.02 1,611.92 <0.01 1,354.61 7,763.41
Underground × Distance 106.69 34.15 <0.01 38.79 174.59
Underground × Gravel −4,687.10 2,241.56 <0.05 −9,143.19 −231.02
R-squared: 0.204
Adjusted R-squared: 0.176
© SAE International
10 ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
FIGURE 6 Map of Location #10. Satellite imagery credit: © 2017 Google, Map Data © 2017 Tele Atlas.
© SAE International
FIGURE 7 Map of Location #5. Satellite imagery credit: © 2017 Google, Map Data © 2017 Tele Atlas.
© SAE International
ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES 11
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
cost plus the mean total capital cost. e error bars represent
the range of expected installation costs.
Because some of the recommended locations expect low
utilization, the uncertainty in the installation cost can aect
their total cost per session considerably. However, as the
number of sessions increases, the eect of installation cost
variation is minimized, and operation costs dominate the
total cost. erefore, when siting DCFC stations, priority
should be placed on choosing a location with potential for
high utilization rather than choosing a location with minimal
installation cost.
Conclusions
EVI-Pro recommended 12 sites for DCFC installations to
support a hypothetical PEV ride-hailing service in Columbus,
Ohio. e total electricity cost at the recommended sites was
estimated to range from $316 to $1,397. Cost per unit of energy
use decreases as sites experience more charging sessions
because xed demand charges are distributed across a greater
number of kilowatt-hours.
Among the recommended sites, the sites with overhead
service lines are recommended for hosting the DCFC as
trenching and boring that are required for underground
service line extension can be a considerable cost driver.
Although the cost of service upgrade generally is a signicant
cost driver, all the recommended sites that are within AEP
Ohio’s territory were found to have enough service capability
to support DCFCs. However, some of the sites have limited
overhead service lines and underground service line extension
may be required.
e uncertainty in the actual installation cost may aect
the total cost; however, as the level of utilization increases, the
operation cost dominates the total cost. erefore, for DCFC
site selection for a ride-hailing service, priority should be
placed siting DCFC at locations with the potential for high
utilization rather than choosing locations based on low cost.
Recommendationsfor
FutureWork
Although this preliminary analysis of the impacts of new
mobility paradigms on alternative fueling infrastructure
requirements has produced several key results, the complexity
of the problem warrants further investigation. Repeating the
ride-hailing emulation process with a larger travel data sample
would increase the probability of matching trips to TNC
vehicles, increasing the overall eciency of the ride-hailing
eet. Simulating a larger data set in EVI-Pro would also shed
light on the ability to share infrastructure as the EV market
for TNC operations grows. Rening the input assumptions
for ride-hailing vehicle operations would add realism to the
proposed process for ride-hailing data emulation. For example,
it would be useful to constrain the rst and last trips of the
day for each driver, as those should start or end either at the
driver’s residence or at a depot where all ride-hailing vehicles
would be parked to charge overnight. In addition, vehicles
completing long out-of-area trips may be excluded from the
FIGURE 8 Total cost of charging infrastructure per site, calculated on a per-session basis. Error bars represent the expected
range of installation cost, which varies depending on the specific location chosen for the charging site.
© SAE International
12 ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES
© 2018 SA E Interna tional; Argonn e National Labo ratory; Natio nal Rene wable En ergy La borator.
ride-matching pool of candidates, as they are unlikely to be
used for ride-hailing services for such trips.
Modifying the algorithm to allow for ride-pooling (i.e.,
shared, multi-passenger ride-hailing) would shed light on the
potential to achieve VMT reductions due to this mobility
option. Developing algorithmic processes for other mobility
paradigms, such as car-sharing or car-pooling, would be inter-
esting additions. A comparison of the ride-hailing data emula-
tion results with real-world ride-hailing data would be invalu-
able for validating our methodology.
It is also important to point out that the site selection
criteria in t his study were solely based upon potential charging
demand: a location with a high level of simulated charging
needs is recommended for the DCFC installation. However,
in realit y, the property owner provides the space for the DCFC
installation, and it is uncertain if the recommended site would
be available for hosting the charging stations. e monthly
energy consumption was estimated based on the simulation
of energy use within a single day in the summer. However,
variability across dierent seasons and between weekdays and
weekends needs to be considered for a more accurate estima-
tion. Additionally, a charging station can be either owned by
a utility or run as a standalone business. Future research can
investigate the dierence in operational cost between a utility-
owned charging station and a charging station operated by a
non-standalone business.
References
1. Schroeder, A. and Traber, T., “e Economics of
FastCharging Infrastructure for Electric Vehicles,”
EnergyPolicy 43:136-144, 2012, htt ps://doi.org /10.1016/j.
enpol.2011.12.041.
2. Madina, C., Zamora, I., and Zabala, E., “Methodology for
Assessing Electric Vehicle Charging Infrastructure Business
Models,” Energy Policy 89:284-293, 2016, doi:10.1016/j.
enpol.2015.12.007.
3. Henao, A., “Impacts of Ridesourcing-Ly and Uber-on
Transportation including VMT, Mode Replacement,
Parking, and Travel Behavior,” accessed August 2017, 2017,
Available at: https://media.wix.com/ugd/c7a0b1_68028ed55e
47a1bb18d41b5a5af4.pd f.
4. Jenn, A. and Clewlow, R., “New Mobility Services and
Vehicle Electrication (No. 17-05269).” Transportation
Research Board 96th Annual Meeting Compendium of Papers,
2017, accessed August 2017, Available at: https://trid.trb.org/
view.aspx?id=1439073
5. Shaheen, S., Chan, N., Bansal, A., and Cohen, A., “Shared
Mobility: A Sustainability & Technologies Workshop:
Denitions, Industry Developments, and Early
Understanding,” accessed August 2017, 2015, Available at:
http://innovativemobility.org/wp-content/uploads/2015/11/
SharedMobility_WhitePaper_FINAL.pdf.
6. Wadud, Z., MacKenzie, D., and Leiby, P., “Help or
Hindrance? e Travel, Energy and Carbon Impacts of
Highly Automated Vehicles,” Transportation Research Part
A: Policy and Practice 86:1-18 , 2 016.
7. Chen, T.D., Kockelman, K.M., and Hanna, J.P., “Operations
of a Shared, Autonomous, Electric Vehicle Fleet:
Implications of Vehicle & Charging Infrastructure
Decisions,” Transportation Research Part A: Policy and
Practice 94:243-254, 2016.
8. INRIX, INRIX Analytics: Trips, 2017, http://inrix.com/
products/trips/.
9. Wood, E., Rames, C., and Muratori, M., “Charging Electric
Vehicles in Smart Cities: A Scenario Analysis of Columbus”,
Ohio, Forthcoming.
10. Rayle, L., Dai, D., Chan, N., Cervero, R. et al., “Just a Better
Taxi? A Survey-Based Comparison of Taxis, Transit, and
Ridesourcing Services in San Francisco,” Transport Policy
45:168-178, 2016, doi:10.1016/j.tranpol. 2015.10.004 .
11. Wood, E., Rames, C., Muratori, M., Raghavan, S. et al.,
“National Plug-in Electric Vehicle Infrastructure Analysis,”
Report from the US DOE Oce of Energy Eciency and
Renewable Energy, September 2017, https://www.nrel.gov/
docs/fy17osti/69031.pdf
12. Wood, E., Raghavan, S., Rames, C., Eichman, J. et al.,
“Regional Charging Infrastructure for Plug-in Electric
Vehicles: A Case Study of Massachusetts,” NREL/TP-5400-
67436, Jan 2017, https://www.nrel.gov/docs/fy17osti/67436.pdf
13. California Public Utilities Commission, What are TOU
rates?, 2017 http://www.cpuc.ca.gov/General.aspx?id=12194
14. Department of Energy, Oce of Electricity Delivery &
Energy Reliability, “Demand Response,” 2017, https://energy.
gov/oe/services/electricity-policy-coordination-and-
implementation/state-and-regional-policy-assistanc-4
15. Ohio Power Company, Columbus Southern Power Rate Zone
Bill Calculation Spreadsheet, 2017, https://www.aepohio.
com/account/bills/rates/AEPOhioRatesTarisOH.aspx
16. Public Utilities Commission of Ohio, “AEP Ohio Standard
Tari,” AEP Ohio, 2017, https://www.aepohio.com/account/
bills/rates/AEPOhioRatesTarisOH.aspx
17. Department of Energy, Oce of Energy Eciency &
Renewable Energy: Vehicle Technologies Oce, “DC Fast
Charging at the Workplace,” 2017, https://www.afdc.energy.
gov/case/2832
Contact Information
Er ic Wo o d
eric.wood@nrel.gov
National Renewable Energy Laboratory
15013 Denver West Pkwy MS 1634
Golden, CO 80401
Yutaka Motoaki
yutaka.motoaki@inl.gov
Idaho National Laboratory
2525 Fremont Ave
Idaho Falls, ID 83402
Zhi Zhou
zzhou@anl.gov
Argonne National Laboratory
9700 S. Cass Ave.
Argonne, IL 60439
ANALYSIS OF FAST CHARGING STATION NETWORK FOR ELECTRIFIED RIDE-HAILING SERVICES 13
Acknowledgments
is report and the work described were sponsored by the
U.S. Depart ment of Energy (DOE) Vehicle Technologies Oce
(VTO) under the Systems and Modeling for Accelerated
Research in Transportation (SMART) Mobility Laboratory
Consortium, an initiative of the Energy Ecient Mobility
Systems (EEMS) Program. e authors acknowledge John
Smart of Idaho National Laboratory for leading the Advanced
Fueling Infrastructure Pillar of the SMART Mobility
Laboratory Consortium. e following DOE Oce of Energy
Eciency and Renewable Energy (EERE) managers played
important roles in establishing the project concept, advancing
implementation, and providing ongoing guidance: David
Anderson, Sarah Olexsak, and Rachael Nealer.
The U.S. Government retains and the publisher, by
accepting t he article for publication, acknowledges that the U.S.
Government retai ns a nonexclusive, paid-up, i rrevocable, world-
wide license to publish or reproduce the published form of this
work, or allow others to do so, for U.S. Government purposes.
Definitions/Abbreviations
AEP - American Electric Power
BEVxx - battery electric vehicle with a range of xx miles
DCFC - direct current fast charger
EVI-Pro - Electric Vehicle Infrastructure Projection tool
L1 - level 1 charging station
L2 - level 2 charging station
MUD - multi-unit dwelling
PEV - plug-in electric vehicle
PHEVx x - plug-in hybrid electric vehicle with a range of
xx miles
SUD - single-unit dwelling
TNC - transportation network company
VMT - vehicle miles traveled
This is the work of a Govern ment and is not subject to copyright protection. Foreign copyrights may apply. The Government unde r which this paper was written
assume s no liability or responsibility for the contents of this paper or the use of this p aper, nor is it endorsing any manufacturers , products, or services cited
hereinand any trade name that may appea r in the paper has be en included only because it is essential to th e contents of the paper.
Positions and opinions advanced in this paper are those of the autho r(s) and not necessarily those of SA E International. The author is solely responsible for the
content of the paper.
ISSN 0148-7191
