IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO.

1, MARCH 2012 341

PEV Charging Profile Prediction and Analysis Based

on Vehicle Usage Data
Ali Ashtari, Member, IEEE, Eric Bibeau, Member, IEEE, Soheil Shahidinejad, Member, IEEE, and
Tom Molinski, Member, IEEE

Abstract—Present-day urban vehicle usage data recorded on security, and emission reduction [7], [19], [22]—have prompted
a per second basis over a one-year period using GPS devices the emergence of plug–in electric vehicles (PEVs). PEVs refer
installed in 76 representative vehicles in the city of Winnipeg, to pure electric and plug–in hybrid electric vehicles. With the in-
Canada, allow predicting the electric load profiles onto the grid
as a function of time for future plug-in electric vehicles. For each creased adoption of PEVs, electric utilities are involved in trans-
parking occurrence, load profile predictions properly take into portation in an unprecedented way [15], [20] which will mean
account important factors, including actual state-of-charge of the new challenges, infrastructure costs, and opportunities for them.
battery, parking duration, parking type, and vehicle powertrain. Understanding PEV load profiles helps utilities to mitigate its
Thus, the deterministic simulations capture the time history of impacts and optimize this new load into their system [9], [17].
vehicle driving and parking patterns using an equivalent 10 000
urban driving and parking days for the city of Winnipeg. These Utilities are interested in alleviating impacts of PEVs on
deterministic results are then compared to stochastic methods power supply, on the need for upgrades to distribution systems,
that differ in their treatment of how they model vehicle driving and on the installation of charging infrastructure. Moreover,
and charging habits. The new stochastic method introduced in electric mobility addresses energy drivers effectively when
this study more accurately captures the relationship of vehicle utilities match new PEV loads with new renewable genera-
departure, arrival, and travel time compared to two previously
used stochastic methods. It outperforms previous stochastic tion—independent of the generation mix. Thereby, utilities
methods, having the lowest error at 3.4% when compared to also need to understand additional restrictions imposed on
the deterministic method for an electric sedan with a 24-kWhr marginal power generation. Thus, PEVs translate to increased
battery pack. For regions where vehicle usage data is not available infrastructure costs for utilities, which requires optimization.
to predict plug-in electric vehicle load, the proposed stochastic However, new PEV loads can potentially improve utilities
method is recommended. In addition, using a combination of
home, work, and commercial changing locales, and Level 1 system by providing opportunities for load shaving, load
versus Level 2 charging rates, deterministic simulations for urban shifting, demand response, storage, and ancillary services for
run-out-of-charge events vary by less than 4% for seven charging distributed generations [11], [16], [24], [25]. In [6] and [8], the
scenarios selected. Using the vehicle usage data, charging scenarios impact of PEVs on generation systems are assessed by studying
simulated have no significant effect on urban run-out-of-charge dispatch, operational reserves, and wind generation integration
events when the battery size for the electric sedan is increased.
These results contribute towards utilities achieve a more optimal in the presence of PEVs. Other studies focus on the impact of
cost balance between: 1) charging infrastructure; 2) power trans- PEVs on the generation mix and emission reductions [1], [2].
mission upgrades; 3) vehicle battery size; and 4) the addition of Impacts of PEVs on distribution systems are studied addressed
new renewable generation to address new electric vehicle loads for in [5], [11]–[14], [24]. New renewable generation to address
addressing energy drivers. the new load remains the optimal strategy to address energy
Index Terms—Plug-in electric vehicles, power generation drivers simultaneously [21].
peaking capacity, power generation planning, signal analysis. Studies on PEV–utility issues provide recommendations for
utilities on how to better optimize the integration of PEVs into
their system. For optimization to be effective, they require real-
I. INTRODUCTION
istic PEV charging profiles. PEV–related decisions by utilities

R ECENT technological developments in the battery in- without reasonably accurate information on PEV charging pat-
dustry, the growing pressure from energy drivers—peak terns can lead to implementation issues and over expenditures.
oil, greenhouse gas emissions leading to climate change, energy Deterministic and stochastic approaches can be used to predict
the charging patterns of PEVs. Our study addresses gaps in the
Manuscript received February 01, 2011; revised May 20, 2011; accepted June literature: predicting PEV charging behavior based on vehicle
24, 2011. Date of publication September 12, 2011; date of current version Feb- usage habits using a relatively large database of driver behavior.
ruary 23, 2012. This work was supported by the AUTO21 Network of Centres
Previous studies on PEV load impacts do not use such extensive
of Excellence. Paper no. TSG-00026-2011.
A. Ashtari is with the Department of Mechanical and Manufacturing En- high-resolution database. However, our proposed approach still
gineering, University of Manitoba, Winnipeg, MB R3T 5V6 Canada (e-mail: samples driving behavior in a particular city using gasoline ve-
[email protected]).
hicles. Furthermore, all simulations currently neglect air-condi-
E. Bibeau and S. D. Shahidinejad are with the Department of Mechanical and
Manufacturing Engineering, University of Manitoba, Winnipeg, MB R3T 5V6 tioning loads and charging restrictions imposed by controllers,
Canada. as these can be added later and can vary considerably.
T. Molinski is with Manitoba Hydro, Winnipeg, MB R3C 2P4, Canada.
Currently, some PEV–utility studies in the literature assume
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. PEV charging profiles dictated by utilities [2]. Those studies
Digital Object Identifier 10.1109/TSG.2011.2162009 commonly use an arbitrary ideal charging profile optimized

1949-3053/$26.00 © 2011 IEEE

342 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 1, MARCH 2012

To best represent vehicle use patterns in the city of Winnipeg,

100 volunteers’ vehicles were randomly selected according to
socioeconomic attributes and data loggers were installed in their
vehicles. After a period of two months, 76 volunteer drivers in-
terested in the research continued their collaboration for another
ten months. Therefore, the current study uses the movements
of a fleet of 76 vehicles recorded on a per second basis over
a period of one year. The volunteers were selected from dif-
ferent income brackets, education levels, and gender and from
different areas of the city to create a statistical population best
representing the drivers within the city of Winnipeg. The fleet
of participating vehicles consists of sedans, both full and mid-
size (67%), and sport utility vehicles and pickup trucks (33%).
The collected data are representative for the activities such as
commuting, shopping, and socializing. Such representativeness
is ensured in this study by the relatively large number of vehi-
cles in the fleet and the long period of data collection.
The database consists of more than 44 million data points and
over 150 000 parking events. The collected data required signifi-
Fig. 1. GPS recording system. cant postprocessing due to weather, buildings, and underground
parking effects on GPS signals and categorizing parking locals
from the utilities’ perspective. Moreover, any controlled PEV as home, work, commercial, or other types of parking like road-
charging profile requires to first know the uncontrolled charging side parking. Details of the data reduction techniques are avail-
profiles. Recent PEV studies try to address this issue by using able in [4], including remapping of longitude and latitudes to
travel survey data [10], [14]. This study demonstrates that protect confidentiality of drivers. This database has been previ-
accurate prediction of PEV charging profiles requires more ously used to construct driving cycles representative of people’s
sophisticated stochastic modeling than available literature driving behavior in Winnipeg [3], [21], [23]. The current study
models. Lack of proper stochastic modeling can hinder op- uses trips’ beginnings, ends, and distance in addition to charging
timized integration of PEVs into electrical grid and lead to time, duration, and location to predict PEV charging behavior.
excess infrastructure. Electrification of the transportation sector The processed dataset used in this study can be downloaded
will achieve the original goal to address energy drivers by from [4].
the minimization of infrastructure costs and substituting the
displaced oil with new renewable generation. III. METHODS
This study analyzes vehicle usage data of 76 vehicles over a This section has three parts. Section III-A explains the as-
one–year period using data obtained by GPS recording devices sumptions used in the model. Section III-B focuses on devel-
on a secondly basis in the city of Winnipeg, Canada. Avail- oping a stochastic PEV charging profile prediction method, and
ability of the 44 million data points accurately predicts PEV Section III-C compares different PEV charging infrastructure
charging profiles. Such data is not commonly available in most with respect to PEV electric range.
regions, with survey data available [10] instead. We therefore
also introduce a stochastic method to use driver-surveyed data to A. System Model
predict PEV charging profiles in different regions and validate In order to predict the impact of PEVs on electricity load,
the new method using deterministic results obtained from ac- one has to identify key parameters and determine their contri-
tual vehicle usage data from a large database. Once proper PEV butions which statistically is a challenging task. Such param-
charging profile prediction methods are developed, the study eters include PEV powertrain (pure electric or plug-in hybrid),
compares different charging scenarios, such as home, work, and the daily usage profile of PEVs, the battery capacity of the vehi-
commercial locales using Level 1 and Level 2 charging rates. cles, and seasonal climate effects on the charging load. Accurate
These simulations provide useful information for optimized in- assessment of all these parameters will not be possible until the
vestments in PEV charging infrastructure to achieve sustainable technology is widely adopted. However, it is required to make
transportation. assumptions as the information is required now by utilities. Fur-
thermore, previous scientific methods developed to obtain duty
II. DATA COLLECTION cycles, like the car chasing technique, are concerned more with
This study uses vehicle data collected by the University of vehicle emissions that do not apply to PEVs in the same manner.
Winnipeg [4]. Data logging devices, equipped with Global As an example, a PHEV engine is decoupled from the accel-
Positioning System (GPS) manufactured by PERSENTECH erator and vehicle acceleration has less of a direct impact on
Inc., collect vehicle displacement data with commercial parking vehicle emissions but affects the vehicle driving range instead.
and road elements each having property attributes identified. This section describes the assumptions used in simulating the
Fig. 1 shows a picture of the data logging system installed in impact of PEVs on the electrical grid and explains why they can
the vehicles. meet the requirements of this study. The present analysis aims at
ASHTARI et al.: PEV CHARGING PROFILE PREDICTION AND ANALYSIS BASED ON VEHICLE USAGE DATA 343

providing the utilities a better insight to predict the forthcoming established using a deterministic or stochastic analysis, as
PEVs electrical loads now and help to recognize the grounds in presented in this study.
which investments are optimized. • PEV driving habits will be the same as current gasoline ve-
Several studies have attempted to predict charging load im- hicles as PEVs need to deliver a comparable service as ICE
pact on the grid using stochastic methods making general as- vehicles. This assumption allows predicting the PEV bat-
sumptions on the daily vehicle usage and charging availability tery state of the charge during simulations. However, it is
profile [1]. The present study uses a comprehensive database important to highlight that reduction in acceleration rates
of driving behavior in the city of Winnipeg that makes the re- improves electric driving range. Thus, driver feedbacks by
sults more representative and more deterministic. This study manufactures or third party products for PEVs may affect
aims at providing utilities like Manitoba Hydro, the provincial driving behavior in the future. Furthermore, implementa-
utility, with better information for decision-making. However, tion of sensors, both in the vehicle and on roads (e.g., on-
the methodology developed herein and the results of the case board over speed indicators and photo radar), may further
study for the city of Winnipeg are of interest to other North affect driver behavior.
American utilities. This is somewhat similar to previous duty
cycles developed for vehicle emission in a particular city where, B. PEV Charging Profile Prediction
although a very small representative measurement sample is
used, these cycles are used extensively in many jurisdictions. In this section, a deterministic and three stochastic methods
The assumptions used as part of the development of the simula- predict the charge profile using a simulation model implemented
tion algorithm shown in Fig. 2 to predict load profiles are: in Matlab. The deterministic approach is novel and uses the
• Only an EV sedan is presently considered. This EV sedan vehicle data directly. The three stochastic approaches use the
is assumed to have similar characteristics as a Nissan Leaf driver data indirectly using a statistical approach. Stochastic
[18]. Nissan Leaf has a relatively large battery pack (24 methods do not require accessing the complete vehicle usage
kWh) compared to the PEVs currently in the market. How- data and are presented for regions where vehicle usage data are
ever, this situation is most likely to change in future. For not available. Stochastic approaches only have access to dis-
the purpose of this study, using the characteristics of a typ- tributions of home departure time, daily traveled distance, and
ical full electric sedan, Nissan Leaf, is of interest to utilities home arrival time, similar to the data found in [10]. The first two
as it potentially represents an extreme PEV configuration of the stochastic approaches modeled are from previous studies.
that demands the most charging infrastructure. The next The third stochastic method is developed in this study and is
step will be to implement a PEV penetration model that shown to adhere to the deterministic exact results more closely
accounts for various powertrains and includes weather im- compared to previous stochastic methods. Implementation of
pacts. these different methods of calculations in Matlab using assump-
• The battery pack has a capacity of 24 kWh. However, dif- tion presented earlier and methods summarized in Table I are
ferent battery sizes are also simulated to assess the sensi- now presented. Fig. 2 details the algorithm for the various math-
tivity to the PEV battery size. ematical transformation required to produce the charging pro-
• Full charging takes 20 h on Level 1 charger and 8 h on files, with some of these transformations already explained in
Level 2 charger [18]. Note that not all charging events previous publications.
during vehicle simulations require a full charge as the state It should be noted that all three stochastic approaches use nu-
of charge (SOC) is tracked. merically obtained probability density functions (pdf’s). Pdf’s
• Electric drive efficiency is 6.7 kilometers per kWh [18]. used in the simulations are not fit to explicit functions. Rather,
A more accurate simulation would include a range of ef- they are used numerically to avoid unnecessary errors. This
ficiency values. However, the forecast of how different means these pdf’s are reproducible by simply obtaining the his-
PEV technologies having varied gas engine efficiencies are tograms of the data. All the driving data and the code is made
going to be adopted is beyond the scope of the present available publically to help the readers understand and replicate
study. Thus, this value is chosen for the simulations. the method.
• Drivers start charging immediately after they arrive at 1) Deterministic Method: Daily traveled distances, home ar-
home and at work. This is expected as because of extreme rival times, and home departure times extracted from the vehicle
cold winters in Manitoba, people currently plug in their usage database is the basis of the deterministic approach, which
block heaters immediately upon home and work arrival. for the purpose of this study requires processing 44 million data
However, time of day pricing of electricity can influence points. Vehicles are first charged in the simulation using a Level
charging profiles. It is hard to predict how different elec- 1 charger at home (120 V) and without utility charge control. For
tricity rate patterns would affect PEV charging profile: this each car in the first iteration of the SOC simulations, the SOC is
aspect can be investigated separately for different juris- assumed to be 100% when leaving home on the first day. SOC
dictions by modifying the simulation. Because Manitoba values for each car are iteratively calculated for the duration
currently has a single-tariff pricing system is in effect, this of the vehicle usage data over one-year duration. On occasions
is not an issue. More important, controllers can modify the when the vehicle is out of charge, its charge is set to zero when
PEV charging profile. This aspect should not be attempted charging begins. During the simulation, tabulated occurrences
until a clear picture of the uncontrolled charge profile is when vehicles are out of charge reflect the necessity of charging
344 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 1, MARCH 2012

Fig. 2. Simulation algorithm from the GPS data recoded by the OTTO system in each vehicle to the load impact profile calculations using a deterministic and
stochastic approach presented in this study.

TABLE I Because the focus of the research is on the methodology, using

PEV CHARGING PROFILE PREDICTION METHODS various powertrains and including air conditioning loads is out-
side the scope of simulations.
The following equation describes the charge cycle:

SOC SOC (2)

where is the charging time in hours and is the charging

infrastructure and impose a restriction on utility control possi- rate in kW. Charging time is the minimum of the parking time
bilities. Section III-C addresses this important aspect. and the required time for full charging considering the SOC at
Let , and be the home departure time, daily trav- the beginning of the charging. With respect to charging rate, ,
eled distance and home arrival time, respectively for the car Level 1 charging is used in this Section. In Section III-C, Level
on the day. Range of varies between 1 and 76 and bew- 2 (240 V) charging is also simulated.
teen 1 and 365 in this study. The following equation describes 2) NI Stochastic Method: This section describes the NI sto-
the battery discharge: chastic method using non-iterative and independent probability
density functions (pdf’s). Assume and denote the
SOC SOC (1) pdf’s of home arrival time and the length of daily traveled dis-
tances. These pdf’s are obtained from the home arrival and daily
where SOC and SOC denote the percentage of
traveled distance data points in the GPS records regardless of
SOC when arriving at home and when departing home respec-
the vehicles. A Monte Carlo simulation with 10 000 points is
tively; is the battery capacity and is 24 kWh in this study,
conducted. For each point, charging times are calculated as fol-
and is the overall electrical drive efficiency assumes 6.7
lows. Home arrival time is selected from and daily
km/kWh which takes into account both charge and discharge
traveled distance is selected from . Charging begins at
efficiencies. If (1) yields a negative value for SOC , this
. Charging end time is calculated as
value is set to zero. Tabulated number of days the battery does
not have enough charge is a measure of the PEV electrical range
and availability of public charging infrastructure requirements. (3)
ASHTARI et al.: PEV CHARGING PROFILE PREDICTION AND ANALYSIS BASED ON VEHICLE USAGE DATA 345

Note that in this stochastic method, and are as- TABLE II

sumed to be independent from each other which is not statisti- SIMULATED CHARGING SCENARIOS
cally accurate. In addition, assuming the vehicle is always fully
charged when leaving home which is also not accurate. Home
departure time and home departure SOC are not included during
the simulations to reflect the assumptions made when applying
the NI stochastic method. [14] is an example where the NI sto-
chastic method used in previous studies.
3) II Method: This section describes the iterative and inde-
pendent pdf II method used in the literature. For this method,
assume denotes the pdf of home departure time. Charging be implemented in Winnipeg with relatively low cost. Because
starts at like the NI method. The charging end time on day , of Winnipeg’s extreme cold winters, many public parking places
, is then calculated as are equipped with block heater chargers which can also be used
for charging PEV batteries. However, many of these outlets are
being cycled on and off to save power and would require to
(4)
be modified. H1W1S1 scenario includes charging at shopping
places as well as charging at home and wile at work. H2W1 uses
where is the home departure time in day picked Level 2 charging at home and Level 1 charging at work. H2W2
from and is the distance traveled in day picked applies Level 2 charging both at home and at work. H2W1Box is
from . Note that unlike the NI method, home departure time the same as H2W1 except that it excludes trips that travel more
is now included in the calculations. Another difference between than 150 km from Winnipeg.
the II method and the NI method is that the II method is iter-
ative, i.e., the order of days affects the simulations. However, IV. RESULTS
, and are still assumed to be independent pdf’s. We first present results for NI, II, and IC stochastic charging
The independence of the pdf’s is the similarity of the NI and II prediction methods for comparisons. Results depicting the ef-
methods. fect of charging infrastructure on utility load and on PEV elec-
4) Proposed IC Stochastic Method: This section describes tric range is then presented. These results are based on anal-
the iterative and conditional pdf IC stochastic method in- ysis methods described in Section III, the dataset reviewed in
troduced by this study. We assume that is the home Section II, and the algorithm depicted in Fig. 2.
departure time on day and is selected from . Traveled
distance in day , , is then chosen from pdf A. Results of PEV Charging Profile Prediction Methods
where is the pdf of the daily distance traveled Section III-B presented NI and II methods and the proposed
of the days where home departure time is within 5 min of IC method. The purpose of these methods is to provide pre-
. Home arrival time on day , is obtained from pdf dictions of PEV charging patterns for regions where vehicle
where is the pdf of the usage data is not available. This section compares these three
home arrival times of those days where home departure time is stochastic methods with respect to the more exact deterministic
within 30 minutes of and daily traveled distance is within method which uses actual driving and parking information of
5 km of used in the II method. Charging beginning will the vehicles infrom the database. The results are also analyzed
then be and charging end time is obtained from (4). Note to explain which method is most more accurate with respect to
that the charging end time is calculated from the same equation the sampled driving data.
for both II and IC stochastic methods. However, the values of Fig. 3 compares the power profiles produced by the three sto-
, , and are selected from conditional pdf’s in chastic methods with the deterministic method as obtained from
the proposed IC method and independent pdf’s are used in the simulations (Fig. 2). All three stochastic methods are able to re-
II method. There are no previous studies that have used this produce the trends when compared in the deterministic exact
method. profile. It can be observed that the new IC method produces the
closest results to the exact deterministic profile. However, the
C. Simulated Charging Infrastructure Scenarios ICnew IC method compares better with the produces the closest
Table II shows the various scenarios simulated in this study. results to the exact deterministic profile while the NI method has
Two charging rates, Level 1 and Level 2, and three charging lo- the worst prediction. Table III provides tabulates the numerical
cals previously extracted from the database [23]—home, work, values of the prediction errors of the three stochastic methods
and shopping—are included in the simulations. Utility–imposed as compared to the deterministic profile. Results in this table il-
charging control strategies are currently not included in this lustrate that the modifications in the stochastic modeling of the
study. Table II shows the different charging scenarios simulated PEV charging profile prediction methods, can result in 12% im-
in this study. provements in the accuracy of the results when compared to the
H1 method assumed that charging is done only at home using deterministic method which itself is an approximation of the ac-
a Level 1 charger. It is currently possible to implement this sce- tual load profile. In addition to improved accuracy, an important
nario in Winnipeg at low cost. H1W1 assumes that the work advantage of the IC method over the NI and II methods is its
place provides Level 1 charging as well. This scenario can also ability to more accurately estimate the peak load, an important
346 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 1, MARCH 2012

Fig. 4. Home departure time histogram.

Fig. 3. Comparing the PEV charging profile prediction of different stochastic

methods.

TABLE III
PREDICTION ERRORS OF THE STOCHASTIC METHODS

aspect for utilities. It can be seen from Fig. 3 that the NI and II
methods yield the highest errors at 5 P.M. when PEV charging
load peak. However, the IC method yields a reasonable accu-
racy at 5 P.M.. Considering that the peak load is an important Fig. 5. Change in daily traveled distance histogram versus departure time.
factor for utilities wanting to reduce infrastructure cost, the IC
method is therefore recommended. The sources of error will are that daily traveled distance correlates with the departure time.
now be quantified below. Therefore, it is not correct for a stochastic method to select num-
Fig. 4 shows the histogram of the home departure time ex- bers from independent distributions of daily traveled distance
tracted from the for the vehicles in the database using the math- and home departure time. Fig. 6 demonstrates the correlation
ematical simulation depicted in Fig. 2. A relatively high con- between the daily traveled distance and home arrival time dis-
siderable percentage of the vehicles leave home in the early in tributions. This figure demonstrates that these two variables are
the hours of the morning. This suggests that in some days, some indeed correlated and cannot be selected from independent dis-
vehicles might may not be fully charged when departing home. tributions. Using conditional pdf’s as in IC method incorporates
The NI method ignores this fact, i.e., it does not include home the correlation between home departure time, home arrival time,
departure time in the PEV charging profile calculations. Implic- and daily traveled distance. This approach results in 7.3% im-
itly, the NI method assumes that the vehicles are 100% charged provements in the accuracy of PEV charging profile prediction.
when leaving home. This is the main source of error in the NI
method compared to the II method. Because the II method is
iterative, it effectively simulates the SOC of the vehicle taking B. PEV Utility Load
into account the home departure time. This section provides an analysis on the effect of PEV
Figs. 5 and 6 demonstrate why the IC method is more accu- charging profiles on Manitoba load using the deterministic
rate compared to the II method. Fig. 5 shows the change in daily method. In order to calculate the PEV load, one needs to
traveled distance histogram versus departure time. Results show predict the PEV adoption rate. Such predictions are beyond the
ASHTARI et al.: PEV CHARGING PROFILE PREDICTION AND ANALYSIS BASED ON VEHICLE USAGE DATA 347

Fig. 8. Average load profile per person with 100% adoption of EV sedans in
Fig. 6. Effect of daily traveled distance on home arrival time distribution. winter.

Fig. 9. Average load profile per person with 100% adoption of EV sedans
under different charging scenarios in summer ( ; ;
).
Fig. 7. Average load profile per person with 100% adoption of EV sedans in
summer.
C. Charging Scenarios Results
scope of this study and thus we chose a different approach by The results in Figs. 9 and 10 are obtained using Level 1 home
calculating the PEV load impacts if all the passenger vehicles charging (H1). The results of other strategies such as Level 2
in Winnipeg were EV sedans. More accurate analysis which in- charging (H2), charging at work or shopping (H1W1S1) are pre-
cludes variable PEV adoption rates and other PEV powertrains sented in Figs. 9 and 10. The results show that if charging is re-
is ongoing. However, an upper limit analysis in the current stricted to home, EV load has a significant peak around 6 P.M.
study provides insight on the effect of large adoption of PEVs. Adding charging at work substantailly smoothens the profile.
According to Statistics Canada, there are currently 643 580 Table IV shows the percentage of the trips that PEVs go out
registered passenger vehicles in Manitoba. Manitoba has a pop- of charge under different charging scenarios. As expected, H1
ulation of 1 213 815. Using these numbers, one can calculate the has the highest percentage of the trips without sufficient bat-
average load profile per person in Manitoba with 100% adoption tery charge. However, this number, 3.69% is still a consider-
of EV sedans. Figs. 7 and 8 shows such results using different ably small number. Using Level 1 chargers at work reduces the
PEV charging profile prediction methods. This is chosen as an run–out–of–charge events to 1.95% and adding Level 1 charger
upper limit of the impacts of PEV load, excludes ac loads and at shopping places further reduces that number to 1.18%. Using
provides information on public infrastructure requirements. a Level 2 charger at home with a Level 1 charger at home and
The results show that PEV load can be potentially significant without charging while shopping, H2W1 is more effective than
to Winnipeg utility system. In particular, PEV charging profiles H1W1S1 scenario. Adding a Level 2 charger at work improves
results in sharper peaks in Winnipeg load. the number by only 0.21%. H2W1Box scenario which removes
348 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 1, MARCH 2012

Fig. 11. Effect of battery size and charging scenario on PEV elec-
Fig. 10. Average load profile per person with 100% adoption of EV sedans trical range reliability ( ; ; ;
under different charging scenarios in winter ( ; ; ).
).
D. Sensitivity Analysis
TABLE IV
PERCENTAGE OF THE TRIPS THAT PEVS GO OUT OF CHARGE UNDER
To measure the stability of the simulation results, it is desir-
DIFFERENT CHARGING SCENARIOS able to show some sensitivity analysis. The sensitivity analysis
can clarify how the electricity consumption changes with dif-
ferent input values and show the limitations of the simulation re-
sults. We used these tests for sensitivity analysis: increasing/de-
creasing the driving speed by 10% and delaying the arrival and
departure time by 1 hour. When driving speed increases by 10%,
peak power increases by 5.8% but the peak location does not
change. When both departure and arrival times change by 1 h,
the peak magnitude does not change. However, its location is
shifted by 1 h.
the trips that are outside a 150 km circle from the center of Win-
nipeg results in only 0.58% of the trips with no sufficient charge. E. Economic Considerations
However, H2W1Box is only 0.05% more reliable than H2W2, For electric utilities, it is of paramount importance to have an
which suggests that excluding out of town trips does not guar- accurate estimation of the additional load resulting from large
antee 100% reliability of the electrical range for EV sedan under scale PEV adoption and the investments required to accustom
the described charging scenarios. the electrical infrastructure with new load conditions. The re-
Results in Table IV suggests that other factors such as battery quired investment may include potential new generation sta-
size will have a greater impact on the reliability of the PEV tions, transmission, distribution systems, and charging infra-
electrical range. In order to understand the effect of battery size structure. For example, if the PEV peak load is significant, util-
on PEV electrical range reliability, simulations were conducted ities may consider investments in smart charging technologies
assuming different battery sizes for an EV sedan. Fig. 11 shows or changes in electricity pricing system. Accurate prediction of
the result of that simulation. For a battery size of 32 kWh, all the PEV load is also necessary for proper policy making decisions.
charging scenarios except for H1 yield relatively similar results. The proposed IC method is 12% more accurate than existing
The effect of charging scenarios is more dramatic for smaller NI method. If a utility uses NI method instead of II method, it
battery sizes. For example, a 16–kWh EV charged under H1 can mistakenly decide to build new generation stations which
scenario has less range than an 8–kWh EV charged under H2W1 involve significant investments. Also, decisions regarding up-
scenario. grades in distribution systems are going to need accurate PEV
Ideally, if collaboration exists between vehicle manufac- load prediction methods.
turers, utilities, governments, and customers, data presented On the issue of charging infrastructure, different parties can
in Fig. 11 can minimize the overall costs of PEV technology be involved such as car manufacturers, utilities, and govern-
and maximize its benefits. For example, such collaboration can ments. The question that was risen and was partly addressed
lead to a correct policy regarding the question of whether to in this paper was to find the most cost effective way to increase
invest on bigger battery sizes or on charging infrastructure and PEV adoption. More specifically, it is of interest to know which
make sure that new electrical loads are always match with new scenario makes the most sense: building charging infrastructure
renewable generation to address energy drivers simultanously with faster rates and in different places or making larger battery
rather than requirements of specific driving cycles. packs. Answer to such a question needs access to real–world
ASHTARI et al.: PEV CHARGING PROFILE PREDICTION AND ANALYSIS BASED ON VEHICLE USAGE DATA 349

high resolution driving data which is made available in this REFERENCES

study. Moreover, the goal of such optimization should be the
service to the public, i.e., increased reliability of PEVs and less
[1] “Environmental assessment of plug-in hybrid electric vehicles
gasoline consumption. A solution that is merely optimized for volume 1: Nationwide greenhouse gas emissions,” EPRI, Tech. Rep.
utilities or car manufacturers will not necessarily be optimum TR–1015325, 2007.
for the costumers or the environment. [2] “Environmental assessment of plug-in hybrid electric vehicles volume
2: United States air quality analysis based on Aeo-2006 assumptions
In this section, a preliminary approach for optimizing PEV for 2030,” EPRI, Tech. Rep. TR–1015326, 2007.
integration is proposed. We model the cost of PEV development [3] A. Ashtari, E. Bibeau, and S. Shahidinejad, “Using large driving
as follows: records and a stochastic approach for real-world driving cycle
construction: Winnipeg driving cycle,” Transp. Sci., submitted for
publication.
[4] A. Ashtari, E. Bibeau, and S. Shahidinejad, WPG01 driving cycle 2010
[Online]. Available: http://dx.doi.org/10.5203/ds_bib_2
(5) [5] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of charging
plug-in hybrid electric vehicles on a residential distribution grid,” IEEE
New infrastructure cost is a function of investments in larger Trans. Power Syst., vol. 25, pp. 371–380, 2010.
[6] P. Denholm and W. Short, “An evaluation of utility system impacts
battery sizes and new charging infrastructure and benefits of optimally dispatched plug-in hybrid electricvehicles,”
National Renewable Energy Lab (NREL), Tech. Rep., 2006.
[7] M. Duvall, “Advanced batteries for electric drive vehicles. A tech-
nology and cost-effectiveness assessment for battery electric vehicles,
(6) power assist hybrid electric vehicles, and plug-in hybrid electric vehi-
cles,” EPRI, Palo Alto, CA, Final Rep. 1009299, 2004.
[8] L. P. Fernandez, T. G. S. Roman, R. Cossent, C. M. Domingo, and P.
Function will vary in different regions. Unreliability penalty Frias, “Assessment of the impact of plug-in electric vehicles on distri-
is proportional to the number of times that PEV battery is out of bution networks,” IEEE Trans. Power Syst., vol. 26, no. 1, pp. 206–213,
charge, which is shown in Fig. 11. Therefore, we can write Feb. 2010.
[9] A. Hajimiragha, C. Caizares, M. Fowler, and A. Elkamel, “Optimal
transition to plug-in hybrid electric vehicles in Ontario, Canada, con-
sidering the electricity-grid limitations,” IEEE Trans. Ind. Electron.,
vol. 57, no. 2, pp. 690–701, Feb. 2010.
(7) [10] P. S. Hu and T. R. Reuscher, “Summary of travel trends 2001 national
household travel survey,” U.S. Department of Energy, Tech. Rep.,
Combining (5)–(7) results in 2004.
[11] A. Ipakchi and F. Albuyeh, “The impact of charging plug–in hybrid
electric vehicles on a residential distribution grid,” IEEE Power Energy
Mag., pp. 52–62, 2009.
[12] L. Kelly, A. Rowe, and P. Wild, “Analyzing the impacts of plug-in elec-
(8) tric vehicles on distribution networks in British Columbia,” in Proc.
IEEE Electr. Power Energy Conf. (EPEC), Oct. 2009, pp. 1–6.
where . [13] A. Maitra, K. S. Kook, J. Taylor, and A. Giumento, “Grid impacts of
plug-in electric vehicles on Hydro Quebec’s distribution system,” in
As an example of the application of this cost model, obtaining Proc. Transm. Distrib. Conf. Expo., Apr. 2010, pp. 1–7.
function can help a municipality to decide whether to invest [14] A. Maitra, J. Taylor, D. Brooks, M. Alexander, and M. Duvall, “Inte-
in charging infrastructure or to subsidize PEV batteries. The grating plug-in-electric vehicles with the distribution system,” in Proc.
20th Int. Conf. Exhib. Electr. Distrib., Part 1, Jun. 2009, pp. 1–5.
analysis given in this study is a preliminary effort to calculate . [15] M. K. Meyers, K. Schneider, and R. Pratt, “Impacts assessment of
Function will vary based on the application and the region. plug-in hybrid vehicles on electric utilities and regional US power grids
Part 1: Technical analysis,” Pacific Northwest National Laboratory,
Tech. Rep., 2007.
V. CONCLUSION [16] P. Mohseni and R. Stevie, “Electric vehicles: Holy grail or fool’s gold,”
in Proc. IEEE Power Energy Soc. Gen. Meet., Jul. 2010, pp. 1–5.
We recorded vehicle usage data for 76 vehicles in a one–year [17] K. Morrow, D. Karner, and J. Francfort, “Plug–in hybrid electric
vehicle charging infrastructure review,” U.S. Department of Energy
period in the city of Winnipeg, Canada, and used this data to pre- Vehicle Technologies Program—Advanced Vehicle Testing Activity,
dict PEV charging profiles and electrical range reliability. The Tech. Rep., 2008.
study shows that proper stochastic modeling, i.e., using an iter- [18] Leaf Owner’s Manual NissanUSA, 2011.
[19] X. Oua, X. Zhanga, and S. Changa, “Scenario analysis on alternative
ative method with conditional pdf’s can improve the accuracy fuel/vehicle for china’s future road transport: Life-cycle energy de-
of the predictions by 12% compared to existing methods, which mand and GHG emissions,” Energy Policy, vol. 38, pp. 3943–3956,
are commonly noniterative and use independent pdf’s. 2010.
[20] M. J. Scott, M. K. Meyers, D. B. Elliott, and W. M. Warwick, “Impacts
Combinations of charging places simulate different charging assessment of plug-in hybrid vehicles on electric utilities and regional
profiles; home, work, and shopping and charging rates; Level 1 US power grids Part 2: Economic assessment,” Pacific Northwest Na-
and Level 2. The effect of charging scenario on electrical range tional Laboratory, Tech. Rep., 2007.
[21] S. Shahidinejad, E. Bibeau, and S. Filizadeh, “Development of a new
reliability is more significant where battery sizes are smaller. duty cycle for plug–in electric vehicles in a typical north american
urban setting,” IEEE Trans. Veh. Technol., vol. 59, pp. 3710–3719,
2010.
ACKNOWLEDGMENT [22] R. Sioshansi and P. Denholm, “Emissions impacts and benefits of
plug-in hybrid electric vehicles and vehicle-to-grid services,” Environ.
The authors would like to thank Prof. D. Blair and Mr. R. Sci. Technol., vol. 43, pp. 1199–2004, 2009.
Smith from University of Winnipeg for collecting the data. Dis- [23] R. Smith, S. Shahidinejad, D. Blair, and E. Bibeau, “Characterization of
urban commuter driving profiles to optimize battery size in light–duty
cussions with Emerging Energy Systems at Manitoba Hydro are plug–in electric vehicles,” Transp. Res. Part D, vol. 16, pp. 218–224,
also acknowledged. 2011.
350 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 1, MARCH 2012

[24] J. Taylor, A. Maitra, M. Alexander, D. Brooks, and M. Duvall, “Eval- Soheil Shahidinejad received the B.Sc., M.Sc., and
uation of the impact of plug-in electric vehicle loading on distribution Ph.D. (with honors) degrees from Sharif University
system operations,” in Proc. Power Energy Soc. Gen. Meet., Calgary, of Technology, Tehran, Iran, in 1993, 1996 and 2002,
AB, Canada, 2009. respectively, all in mechanical engineering.
[25] E. Ungar and K. Fell, “Plug in, turn on, and load up,” IEEE Power He is currently a Research Associate with the
Energy Mag., vol. 8, no. 3, pp. 30–35, May 2010. Department of Mechanical and Manufacturing
Engineering, University of Manitoba, Winnipeg,
Ali Ashtari earned the B.Sc. and M.Sc. degrees in MB, Canada. His research deals with the study and
electrical engineering from the Sharif University of analysis of energy-efficient systems and renewable
Technology, Tehran, Iran, in 2001 and 2004, respec- energies.
tively, and the Ph.D. degree in electrical engineering Dr. Shahidinejad is a member of the Society of Au-
from the University of Manitoba, Winnipeg, MB, tomotive Engineers.
Canada, in 2010.
He currently holds an industrial postdoctoral
fellowship at the University of Manitoba, sponsored
by both Mathematics of Information Technology and Tom Molinski received the B.Sc. degree in elec-
Complex Systems (MITACS) and Manitoba Hydro. trical engineering and the M.S. degree in electrical
His current research focuses on the application of engineering degree from the University of Manitoba,
signal processing and analysis techniques in alternative energy systems. Winnipeg, MB, Canada.
He has worked for Manitoba Hydro, Winnipeg,
for the past 36 years and is currently the Section
Head of Emerging Energy Systems in the Power
Eric Bibeau received the B.Sc. degree from McGill Planning Division. He is responsible for research,
University, Montreal, QC, Canada, in 1986 and the planning, concept development, and recommending
M.Sc. and Ph.D. degrees from the University of emerging energy technology projects (like wind,
British Columbia, Vancouver, BC, Canada, in 1988 small hydro, bioenergy, energy storage, and various
and 1993, respectively. solar technologies) suitable for Manitoba Hydro to implement now or in the
He is a Mechanical Engineer and is currently future. He works with government, academia, various research groups, and
an Associate Professor with the University of several learned societies to determine what are the available emerging energy
Manitoba, Winnipeg, MB, Canada, and holds the technologies that Manitoba Hydro can use now and in the future. He has
Natural Science and Engineering Research Council collaborated on research projects with several professors at the University of
of Canada/Manitoba Hydro Chair in Alternative Manitoba for his entire career (36 years in total and 8 years with Dr. Bibeau)
Energy. He specializes in developing power systems and has given many guest lectures there and at other universities.
using biomass feedstocks, industrial waste heat, and kinetic energy from rivers
for distributed-energy applications. He is also involved in developing blade
deicing mitigation strategies for wind turbines and expertise in electric vehicles
and district energy systems as a means to increase the renewable energy ratio in
Canada to simultaneously address climate change and peak oil energy drivers.
Dr. Bibeau is a Registered Professional Engineer in the Province of Manitoba.