You are on page 1of 8

1

Electric Vehicle Charging: Transformer Impacts and


Smart, Decentralized Solutions
Alexander D. Hilshey, Student Member, IEEE, Pooya Rezaei, Student Member, IEEE, Paul D. H. Hines, Member,
IEEE, Jeff Frolik, Senior Member, IEEE

AbstractThis paper compares distribution transformer aging small as 10% could induce additional distribution transformer
impacts resulting from plug-in electric vehicles charging under overloading beyond planned overloading as defined in [14].
AC Level 1 versus AC Level 2 charging conditions. Additionally, Reference [15] estimates that distribution infrastructure costs
we propose an algorithm for PEV smart charging and evaluate
its effectiveness on transformer aging. We use a Monte Carlo could increase by 19% and energy losses could increase by
simulation of a 25kVA distribution transformer, with ambient 40% with substantial (60%) PEV deployment.
temperature data from Burlington, VT and Phoenix, AZ, to Recent publications suggest methods for coordinating PEV
calculate transformer aging under both uncoordinated and smart smart charging for distribution system efficiency improvement.
charging conditions. The results indicate more substantial aging For instance, the results of [9] and [16] suggest that the
as a result of AC Level 2 charging compared to AC Level 1.
Smart charging can significantly mitigate these effects. We also impacts of PEV charging on components of residential feeders
present a more decentralized approach to smart charging and could be minimal if intelligent charging schemes are used.
compare two distributed automaton-based charge management Reference [17] and [18] solve an optimization problem con-
strategies, which both prevent the transformer from becoming sidering all electric vehicles serviced by the grid to minimize
overloaded. These methods give vehicle owners the ability to power losses. However, these solutions require individual PEV
select among charging priorities in an environment in which the
vehicles manage their charging autonomously. battery charge status data and driver departure time data to be
sent to a central location. This approach may not be feasible,
Index TermsPlug-in electric vehicles, transformer aging, as it increases existing concerns regarding data privacy in a
smart charging
Smart Grid environment. Also, as PEV adoption increases, the
problem may not be solvable in real time. Two strategies are
I. I NTRODUCTION proposed in [13] to reduce the peak power of the grid. The first
With several mass-market plug-in hybrid and battery electric strategy limits the transformer load by a pre-defined value; the
vehicles (collectively plug-in electric vehicles, PEVs) currently second, a household load control strategy, causes non-critical
for sale or scheduled to go on sale in the 2012 and 2013 loads to shed when the PEV is charging. The decentralized
model years, there is a growing need to assess the impact demand side management problem is solved in [19], and two
that PEV charging load will have on electricity distribution solutions are proposed for charge coordination. Our previous
infrastructure. Substantial research exists regarding regional publication shows the potential benefits of smart charging
and national impact of PEVs on gasoline consumption [1], in dramatically reducing accelerated transformer aging and
[2], [3], power-plant emissions [1], [2], [4], [5], electricity in reducing transformer aging uncertainty caused by unique
costs [6], [7], [8], transmission adequacy [9], and generating PEV charging profiles derived from diverse PEV driver travel
supply adequacy [8], [10]. One area of particular interest behavior [20].
is quantifying the impact of PEVs on the medium and low Estimating the impact of PEV charging, as well as evaluat-
voltage distribution infrastructure; PEV charging loads will ing the utility of smart charging strategies, on components of
likely impact local distribution infrastructure well before the the distribution infrastructure requires not only good physical
impacts on transmission or generation infrastructure is sig- models of circuit components but also good estimates of
nificant. If these impacts are severe, as is suggested in [11] the electric power demand due to PEV charging. Early PEV
and [12], distribution utilities may need to make significant research assumed very simple methods used to model vehicle
infrastructure investments in high-adoption locations. Accurate behavior and associated charging profiles [2], [21], [8], [22].
information describing PEV impacts is essential to ensure that Recent work, including that by the authors, has proposed more
these investments are made in an effective manner. empirical approaches [23], [24], [25], [26], [20], [27].
Previous studies show evidence of PEV impact on the One objective of this paper, which is an extension of [27],
residential distribution infrastructure. For example, reference [20], is to compare distribution transformer impacts of slow
[11] uses a time-series model of transformer aging and argues (AC Level 1) versus fast (AC Level 2) PEV charging under
that PEV charging could decrease transformer life by 93%. both uncoordinated and smart charging conditions (Section II).
Another study [13] suggests that a PEV penetration level as Having further explored the utility of smart charging, we then
discuss decentralized methods for coordinated PEV charging
This work is supported in part by the Transportation Research Center at the strategies (Section III). Two automaton-based strategies are
University of Vermont, through a grant from the U.S. Dept. of Transportation.
The authors are with the School of Engineering, University of Vermont, proposed to coordinate PEV charging, both of them limiting
Burlington, VT (e-mail: {ahilshey,pooya.rezaei,paul.hines,jfrolik}@uvm.edu). the power of the transformer to avoid power peaks. The first

978-1-4673-2729-9/12/$31.00 2012 IEEE


2

strategy (Method 1) uses a first-come, first serve approach, We use data from the 2009 National Household Travel
and the second (Method 2) uses a probabilistic approach. The Survey (NHTS) database [30] to build a stochastic model
probabilistic method (Method 2), which extends the approach of PEV charging loads (Lv ). The goal of the PEV charging
used in [28], is a result of the analogy between power distribu- model is to estimate the additional time-varying load that
tion system and shared channel access in wireless communi- would result from nv vehicles charging at a certain point in a
cation systems, including limited communication bandwidth, power grid. For each one-year run of our model we randomly
for which random characteristics of demand exist. This charge select nv vehicles from among the vehicles in the Vermont
coordination strategy is a modified version of a technique subset of the NHTS data that both start and end the day at
proposed for random access communication channels in [29]. home [30]. Each vehicle is assumed to have the charging and
Section IV details our conclusions. efficiency characteristics of the Chevrolet Volt [32], a series
hybrid vehicle which operates in straight charge depleting
mode until the limit of its all-electric range (AER) and then
II. I MPACT OF C HARGING ON T RANSFORMER AGING
switches to charge sustaining mode. Additionally, we assume
This section describes our method for estimating the impact that all PEVs charge exclusively from a home charging station.
of PEV charging on distribution transformers (Sub-Sections Without a smart charging system, we assume charging begins
II-A & II-B) followed by a description of our smart charging immediately upon the vehicles arrival at home and continues
algorithm (Sub-Section II-C). We then present the results of until either the battery reaches full capacity or the vehicle
a full year model which investigates impacts on distribution begins a new trip. We examine two PEV charging levels, as
transformers from slow (AC Level 1) and fast (AC Level established in [33]: 1) AC Level 1 (1.4 kW); and 2) AC Level
2) PEV charging using both uncoordinated charging and 2 (7 kW). We assume 85% charge efficiency as reported in
the smart charging algorithm (Sub-Section II-D). Finally, we [34]. Our model simulates a 10kWh useable battery capacity
conclude Section II with a discussion of results (Sub-Section and 3.8 mi/kWh electric drive efficiency, to match that of the
II-E). Chevrolet Volt [32].
We organize the NHTS travel data into two dependent
sub-profiles, which we term the demand profile D and the
A. Modeling transformer loads given demand for PEV charg- charging availability profile A. For time period t the charging
ing availability of vehicle v is a binary value (Av (t)) indicating
In this paper, residential load profiles are comprised of two whether vehicle v is parked at its home charging station. The
components: residential baseline load (Lh ) and load from PEV demand profile for vehicle v at time t (Dv (t)) indicates the
charging (Lv ). As we are primarily focused on the effect of amount of energy required to fill vehicle vs battery at a
variations in vehicle charging patterns, we assume that each particular point in time. The time required to fully charge the
home connected to a distribution transformer has identical, battery at charge rate Pv is:
deterministic baseline load. However, in order to study travel
pattern variation, we sample from empirical travel data [30] Dv (t)
Tv (t) = (1)
to develop a Monte-Carlo model of the PEV portion of the Pv
residential load profile.
As in our past work, we use the baseline load profile in [31] If the time step size is t (in hours) the load (in kW) from
for Lh , which is shown for a 25kVA transformer servicing 12 vehicle v during time interval [t, t + t) is:
homes in Fig. 1.
Av (t)Pv , if Tv (t) t
Dv (t)
Lv (t, t) = A (t) t if 0 < Tv (t) < t (2)
v
0, if Tv (t) 0

For the purposes of this model, we randomly select a weekday


and a weekend driving profile for each v, and reproduce the
two Lv trajectories that form a one-year charging pattern for
each vehicle.
The total load on the transformer is the combination of the
nh residential loads where h indicates the quantity of homes
and loading from nv randomly selected PEVs:

nv

L(t) = nh Lh (t) + Lv (t) (3)
v=1

Figure 1 illustrates the function of the model by showing the


Figure 1. Daily baseline load profile with and without PEVs for a 25kVA additional load due to 6 PEVs added to the load for 12 homes.
overhead distribution transformer servicing 12 homes and 6 randomly selected PEV load is shown at AC Level 1 and AC Level 2 charging
PEVs charging at AC Level I and AC Level 2. Both charging levels correspond rates.
to the same sample set of PEV charging behavior.
3

B. Modeling Distribution Transformer Aging minimally inconveniencing the PEV owner, while mitigating
The model for estimating distribution transformer aging the negative impacts of high loads on the electricity infrastruc-
simulates the thermal performance of an overhead distribution ture. Our algorithm seeks to mitigate the accelerated aging
transformer, installed in a location with a known trajectory of of a distribution transformer. The smart charging algorithm
ambient temperatures (TA (t)), and serving a set of nh homes proposed here continuously (or discretely at time specific
and nv electric vehicles. The ratio of PEVs per home (nv /nh ) check points) determines the number of PEVs connected to
can be chosen to reflect likely PEV adoption rates in particular a specific overhead distribution transformer that may charge
neighborhoods. The output of the model is an estimate of the at a given time without pushing the distribution transformer
accelerated aging of the transformer insulation material, which into sustained, rapid accelerated aging. We assume that smart
results from the thermal properties of the transformer, the time- meters (AMI - Advanced Metering Infrastructure) are installed
varying transformer load (L(t)) and ambient temperatures. at each home, which would allow a charge management device
The primary input variable for the model is L(t) from at the transformer to monitor instantaneous loads and send
(3). The second input variable, TA (t), corresponds to the a signal to vehicles connected to the transformer to forgo
geographic location to be modeled and can span any desired charging for a specified time period.
time period. We produced one-minute resolution temperature Our smart charging algorithm requires two inputs: the
data using a cubic spline from 15 minute temperature data transformer aging status, comprised of FAA (t) and FEQA (t),
from the National Climatic Data Center1 . Finally, the third which are derived from the aging calculations in II-B, and
input defines the rated specifications and parameters of the the quantity of PEVs requesting charge, qr (t). The algorithm
transformer to be modeled, which can be obtained from the yields one output: the quantity of PEVs that may charge at time
transformer manufacturer. With these inputs, the transformer t: q(t) qr (t). When implemented in a charge management
thermal model estimates transformer temperatures, most no- device associated with a transformer, the algorithm operates
tably the transformer hottest spot temperature, THS (t). Our in two steps. Step one determines q(t), qr (t) and the aging
transformer thermal model estimates THS (t) using the Annex factors. The second step dispatches a signal to smart meters,
G method of IEEE C57.91-1995 [14]. Our previous work [20] which subsequently signal each vehicle to either continue or
shows details of our implementation of the Annex G method. discontinue charging. Step two is performed by random allo-
Excessively high hottest spot temperatures damage a trans- cation, which has the advantages of not requiring the exchange
formers insulation through the destructive process of pyrolysis of information pertaining to battery level and avoiding the need
[35]. Therefore, the calculated THS (t) are fed into a damage to decide which PEV deserves charging precedence.
function [14] that estimates the instantaneous accelerated To determine the modeled transformer aging status, we
aging of the transformer (FAA (t)), which can be integrated assume that smart meters report instantaneous household load
to compute the total transformer thermal aging over a time to the transformer, as well as the number of vehicles available
horizon, FEQA (4): for charge management, qr (t). The aggregated L(t) and a
  measured value for TA (t) are fed into the transformer thermal
 model (Section II-B), which yields FAA (t) and FEQA (t)
1 t 15,000 15,000
THS,R +273 THS ( )+273
FEQA (t) = e d (4) averaged over a period of time. Numerical experimentation
T tT
with our model indicates that an FEQA averaged over the
where T is the length of time over which the average is previous 12 hours produces good results.
taken. FAA and FEQA are the primary outputs of the model After calculating FAA (t) and FEQA (t), the algorithm com-
for estimating distribution transformer aging and are also the pares the modeled transformer aging status against four ag-
required inputs for the smart charging algorithm described in ing thresholds (HEQA , Hmin , Hmed , and Hmax ) to determine
Section II-C. whether q(t) should be increased, decreased, or held constant
in the next time period. Equation (6) is used to choose the
C. Smart Charging Algorithm for Transformer Aging change in q(t) from the previous time period:
If a distribution transformer is overloaded due to PEV q(t) = q(t t) + q(t) (5)

charging, it can either be replaced with a larger unit, or +1, if (FAA < Hmin ) or


the PEV charging can be managed to forestall the need for
(FAA < Hmed & FEQA < HEQA )
a transformer upgrade. In [20] we proposed a smart charg- q(t) = 1, if FAA > Hmed (6)
ing algorithm which successfully demonstrated mitigation of


2, if FAA > H max
transformer accelerated aging and a dramatic reduction in
0 otherwise
uncertainty associated with diverse PEV charging profiles de-
rived from unique PEV driver travel behavior. In our previous Table 1 provides suggested aging threshold values, as deter-
work, we only considered AC Level 1 charging. This section mined from our extensive numerical experimentation. Unless
provides summary of our previously proposed smart charging Table I
algorithm. A GING T HRESHOLDS USED FOR S MART C HARGING A LGORITHM
COMPARISON AND DECISION PROCESS
A successful smart charging algorithm should ensure that
HEQA Hmin Hmed Hmax
all PEVs receive as close to a full charge as possible, thus 2.5 3.5 4 4.75
1 http://www.ncdc.noaa.gov q(t) is greater than qr (t), in which case all requesting PEVs
4

may charge, the smart charging algorithm randomly chooses


q(t) vehicles from the set of vehicles that are currently
requesting charge, qr (t), and signals the smart meters to allow
or forgo charging to their respective PEVs.
Figure 2 highlights the differences between AC Level 1
uncontrolled and smart charging using a transformer operating
during a high-temperature, 48 hour period for Phoenix, AZ
and serving six PEVs. The smart charging algorithm delays
charging for several vehicles, away from the hottest hours or
heaviest load periods of the day.

Figure 3. Comparison of AC Level 1 and AC Level 2 charging results data of


Monte Carlo simulation (500 iterations) for Burlington, VT & Phoenix, AZ
annual factor of equivalent aging (FEQA ) corresponding to uncoordinated
PEV charging, smart PEV charging, and no PEVs. The graph displays average
and 10th & 90th percentile FEQA values for each location and charging
scenario combination. A factor of equivalent aging of 1 represents aging at
rated loading and rated ambient temperature.

equivalent aging calculated over the course of one calendar


year as the metric for comparison. When serving 12 homes,
the baseline load, prior to PEV charging, approaches a daily
average load of approximately 25kVA, or near rated capacity.
We model a transformer serving nv = 6 PEVs, each with
unique travel patterns resulting in a unique load profile for
Figure 2. Comparison of uncontrolled PEV charging vs. PEV charging with each model run.
smart charging algorithm. Data corresponds to a 25 kVA transformer serving Figure 3 shows the simulation results for each of the 5 cases
12 homes and 6 PEVs during a 48 hour period in Phoenix, AZ in June 2010.
and for each geographic location. For cases of uncoordinated
charging, there is a noticeable difference between AC Level 1
D. Full Year Model Results and AC Level 2 charging. The aggregate FEQA for Burlington
As concluded in our previous work [27], the annual ambient increases from 0.39 to 1.08 with the introduction of AC Level
temperature of the overhead distribution transformer con- 2 charging; Phoenix increases from 3.24 to 8.15. This result
tributes significantly to the PEV charging impacts. Therefore demonstrates the increased impact from AC Level 2 charging
in this study we examine PEV charging impacts in two compared against AC Level 1 charging.
climatically distinct U.S. cities: Burlington, VT and Phoenix, With the introduction of smart charging, the Burlington
AZ. Our previous work [20] introduced and demonstrated our FEQA reduces to 0.34 for AC Level 1 and to 0.55 for AC
initial smart charging algorithm but only considered AC Level Level 2; The Phoenix FEQA reduces to 1.64 for AC Level
1 PEV charging. In this study, we extend the previous work 1 and to 2.056 for AC Level 2. For all cases, a two-sample
to include a comparison of AC Level 1 and AC Level 2 Kolmogorov-Smirnov test shows statistical significance of the
PEV charging under both uncoordinated and smart charging reduction in aging from smart charging (p < 103 for all
conditions. For baseline purposes, we compare these results to cases). This result demonstrates the utility of smart charging,
a case with no PEV load. especially in the warmer climate and in situations were AC
We use a Monte Carlo simulation to aggregate annual Level 2 is used.
transformer loss-of-life data for 500 distinct sets of travel Examination of the uncoordinated charging 10th and 90th
patterns, under the following five test conditions for each percentiles in Fig. 3 shows substantial variance in transformer
geographic location: 1) no PEV charging; 2) AC Level 1, aging among the model runs. Specifically, uncoordinated
uncoordinated PEV charging; 3) AC Level 1, smart PEV charging in Burlington possesses variances of 0.040 and 2.02
charging; 4) AC Level 2, uncoordinated PEV charging; and for AC Levels 1 and 2 respectively; in Phoenix the variances
5) AC Level 2, Smart PEV charging. We use the factor of were 2.5 and 93.23 for AC Levels 1 and 2 respectively. This
5

result highlights the system stress that may be caused by aging, but also reduces uncertainty in transformer life, in the
AC Level 2 charging as well as the combination of high face of uncertain PEV travel behavior.
temperatures (in the Phoenix case) and highly unpredictable
PEV driver travel behavior. With the introduction of smart
III. D ECENTRALIZED S MART C HARGING A LGORITHM
charging, the Burlington variance decrease to 0.013 and 0.041
for AC Levels 1 and 2 respectively, and in Phoenix the In Section II-C, the smart charging algorithm determines
variances decreased to 0.043 and 0.066 for AC Levels 1 and 2 the maximum number of possible PEVs that may be charged
respectively. Decreases are more dramatic in AC Level 2 cases simultaneously during each discrete time period, and then
and especially in the warmer Phoenix climate. As evident in randomly allocates which PEVs receive charge. This section
Fig. 3, smart charging reduces the FEQA variance, demon- describes two alternative approaches, which are distinct from
strating the utility of smart charging for reducing uncertainty the method presented in Section II-C in that they use discrete
in transformer life expectancy. allocations of PEV charging and allow for decentralized im-
It is also important to note that an algorithm that reduces plementation.
aging but does not provide adequate charging to the PEV is Method 1 uses first-come, first served (FCFS) PEV charge
not desirable. To ensure that our simulated smart charging allocation. We assume that each PEV contains a charge-
resulted in adequate PEV charging, we measured the number management automaton which sends charge requests to the
of cases in which vehicles fully charged before beginning their transformer at discrete time intervals. If there is sufficient
next outing, after having been parked at home for an extended capacity to support charging, the PEV will charge during that
period of time. Specifically, we define a successful charge epoch (discrete interval of time) on a first-come, first served
to be a period in which the PEV was charged to at least basis. Otherwise, the request will be denied.
95% of capacity after being located at home long enough to Method 2 also assumes a network of simple charge-
have received a full charge with no charge management. In management automatons, one per PEV. However, in this
evaluating the success of the smart charging algorithm, we method each PEV sends a charge request with a certain prob-
found that in the case of AC Level 1 charging that 99.1% and ability, which is determined autonomously. Method 2 shows
98.7% of extended home stays in Burlington and Phoenix, re- advantage over Method 1 by considering owner determined
spectively, resulted in successful charges, using this definition; urgency in PEV charging, which assumes a pricing structure
the numbers were reasonable for AC Level 2 charging (98.7% with increased cost for urgent charging. Section III-A explains
in Burlington and 95.7% in Phoenix). These figures indicate these algorithms further and simulation results are shown in
that the proposed smart charging algorithm would have almost Section III-B.
no noticeable effect on most vehicle owners.
A. Algorithm description
E. Discussion of full year model results Methods 1 and 2 assume PEV charging conducted over
We find that for both cases, AC Levels 1 and 2, the warmer many discrete time intervals (e.g. 5 minutes). The charge
climate of Phoenix, AZ results in notably more transformer given during each interval is referred to as a charge packet
aging from PEV charging, relative to the cooler climate in deference to communication systems data packets [29],
of Burlington, VT. Introduction of AC Level 2 drastically upon which this random access technique finds inspiration.
increases aging compared to AC Level 1 for both locations; In Method 1, charge packets are given based on a first-
the results indicate that a moderate amount of PEV charging come, first-served (FCFS) approach. All requests are treated
may not substantially decrease the life of transformers located equally and the PEVs regularly compete for charging capacity.
in cooler climates until AC Level 2 charging is introduced. Additionally, all PEV automatons send requests at every epoch
For the Phoenix, AZ cases, the average transformer aging and the requests are accepted until the overload limit of the
approximately triples against the test case with no PEV transformer is reached.
charging at AC Level 1 and is nearly eight times greater at Figure 4 shows a simple probabilistic 2-state automaton
AC Level 2. These results highlight the need to use location- state diagram to illustrate Method 2. The PEVs adjust their
specific ambient temperature data when evaluating the impact state autonomously using this method based only on the
of PEV charging on transformers. Additionally, because of feedback provided by the charge manager. This approach
the variability in driver behavior, PEV charging introduces enables PEV owners to indicate charge requirement urgency.
enormous variance in transformer aging, particularly for the It is assumed that the PEV owner may choose to receive
hot climate. more charge priority if the need is immediate. Pricing structure
Smart charging has a greater effect on the hot climate and policy may establish that the PEV owner must pay a higher
in both climates when AC Level 2 is used. In addition to this price rate for increased charge priority during the peak load
average reduction in aging effect, the introduction of charge duration. The cost allocation is not considered in this work.
management removes much of the variability in aging. This When the PEV arrives at the home charging station, the
reduction comes at the cost of leaving vehicles with less than automaton sends charge requests in every epoch (for example,
95% of a full charge in only about 1% of cases except the 5 minutes in the case of this study). If any request is denied
Phoenix AC Level 2 case, in which the percentage is closer by the transformer, as a result of overloading (congestion =
to 4%. Thus, smart charging not only decreases transformer 1), and the PEV owner requested normal charging by a 2-state
6

28
Baseline load w/out PEVs
Load w/ PEVs (uncoordinated charging)
26
Load w/ PEVs (Method 1 charging)
Load w/ PEVs (Method 2 charging)
24

22

Load (kW)
20

18

16
Figure 4. Two-state automaton where P1 corresponds to a higher probability
for charge request and P2 to a lower one. In case of overload (congestion = 1) 14
and lack of charge urgency, charge requests are sent with less probability. If
the PEV owner requires urgent charging (urgency = 1) the automaton makes 12
requests during every epoch of time.
10
12 14 16 18 20 22 0 2 4 6 8 10 12
Time (hours)
knob (urgency = 0), the automaton moves to the lower state,
Figure 5. Daily load curve during 24 hours with and without PEVs when
P2 , which is set to a lower value than P1 . two automaton-based charge strategies are utilized (Level 1 charging)
If a PEV owner requires charge immediately, the urgent
option is selected (urgency = 1), wherein the automaton
remains in state P1 and sends charge requests in every epoch, three hours. If the transformer load is limited to 25 kW, the
repeatedly. In this way, PEVs requiring immediate charge will overload problem is solved by both strategies. Moreover, a
receive more charge, on average, during the peak load period. large number of simulations (500 iterations) show that when
probabilistic charging (Method 2) is used, the PEV requiring
immediate charge obtains 83% of the total charge received
B. Simulation Results from uncoordinated charging, whereas Method 1 results in
To illustrate both Method 1 and Method 2, we present only 69% of the total charge from uncoordinated charging.
three case studies. In Cases 1 and 2, the low voltage (LV) Additional observations show that all PEVs receive full charge
transformer is assumed to be the charge manager, and a 25 by the next morning in 97% of instances using both Methods
kVA transformer is considered to serve 7 customers with a 1 and 2.
daily load curve used in Section II-A. We assume 4 PEVs 2) Case 2: LV transformer is the charge manager, Level 2
require charge, and one PEV requests urgent charging. AC charging: This simulation uses the same criteria used in Case
Level 1 and AC Level 2 charging is assumed for Cases 1 and 1; however, PEVs charge at AC Level 2 (3.3 kW). Figure 6
2 respectively. In Case 3, all the charge requests are sent to a shows the results of this example, which indicates that Level
500 kVA high voltage (HV) transformer, which is assumed to 2 charging imposes a higher peak load and the transformer is
serve 100 PEVs submitting charge requests. overloaded for a longer period. Also, Method 2 observations
1) Case 1: LV transformer is the charge manager, Level 1 indicate that the PEV with urgent need of charge will receive
charging: Case 1 assumes that the requests are sent to the more of the total possible charge during the plugged-in period,
LV transformer, which is the first transformer upstream from as compared to Method 1.
the end-users. In this example, the battery capacity is selected
randomly from two available PEVs which are available as of
the 2011 market year. The Chevrolet Volt has a 16 kWh battery 35 Baseline load w/out PEVs
Load w/ PEVs (uncoordinated charging)
of which 10.4 kWh is available, and the Nissan Leaf has a 24 Load w/ PEVs (Method 1 charging)
Load w/ PEVs (Method 2 charging)
kWh battery capacity. The required charge for each PEV is 30
assumed to be normally distributed between 70% and 90%
of battery capacity. AC Level 1 charging is assumed to be
available for all the PEVs, and the charge efficiency is assumed 25
Load (kW)

to be 90%. In our simulations, we assume that the plug-in time


of the PEVs are normally distributed between 1700 and 1900 20
hrs. Also, the plug-out times are normally distributed between
0600 and 0800 hours of the next day. Additionally, one PEV is
15
selected randomly to require urgent charge; its plug-out time
is normally distributed between 2000 and 2200 hrs of the same
day of arrival. Later, this PEV will again be plugged into the 10
12 14 16 18 20 22 0 2 4 6 8 10 12
grid after a normally distributed random time between 1 and Time (hours)

3 hours. Fig. 5 shows the result of one sample simulation. In Figure 6. Daily load curve during 24 hours with and without PEVs when
this example, P2 is set to be 0.4. two automaton-based charge strategies are used (AC Level 2 charging)
Figure 5 shows that uncoordinated PEV charging can cause 3) Case 3: HV transformer is the charge manager, Level 1
the 25 kVA transformer to be overloaded for approximately charging: A final example shows the robustness of Method
7

2. The requests are sent to a 500 kVA HV transformer that the customers and treats all the requests equally. As long
is located at the beginning of the primary distribution system as there is enough available capacity, all PEVs may charge;
feeder. Overloading this transformer is riskier and must be Otherwise, the transformer limits the charging. Moreover, the
avoided. We assume that 100 PEVs are charging on the feeder PEV automatons send their requests independently based on
and 20 of these choose urgent charging. P2 is assumed to be the feedback they receive regarding the congestion of the
0.2 here. Figure 7 shows the results of the third example. distribution feeder. Also, by setting urgency, customers are
For best visualization, it is assumed that the last 20 PEVs of able to obtain more charge during the peak load times if they
the total 100 set urgency = 1 and will depart from the home determine their need to be urgent and are willing to pay for
charging station later in the evening. As in Case 1 and 2, the priority.
these PEVs are plugged into the grid later at night, and are
fully charged at the next morning.
IV. C ONCLUSIONS

This paper describes a method for estimating the impact


600 of PEV charging on overhead distribution transformers by
Baseline load w/out PEVs
Load w/ PEVs (uncoordinated charging) combining a transformer thermal aging model with empirical
500 Load w/ PEVs (Method 1 charging)
travel behavior. We use Monte Carlo simulation to estimate
Load (kW)

Load w/ PEVs (Method 2 charging)

400 the aging in overhead distribution transformers with ambient


temperature data from Phoenix, AZ and Burlington, VT.
300
We investigate and compare cases with AC Levels 1 and 2
200
17 19 21 23 1 3 5 7 9
charging. When AC level 1 charging is replaced with AC level
Time (hours) 2 charging for the uncoordinated charging case, transformer
average annual factor of equivalent aging could increase from
20
0.39 to 1.08 in the colder climate (VT) and from 3.24 to 8.15
in the warmer climate (AZ). This increase indicates a greater
PEV IDs

40
impact from AC Level 2 charging on the warmer climate. We
60
also evaluate our initial smart charging algorithm designed to
80 mitigate the impacts of PEV charging on overhead distribution
100 transformers. For AC Level 1 charging, the reduction in aging
17 19 21 23 1 3 5 7 9
Time (hours) for smart charging compared to uncoordinated charging is
12.8% and 49.4% for VT and AZ respectively. The reduction
in aging as a result of smart charging is more dramatic for
20
AC Level 2 cases; aging decreases by 48.9% and 74.8% for
PEV IDs

40
Urgent VT and AZ respectively. The introduction of smart charging
Charging
60 also reduces uncertainty in transformer aging at both climatic
80 regions and for both AC Levels 1 and 2 charging.
100 Additionally, we introduce two decentralized automaton-
17 19 21 23 1 3 5 7 9
Time (hours)
based strategies for charge management coordination, which
limit the transformer power to its nominal value (or other
Figure 7. Method 1 and Method 2 approach to charge 100 PEVs when
requests are sent to the 500 kVA HV transformer. Top: Daily load curve with acceptable limit). We have combined the independent charge
uncoordinated and coordinated charging strategies. Middle and Bottom: charge management strategy with a pricing structure that allows
activity over the 16-hour window with Method 1 and Method 2 respectively. user choice of charge urgency requirement by using the
White indicates vehicle is charging. Colored indicates vehicle is not charging
(Red: PEV not connected to charging station, Black: PEV does not receive probabilistic charge request submission method (Method 2).
charge packet, Blue: PEV fully charged). The results indicate that if the user requires charging more
urgently, there will be a greater probability of receiving charge
Figure 7 shows that if uncoordinated charging is utilized,
compared against a PEV with less immediate need. If charge
the transformer may incur a peak load of approximately 600
urgency is to be considered, Method 2 shows superiority over
kVA. By using Method 1 and Method 2 approaches the daily
Method 1 and the algorithm in Section II-C. In future work we
load curve is limited to 500 kVA. These two methods have the
will improve upon and apply the empirically derived vehicle
same effect in terms of daily load curve. However, the Method
demand and availability profiles of Section II and the impacts
2 approach (See ellipse in Fig. 7- Bottom) provides much
of ambient temperature to simulations of the decentralized
more charge to the cars which need it immediately (PEV IDs
charging strategies in Section III.
81 to 100), whereas the Method 1 approach (Fig. 7- Middle)
provides equal charge to all of the PEVs. Also, it can be
seen that using the probabilistic method, all of the other PEVs ACKNOWLEDGMENT
obtain full charge by the next morning, when they are needed.
The authors gratefully acknowledge financial support from
It should be noted that the probabilistic automaton-based the U.S. Department of Transportation, through the Trans-
approach does not require specific information about all of portation Research Center at the University of Vermont.
8

R EFERENCES [23] A. Ashtari, E. Bibeau, S. Shahidinejad, and T. Molinski, Pev charging


profile prediction and analysis based on vehicle usage data, Smart Grid,
[1] J. Kliesch and T. Langer, Plug-in hybrids: An environmental and IEEE Transactions on, vol. PP, no. 99, p. 1, 2011.
economic performance outlook. American Council for an Energy- [24] Q. Gong, S. Midlam-Mohler, V. Marano, and G. Rizzoni, Study of pev
Efficient Economy, Washington, DC, Tech. Rep., 2006. charging on residential distribution transformer life, Smart Grid, IEEE
[2] EPRI, Environmental assessment of plug-in hybrid electric vehicles: Transactions on, vol. PP, no. 99, pp. 1 9, 2011.
Volume 1: Nationwide greenhouse gas emissions, Electric Power Re- [25] D. Wu, D. C. Aliprantis, and K. Gkritza, Electric energy and power
search Institute, Tech. Rep., 2007. consumption by light-duty plug-in electric vehicles, IEEE Transactions
[3] J. Gonder, T. Markel, M. Thornton, and A. Simpson, Using global on Power Systems, vol. in press, pp. Paper number: TPWRS00 890
positioning system travel data to assess real-world energy use of plug- 2009, 2010.
in hybrid electric vehicles, Transportation Research Record, 2007. [26] M. Kuss, T. Markel, and W. Kramer, Application of distribution
[4] C. Samaras and K. Meisterling, Life cycle assessment of greenhouse transformer thermal life models to electrified vehicle charging loads
gas emissions from plug-in hybrid vehicles: Implications for policy, using monte-carlo method, 25th World Battery, Hybrid and Fuel Cell
Environmental Science and Technology, vol. 42, no. 9, pp. 31703176, Electric Vehicle Symposium and Exhibitionl, 2010.
2008. [27] A. Hilshey, P. Hines, and J. Dowds, Estimating the acceleration of
[5] R. Sioshansi and P. Denholm, Emissions impacts and benefits of plug- transformer aging due to electric vehicle charging, Summer 2011 IEEE
in hybrid electric vehicles and vehicle-to-grid services, Environmental PES General Meeting, 2011.
Science & Technology, vol. 43, no. 4, pp. 11991204, 2009. [28] J. Frolik and P. Hines, Random access, electric vehicle charge man-
[6] W. Kempton and J. Tomic, Vehicle-to-grid power fundamentals: Cal- agement, in IEEE International Electric Vehicle Conference, accepted,
culating capacity and net revenue, Journal of Power Sources, vol. 144, March 2012.
no. 1, pp. 268279, 2005. [29] J. Frolik, Qos control for random access wireless sensor networks, in
[7] M. J. Scott and M. Kintner-Meyer, Impact assessments of plug-in Wireless Communications and Networking Conference, 2004. WCNC.
hybrid vehicles on electric utilities and regional u.s. power grids part 2004 IEEE, vol. 3, march 2004, pp. 1522 1527 Vol.3.
ii: Economic assesment, Pacific Northwest National Laboratory, Tech. [30] Oak Ridge National Laboratories. (2009) 2009 nhts user
Rep., 2007. notes. oak ridge national laboratories. [Online]. Available:
[8] S. Hadley and A. Tsvetkova, Potential impacts of plug-in hybrid electric http://nhts.ornl.gov/2009/pub/usernotes.pdf
vehicles on regional power generation, Oak Ridge National Laboratory, [31] NEMS. (2001) Reload database documentation and
Tech. Rep., 2008. evaluation and use in nems. [Online]. Available:
[9] C. Gerkensmeyer, M. Kintner-Meyer, and J. DeSteese, Technical chal- http://www.onlocationinc.com/LoadShapesReload2001.pdf
lenges of plug-in hybrid electric vehicles and impacts to the us power [32] G. Motors, Chevrolet volt leads general motors into it second century,
system: Distribution system analysis, Prepared for U.S. Dept. of Energy, Sept. 2008.
Pacific Northwest National Laboritory, Tech. Rep., 2010. [33] Society of Automotive Engineers, SAE charging configurations and
[10] M. Kintner-Meyer, K. Schneider, and R. Pratt, Impact assessments of ratings terminology, SAEJ1772, 2011.
plug-in hybrid vehicles on electric utilities and regional u.s. power grids [34] K. Morrow and D. Karner, Plug-in hybrid electric vehicle charging
part 1: Technical analysis, Pacific Northwest National Laboratory, Tech. infrastructure review, U. S. Dept. of Energy, Idaho National Laboratory.,
Rep., 2007. Tech. Rep., 2008.
[11] C. Roe, F. Evangelos, J. Meisel, A. Meliopoulos, and T. Overbye, Power [35] W. McNutt, Insulation thermal life considerations, Power Delivery,
system level impacts of phevs, in Proceedings of the 42nd Hawaii IEEE Transactions on, vol. 7, no. 1, pp. 392401, january 1992.
International Conference on System Sciences, Waikoloa, HI, 2009.
[12] Jeff LeBrun, EVs a tale of 3 cities, Presentation at the Intelligent
Utility Webinar, Jan 2011. AUTHOR BIOGRAPHIES
[13] S. Shao, M. Pipattanasomporn, and S. Rahman, Challenges of phev
penetration to the residential distribution network, in Proceedings of
the IEEE Power & Energy Society General Meeting, Minneapolis, July
2009.
[14] Transformers Committee of the IEEE Power Engineering Society,
IEEE Std C57.91-1995: IEEE Guide for Loading Mineral-Oil-Immersed Alexander D. Hilshey (S10) is a Graduate Research Assistant with the
Transformers. IEEE, 1995. Transportation Research Center at University of Vermont and is working
[15] L. P. Fernndez, T. G. S. Romn, R. Cossent, C. M. Domingo, and toward a M.S. in Electrical Engineering from the University of Vermont.
P. Fras, Assessment of the impact of plug-in electric vehicles on
distribution networks, IEEE Transactions on Power Systems, vol. in
press, no. TPWRS-00985-2009, 2010.
[16] M. Rutherford and V. Yousefzadeh, The impact of electric vehicle bat-
tery charging on distribution transformers, in Applied Power Electronics
Conference and Exposition (APEC), 2011 Twenty-Sixth Annual IEEE, Pooya Rezaei (S11) received the M.Sc. degree in Electrical Engineering
march 2011, pp. 396 400. from Sharif University of Technology, Tehran, Iran in 2010. Currently, he is
[17] K. Clement-Nyns, E. Haesen, and J. Driesen, The impact of charging pursuing the Ph.D. degree in Electrical Engineering at University of Vermont.
plug-in hybrid electric vehicles on a residential distribution grid, IEEE
Transactions on Power Systems, vol. 25, no. 1, pp. 371380, Feb. 2010.
[18] S. Deilami, A. Masoum, P. Moses, and M. Masoum, Real-time
coordination of plug-in electric vehicle charging in smart grids to
minimize power losses and improve voltage profile, Smart Grid, IEEE Paul D. H. Hines (S96, M07) received the Ph.D. degree in Engineering
Transactions on, vol. 2, no. 3, pp. 456 467, sept. 2011. and Public Policy from Carnegie Mellon University in 2007, and the M.S.
[19] S. Vandael, N. Bouck, T. Holvoet, and G. Deconinck, Decentralized degree in Electrical Engineering from the University of Washington in 2001.
demand side management of plug-in hybrid vehicles in a smart grid, in He is currently an Assistant Professor in the School of Engineering at the
Proceedings of the First International Workshop on Agent Technologies University of Vermont.
for Energy Systems (ATES 2010), 2010, pp. 6774.
[20] A. D. Hilshey, P. D. H. Hines, and J. R. Dowds, Estimating the effect of
electric vehicle smart charging on distribution transformer aging, IEEE
Transactions on Smart Grid - In Review, 2011.
[21] K. Parks, P. Denholm, and T. Markel, Costs and emissions associated
with plug-in hybrid electric vehicle charging in the xcel energy colorado Jeff Frolik (S85, M95, SM11) received the B.S.E.E. degree from the
service territory, National Renewable Energy Laboratory, Golden, CO, University of South Alabama in 1986, the M.S.E.E. degree from the University
Tech. Rep., 2007. of Southern California in 1988 and the Ph.D. degree in Electrical Engineering
[22] D. Lemoine and D. Kammen, Effects of plug-in hybrids electric Systems from the University of Michigan in 1995. He is currently an Associate
vehicles in california energy markets, in Proceedings of the TRB 86th Professor in the School of Engineering at the University of Vermont.
Annual Meeting, 2007.

You might also like