Applications of Probability Model To Analyze The Effects of Electric Vehicle Chargers On Distribution Transformers
Applications of Probability Model To Analyze The Effects of Electric Vehicle Chargers On Distribution Transformers
Applications of Probability Model To Analyze The Effects of Electric Vehicle Chargers On Distribution Transformers
Abstract—Society’s increased concern over green house gas and [4]. The effects on electric distribution systems were ana-
emission and the reduced cost of electric vehicle technologies has lyzed in [1], with major emphasis placed on distribution trans-
increased the number of electric vehicles (EV) and plug-in hybrid formers, cable systems, and switches. The increased burden due
vehicles on the road. Previous studies into the effects of electric
vehicles on the electric system have focused on transmission, to EV chargers causing faster loss of life in distribution trans-
generation, and the loss of life of distribution transformers. This formers is addressed in [5]. While the reduced loss of life of a
paper focuses specifically on identifying distribution transformers distribution transformer due to EV chargers is a long-term ef-
that are most susceptible to excessive loading due to the implemen- fect, in the short term, EV chargers may result in a high number
tation of electric vehicles. The authors use a binomial probability of blown transformer fuses due to the excessive loading that
model to calculate the probability that a specific distribution
transformer will experience excessive loading. Variables to the may occur during peak periods. This can result in an increased
function include the existing peak transformer demand, number number of customer outages.
of customers connected to the transformer, and the most common The North American Standard SAEJ1772 categorizes EV
EV charger demand. Also included in the paper is an optimization chargers into three different operating groups, commonly
approach that utilizes the results from the binomial function referred to as Level 1, Level 2, and Level 3 [6]. The power
to determine the optimal replacement strategy to minimize re-
placement costs. An extension of the approach is also utilized to drawn from these chargers can range between 1 kW and 240
explore the effectiveness of EV targeted demand side management kW for periods between 30 min and 7 h, although most Level
programs. The authors apply the described algorithms to 75 000 1 and 2 chargers will not exceed a peak demand of 10 kW [6].
distributions transformers within a distribution system located in Studies have shown the peak charging period of EV chargers,
Denver, Colorado. if not controlled by the utility company, will likely occur after
Index Terms—Binomial distribution, demand side management, 5 p.m. and may coincide with existing customer load peaks,
distribution planning, distribution transformers, electric vehicles. thereby significantly increasing the maximum power draw of
the distribution transformer [1], [5], [7]. For transformers that
have a peak demand that is near or above its rated capacity, the
I. INTRODUCTION addition of several chargers can push the total demand above
its fuse ratings. These ratings typically range between 180%
and 300% depending upon transformer size, available fuse size,
T HE increased concern about reducing green house gas and protection scheme [8]. Therefore identifying transformers
emissions has led to the continued increase in the number that are susceptible to excessive loading due to EV can have an
of pure electric vehicles and plug-in hybrid electric vehicles on impact on preventing future customer outages.
the road. These vehicles, which the authors will refer to as elec- In the late 1990s, Xcel Energy experienced a high number
tric vehicles (EVs), can increase mileage by 60% given the same of distribution transformer failures caused from unexpected
amount of primary energy [1]. The number of electric vehicles loading that was due to widespread implementation of air
will continue to grow over the next 20 years, at which time EVs conditioning units. Prior to this period most homes were not
may comprise as much as 50% of new vehicle purchases [2]. equipped with central air conditioning units and relied upon low
Studies have been performed over the last decade to deter- energy demand swamp coolers. During this AC implementa-
mine the effects of EV chargers on the electric power system. tion period, the number of failed transformers due to excessive
EV effects on generation and transmission were studied in [3] loading averaged approximately 600 (1% of all transformers)
annually, which also contributed to approximately 84 000
customer outage minutes annually.
Manuscript received October 28, 2011; revised February 17, 2012 and April Considering that transformers can serve as many as 20 house-
24, 2012; accepted May 30, 2012. Paper no. TPWRS-00981-2011. holds, the probability of a transformer being overloaded or re-
J. M. Sexauer and K. D. McBee are with the Colorado School of Mines,
sulting in an unplanned outage due to fuse-blowing during peak
Golden, CO 80401 USA and Xcel Energy, Denver, CO 80223 USA (e-mail:
[email protected]; [email protected]). hours can be significant if the utility does not incorporate de-
K. A. Bloch is with Xcel Energy, Denver, CO 80223 USA (e-mail: Kelly. mand side management programs to control charging times. An-
[email protected]).
other approach utilities can employ to avoid excessive distribu-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. tion transformer loading due to EV penetration is to identify the
Digital Object Identifier 10.1109/TPWRS.2012.2210287 transformers that have the highest probability of overload based
on existing peak demand, number of connected customers, and The most accurate means to forecast load growth is through
the most common EV charger size. simulations-based methods that attempt to predict the size of
In this paper, the authors describe a novel load forecasting the increased demand along with the causes of the increased
approach that utilizes a binomial distribution to identify trans- demand that include economic growth, historical data, and
formers that have a high probability of being overloaded with customer type. Because of the accuracy associated with this
the addition of one or several EV chargers. Variables to the func- approach if applied correctly and the intensive effort required
tion include the peak transformer demand, the number of cus- to apply, the simulation-based approach is typically applied
tomers served by the transformer, and the most common size to long-term planning. Unfortunately the simulation approach
of an EV charger. Existing load growth forecasting approaches heavily relies upon historical data and information regarding
along with their validity for application to EV demand growth the causes of load growth, such the methods described in [13],
are discussed in Section II. The binomial model is described which in the case of the paper would be increased EV pene-
in detail in Section III. Section IV illustrates the application of tration. Aspects that affect EV penetration include customer
the binomial function to determine the total number of suscep- acceptance, purchase costs, and government subsidies, all of
tible distribution transformers located in Denver, Colorado. This which are difficult to predict for long-term planning on small
section also illustrates the use of a binomial function to opti- spatial levels [2], [9]. Because of the lack of information re-
mize replacement strategy and evaluate demand side manage- garding the attributes that drive EV penetration growth, the load
ment (DSM) programs. demanded by future EV chargers can be considered “causal”
event loads in that more information regarding the driving
II. EXISTING APPROACHES TO FORECASTING factors are required to make accurate long-term forecasts for
Evaluating the impact of EVs on the existing distribution in- specific areas.
frastructure will require planning engineers to forecast the load The approach described in the paper, which would be consid-
growth associated with EV charger demand. As always, the ered a high resolution spatial forecasting analysis, is sufficient
three most important questions that planning engineers must an- for a stand alone analysis or as an input to the simulation-based
swer are [9]: approach method, where historical growth information is un-
1) “Where” will the system be impacted by the increased load known or nonexistent. Instead of trying to predict growth based
demand? on historical growth data, the authors have developed an ap-
2) “When” will the impact occur? proach that calculates the probability of impact on utility de-
3) “What are the costs” associated with the impact? vices based on their existing peak demand, number of connected
To answer these questions, planning engineers must rely upon customers, which accounts for the possibility of growth, and a
load growth forecasting techniques. Such prediction techniques given EV penetration rate, which if applied to future EV growth
can consist of short-term forecasting, which extends between 1 predictions can determine “when” a device is impacted. Once
and 6 years, and long-term forecasting, which can span over the probability of impact is derived through the binomial distri-
20 years [9], [10]. Because the EV penetration growth para- bution function described in the following section, an economic
digm is being developed, it is unknown if evaluating the im- impact study is performed to the answer the question of “how
pact of EVs requires long-term or short-term forecasting. Vari- much” will the load growth cost. However, before evaluations
ables such as customer acceptance and available capacity of are performed to answer “when” and “how much,” distribution
the existing system infrastructure will dictate whether short- engineers must first determine “where” will the impacts occur.
or long-term forecasting is required for a specific geographical
location. However, because there is so little historical data re- III. PROBABILISTIC MODEL
garding a continuously increasing EV penetration, existing tech- For a distribution planner to answer the question of “where”
niques for forecasting, long-term or short-term, may not be ap- will the system be impacted, they must first determine if
plicable. specific devices are susceptible to overloading due to the
Trending is the most common approach to short-term load increased demand. Although there are many devices on the
growth forecasting [9]. Utilizing a spatial approach, which di- distribution system that may be impacted by EV penetration
vides larger service areas into smaller and discreet areas based growth, the authors are focusing on evaluating the suscepti-
on customer class or type, city, subdivision, or geographical lo- bility of distribution transformers, or in layman’s terms, which
cation, distribution planners match historical load growth infor- distribution transformers on the existing system are susceptible
mation to growth curves that predict future growth [10]–[12]. to overloading based on their existing available capacity and
The approach typically utilizes multiple regression curve-fit- number of residential customers, which represents growth
ting techniques that depend on historical data just to initiate. potential. Load growth is ignored in the calculations because
Because so little information is known about EV penetration this occurrence is typically the result of new subdivision de-
growth, even if these approaches are applied, the results would velopments and/or new business, which require the installation
be subject to inaccuracies. Another problem with applying these of new distribution transformers. The authors have chosen to
trending approaches is that they can be significantly inaccurate identify these susceptible transformers by utilizing a binomial
at high resolution spatial forecasting, which hinders the eval- distribution function.
uation of the impact of small distribution areas such as sec- The probabilistic model considered for this paper is founded
ondary service conductors and distribution transformers, which in binomial distribution theory. The model describes the prob-
may only consist of several customers [9]. ability of a distribution transformer overloading due to connec-
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
tion of new EV. The authors define overload as the limit in the transformer, a peak demand that blows protection fuses
which the loading on a transformer exceeds acceptable levels, will result in added operational cost that a utility company
which could be well above the nameplate rating. After a brief in- will typically mitigate by upgrading the transformer to prevent
troduction to the binomial random variable, the specific methods future costs and prevent customer outages [9], [15]–[19]. Many
needed to use the distribution are explored. Finally, the assump- utility companies utilize a predetermined loading limit with
tions taken to use this distribution are enumerated. and without a duration threshold to determine if a transformer
requires upgrading due to load [20]. Although these limits may
A. Mathematical Introduction to the Binomial Distribution not be able to identify that a specific transformer will have a
A binomial random variable describes the number of “suc- shorten life due to loading above threshold limits, for a large
cesses” expected from a series of trials. An experiment such as population size they have proven to minimize the customer
“What are the chances of seeing three heads when flipping five impact to failed transformers and outages due to repeated
fair coins?” is described using a binomially distributed random blown fuses.
variable [8], [14]. An additional option for establishing the peak demand is
The probability mass function of a binomial random variable based on the protection scheme of distribution transformer.
is as follows: Transformer fusing is generally based on two principles. The
first is to protect the transformer from experiencing loading
(1) that can significantly reduce the life of the transformer. Fuses
rated for this type of protection typically ranges between
where is the probability of successes in trials, 150%–250% of nameplate [21]. In the last 15 years, many
each with a probability of [14]. utility companies have incorporated a second type of pro-
tection philosophy which is more focused on protecting the
B. Creation of Transformer Overloading Model
entire distribution system from transformer failures instead of
To describe the probability of several EV being connected preventing the transformer from failing [20]. These fuses are
to a transformer, a binomial random variable as in (1) is used. typically sized between 180%–300% depending upon available
The number of trials is the total number of households con- sizes. The one thing that both of these approaches have in
nected to a given transformer. The probability of a single cus- common is that they result in unplanned customers outages and
tomer choosing to connect an EV to the transformer is . then the replacement of distribution transformers if loading limits
describes the probability that electric vehicles are on the trans- are exceeded [7].
former. From a distribution system planning perspective, either one of
However, interest is not in the probability of electric vehi- these limits would be sufficient for application to the acceptable
cles being on the transformer, but the probability that the trans- peak loading of transformer. Because the probability density
former experiences excessive loading due to electric vehicles. function is utilized to predict which transformers are likely to
To find this, the parameters of the distribution function need to experience adverse loading that impact customer reliability and
be adjusted so that the number of electric vehicles that can be operational costs, employing the existing company guidelines
placed on the transformer before it overloads can be found. This for identifying overloaded transformers is paramount. Although
number is a function of , the peak demand of the transformer; these limits may not necessarily reflect the actual aging of the
, the acceptable limit to which a transformer can be loaded; distribution transformer, they will reflect the loading limits that
and , the peak demand of an electric vehicle charger. Let the typically impact customers and operational costs.
capacity for the number of EV that can be connected without The only exception would be in cases where fuse sizes are
overloading the transformer be : near 300% of the nameplate rating. Although the fuses will not
operate until loading reaches these high levels, liquid immersed
(2) transformers began to experience oil degradation at much lower
loads, which can ultimately reduce the operating capacity of
where the transformer and make them more susceptible to failure due
maximum number of EVs that can be connected to EV charger loading at lower loading levels [21]–[23]. The
to a transformer before acceptable loading limit is reduced capacity can result in failures although loading never
exceeded; reached 300%. Therefore utilizing this as an upper threshold
acceptable peak demand on transformer; would lead to a forecast that identifies fewer transformers than
existing peak demand on transformer; actual. The level in which oil degradation occurs is a function
of the oil type, loading, and environmental conditions, therefore
average peak demand of an EV charger.
the acceptable peak loading that reflects this oil characteristics
The peak acceptable demand on a distribution transformer is must be location dependant.
an arbitrary value that represents the loading on a transformer The demand of the most commonly utilized EV chargers has
that “may” lead to blown fuses, distribution transformers with yet to be determined due to the current low implementation rate.
shortened lives, and customer outages if not addressed. Al- As mentioned previously, EV chargers can demand anywhere
though the loading of EV chargers may only result in adverse between 1 kW and 240 kW depending upon the level of charger
transformer loading during a short period of time, such as in [5]. When considering driving distance and the required home
the summer, which may not significantly reduce the age of electric upgrades for Level 3 chargers, the current consensus is
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
(5)
where
is still optimal to let most transformers fail and replace them,
total costs of replacement for entire system ] retroactively.
expected value for the th transformer’s random
variable, per (4); C. Electric Vehicle Targeted Demand Side Management
number of transformers in population; Because of the long charging time of EVs and the temporal
number of reactive replacements; overlap between daily commutes and load peaks, it is likely
number of proactive replacements; that EVs will be coincident with the peak loading. However,
if a DSM program is implemented so that EV owners are incen-
reactive cost to replace a transformer;
tivized to not charge their vehicle during peak loading most of
value realized from a proactive replacement the capacity issues related to EVs could be minimized. To help
(planned and reduced outage time, salvage or establish what effects DSM programs might have, the baseline
redeployment value, etc.). model has been further refined to reflect DSM participation.
For a given overload probability (y-axis of Fig. 2), the cor- Recall that for each transformer, a binomial probability is cal-
responding transformer ordinal (x-axis of Fig. 2) provides the culated to determine if enough homes will get EVs to cause an
value of . To find , the curve is “integrated” (i.e., summed) overload. We now introduce a second, nested series of binomial
on the bounds to . In order to avoid a curve fit of Fig. 2 experiments to see, for households which have an EV, if they
and the nonlinearities that would result, the authors have opted participate in a DSM program. This model can then be used to
to bin in 1% increments the overload probability, find the cor- see to what degree curtailment from DSM will avert an over-
responding ’s and ’s, and solve the optimization numeri- load.
cally. Costs have also been per-unitized to where and Take for example a transformer with five homes on it which
is varied between 5% and 15% of . As such, any total also only has capacity for one electric vehicle, as described in
replacement value less than 1.0 is beneficial, though possibly Table II. Assume an EV penetration of 15% and a DSM par-
sub-optimal. The results are shown in Fig. 3. Table I shows the ticipation rate of 10%. To analyze the effectiveness of DSM, a
replacement rates for various as a percentage of replacement second series of four binomial probabilities must now be for-
costs and the per-unitized costs experienced at —the op- mulated, one for each event which causes an overload (i.e., for
timal replacement threshold. As can be seen, for even the most the events that 2 (event A), 3 (B), 4 (C), and 5 (D) households
aggressive scenario with a large proactive replacement value, it connect an EV):
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
TABLE II
EXAMPLE OF DSM OUTCOMES FOR A TRANSFORMER WITH FIVE HOUSEHOLDS AND CAPACITY FOR ONE EV
, , .
Probability of specific event, i.e., , , etc. The sum of non-overloading events (i.e., the unlabeled events) is the probability of no overload because
of sufficient capacity, i.e., .
Probability of any DSM event occurring which averts an overload, i.e., , , etc.
(11)
(7)
TABLE III [12] E. P. Cody, “Load forecasting methods cuts time, cost,” Elect. World,
NUMBER OF TRANSFORMERS REQUIRING REPLACEMENT FOR VARIOUS EV p. 87, Nov. 1987.
PENETRATIONS AND DSM PARTICIPATION RATES [13] H. L. Willis, M. V. Engel, and M. J. Buri, “Spatial load forecasting,”
IEEE Comput. Appl. Power, vol. 8, no. 2, pp. 40–43, Apr. 1995.
[14] W. Navidi, Statistics for Engineers and Scientists, 3rd ed. New York:
McGraw-Hill.
[15] K. D. McBee and M. G. Simoes, “Utilizing a smart grid monitoring
system to manage distribution transformers,” in Proc. North American
Power Symp. (NAPS), 2009, Oct. 2009.
[16] IEEE Guide for Loading Mineral-Oil-Immersed Transformers, IEEE
Std. C57.91-1999, 1999.
[17] IEEE Standard Test Procedure for Thermal Evaluation of Liquid-Im-
susceptible transformers before they result in a high number mersed Distribution and Power Transformers, IEEE Std. C57.100-
of customer outages. The authors have presented a method that 1999, 1999, p. i.
utilizes a binomial distribution to predict the probability that a [18] Loading Guide for Oil-Immersed Power Transformers, IEC 354, Sep.
1991.
distribution transformer will be loaded above a specific level [19] IEEE Recommended Practice for Establishing Liquid-Filled and Dry-
based on existing peaking demand, the number of customers Type Power and Distribution Transformer Capability When Supplying
served by said transformer, and average size of EV charger. Also Nonsinusoidal Load Currents, IEEE Std. C57.110-2008 (Revision of
IEEE Std. C57.110-1998), pp. c1-44, 15 2008.
presented in the paper is an optimization approach that utilizes [20] J. Lunsford and T. Tobin, “Detection of and protection for internal
data from the binomial distribution to determine the replace- lowcurrent winding faults in overhead distribution transformers,” IEEE
ment strategy in regards to economic feasibility. The authors Trans. Power Del., vol. 12, no. 3, pp. 1241–1249, Jul. 1997.
[21] C. McShane, J. Luksich, and K. Rapp, “Retrofilling aging transformers
finally present a DSM strategy that utilizes information from with natural ester based dielectric coolant for safety and life extension,”
binomial function. in Proc. IEEE-IAS/PCA 2003 Cement Industry Technical Conf., 2003,
May 2003, pp. 141–147.
[22] A. Marulanda, M. Artigas, A. Gavidia, F. Labarca, and N. Paz, “Study
ACKNOWLEDGMENT of the vegetal oil as a substitute for mineral oils in distribution trans-
former,” in Proc. 2008 IEEE/PES Transmission and Distribution Conf.
The authors would like to thank Xcel Energy for their aid Expo.: Latin America, Aug. 2008, pp. 1–6.
in this project, C. Lorenzini for his advice, and the Colorado [23] T. Oommen and S. Lindgren, “Bubble evolution from transformer
School of Mines. overload,” in Proc. 2001 IEEE/PES Transmission and Distribution
Conf. Expo., 2001, vol. 1, pp. 137–142.