842 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 38, NO.

2, APRIL 2023

Distribution Transformer Loading: Probabilistic

Modeling and Diversity Factor
Gerald Thomas Heydt , Life Fellow, IEEE

Abstract—For planning purposes, the loading of assets in electric feeder. Let the diversity factor be given by d,
power distribution circuits is often modelled in terms of a diversified
load. The temporal and probabilistic nature of the load is a sum max (p1 (t)) + max (p2 (t)) + . . . + max (pN (t))
of several individual loads of various levels and characteristics. d = (1)
sup (p1 (t) + p2 (t) + . . . + pN (t))
The diversity factor of such circuits is the ratio of the sum of
the individual non-simultaneous load asset ratings energized by where the several served loads are p1 , p2 , …, pN , and the
that circuit to the total peak simultaneous demand. The diversity
factor is temporal and probabilistic in nature. In this paper, the numerator is the sum of the non-simultaneous maxima of these
probability density of the load factor is described and modelled for loads and the supremum function in the denominator refers to the
power distribution loads. The model used is the ratio of two depen- sum of the total load. Equation (1) may be viewed as the ratio of
dent bivariate normal random variables. The resulting probability the sum of the individual peak demands to the group peak. The
density is termed a ratio density and a recent algebraic formulation diversity factor is the ratio of the total non-coincidental demand
of the ratio density is used. The probabilistic model is compared to
results obtained by others. The potential impact of electric vehicle to the maximum diversified demand. Renewable resources, if
charging loads is applied to the sizing of distribution power supply they are present, are included in the numerator and denominator
assets using diversity loading strategies for Level 1 and single phase of (1). The number of residential services in this configuration,
Level 2 EV chargers. The probabilistic method is offered as an N, in the United States is in the general range 3≤ N ≤ 10. It is
added tool to the conventional, fixed diversity factor method in convenient to view the diversity factor as a ratio,
distribution engineering.
X
Index Terms—Power distribution engineering, diversity d = (2)
factor, load factor, probability, distribution transformers, ratio Y
probability density, electric vehicle charging.
where the properties of the numerator X and the denominator
Y shall be exploited presently. In the literature, the term co-
I. LOADING OF DISTRIBUTION SYSTEM ASSETS incidence factor is occasionally found and this factor is the
OST power distribution systems worldwide are radial reciprocal of d.
M and tree-like in nature. This means that energy is dis-
tributed from an overarching subtransmission system (typically II. DIVERSITY FACTOR
a networked system), to mostly radial distribution primaries The use of the foregoing diversity factor has been known
which energize branch circuits in an outward reaching tree-like and used in power engineering for over 100 years [1]. To be
configuration. Assets in this structure are designed to accom- concise, typical values of diversity factor in various distribution
modate loading according to the assumption of load diversity. applications usually lie in the range (1.0 ≤ d ≤ 8.0). In applica-
The term ‘load diversity’ refers to the characteristic that the tions in which the several loads served are diverse temporally,
many connected loads are not all operative simultaneously at the denominator of (1) will be smaller than the sum of the
their rated levels. Thus the temporal and probabilistic nature nonsimultaneous peaks (the numerator), and the diversity factor
of the loads allows the sizing of distribution assets (mainly will be large. The denominator Y may be viewed as an uncertain
transformers) less than the requirements if all loads were ‘turned quantity because of the diversity in the time of the loads, and
on’ simultaneously. also in the operational levels. Let Y be taken to be a sum of
The accepted analysis approach in power distribution engi- random variables. Similarly, the numerator may be regarded as
neering is to utilize diversity factor to capture the probabilistic a random variable and as the result of a summation of load
and temporal nature of many loads connected to a distribution levels. Especially for large N, there is a motivation to appeal
to the central limit theorem which states that for the case that
Manuscript received 26 January 2022; revised 27 June 2022; accepted 10 the individual load levels pi (t) being statistically independent,
August 2022. Date of publication 18 August 2022; date of current version 24 the sums X and Y tend to a normal (gaussian) probabilistic
March 2023. Paper no. TPWRD-00118-2022.
The author is with the School of Electrical, Computer and Energy En- distribution [2]. This is the case even if the random variables
gineering, Arizona State University, Tempe, AZ 85287-5706 USA (e-mail: in the numerator and denominator are not normally distributed
[email protected]). themselves. While it is clearly an assumption that X and Y are
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TPWRD.2022.3199999. normally distributed, we shall see what results are obtained with
Digital Object Identifier 10.1109/TPWRD.2022.3199999 this assumption, and subsequently we shall make a comparison

0885-8977 © 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 HEYDT: DISTRIBUTION TRANSFORMER LOADING: PROBABILISTIC MODELING AND DIVERSITY FACTOR 843

with other characterizations of the probabilistic nature of the and μx , μy are the mean values of X and Y. The notation K2 and
diversity factor d. θ2 is consistent with that in 14], ρ is the correlation coefficient
The diversity factor of distribution system loads has been between X and Y, ‘exp’ denotes exponentiation ex , and 1 F1 (•;•;•)
estimated from operational and research data (e.g., [3], [4]). is the confluent hypergeometric function of the first kind [15].
Operating experience for over 100 years has resulted in industry To avoid confusion with other definitions of the hypergeometric
accepted designs as exemplified by [5], [6], [7]. Operational data function, note that in (3),
may be used in a Monte Carlo study [8] to estimate d for given ∞
 Γ (a + k) ck Γ (b)
applications. Appealing to load models [9] themselves may be 1 F1 (a, b, c) = (6)
used to estimate the diversity factor of distribution feeders, but Γ (a) Γ (b + k) k!
k=0
there appears to be limitations to this approach because of the
with Γ (.) denoting the gamma function. Modern software tools
large number of load types and operating conditions. Zhu and
generally have simple syntax for 1 F1 . For example, in Matlab,
Mather show some techniques using time series simulations
[10]. Perhaps the overarching operating and design strategies 1 F1 is coded as f = hypergeom(a, b, c)
The expressions (3, 4, 5) may appear formidable, but they are
come from the industry standards, e.g., [11], [12]. Modeling the
straightforward, non-iterative algebraic calculations in terms of
transformer diversity factor probabilistically allows the distribu-
the statistics of X and Y. Reference [14] includes discussions of
tion planner to estimate the probability of loading scenarios, a
cases in which the ratio probability density function is unimodal
capability not possible in deterministic formulations. This offers
as well as bimodal (single peak or two peaks). Most other
the potential to conveniently compare alternative design choices.
simplified probabilistic models of d are unable to capture this
flexibility, but Monte Carlo simulations and Gaussian mixture
III. THE PROBABILISTIC MODEL
models can do so. Gaussian mixtures have been used for imple-
Attention now turns to the modeling of d in (2) as a ratio mentation of dynamic loading regimes for power transformers
of two normally distributed random variables. Obviously, these [16]. One factor that favors the model in (3, 4, 5) is the paucity of
variables are correlated. The assumption, then, is that X, Y data and very rapid calculation needed to obtain fd (d) from (3).
are bivariate normal with mean values μX , μY and variances Examination of load data for over fifty distribution transformers
2
σX , σY2 and correlation coefficient ρ. The probability distribu- in-service also leads one to favor the indicated model because
tion of the ratio of two bivariate normal random variables is not of its faithfulness to reproduce observed load diversity.
normal in general. Chatlani et al. [13] examined this ratio in some
detail using distribution feeders and loads in the U. S. southwest, IV. IMPLICATIONS FOR POWER DISTRIBUTION ENGINEERING
and they found that modeling d as a gamma distributed random
variable yields accuracy in statistically modelling the load on The discussion above has implications for assessing loading of
distribution feeder transformers. The application in this case is power distribution transformers that serve multiple loads. To size
that several diverse loads (N loads) are served from a common the distribution transformer, the loads are assumed to be diverse
distribution transformer. Their end objective appears to have and characterized by a diversity factor. Anecdotally, the diversity
been in the estimation of loss of transformer life depending on factor, as a ratio of two assumed gaussian random variables, was
load diversity. However, it is possible to apply additional results assumed by some to be itself gaussianly distributed. Chatlani
on the assessment of the probability density function of the ratio et al. [13] demonstrated that a better approximation for the
of two bivariate normal random variables. The present objective probability density of d (at least in some cases) is the gamma
is to obtain a more accurate probabilistic model for the diversity probability density,

factor d. The ratio of two normal random variables has occupied
dα−1 exp − βd
considerable attention of mathematicians for the last 100 years. fd (d) = (7)
Pham-Gia, Turkkan and Marchand in 2006 have definitively β α Γ (α)
solved the problem of the calculation of the probability density where α and β are shape parameters and Γ (.) denotes the gamma
of the ratio of two bivariate normal variables for all possible function. The basis of the claim of a ‘better approximation’ to
cases [14], and their formula for the probability density of d is the probability density of d is that the gamma density is able to
fd (d), capture a skew to the right (i.e., larger values of d). This cannot be
  2 2
2K2 1 − ρ2 σX σY 1 F1 (1; 0.5; θ2 ) accommodated by the simple assumption that d has a gaussian
fd (d) = (3) probability density function. The density in (7) is easily fit to
σY2 d2 − 2ρσX σY d + σX 2
measured data using the mean and variance of d,
2
where the variances of X and Y are σX , σY2 and
E (d) = μd = αβ σd2 = αβ 2 (8)
1
K2 = 
2πσX σY 1 − ρ2 (note: E(.) denotes expectation). Like the gaussian density, the
  gamma density function is determined by the mean and variance
σ 2 μ2 − 2ρσX σY μX μY + μ2Y σX 2
in the univariate case. Dividing the expressions for μd and σd2
× exp − Y X 2 σ2 (4)
2 (1 − ρ2 ) σX Y in (8) immediately gives β, and subsequently α found. This is
 2 
2 2 a convenient advantage of the gamma formulation. However, in
−σY μX d + ρσX σY (μY d + μX ) − μY σX
θ2 = 2 σ 2 (1 − ρ2 ) (σ 2 d2 − 2ρσ σ d + σ 2 ) (5) view of the development above, namely (3, 4, 5), the probability
2σX Y Y X Y X density of d is not the gamma probability density; instead, fd (d)
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 844 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 38, NO. 2, APRIL 2023

is given by (3), namely a ratio probability density. It appears to be

more consistent with assumptions to probabilistically model the
diversity factor in this way, and to size distribution transformers
using the ratio density model to calculate the probability of
transformer overload.
Let a given group of N loads be energized from a power
distribution transformer. And further assume that the loads so
serviced exhibit a diversity factor d. Utilizing measured and
assumed data, the statistics of the individual loads and the
group (total) load yield the development in (3)–(5), namely the
probability density of the diversity factor. For purposes of this
discussion, let the electric current demands in the several loads
be i1 , i2 , … and let these currents have associated apparent power
levels (e.g., kVA) denoted p1 , p2 , …. Further assume that the
single supply transformer for this load center be evaluated for
overload by the total apparent power which is estimated as the
simple sum of the several apparent power values p1 , p2 , …. The
probability of overload is the area under the curve of fd (d) from
zero (the lowest possible value of d) to unity (d = 1). This is
the case since this is the operating region for which the total
load exceeds the total connected load which was assumed in the
design phase (i.e., the rating of the transformer). Thus,
1
Pr (d ≤ 1) = fd (d) dd (9)
0 Fig. 1. An outline of distribution transformer sizing strategy based on diversity
loading.
with ‘Pr’ denoting probability. If it were desired to find the
probability of overload by 20%, of rating (e.g., a short term
overload), the probability Pr(d ≤ 1/1.2) = Pr(d ≤ 0.833) would TABLE I
be calculated. APPLICATION EXAMPLE PARAMETERS
The basic strategy of distribution transformer sizing offered
in this paper is shown in Fig. 1. The concept is to base sizing
on the probabilistic load model and the acceptable level of
probability of transformer overloading. The suggestion is made
as an added alternative to alternative sizing methods, not as a
replacement to those methods. The approach taken in this design
procedure is to accommodate maximum transformer loading
probabilistically rather than to ‘overdesign’ by an arbitrarily
chosen ‘safety factor’.

V. EXAMPLES OF APPLICATIONS
study of the impact of wider statistical variation of the total
A. An Example From Contemporary Distribution Engineering
transformer load. The three cases studied are explained below.
In this section, an illustration is given for the impact of Example A listed in Table I is a base case study. The statistical
sizing a distribution transformer that serves eight distribution data shown for Case A is used in the probabilistic model for the
secondary services. The artifact transformer is a single phase, transformer diversity factor (3, 4, 5), and the resulting probability
60 Hz, residential unit with voltage ratings 4160/240/120 volts. density of the diversity factor is shown in Fig. 2. The solid line
Historical measured data are used to obtain the statistical in- in Fig. 2 corresponds to the case of a 300 kVA distribution
formation in Table I. The examples shown were selected to transformer, and the dashed line corresponds to a 350 kVA
illustrate the probabilistic formulation in this paper, and the cases transformer. If the higher rating is used, the probability density
are considered ‘typical’. Note that the correlation coefficient ρ moves to the right (i.e., higher diversity factor due to the larger
is obtained from historical data (where available) in the cases values of X in (2)).
studied, Column B in Table I corresponds to the application of a higher
E ((X − μX ) (Y − μY )) rated transformer, namely 350 kVA. The probability density of
ρ = . the diversity factor is plotted for this case as the dashed line in
E((X − μX )2 (Y − μY )2 ) Fig. 3.
Note that the density characteristic appears to ‘lean’ to the
The three columns in Table I correspond to: (A) a base case right. This characteristic was observed by Chatlani et al. in
study; (B) consideration of a higher rated transformer; (C) a [13], and in that work, the authors fit a gamma probability
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 HEYDT: DISTRIBUTION TRANSFORMER LOADING: PROBABILISTIC MODELING AND DIVERSITY FACTOR 845

Fig. 4. Increase in the probability of operating a distribution transformer over

Fig. 2. The probability density of the transformer diversity factor d in Cases A rating with increase in the standard deviation of peak total load (data from
and B. The solid line is for the case A of a 300 kVA distribution transformer, and Example C).
the dashed line corresponds to an upgraded 350 kVA transformer denominated
as Case B. The probability density function was numerically calculated using
the model (3, 4, 5).
mostly in the tails of the density functions. Therefore one would
expect inaccurate agreement between the probability of overload
of the supply transformer since that probability is mainly the
integral of the probability density function in the zero to unity
region, i.e., in the tail of the density.

B. Increase of Transformer Life By Decreasing the Probability

of Overload
Column C in Table I corresponds to various levels of total
peak circuit loadings (i.e., levels of Y in (2)). Noting the in-
verse characteristic between Y and d, one would expect that the
probability of transformer peak load above the distribution trans-
former rating will decrease as the peak load standard deviation
decreases. Using the data in Column C of Table I, the probability
of transformer overload is calculated by integrating fd (d). The
results are shown in Fig. 4. The cited integration is done using the
ratio probability density model and not the gamma probability
density approximation. In this example, the integration of fd is
done numerically using the trapezoidal rule with Δd equal to
0.001 The portion of Fig. 4 to the right is in the region of lower
reliability and higher unit transformer loading, and the region to
Fig. 3. Probability density function of the diversity factor fd (d) (dashed line) the left is high reliability and lower per unit transformer loading.
in case B for the example study. Also shown is a best fit gamma density
function (dotted line) to illustrate the use of the gamma density rather than For this particular transformer and example, the probability of
the ratio density. The difference between the two plots is mainly in low and high over rating operation for σ Y below about 35 kVA is very small
probability density regions, but these are the regions that are used to calculate the and not conveniently visible in Fig. 4.
probability of transformer overload. The probability density function illustrated
by the dashed line was numerically calculated using the model (3, 4, 5). In view of the very small numbers for the probability of
transformer overload, it may be convenient to cite the probability
of ‘within rating’ as a measure of reliability and this may be
characterized as a ‘number of nines.’ For example, being within
density function to the statistics of d. The result of such a fit is rating 0.9999 fraction of the time is termed ‘four nines’, and
shown in Fig. 3. Since the gamma density is determined by two 0.99999 is five nines and so forth. A comparison of N9 values
scalar quantities as in (8), it is very easy to fit the best gamma in several engineering applications appears in [17]. Obviously,
probability density to the given data. However, note that the
gamma probability density and the ratio density in Fig. 3 differ N9 = −log (Pr (overload)) (10)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 846 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 38, NO. 2, APRIL 2023

TABLE II
A COMPARISON OF THREE LEVELS OF EV CHARGING∗

social, engineering and political decisions may, in fact, finally

accelerate EV growth. The two main types of EVs are: fully
Fig. 5. Increase in the probability of operating a distribution transformer over electric vehicles which are charged from a distribution supply;
rating with increase standard deviation of the total feeder load (example C). The
cited probability is depicted using ‘number of nines’ N9 . The right portion of and plug-in hybrid EVs which are partially charged from the
this characteristic (near N9 = 3, probability of operating within rating = 0.999). vehicle on-board internal combustion engine and partly from the
The left portion near N9 > 5 corresponds to the highly reliable case and the electric distribution system. There are a wide range of charging
probability of operation within rating of very close to unity.
power levels, as shown in Table II (adapted representative data
from [24]). Over the three indicated levels of charging, and
across the wide range of alternative EV technologies that are
where N9 is the ‘number of nines’, and ‘log’ is the common
presently commercialized (or proposed), there is an extensive
logarithm. When the example in Case C and Fig. 4 is replotted
spectrum and permutation of power levels, voltages, charging
in terms of the number of nines, Fig. 5 is obtained.
schedules, and details of the delivery of electric energy to EV
batteries. As indicated in Table II, the higher power levels of
C. Loss of Transformer Life and Hot Spot Temperature L2 chargers, and all L3 chargers are likely to entail low load
The discussion in this paper relates mainly to a probabilistic diversity owing to the use of dedicated distribution transformers.
‘high level’ model of distribution transformer loading. And that The concept of diversity loading of distribution transformers is
model is used at the service design procedure to render a design largely limited to L1 and low power levels of some L2 chargers.
strategy with acceptable life. Transformer life and performance, The L1 charger is really a ‘trickle’ charge that is at a low power
in general, are more complex issues that are largely connected level that is not very much different in temporal diversity than
with the insulation and cooling systems used in the transformer, other power distribution system loads. The L1 and lower power
and, specifically, hot spot temperature. Reference [18] provides rating L2 chargers are the focus areas in the discussion below.
background. The classical literature documents detailed ther-
mal models for transformers and their use in estimating life. A. Inclusion of EV Charging Into the Probabilistic
References [19], [20] are further documentation of the detailed Formulation
approach. Some relatively recently reported approaches use de-
Dixon et al. in [25] suggest a probabilistic formulation of
tailed models of specific load types, new transformer instrumen-
the diversity factor. References [26], [27] are a sampling of
tation (e.g., fiber optics), and innovative data processing appear
the possibilities of resulting load factor characteristics. Some
to give worthwhile results (as a sampling of these approaches,
actual EV deployment tests have been reported, and these have
references [21], [22], [23] are cited). While these methods may
been augmented with a number of simulation studies. These
give improved accuracy and information on transformer life,
studies give a picture of expected statistics of EV charging loads
they require commensurate detail relating to loading, investment
and concomitant distribution transformer loading. Simulation
in instrumentation, and construction of the transformer.
studies not based on probabilistic methods are reported in [28],
[29], and a system theoretic approach is reported in [30]. It is
VI. THE IMPACT OF ELECTRIC VEHICLE CHARGING ON fair to speculate as EV deployment progresses in large scale,
DISTRIBUTION TRANSFORMER LOAD DIVERSITY statistical data for L1 and smaller L2 transformer loads will
Perhaps the greatest change in expected future diversity fac- become more readily available. The relatively low cost of L1
tor loading characteristics of power distribution transformers type chargers (e.g., in the order of about 60 to 120 USD per
in the residential and commercial sectors is the proliferation kVA in 2021, in the 1.2 to 3.8 kW range [29]) suggests that
of electric vehicle (EV) charging. While EV loads have been this charger may be an attractive choice for some owners of
expected to grow rapidly for the past 100 years, contemporary passenger cars [31]. The faster charging of L2 chargers suggests

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 HEYDT: DISTRIBUTION TRANSFORMER LOADING: PROBABILISTIC MODELING AND DIVERSITY FACTOR 847

that many applications will entail these higher power level TABLE III
EXAMPLE D, INCLUDING EV CHARGING
units. Straightforward calculation of the statistical moments of
transformer loads is assumed here.

B. Accommodation of a Bimodal Transformer Loading Model

In the planning stage, it is assumed that the distribution
engineer will have an estimate of the connected load, peak
demand, diversity factor, and mean and standard deviation of
distribution transformer loads, including EV charging. Assum-
ing L1 chargers and single phase L2 units, estimates suggest
that EV charging will occur mid-day at commercial locations
and after the business day at residential locations. Studies reveal
that transformer loads may exhibit bimodal characteristics (i.e.,
two peaks daily). The bimodality occurs both temporally (i.e.,
two peak demands occurring at different times) and bimodal
occurrence at two different power demand levels corresponding
to the dispersion of EV loads ‘plugged in’. Directly, this im-
plies that statistical modeling of transformer loads as normal or
gamma probability densities is not possible since normal and
gamma densities are unimodal. The ratio probability density in
(3, 4, 5) allows bimodal probability as shown (and illustrated)
by Pham-Gia, Turkkan, and Marchand in [14]. The cited authors
suggest a test of assumed load statistics for bimodal versus uni-
modal probability density. Their test is based on two statistical
parameters,
 
1 μx ρμy μy
a = − b = (11)
1−ρ 2 σ x σ y σy
These two parameters may be used in a fairly complicated test
for bimodality as shown in [14]. Or, the planning engineer may
simply substitute estimates of mean and variance of the loads,
and an estimate of the correlation coefficient ρ between x and y
in (1); then numerical calculation of the statistical model (3, 4,
5) will show whether the diversity factor d exhibits bimodality.

C. An Example of the Bimodal Density Model and Calculation

The indicated bimodal probability density function occurs
without any special attention directly from the model (3, 4, 5).
Table III shows a test case to augment the previously cited test Fig. 6. Probability density of the total transformer load diversity factor, Case
D. In this case, the probability density function is bimodal. The probability
cases, and this additional case is denominated as case D. Data density function was numerically calculated using the model (3, 4, 5).
for this example come from a distribution feeder in Turkey in
which a substantial portion of the primary feeder load is due to
EV charging using mixed L1 and L2 chargers. In Case D, about
X and Y in (1) was very high, calculated as 0.95 as indicated in
16% of the energy passing through the substation is due to EV
Table III.
charging at ‘auto parks’. In this example, the diversity factor of
the entire feeder is used to size the substation transformer, a 600
kVA unit in this case. The Turkish study is documented further D. Observations Relating to EV Charging Loads
in [32]; however, since some data are not specified in that study, The main observation from the Case D example is that the ratio
the substitutions /assumptions were made as shown in Table III. probability density model is suitable for cases of EV charging
The unitless bimodality test parameters a and b in Case D are loads in which bimodal probability density occurs. The use of
2.643 and 0.5522 respectively. load diversity in the case of high EV charging is mainly for
Results for Case D are shown in Fig. 6. Note the low levels of L1 and single phase L2 chargers. In the United States, there
diversity factor d in Fig. 6 (e.g., in the range 1 < d ≤ 4). This is are about 43000 public EV charging stations (2021) and about
representative of what one may expect for EV charging loads. 120000 charging ports, according to U.S. Department of Energy
As far as could be determined from the captured test data at the data [32], and so there appears to be an application for diversity
automobile charging station, the correlation coefficient between loading in this load sector. The appearance of EV charging

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 848 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 38, NO. 2, APRIL 2023

lowers the diversity factor (an observation which agrees with normal random variables. The resulting probability density is
[31]), and one would expect that diversity loading applications termed a ratio probability density and this may be used as a
are limited to the lower EV charging power levels where the consistent consideration in sizing distribution transformers that
circuit diversity factors are the highest. Additional comments serve several individual loads. The relationship between the
on the EV charging impact on distribution loading appear in the statistical model of the diversity factor and expected transformer
literature (e.g., [29], [33], [34], [35]). reliability is discussed, and examples are shown.
Note that at lower power EV charging levels (e.g., L1 charg- It is expected that the proliferation of electric vehicle charging
ing) are likely to use controlled rectifiers, and these devices may loads will result in higher system wide load factor but lower in-
have harmonic load current impact on the transformer load. dividual distribution transformer diversity factor. This will need
Transformer core losses can be high owing to harmonic load to be accounted in sizing distribution transformers. Diversity
currents, but the IEEE C57.110 [36] requirements are conserva- factor sizing distribution transformers for services to electric
tive as indicated by laboratory tests [37]. And the contributory vehicle charging loads appears practical for Level 1 charging and
heating due to rectifier EV charger loads is not likely to be a single phase, low power Level 2 charging. The ratio probability
significant factor in the application of diversity factor sizing density model has the ability of modelling bimodal (two peaks)
of distribution transformers. Higher power EV charging (e.g., probability density.
L2 and L3) is likely to utilize three phase and pulse width
modulation technologies in the AC/DC conversion, and these
technologies are unlikely to result in excess transformer heating. REFERENCES
Nonetheless, these issues are further motivation to use diversity [1] H. B. Gear, “Diversity factor,” Proc. Amer. Inst. Elect. Engineers, vol. 29,
factor sizing in parallel (i.e., in addition to) other sizing tech- no. 8, pp. 1365–11374, Aug. 1910.
niques in the presence of high percentage of nonlinear loads. [2] P. Billingsley, Probability and Measure, 3rd ed. New York, NY, USA:
Wiley, 1995.
[3] A. Sargent, R. Broadwater, J. Thompson, and J. Nazarko, “Estimation
VII. THE CONNECTION BETWEEN DIVERSITY FACTOR AND of diversity and kWh-to-peak-kW factors from load research data,” IEEE
LOAD FACTOR Trans. Power Syst., vol. 9, no. 3, pp. 1450–11456, Aug. 1994.
[4] R. Broadwater, A. Sargent, A. Yarali, H. Shaalan, and J. Nazarko, “Es-
The load factor, L, of an electric power system is given by timating substation peaks from load research data,” IEEE Trans. Power
Del., vol. 12, no. 1, pp. 451–4456, Jan. 1997.
Average load [5] W. H. Kersting, Distribution System Modeling and Analysis. Boca Raton,
L = . (12) FL, USA: CRC Press, 2002.
P eak load [6] H. L. Willis, Power Distribution Planning Reference Book. New York, NY,
USA: Marcel Dekker, 2004.
While the load factor bears some resemblance to the inverse [7] Westinghouse Central Station Engineers, Electrical Transmission and
of the diversity factor, note that L is generally used in connection Distribution Reference Book. Pittsburgh, PA, USA: Westinghouse Electric,
with large scale systems including residential, commercial, and 1964.
[8] D. H. O. McQueen, P. R. Hyland, and S. J. Watson, “Monte Carlo
industrial loads whereas d is usually applied only to a single simulation of residential electricity demand for forecasting maximum
feeder for residential loads. The temporal and size diversity of demand on distribution networks,” IEEE Trans. Power Syst., vol. 19, no. 3,
residential loads is generally greater than loads in other sectors, pp. 1685–1689, Aug. 2004.
[9] J. Dickert and P. Schegner, “Residential load models for network planning
and for this reason, d usually lies in the range (2.4 < d < 10) purposes,” in Proc. Int. Symp. Modern Electric Power Syst., Wrocław,
while L is often in the range (0.4 < L < 0.6). The comparison to Poland, Sep. 2010, pp. 1–6.
be made here is between d and 1/L. A further difference between [10] X. Zhu and B. Mather, “Data-driven load diversity and variability modeling
for quasi-static time-series simulation on distribution feeders,” in Proc.
the load factor and diversity factor relates to the time span over IEEE Power Energy Soc. Gen. Meet., Atlanta, Georgia, Aug. 2019, pp. 1–5.
which Eqs. (2) and (11) are applied. Since the diversity factor is [11] Recommended Practice for Electric Power Systems in Commercial Build-
used in sizing transformers over their life, the time span for the ings, IEEE Std. 241-1990, Piscataway, NJ, USA, 1990.
[12] Standard Requirements for Liquid-Immersed Distribution Substation
data used in estimating d is over the lifespan of a transformer. Transformers, IEEE Std. C57.12.36-2007, Piscataway, NJ, USA.
On the contrary, the load factor of a large scale power system is [13] V. P. Chatlani, D. J. Tylavsky, D. C. Montgomery, and M. Dyer, “Statistical
often quoted as a measure of the efficiency or efficacy in utilizing properties of diversity factors for probabilistic loading of distribution
transformers,” in Proc. North Amer. Power Symp., Tempe, AZ, USA,
system assets. The load factor of a system may be quoted over Oct. 2007, pp. 555–561.
a season or perhaps a calendar year. [14] T. Pham-Gia, N. Turkkan, and É. Marchand, “Density of the ratio of two
Experiments with research data for the statistics of average normal random variables and applications,” Commun. Statist. - Theory
Methods, vol. 35, no. 9, pp. 1569–1591, Sep. 2006.
system load and peak load in (11) indicate that the statistical [15] J. W. Pearson, S. Olver, and M. A. Porter, “Numerical methods for
model in (3, 4, 5) accommodates load factor modelling very the computation of the confluent and Gauss hypergeometric functions,”
well as a ratio density. The numerical values in modelling L are Aug. 2018, arXiv open-access repository, [Online]. Available: https://
arxiv.org/pdf/1407.7786
quite different from those in modelling d however. [16] M. Dong, “A data driven long term dynamic rating estimating method for
power transformers,” IEEE Trans. Power Del., vol. 36, no. 2, pp. 686–697,
VIII. CONCLUSION Apr. 2021.
[17] G. T. Heydt, “Improving distribution reliability (the N9 problem) by the
The main conclusion of this paper is that the diversity factor of addition of primary feeders,” IEEE Trans. Power Del., vol. 19, no. 1,
pp. 434–435, Jan. 2004.
distribution transformer loading may be modeled as a probabilis- [18] M. T. Isha and Z. Wang, “Transformer hotspot temperature calculation
tic quantity, and the diversity factor of distribution transformer using IEEE loading guide,” in Proc. Int. Conf. Condition Monit. Diagnosis,
load is conveniently modelled as the ratio of correlated bivariate Beijing, China, 2008, pp. 1017–1020.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.
 HEYDT: DISTRIBUTION TRANSFORMER LOADING: PROBABILISTIC MODELING AND DIVERSITY FACTOR 849

[19] N. Saleh, N. Azis, J. Jasni, M. Zainal A. Ab Kadir, and M. Aizam Talib, [31] J. Waddell, M. Rylander, A. Maitra, and J. A. Taylor, “Impact of plug in
“Prediction of a transformer’s loading and ambient temperature based on electric vehicles on Manitoba Hydro’s distribution system,” in Proc. IEEE
SARIMA approach for hot spot temperature and loss of life analyses,” in Elect. Power Energy Conf., 2011, pp. 409–414.
Proc. IEEE Int. Conf. Properties Appl. Dielectric Mater., Malaysia, Johor [32] I. G. Tekdemir, D. Gunes, B. Alhoyaci, and M. Sengul, “A probabilistic
Bahru, 2021. approach for evaluation of electric vehicles’ effects on distribution sys-
[20] R. E. Rood, “Method for estimating the thermal life expectancy of dis- tems,” in Proc. 4th Int. Conf. Elect. Electron. Eng., Dubai, UAE, May,
tribution transformers,” IEEE Trans. Power App. Syst., vol. 82, no. 69, 2017, pp. 143–1147.
pp. 1015–1018, Dec. 1963. [33] U.S. Department of Energy, “Maps and data - electric vehicle registrations
[21] C. Gezegin, O. Ozgonenel, and H. Dirik, “A monitoring method for by state,” WA DC, USA, 2021, [Online]. Available: https://afdc.energy.gov
average winding and hot spot temperatures of single phase, oil immersed [34] J. Antoun, M. E. Kabir, B. Moussa, R. Atallah, and C. Assi, “Impact
transformers,” IEEE Trans. Power Del., vol. 36, no. 5, pp. 3196–3203, analysis of level 2 EV chargers on residential power distribution grids,” in
Oct. 2021. Proc. IEEE 14th Int. Conf. Compat., Power Electron. Power Eng., Setubal,
[22] R. Duan, “Real time hotspot tracing and model analysis of a distributed Portugal, vol. 1, 2020, pp. 523–529.
optical fiber sensor integrated power transformer,” IEEE Access, vol. 10, [35] S. Habib, M. M. Khan, K. Hashmi, M. Ali, and H. Tang, “Comparative
pp. 57242–557254, 2022. study of electric vehicles concerning charging infrastructure and power
[23] Y. Deng et al., “A method for hot spot temperature prediction of a 10 kV oil levels,” in Proc. Int. Conf. Front. Inf. Technol., Islamabad, Pakistan, 2017,
immersed transformer,” IEEE Access, vol. 7, pp. 107380–1107388, 2019. pp. 327–332.
[24] SAE electric vehicle and plug-in hybrid electric vehicle conductive charge [36] Recommended practice for establishing liquid-immersed and dry-type
coupler, SAE Standard J1772, Warrendale, PA, USA, Jan. 2010. power and distribution transformer capability when supplying nonsinu-
[25] J. Dixon, I. Elders, and K. Bell, “Electric vehicle charging simulations on soidal load currents, IEEE Standard C57.110-2018, Piscataway, NJ, USA,
a real distribution network using real trial data,” in Proc. IEEE Transp. Oct. 2018.
Electrific. Conf. Expo, Asia-Pacific, 2019, pp. 1–7. [37] A. W. Galli and M. D. Cox, “Temperature rise of small oil-filled distribution
[26] L. A. Pecorelli, J. F. M. P. Peres, A. G. Garcia Lima, and W. Braga Pereira, transformers supplying nonsinusoidal load currents,” IEEE Trans. Power
“Electric vehicle charging diversity factors,” in Proc. IEEE 14th Int. Conf. Del., vol. 11, no. 1, Jan. 1996, pp. 283–291.
Compat., Power Electron. Power Eng., 2020, pp. 397–401.
[27] U. C. Chukwu, O. A. Nworgu, and D. O. Dike, “Impact of V2G penetration
on distribution system components using diversity factor,” in Proc. IEEE
SOUTHEASTCON, Lexington, KY, USA, Mar. 2014, pp. 1–8.
[28] L. Xue and J. Xia, “Simulator to quantify and manage electric vehicle Gerald Thomas Heydt (Life Fellow, IEEE) is from Nevada. He received the
load impacts on low-voltage distribution grids,” Washington, DC, USA: B.E.E.E. degree from The Cooper Union, New York, NY, USA, and the M.S.E.E.
Technical note, World Resources Institute, Dec. 2020, [Online]. Available: and Ph.D. degrees from Purdue University, West LaFayette, IN, USA, in 1965
https://www.wri.org/publication/ and 1970, respectively. He has industrial experience with Commonwealth Edison
[29] J. M. Sexauer, K. D. McBee, and K. A. Bloch, “Applications of probability Company, Chicago, IL, USA, and E. G. and G., Mercury, NV, USA. Most of his
model to analyze the effects of electric vehicle chargers on distribu- academic career has been with Purdue University, Arizona State University,
tion transformers,” IEEE Trans. Power Syst., vol. 28, no. 2, May 2013 Tempe, AZ, USA, and the University of Nevada – Reno, Reno, NV, USA.
pp. 847–8854. In 2018, he retired from Arizona State University, where he was a Regents’
[30] S. Ling, Y.-J. Zhou, and X.-Y. Yu, “Analysis of multi-dimensional factors Professor and Professor of Advanced Technology. He is currently a Regents
sensitive to comprehensive utilization of power distribution equipment,” Professor Emeritus with ASU and working on infrastructural Projects in Latin
in Proc. 7th Int. Forum Elect. Eng. Automat., Hefei, China, Sep. 2020, America and the accreditation of engineering educational programs. He is also
pp. 438–444. a Member of the National Academy of Engineering.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on August 28,2025 at 12:22:52 UTC from IEEE Xplore. Restrictions apply.