Sensitivity Analysis of Electric Vehicle Impact

on Low-Voltage Distribution Grids

Jochen Stiasny∗ , Thierry Zufferey† , Giacomo Pareschi‡ , Damiano Toffanin§ ,
Gabriela Hug† and Konstantinos Boulouchos‡
∗
ETH Zurich, Switzerland, jochen.stiasny@alumni.ethz.ch
†
Power Systems Laboratory, ETH Zurich, Switzerland, thierryz@eeh.ee.ethz.ch
‡ Aerothermochemistry and Combustion Systems Laboratory, ETH Zurich, Switzerland, pareschi@lav.mavt.ethz.ch
§ Adaptricity AG, Zurich, Switzerland, dtoffanin@adaptricity.com

Abstract—The presented work identifies the dominating in- lead to under- or over-dimensioning of the grid components
fluencing factors in electric vehicle (EV) modelling on low- and potentially to higher costs for grid operators.
voltage distribution grids to establish guidance for reliable impact
assessments of increasing EV penetration. Seven aspects are
distinguished with respect to the modelling of the load of EVs
that influence the flows and voltages in the grid. For each of
these aspects sensitivity analyses are carried out by running
power flow simulations in a Monte-Carlo fashion to account
for the stochasticity in the model parameters. The impacts are
analysed using a variety of metrics including transformer and line
loadings. The highest sensitivities are observed for the number
of vehicles in the grid, the used charger power rating and the
modelling of driving patterns. The grid configuration as well
as locally higher EV shares gain significance for line loading Fig. 1. Transformer load duration curve of household loads in January.
assessments. Car modelling and people’s charging behaviour play
minor roles.
A common approach to evaluate these peak power increases
Index Terms—electric vehicle, low-voltage distribution grid, caused by EVs is based on measured or modelled charging
Monte-Carlo simulation, sensitivity analysis load profiles which are added to the existing loads [3], [4].
However, the rapid developments in the EV market question
the future applicability of these patterns. A main limitation
I. I NTRODUCTION concerns the up-rise of pure battery EVs in contrast to plug-in
To fulfil the commitments of the Paris Agreement 2015 [1] hybrid EVs. Moreover, the ever increasing battery capacities
the transportation sector is obliged to drastically reduce its and hence EV driving ranges as well as the EV usage
CO2 emissions. The increased efforts towards electric vehicles habits need to be considered in future assessments. Finally,
(EVs) show a route forward to achieve a decarbonisation. As the emerging higher charger power ratings can substantially
a consequence, the power grid infrastructure is expected to change the charging patterns [2].
face increased loading. Private and uncontrolled charging at Successful attempts to model charging profiles often employ
home, presumably the most common charging method [2], a modular approach as reviewed in [5]. They determine
impacts especially low-voltage distribution grids. Not only the energy and power demand based on travel data, EV
the additional energy demand poses a challenge, but the peak specifications, and available charging infrastructure [6]–[12].
power increase emerges as the crucial factor for grid planning Such a modular approach offers the advantage of flexible
and risk assessment. Fig. 1 depicts the load duration curve modelling for future assessments and it allows for stochastic
(LDC) of the transformer for the household load in the selected modelling in form of Monte-Carlo simulations [9], [10].
grid. Although the transformer peak loading (∼55%) in the However, the sensitivities of used modelling assumptions, e.g.
studied grid is well below operational limits, the LDC still EV specifications or people’s charging behaviour, have not
demonstrates how rare load cases can substantially increase been comprehensively investigated yet. While some studies
the required operating range. The effect of EVs on the LDCs explore the influences of varying single factors such as pen-
and especially on peak loads requires therefore a thorough etration levels or charger ratings [11], [12], the presented
modelling and analysis to avoid ill-informed decisions that work integrates seven influencing factors, here referred to as
dimensions, into a single framework. These dimensions split up
EV charging into a handful of tangible characteristics, which
This research is part of the activities of the Swiss Centre for Competence could be expressed as questions like ‘when do people charge’
in Energy Research on the Efficient Technologies and Systems for Mobility
(SCCER Mobility), which is financially supported by the Swiss Innovation or ‘how often do people charge’. By closely investigating each
Agency (Innosuisse - SCCER programme). dimension and the interaction between them one can obtain

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
a better understanding of their sensitivities with respect to TABLE I
the overall grid load. This allows to break down complex OVERVIEW OVER EV MODELLING DIMENSIONS AND MODELS .
models into the influences they have on the basic charging Dimension Default model Additional models
characteristics. Combining this insight with the sensitivities
found in this paper will improve the quality and reliability of independent GMM (A1)
EV modelling since overly complex or simplistic models can Driving patterns MZMV sampling joint GMM (A2)
be identified. Furthermore, this work improves the ability to availability based (A3)
judge whether assumptions and results may generalise to other EV type sales based average sales based sampling (B1)
environments or not. In this work, the framework incorporates uniform threshold (C1)
Charging behaviour Gaussian threshold
a real-world case in Switzerland by using a residential low- always charging (C2)
voltage grid model, metered household loads, and a large 25% 11kW (D1)
Charger type 0% 11kW 50% 11kW (D2)
scale mobility survey. In this set-up, multiple models, which
100% 11kW (D3)
integrate regularly used modelling techniques and assump-
40% (E1)
tions, are compared against one another for each dimension.
Penetration level 20% 60% (E2)
The resulting assessment of the dimension’s relevance may
80% (E3)
provide guidance for future work on EV integration and system
mildly clustered (F1)
planning. EV placement even
strongly clustered (F2)
The remainder of the paper is structured as follows: Sec-
Grid configuration radial meshed (G1)
tion II introduces the modelling of the seven dimensions,
Sec. III presents the simulation framework and Sec. IV dis-
plays the results from which Sec. VI draws the conclusions.
A. Available data
II. M ODELLING FRAMEWORK
The modelling framework follows a bottom-up approach, Segment with X buses and X-1 lines in a
as depicted in Fig. 2. First, the charging load of individual X (Y) buses purely radial configuration
T
cars over time is generated by determining the charging Z customers
Z customers (potential EV owners), placed at Y buses.

demand from driving patterns and the type of EV. People’s

A B
charging behaviour and the used charger type then govern
the resulting load of individual vehicles in the examined grid. 9
29 (15) buses 36 (17) buses 8 (3) buses 41 (20) buses 26 (11) buses
The penetration level and EV placement specify how many
C
cars are used as well as their connection points in the grid. 142 63 21 128 64
The total EV load is added on top of the base load, i.e. the 7
metered household loads in the area, and then added to the 11 (4) buses 14 (8) buses 15 (9) buses
grid in a radial or meshed configuration.
23 60 68
In Sec. II-A the underlying data sources are introduced,
while Sec. II-B to II-H provide details for each dimension. Fig. 3. Low voltage grid with customer location.
Table I gives an overview over all models.
For the study, a low-voltage distribution grid operated by
IWB, the distribution system operator of the City of Basel,
Total grid load Scenario realisation
was chosen. Fig. 3 shows its main topological characteristics.
Total EV load The transformer rated at 630 kVA connects the 400V level to
the higher voltage level (11kV). According to the geographical
Single EV load Grid configuration information system (GIS) of the grid model, the 196-bus
Driving pattern system hosts in total 585 households at 89 connection points
which are eligible as EV connection points in this study. In
Charger type EV type
AC power flow general, the main strand lines are rated at about 350A and the
Charging behaviour household connections at 140A. In addition, a smart meter
data set with 15-minute resolution for the considered grid is
EV placement Penetration level used as base load. Since the measured base load reaches its
Metric analysis
highest values in January the work focuses on this month only.
Base load
Fig. 1 shows the LDC of the base load.
The outcome of the 2015 Swiss national travel survey
‘Mikrozensus Mobilität und Verkehr’ (MZMV) [13] forms the
Scenario statistics from realisation metrics
basis for the driving pattern modelling. The data set consists
Fig. 2. Modelling framework used to assess the impact of EVs on distribution of about 40’000 1-day travel surveys, 20’000 of which contain
grids, all dimensions marked in bold. information about car movements. The data set is processed to

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
yield a table of round trips that start and end at home. Stops of the most sold EV models in Europe in 2018 which forms
during a round trip, e.g. for shopping, are possible but are the basis for the analysis [15], [16]. At a later stage, increases
not included in the modelling. Between two round trips the in battery capacity by 25% and 50% as well as consumption
car is per definition at home. Each trip is associated with a variations of ±10% are considered.
departure time, an arrival time, a travel distance and the day of
the week on which it occurred. While each weekday is treated TABLE II
separately, seasonal variations as well as geographic properties E NERGY CONSUMPTION AND BATTERY CAPACITIES OF THE 10 MOST
SOLD EV S IN E UROPE IN 2018 [15], [16]
are not considered since further splits of the data set would
weaken the representativity of the results. As the survey is Model
Consumption Battery capacity Sales
primarily based on combustion cars, it has to be assumed that [kWh/100km] [kWh] [-]
the mobility behaviour does not change with the switch to EVs Nissan Leaf 16.5 40 40609
until similar surveys with sufficient EVs exist. Renault Zoe 15.7 41 38538
BMW i3 16.4 42.2 24432
VW e-Golf 16.8 35.8 21252
B. Driving pattern Tesla Model S1 18.4 100 16682
The driving pattern models people’s car usage. The first Tesla Model X1 21.4 100 12694
model is MZMV sampling, which directly uses trips from the Hyundai IONIQ 14.4 30.5 9605
Smart fortwo 15.9 17.6 8688
MZMV data set by concatenating them into a month-long Kia Soul EV 17.1 33 6641
time series. In contrast, the following three models rely on Jaguar i-Pace 22.3 90 6319
a parametric description that aims at resembling the nature Fleet average 16.9 49.4 -
of the MZMV data set. Such models can be used if no
travel survey is available. The independent GMM model fits
a Gaussian mixture model (GMM) to the probability density D. Charging behaviour
functions (PDF) of the departure time in the MZMV data set
While driving patterns define the times a car is available
and another one to the arrival time. The joint GMM model
for charging, the charging behaviour defines below which
approximates the joint PDF of the two time variables. In
threshold of state of charge (SOC) people start to charge.
contrast, the availability-based model uses the share of cars
A SOC of 0% represents a fully depleted and 100% a
at home, i.e. the car availability. The share varies during
fully charged battery. The first of three models follows the
the day and from its rate of change transition probabilities
analysis in [17] by approximating people’s personal threshold
between being and not being at home are calculated. Based
by a Gaussian distribution N (45%, 19%) truncated between
on the transition probabilities car trips are sampled. Fig. 4
[10%, 100%]. As a second model, a uniform threshold distri-
illustrates for all four models the resulting PDFs of the arrival
bution between [10%, 100%] is tested. In the always charging
time as well as the availability during the day. For the latter
model everybody starts to charge as soon as the SOC drops
three parametric models a logarithmic normal distribution
below 100%, i.e. people charge after every drive. Regardless
determines the driving distance for each trip. All models,
of the model, the charging process is assumed to commence
including the fitted parameters, are described in detail in [14].
upon arrival without delay to avoid complexity. The analysis
of the survey ”My Electric Avenue” [18] indicates that about
MZMV sampling Independent GMM.
Driving pattern model 65% of EV owners charge within 15 minutes upon arrival.
Joint GMM Availability−based

0.5 100% E. Charger type

Share of cars at home
Arrival time density

0.4
75% Two charger types with power ratings of 3.7kW and 11kW
0.3 are considered for home charging. While the power ratings
50%
0.2 themselves are known, their share of installation is subject to
0.1
25% the sensitivity analysis. Starting with only 3.7kW chargers,
i.e. 0% 11kW chargers, being installed, the share of 11kW
0.0 0%
0 6 12 18 24 30 stations is increased to 25%, 50% and ultimately to 100%.
Hour of the day For all chargers active power factor correction, hence a power
Fig. 4. Availability curve (dashed) and arrival time distribution (solid) of factor of unity, is assumed [19].
driving pattern models on Mondays.
F. Penetration level
C. Electric vehicle type The number of EVs in the grid is expressed by four
penetration levels which are defined as the share of households
The EV characteristics of interest are the car’s battery capac-
that own an EV. Starting from the load case with no EVs,
ity and the energy consumption per kilometre. Two approaches
penetration levels of 20%, 40%, 60% and 80% are tested.
are tested: 1) modelling a diverse fleet with different vehicle
classes [6], [11], [12] or 2) assuming a single average vehicle 1 All models in 100kWh configuration for simplicity and in light of overall
using the sales-based average. Table II shows a compilation increasing battery capacities.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
G. Electric vehicle placement EVs added. The Box-Whisker-Plots (1,5 IQR) represent the
The placement of EVs may become critical if many cars are distribution of the 400 scenario realisations. This first analysis
charged within a neighbourhood connected by the same line to suggests that, not surprisingly, variations in the penetration
the transformer. Therefore, an evenly spread placement, where level and the charger type clearly cause the biggest impact,
each household’s probability of owning an EV is equal across further investigated in Section IV-A.
the grid, is compared with two levels of EV clustering. While

Charging behaviour
Reference scenario
the mild clustering results in some local hotspots, the strong

EV penetration
Driving pattern

EV placement
Charger type

Grid model
clustering places all EVs directly next to each other.

Car type
H. Power grid configuration
Three circuit breakers in the grid (like element C in Fig. 3) 80%

Transformer peak load increase

Absolute transformer peak load

can be switched to change between a radial and a meshed grid 40%

configuration which are both used by the grid operator.

70%
III. S IMULATION
20%
This section treats the combination of dimensions with the
respective models and the metrics employed for the analysis of 60%
the power flow results. In the following, a scenario is referred
0%
to as the selection of a single model for each dimension. For
0 A1 A2 A3 B1 C1 C2 D1 D2 D3 E1 E2 E3 F1 F2 G1
each dimension a default model is selected as indicated in Single dimension variation
Table I. Starting from a reference scenario, which is based
purely on default models, the variation of a single dimension at Fig. 5. Impact of single dimension variations on the transformer peak load,
values relative to the scenario without EVs (dashed line) on the left y-axis
the time provides a first indication for the dimensions’ impacts. and in absolute terms on the right y-axis, see Table I for abbreviations.
Subsequently, scenarios with multiple variations are simulated.
All variations are stated explicitly alongside figures, for all
other dimensions the default models apply. A. Penetration level and charger type
To incorporate the stochastic nature of some models, each Several combinations of these two dimensions are simulated
scenario is simulated 400 times with random samples of the to investigate their interaction. The resulting transformer peak
selected models. For each of these 400 scenario realisations load increases are shown in Fig. 6. In a first analysis it appears
a time-series power flow computation with a 15 minute that the transformer peak load increases proportionally to the
resolution is performed with MATPOWER [20]. The results penetration level, i.e. to the number of EVs, hence suggesting
of each realisation are then evaluated with respect to a set the lack of a correlation between the two dimensions.
of metrics. This Monte-Carlo style approach is performed
for each scenario separately and yields the basis for the Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger
sensitivity analysis in which the metrics of the realisations are
Transformer peak load increase

120%

Absolute transformer peak load

120%
displayed. Since the simulation of a single scenario requires 100% 110%
about five hours on a standard machine, multiple scenarios 80% 100%
were simulated in parallel on a 36-core machine with 479 GB 90%
60%
RAM. 80%
40%
The primary metrics of interest are the relative changes in 70%
20%
the loading of the transformer and lines. Therefore, relative 60%
LDC changes for each scenario realisation are computed at 0%
20% 40% 60% 80%
selected percentiles of the LDC according to EV penetration level

LDCwith EVs − LDCwithout EVs Fig. 6. Penetration and charger type scenarios.
relative LDC change = . (1)
LDCwithout EVs
Besides, the aggregated charging load over time is evaluated. By scaling these values by the EV penetration (the number
of cars) the additional power demand per car is calculated.
IV. R ESULTS The upper plot in Fig. 7 depicts this re-scaling. At first, it
Fig. 5 depicts the results of the simulations varying a single again appears as if the two dimensions influence the results
dimension with respect to the default case to enhance the independently of each other. However, when analysing only
intuition of each dimension’s impact. The abbreviation used the added EV load (lower plot in Fig. 7) a different picture
for the varied models can be found in Table I. The relative emerges. Firstly, an aggregation effect occurs, meaning that
transformer peak load increase is indicated on the left y-axis with an increasing number of EVs the average load per car
while the absolute values are plotted on the right one. The decreases since a superposition of the peak load of all EVs,
dashed line marks the transformer peak load (55.4%) with no i.e. simultaneous charging of all EVs, becomes statistically less

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
likely. Secondly, the deviation between the two plots differs Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger
depending on the charger type. Two effects are at play. On

Transformer peak load increase

Absolute transformer peak load

100% 110%
the one hand, higher charger loads cause not only higher peak
EV loads (equivalent to the EV load per car) but also shift the 80% 100%

peak more towards the base load peak around 6 o’clock, hence 60% 90%
EV and base load peak coincide more. On the other hand, the
80%
penetration level increase itself causes a mere scaling of the 40%

EV load including the mentioned aggregation effect without 20%

70%

shifting the timing. However, the peak of the total load (sum 60%
0%
of base and EV load) is affected in timing which counters the
Gaussian threshold Uniform threshold Always charging
aggregation effect. Fig. 9 shows an exemplary case of the base Charging behaviour
and EV load against time. The plot is discussed in detail in
Fig. 8. Charging behaviour impact for 60% EV penetration.
Sec. IV-C related to the driving pattern results.

Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger C. Driving patterns
Based on transformer peak load Since the driving patterns and more specifically people’s ar-
1.50 rival times govern the charging period, a comparison between
1.25
the distribution of EV charging load on Mondays and Satur-
1.00
0.75
days illustrates the sensitivity with respect to driving patterns
Additional load per EV [kW]

0.50 well. On both days the number of trips (including no trips)

0.25 per person is very similar but the arrival time distributions
0.00 differ notably. This leads to a wider spread charging period on
Based on peak EV charging load
Saturday and lower peak loads, especially for higher charging
1.50
powers as depicted in Fig. 9.
1.25
1.00
0.75 Realisation 0−100% 10−90% 20−80% 20%
EV penetration
0.50 percentiles 30−70% 40−60% 60%
0.25
Monday Saturday
0.00
20% 40% 60% 80%
Penetration level 400

0% 11kW charger
Fig. 7. Additional peak load normalised by the number of cars. 300

200

B. Charging behaviour
100
When comparing different charging behaviours, the case of
Load [kW]

people charging after every trip creates a significantly higher 0

peak load increase than the other two models as shown in

400
Fig. 8. In addition, covering the daily average energy use

100% 11kW charger

(MZMV average driving distance ∼30km, average energy 300
consumption from II-C ∼0.17kWh/km) with a 11kW charger
200
requires around half an hour of charging. The longer charging
time of slower chargers leads to more overlapping of people’s 100
charging periods. The resulting higher coincidence factor
causes the relatively strong impact of the always charging 0
0 6 12 18 24 0 6 12 18 24
scenario for slow chargers; its peak load increase amounts to Hour of the day
nearly double the one of the other two charging behaviours.
Fig. 9. Distribution of realisations of EV load (coloured areas) including
For higher charger power ratings the peak load increase is their median (solid black line), the base load (dashed line) and the resulting
more similar across the three behaviours. The Gaussian and median of the total load (solid coloured lines) during a day.
uniform threshold models show no major difference amongst
each other. People charge in both cases every other or third With this example in mind, it becomes apparent why the
day, hence less often but over a longer period. This leads to modelling of driving patterns needs special attention. The
lower coincidence factors than for the always charging model. proposed GMMs match the peak load metric obtained with
The slightly higher increase seen for the uniform distribution is the MZMV sampling model within small margins, illustrated
explained by a bit more frequent charging and thereby higher in Fig. 10. Revisiting the availability curves of both models
coincidence factors. in Fig. 4, the joint GMM model comes closer to the MZMV

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
sampling model and hence more cars are available around E. Grid configuration
noon. However, the sharp increase around the evening is Up to this point the analysis focused on the transformer
captured better by the independent GMM model. The model peak load. The grid configuration as well as the placement
choice is therefore a trade-off and is dependent on the timing of EVs in the grid do not affect the total load (except for
of the critical base load. The availability-based model even losses) nor the transformer-based metric. However, the peak
further pronounces the arrivals at noon at the cost of the load changes strongly for lines A and B in Fig. 3 when
accuracy of the evening arrival times. Overall, a carefully switching from the radial to the meshed configuration. The
chosen model yields satisfactory results. latter creates a levelling effect such that the previously higher
loaded line A sees lower peak line loading at the expense of
Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger a higher peak load for line B, as shown in Fig. 12. Not only
for this case but the grid overall, the meshed configuration
Transformer peak load increase

Absolute transformer peak load

80% 100% provides a beneficial line loading situation. Even very high
90%
penetration levels of 80% can be accommodated without major
60%
peak load increases when considering the power line ratings
80%
40% for the meshed grid. As mentioned before, the transformer
70% peak load still increases substantially under such a scenario.
20%
This emphasises the importance of assessing the affected grid
60%
0%
elements separately and identifying the critically loaded ones.
MZMV sampling Independent GMM Joint GMM Availability−based
Driving pattern models
Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger

Fig. 10. Impact of driving pattern modelling on the transformer peak load Radial grid configuration Meshed grid configuration
for a 60% EV penetration.
100%

D. Electric vehicle type 75%

Line A
The car type modelling influences the battery discharge and 50%
Absolute line peak load

hence the charging frequency and duration of each individ- 25%

ual EV. However, Fig. 11 shows that the model differences 0%

between a diverse EV fleet and a single average car barely
influence the EV loads and the transformer peak load metric. 100%

Increases in battery capacity and variations of the EV’s energy 75%

Line B
consumption yield notable differences, especially for higher 50%
charging power. Generally speaking, with larger batteries and
25%
lower consumption people charge less often and hence the
0%
coincidence factor decreases (this effect would also be seen 40% 80% 40% 80%
if the average travel distance decreased). Consequently, this EV penetration level

observation gives reason to use a single average car model to Fig. 12. Impact of the grid configuration on peak loading for lines A and B.
represent a diverse fleet. Furthermore, it underlines that driving
patterns and charging behaviour already create diversity among
individual EVs such that the diversity of EV types becomes F. EV placement
irrelevant. Peculiar aspects of this dimension require an adjustment of
the plotting functions. Fig. 13 compares the evenly distributed
Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger placement with a mild and an extreme clustering for lines A
Transformer peak load increase

Absolute transformer peak load

110% and B. Due to the clusters either being located at a branch or

80% 100% not, the realisations yield a non-normal distribution, illustrated
90%
as dots for single realisations. Line B experiences substantial
60%
peak load increases for the realisations when it happens to
80%
40%
host an EV cluster, but low impact in the other situations.
70%
20% The global penetration level increases primarily the likelihood
60% of higher loads and secondarily the peak itself for clustered
0%
Sales− Sales− +25% +50% +10% −10%
scenarios. The clustering can hence be seen as a locally higher
based based battery battery consumption consumption
average sampling capacity capacity penetration level. Since line A hosts more customers, it is
Car type model exposed to higher base loads. Therefore, the relative impact
Fig. 11. Impact of the car type modelling on the transformer peak load for of clustering is less extreme than for Line B. Moreover,
a 60% EV penetration. aggregation effects come into play causing lower peak load
increases for line A.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 Charger type 0% 11kW charger 50% 11kW charger 100% 11kW charger V. D ISCUSSION
20% EV penetration 40% EV penetration This work investigated the impact of EV expansion in
250%
a specific urban environment and models with mostly lim-
200%
ited complexity. Nevertheless, the presented results provide
Relative increase of peak line loading

guidance for other set-ups and more sophisticated modelling

Line A
150%

100%
approaches. For the generalisation of the results, the essential
50%
question to ask is how the change of set-up or model can
0%
be interpreted within the given framework. If, for example,
EV owners use to charge at work, they will cover only a
250%
fraction, e.g. 75%, of their daily energy demand at home,
200%
hence at the grid area of interest. Instead of introducing a new

Line B
150%
dimension to incorporate charging at work, one can resemble
100%
the effect by reducing the energy consumption per kilometre
50%
accordingly, in this example by 25%, and hence the daily
0%
Even Mild Strong Even Mild Strong energy demand which needs to be covered at home. The
clustering clustering clustering clustering
EV placement provided sensitivity analysis will then give an indication on
the impact that charging at work most likely has on the grid.
Fig. 13. Impact of clustered EV placement on peak line loading for lines A
and B using single realisations (dots) and their summary (boxplots). Y-axis
The possibility to estimate effects of such additional aspects
according to (1) for peak loads. of EV charging by expressing them in terms of the proposed
dimensions highlights the flexibility of this work. Another
example is to consider a rural, less-densely-populated area
G. Load duration curves with a different grid infrastructure. As long as the transformer
The primary focus of the study lies on the peak load, is the critical component, the overall picture would remain
nevertheless a brief analysis of other points on the LDC similar to the shown results; however, when individual line
provides further valuable insights. Fig. 14 shows that the loading becomes important the applicability of the provided
absolute transformer load quickly decreases when considering observations becomes limited. Analysing the effect of differing
non-peak hours. This means that for grids with EVs still national driving patterns serves as a third illustration of how
only a small number of hours determines the power rating the presented results can generalise. By analysing how the
requirements if no smart charging is employed. Furthermore, arrival time distributions and travel distances vary across
the spread of the results reduces drastically for non-peak hours different countries, first conclusions on the EV impact can
which increases their trustworthiness. This observation holds be drawn. Since national travel surveys are readily available
throughout the scenarios and also for line loadings. EV impact such a first analysis requires no additional simulations. Lastly,
analyses should therefore also consider non-peak load hours the framework offers the opportunity to study the effects of
to provide backing for peak load assessments governed by complex models even before simulating a single power flow.
extreme cases with large uncertainties. Investigating, for example, how agent-based traffic modelling
compares to a simple GMM in terms of the arrival time
distribution, can lead to conclusions such as: The models yield
0−100% 10−90% 0% 11kW charger
Realisation a very similar arrival time distribution hence either models
20−80% 30−70% Charger type 50% 11kW charger
percentiles could be used, potentially favouring the simpler one; or the
40−60% 100% 11kW charger
models show significant differences which would justify to
test both models. Such a comparison not only improves the
100%
understanding of the applied models but exposes the use of
Absolute transformer load

overly-complex or too simplistic ones. Although a number of

75%
simulations is inevitable for each set-up and model choice in
order to quantify the EV impact, the framework and results
50%
shown, clearly indicate which aspects of EV modelling should
be considered first in order to keep the computational burden
25%
as low as possible. This ultimately will lead to more effective
resource allocation as well as more reliable results.
0% VI. C ONCLUSION
0% 10% 20% 30% 40% 50%
Load duration In light of the previous analyses, the first step in EV
modelling is to determine whether the transformer or certain
Fig. 14. LDC of absolute transformer loads for an 80% EV penetration as a
distribution of realisations (coloured areas and solid line as median) and the power lines are the critical grid elements. If the transformer
base LDC (dashed line). is the limiting component, penetration level and charger types
are the dominating dimensions. If a line is at risk, the grid

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
configuration strongly influences the picture and very clustered [5] N. Daina, A. Sivakumar, and J. W. Polak, “Modelling electric
EV placement can cause a further load increase. vehicles use: a survey on the methods,” Renewable and Sustainable
Energy Reviews, vol. 68, pp. 447 – 460, 2017. [Online]. Available:
Driving patterns constitute a major influencing factor on http://www.sciencedirect.com/science/article/pii/S1364032116306566
the results. Careful modelling, as for the presented models, [6] J. Tan and L. Wang, “Integration of plug-in hybrid electric vehicles into
can resemble the results obtained with the original driving residential distribution grid based on two-layer intelligent optimization,”
IEEE Transactions on Smart Grid, vol. 5, no. 4, pp. 1774–1784, 2014.
pattern data set within satisfactory margins, especially when [7] Q. Yan, C. Qian, B. Zhang, and M. Kezunovic, “Statistical analysis
focusing on the arrival time distribution. EV characteristics and modeling of plug-in electric vehicle charging demand in distribu-
and charging behaviour reveal a low sensitivity, unless they tion systems,” in 19th International Conference on Intelligent System
Application to Power Systems, San Antonio, Texas, USA, 2017.
significantly impact people’s charging frequency and hence [8] N. H. Tehrani and P. Wang, “Probabilistic estimation of plug-in electric
the coincidence factor of EVs. vehicles charging load profile,” Electric Power Systems Research, vol.
The peak EV charging load and its timing with respect to 124, pp. 133–143, 2015.
[9] F. J. Soares, J. A. Peas Lopes, and P. M. Rocha Almeida, “A monte carlo
the base load peak mainly determine the impact assessment method to evaluate electric vehicles impacts in distribution networks,”
results. Since charging power, penetration level and driving in 2010 IEEE Conference on Innovative Technologies for an Efficient
patterns are the governing factors associated with modelling and Reliable Electricity Supply, 2010, pp. 365–372.
[10] M. ElNozahy and M. Salama, “A comprehensive study of the impacts
uncertainty, these dimensions should be subject to parameter of phevs on residential distribution networks,” Sustainable Energy, IEEE
variations in any work that aims at assessing the impact of Transactions on, vol. 5, pp. 332–342, 07 2014.
[11] Z. Darabi and M. Ferdowsi, “An event-based simulation framework to
EVs reliably. Further off-peak load analyses can add value for examine the response of power grid to the charging demand of plug-in
lifetime and risk assessments due to higher confidence. hybrid electric vehicles,” IEEE Transactions on Industrial Informatics,
The analysis of how other set-ups and models affect each vol. 10, no. 1, pp. 313–322, 2014.
[12] H. Hashemi-Dezaki, M. Hamzeh, H. Askarian-Abyaneh, and H. Haeri-
dimension of this framework presents the opportunity to gen- Khiavi, “Risk management of smart grids based on managed charging
eralise the shown results. These insights not only improve of PHEVs and vehicle-to-grid strategy using Monte Carlo simulation,”
the understanding of model assumptions but also allow for Energy Conversion and Management, vol. 100, pp. 262–276, aug 2015.
[13] Swiss Federal Statistical Office. (2018) MZMV 2015. [Online].
more reliable impact assessments. With the future introduction Available: https://www.bfs.admin.ch/bfs/de/home/statistiken/mobilitaet-
of control strategies additional dimensions occur, e.g. the verkehr/erhebungen/mzmv.html
control strategy type and how many people apply it. While [14] J. Stiasny. (2019) Sensitivity analysis of EV impact on distribution
grids based on Monte-Carlo simulations. [Online]. Available:
the dimensions’ impact on the effectiveness to shift and reduce https://psl.ee.ethz.ch/Publications/SAMA.html
peak EV loads remains to be determined, controlled charging [15] EVdatabase. (2019) Electric vehicle database. [Online]. Available:
is expected to be a promising way forward for grid operators https://ev-database.org/
[16] J. Pontes. (2019) Europe EV sales: Renault Zoe wins In
to foster EV integration into the grid. december, Nissan LEAF takes 2018 title. [Online]. Avail-
able: https://cleantechnica.com/2019/02/03/europe-ev-sales-renault-zoe-
R EFERENCES wins-in-december-nissan-leaf-takes-2018-title/
[1] United Nations. (2015) Paris Agreement. [Online]. Available: [17] T. Franke and J. F. Krems, “Understanding charging behaviour of electric
https://unfccc.int/sites/default/files/english paris agreement.pdf vehicle users,” Transportation Research Part F: Traffic Psychology and
[2] Ernst Basler + Partner. (2018) Szenarien der Elektromo- Behaviour, vol. 21, pp. 75–89, 2013.
bilität in der Schweiz Update 2018. [Online]. Avail- [18] EA Technology. (2016) My Electric Avenue. [Online]. Available:
able: https://www.ebp.ch/sites/default/files/unterthema/uploads/2018-03- http://myelectricavenue.info/
05 EBP CH EmobSzen PKW 2018 1.pdf [19] A. J. Collin, S. Z. Djokic, H. F. Thomas, and J. Meyer, “Modelling of
[3] J. Quiros-Tortos, A. Navarro-Espinosa, L. F. Ochoa, and T. Butler, electric vehicle chargers for power system analysis,” in Proceeding of the
“Statistical representation of EV charging: Real data analysis and International Conference on Electrical Power Quality and Utilisation,
applications,” in 20th Power Systems Computation Conference, Dublin, EPQU, 2011, pp. 476–481.
Ireland, 2018. [20] R. D. Zimmerman, C. E. Murillo-Sánchez, and R. J. Thomas, “MAT-
[4] A. Navarro-Espinosa and L. F. Ochoa, “Probabilistic Impact Assess- POWER: Steady-state operations, planning, and analysis tools for power
ment of Low Carbon Technologies in LV Distribution Systems,” IEEE systems research and education,” IEEE Transactions on Power Systems,
Transactions on Power Systems, vol. 31, no. 3, pp. 2192–2203, 2016. vol. 26, no. 1, pp. 12–19, 2011.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020