You are on page 1of 12

Electrical Power and Energy Systems 147 (2023) 108868

Contents lists available at ScienceDirect

International Journal of Electrical Power and Energy Systems


journal homepage: www.elsevier.com/locate/ijepes

Effects of long-range electric vehicles on distribution system using


probabilistic analysis
Jorge Henrique Angelim *, Carolina M. Affonso
Faculty of Electrical and Biomedical Engineering, Federal University of Para, Belem, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: The number of electric vehicles with larger battery capacity that charges at higher power levels is increasing in
Electric Vehicles residential areas. This scenario directly impacts users’ charging behavior and modifies the demand profile of
Voltage Quality electric vehicles. This paper evaluates the effects of long-range electric vehicles on total charging demand,
Transformer loading condition
transformer loading and voltage quality in a real distribution system. Electric vehicles demand is modeled based
Monte Carlo Simulation
Distribution Network
on real-world data obtained from a large-scale project developed in the United Kingdom. This study proposes
performance indices to perform a qualitative analysis of the system and uses Monte Carlo Simulation to consider
uncertainties in residential demand and vehicle charging behavior. The results show that the current trend of
adopting long-range vehicles that charge at higher power levels modifies vehicles charging frequency and
duration. It increases transformer peak demand, and the number of hours the transformer operates overloaded.
Voltage quality violations occur less frequently when three-phase connectors are used to charge vehicles, as they
split demand between phases.
© 2017 Elsevier Inc. All rights reserved.

1. Introduction UK, and the results show EV’s charging frequency is directly associated
with battery capacity [5]. EVs with larger battery capacity are plugged
The transportation sector is responsible for almost 16.2 % of global in less often (2.79 sessions per week) than EVs with smaller battery
greenhouse gas emissions, and the use of plug-in electric vehicles (EVs) capacity (3.31 sessions per week). Also, the My Electric Avenue project
is essential to decarbonize this segment [1]. Electric vehicles market (MEA) collected EV charging data from 2013 to 2015, and results show
sales have grown rapidly in the past decade, and this trend is expected to that even vehicles with lower battery capacity do not charge every day
continue in future years [2]. The battery industry is investing in tech­ [6]. This behavior directly affects the number of vehicles charging
nology, reducing costs, and improving storage capacity. To better suit simultaneously and, therefore, EV peak load. Thus, considering an
customer requirements, EV manufacturers are launching models with everyday charging behavior can provide biased results and overestimate
higher battery capacity, capable to accept higher charging power levels. EVs impact on grid. Besides, EV’s battery size and charging power level
In 2014, most EV models had batteries with 24 kWh and charging power define the charging section duration, modifying the charging profile and
limited to 3–4 kW. Nowadays, EVs usually have 40 kWh or larger battery their impact on the distribution system.
pack, and charge at 6–8 kW. Some models go further, such as Tesla 3 Several papers have been developed analyzing the technical impacts
Performance with 75 kWh and charging rate of 11 kW [3]. of EVs on the grid. However, most of them model EVs demand based on
This next generation of EVs models with long-range batteries is ex­ travel surveys obtained from internal combustion vehicles, which may
pected to be followed by an increase in charging power levels, which not be feasible to EVs due to its limited driving range. Besides, the
will affect EV users charging behavior and consequently, the total en­ adoption of an electric vehicle may change driver behavior. In [7], au­
ergy consumed [4]. Some EV trials have been conducted to better un­ thors estimate EVs hosting capacity of real distribution systems using
derstand the impact of these changes on EVs charging demand. As an Monte Carlo simulation considering the impacts on voltage magnitude,
example, the Electric Nation Project (ENP) collected EVs charging data voltage unbalance, line congestion and transformer capacity. EVs
from participants running different EVs models from 2016 to 2019 in charging pattern is modelled based on assumptions of user’s driving

* Corresponding author.
E-mail addresses: jorge.angelim@itec.ufpa.br (J. Henrique Angelim), carolina@ufpa.br (C.M. Affonso).

https://doi.org/10.1016/j.ijepes.2022.108868
Received 22 February 2022; Received in revised form 14 October 2022; Accepted 27 November 2022
Available online 16 December 2022
0142-0615/© 2022 Elsevier Ltd. All rights reserved.
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Table 1
Literature review of recent papers analyzing the impacts of EVs on the grid.
Reference Year Real EV database Week-long simulation Long-range batteries Technical Impacts
(Trial projects) (Not everyday plug-in) Transformer Loading Undervoltage Voltage Unbalance

[7] 2020 – – – ✓ ✓ ✓
[8] 2021 – – – ✓ ✓ –
[9] 2022 – – – – ✓ ✓
[10] 2021 – – ✓ ✓ – –
[11] 2021 ✓ – ✓ – – ✓
[12] 2022 ✓ – ✓ – – –
[13] 2020 – ✓ ✓ – – –
[14] 2022 – ✓ ✓ – – –
[15] 2021 ✓ ✓ ✓ – – –
This paper 2022 ✓ ✓ ✓ ✓ ✓ ✓

behavior, considering battery capacity of 20kWh and charging power of


2.0/3.0 kW. Reference [8] investigates the impact of EVs charging on
the Colombian electricity grid using Monte Carlo simulation, consid­
ering EV projection scenarios and local surveys. The study adopts EVs
for private transportation with battery capacity of 11kWh/28kWh and
charging power of 3.7 kW/7.6 kW, and analyzes the impact on system
voltage profile, transformer, and transmission line overload. In [9],
authors analyze power quality impacts of different EV penetration levels
in a distribution system using Monte Carlo simulation. EVs are modeled
using local and regional surveys, considering EVs with battery capacity
of 30 kWh and charging power of 7.2 kW. Reference [10] analyzes the
influence of several factors in EV modelling using Monte Carlo simula­
tion. EVs charging pattern are modelled based on Swiss National Travel
Survey and considers different batteries capacities (33kWh-100kWh)
and charging power (3.7 kW, 11 kW). In [11], authors evaluate the
impact of nonlinear loads and EVs on voltage unbalance in a low voltage
residential system. EVs profiles are generated based on real demand
profile measurements, using different battery capacities (16 kWh − 54
kWh) and charging power rates (1.44 kW − 3 kW). Reference [12]
evaluates the coincidence factor of EVs combining data sources from
travel surveys and recorded EV charging data. The study adopts vehicles
with different battery capacities, from 24 kWh to 60 kWh, charging with
power rates from 3.7 kW to 22 kW.
All works mentioned above assume an everyday charging behavior,
which is not a valid assumption, especially with long-range vehicles.
Few studies have started to consider in their models the fact that EV
users do not charge every day. Reference [13] investigates the impact
that different battery capacities, charging power rates and economic
aspects have on electricity demand, including long-range batteries with
high charging power. The study model EVs charging pattern for a week-
long period. However, the study is based on UK National Travel Survey
Fig. 1. Flowchart of the proposed methodology.
dataset and authors do not evaluate technical impacts on the distribu­
tion system. Reference [14] investigates the potential of EVs operating
and voltage quality in a real distribution system. EVs charging behaviour
in V2G mode to provide system flexibility. EVs charging patterns are
is modelled considering a week-long period of analysis based on real
based on week-long travel diaries using different battery capacities (24
data from Electric Nation Project developed in the United Kingdom [16].
kWh − 75 kWh) charging at 7.4 kW. However, technical impacts on the
Five EVs penetration levels are considered. Performance indices are
grid are not investigated, and EVs charging pattern are based on UK
proposed based on international standards, and a qualitative analysis is
National Travel Survey. In [15], authors consider EV does not charge
employed to better identify when the system operates inappropriately.
every day and evaluates its impact on EVs peak demand using a real
Monte Carlo Simulation is applied to consider uncertainties through a
database from the Electric Nation Project. The analysis considers small,
weekly Quasi-Static Time Series (QSTS) simulation with a 10-minute
medium, and large battery sizes and three charging power levels.
time interval using OpenDSS software [17], using real household de­
However, voltage impacts are not investigated.
mand curves provided by UK Data Service [18]. The key contributions of
Table 1 provides an overview of the most recent studies that analyzes
this study are highlighted as follows.
the impacts of EVs on the grid. To the best of the author’s knowledge, no
paper has been reported in the literature that investigates the effects of
• Models EVs demand based on real data from EV trial project instead
long-range EVs on distribution systems based on a real EV database and
of surveys or projection scenarios.
considering that users do not charge their vehicles every day. Their
• Does not assume an everyday plug-in charging behavior, but models
possible impacts on distribution system are unknown.
EVs charging pattern based on week-long travel diaries.
This paper tries to address this knowledge gap in the literature by
• Proposes performance indices to analyze qualitatively transformer
probabilistically analysing how vehicles with different battery sizes and
loading and voltage quality in a real distribution system with long-
charging at different power levels affect distribution transformer load
range vehicles charging with higher power rates.

2
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

the flowchart of the proposed methodology. Based on probabilistic


models, random values are sampled to generate residential and EV de­
mand profiles with a 10-minute resolution for one week, considering
different EVs battery capacities and charging power rates. Then, a quasi-
static time series (QSTS) power flow analysis is conducted to obtain
weekly power flow results and quantify probabilistic performance
indices to perform a qualitative analysis evaluating transformer loading
and voltage quality in a distribution system. These probabilistic indices
are stored, and the Monte Carlo Simulation is repeated until the
convergence is reached, which occurs when the Monte Carlo coefficient
of variation reaches the maximum threshold specified, or when the
number of iterations is 2,000, as a trade-off between accuracy and
computation-time cost. To guarantee simulation’s reproducibility,
random numbers are generated with Mersenne Twister with a seed equal
to 3,000 [19]. Power flow simulations are conducted using OpenDSS in a
MATLAB® environment.
The proposed methodology is applied in a low voltage distribution
system, representative of a residential network from the United
Kingdom. The system is supplied by a 200 kVA 11 kV/0.400 kV distri­
bution transformer with 200 single-phase consumers distributed among
the phases as follows: 74 on phase A, 68 on phase B, and 58 on phase C
[20]. Fig. 2 shows the single-line diagram of the low-voltage system
used.

2.2. Residential load modeling

Residential energy consumption is probabilistically modeled based


on records from UK Data Service and comprehends real electricity
measurements from 22 dwellings in the East Midlands, United Kingdom,
with a one-minute resolution over two years [18]. The winter season
period is chosen to be analyzed since it has higher demand. It goes from
December to February [21] (90 days) and is organized in 12 weeks,
starting at 00:00 on Monday and finishing at 23:59 on Sunday. This
Fig. 2. European low-voltage distribution system.
dataset is initially preprocessed to further obtain a probabilistic model,
as described below.
This work is organized as follows. Section 2 presents the proposed
methodology, the distribution system, and the probabilistic models. 2.2.1. Pre-processing data
Section 3 shows the performance indices adopted to evaluate EVs impact Usually, data collected from real measurements are incomplete, have
on distribution system. The results are presented in Section 4, followed noise, and need to be pre-processed before doing any further analysis.
by main conclusions in Section 5. Then, adequate treatment for missing values and outliers is needed. One
way to deal with missing values is to use the imputation technique,
2. Materials and methods replacing the missing value with the mean, median, or the most frequent
value [22]. In this paper, missing values are replaced by an equivalent-
2.1. Proposed methodology nearest measurement to maintain the seasonality property. For example,
if a week has missing values on Tuesday, they are replaced by data from
This paper applies a probabilistic approach to evaluate distribution the nearest Tuesday with valid measurements. Outliers are identified by
transformer loading and voltage quality using a non-sequential Monte the interquartile range (IQR), a commonly used rule that equally divides
Carlo Simulation to consider inherent uncertainties of data. Fig. 1 shows the distribution into four equal parts called quartiles. Then, values that

Fig. 3. Boxplot of residential load. a) Before pre-processing. b) After pre-processing.

3
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 4. Mean value from measured residential load and data generated with the proposed probabilistic model.

data generated with the proposed probabilistic model. Results indicate a


good fit for original data.

2.3. Electric vehicle load modeling

EVs probabilistic modeling is performed according to the flowchart


in Fig. 5. The parameters considered are: (a) initial State-of-Charge
(SOCinitial), (b) start charging time, (c) battery capacity, (c) charging
power, and (e) charging duration. Initial SOC and start charging time are
Fig. 5. Electric Vehicle charging demand profile calculation. randomly sampled following probabilistic distributions, obtained based
on a real large-scale database from Electric Nation Project (ENP). The
lie more than 1.5 IQR below the first quartile or 1.5 IQR above the third charging section duration is evaluated considering four charging power
quartile are considered outliers and removed from data [23]. The box­ rates. Thus, several EVs charging demand profiles are obtained using
plot of residential demand before and after pre-processing is presented Monte Carlo simulation.
in Fig. 3. According to international standards, voltage quality is The ENP project collected data from 673 EV smart chargers installed
analyzed with a 10-minute sampling rate. Then, all load profiles are at participants’ homes in the United Kingdom from January 2017 to
converted to this time resolution. December 2018 [24]. This dataset has information from different EV
models with a variety of battery capacities, as shown in Fig. 6. Two
2.2.2. Probabilistic modelling
The probabilistic model can be obtained by fitting the dataset to a Table 2
distribution function. The goodness-of-fit must be evaluated to measure Frequency of charging per battery capacity from Electric Nation Project Report.
the reasonableness of the parameters. Since no parametric distribution Battery Capacity (kWh) Daily average number of charging sessions
fits the original data to all 10 min sampling intervals, a non-parametric
10 to 25 0.63 (around every other day)
Gaussian kernel with 0.01 bandwidth is considered. The goodness-of-fit 25 to 35 0.39 (around every-three days.)
is evaluated employing the Kolmogorov-Smirnov statistic test (KS-test) >35 0.31 (around every-four day)
[23]. Fig. 4 compares the mean value from original load measures and

Fig. 6. Number of EVs according to battery capacity from ENP report.

4
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 7. Initial SOC histogram and PDF model for different battery capacities. a) 24 kWh; b) 75 kWh.

Fig. 8. EVs charging station connection. a) Single-phase; b) Three-phase.

distinct EVs groups can be observed: EVs with smaller battery capacity range models) charge less frequently than vehicles with smaller battery
(11.2 – 24.2 kWh), and EVs with larger battery capacity (35.8–100 capacity (short-range), as shown in Table 2. Therefore, two EV models
kWh). This report shows that vehicles with larger battery capacity (long- are considered in this study: 24kWh Nissan Leaf representative of small

Fig. 9. Start charging time histogram and PDF model. a) Weekday. b) Weekend.

5
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 10. Random samples of residential demand and EV demand generated for different battery capacities during a week. a) 24 kWh; b) 75 kWh.

EVs fleets and commonly adopted in several papers, and 75 kWh Tesla duration evaluated in (1) and the start charging time obtained with the
representative of long-range EVs fleets. These vehicles are selected for probabilistic model, on a week basis with 10 min intervals. Each sample
being the most selling models to each battery capacity range, according is randomly assigned (uniform distribution) to each residential node and
to the Electric Nation Report [24]. is assumed that each vehicle starts charging as soon as it is connected to
The ENP report [24] also shows that vehicles with smaller battery the grid and remains connected until full charge (SOCfinal = 100 %). EVs
capacity usually have lower SOC when they start charging. To consider demand curves are modeled individually, and the power demanded by
this specificity, two probabilistic models are used to generate the initial the EV fleet is the sum of all individual EVs demand. Fig. 10 illustrates a
SOC, each one associated with a battery capacity (24 kWh and 75 kWh). random sample of residential demand, and two random samples of EV
A non-parametric Gaussian Kernel with 0.01 bandwidth is employed and charging profiles, each one associated with a battery capacity. Note that,
the goodness-of-fit is evaluated with KS-test. Fig. 7 shows the real data weekly, vehicles with higher battery capacity charge less frequently
histogram and the proposed probability density function (PDF). than vehicles with lower battery capacity.
According to [4], most electric vehicles on the market today charge
below 10 kW in residential areas, between 3 and 4 kW or 6–8 kW. 3. Performance indices
However, as EVs battery capacity increases, there must be a corre­
sponding increase in the charging power level as larger batteries require This study presents a methodology for probabilistic evaluation of
more time to fully charge. For this reason, this study adopts four distribution transformer loading and voltage quality. Transformer
charging power levels: 3.7 kW and 7.4 kW with single-phase connection loading is evaluated by computing transformer peak demand (Speak
weekly )
which are being mostly adopted today, and 11 kW and 17 kW with three- overload
and overload duration (Tweekly ), both verified over one week period. It is
phase connection as a future trend. On single-phase charging stations,
the EV charger is connected at the same phase the residence is served. important to assess transformer overload duration because short time
On three-phase charging stations, EV charger is connected in Y config­ overloading may be acceptable during emergency conditions, however,
uration, as shown in Fig. 8. when exposed for long periods, transformer insulation may deteriorate
Using the probabilistic model, random values of initial state-of- and cause equipment loss of life, especially during high temperatures
charge are generated, and adopting different battery capacities [25].
(24kWh, 75 kWh) and charging power (3.7 kW, 7.4 kW, 11 kW, 17 kW), Voltage quality is evaluated in terms of voltage magnitude, one of the
EV charging duration (ΔTcharg ) is estimated according to (1) [25]. main parameters that determine the quality of electricity supplied to
( ) consumers, and voltage unbalance factor (VUF), which occurs due to the
Cb SOCfinal − SOCinitial uneven distribution of loads and generation between the three phases of
ΔTcharg = (1)
Pch *η the system, causing unsymmetrical voltage drops. The operational limits
for voltage magnitude and VUF are strictly defined by various national
where Cb is the battery capacity (kWh), Pch is the charging power
and international regulations. According to Standard EN50160, at least
(kW), SOCfinal and SOCintial are the final and initial SOC, respectively, and
95 % of voltage measurements must be within the range ± 10 % of
η is the charger’s efficiency assumed as 99 %.
nominal voltage, and all measurements must be within the range of +
The start charging time defines the instant in which the charging
10 % / − 15 %, considering measurements collected during one week
session starts. Since no parametric distribution fits the original data, this
with a 10-minute sampling rate under normal operating conditions [26].
parameter is modeled using a non-parametric Gaussian Kernel with 0.01
Voltage unbalance factor is evaluated in terms of symmetric components
bandwidth. K-S test is applied to validate the probabilistic model. Fig. 9
as in (2), and EN50160 recommends that at least 95 % of all measures
shows the start charging time histogram from real data and the proposed
must be below 2 %.
PDF model.
EVs charging profiles are obtained by combining the EV charging

6
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 11. Graphical definition of V5th and VUF95th. a) Attending EN 50,160 Standard. b) Not attending EN 50,160 Standard.

V−
VUF% = × 100 (2) Table 3
V+
Qualitative Categorization of Events Violation.
where V - is the voltage negative sequence and V + is the voltage
positive sequence.
Based on these regulations, two performance indices are proposed
for voltage quality evaluation considering the probability of occurrence,
which are the 5th percentile of node voltages (V5th) and the 95th
percentile of VUF (VUF95th). The nth percentile is a number where a
certain percentage of ordered data falls below that number, being data
divided into 100 parts [23]. As an example, Fig. 11 shows two cases of
cumulative distribution functions (CDF) for node voltage and VUF. In
both cases, 100 % of voltage falls between 0.85 pu − 1.1 pu, and 100 %
of VUF falls between 0 % − 3 %. In the first case, which attends EN
50,160 standard, the V5th is above 0.9 pu, which means that more than
The violation of V5th
min
and VUFmax
95th may affect system operation and is
95 % of voltages are between 0.9 pu − 1.1 pu. The VUF95th is below 2 %,
considered as the occurrence of an undesired event, which should be
meaning that more than 95 % of VUF are below 2 %. In the second case,
easily identified during system planning. Due to the large amount of
which does not attend EN 50,160 standard, V5th is below 0.9 pu, which
data, the interpretation of results is more difficult, and qualitative
means that less than 95 % of voltages are between 0.9 pu − 1.1 pu. The
analysis is employed. The probability of not attending these indices is
VUF95th is above 2 %, meaning that less than 95 % of VUF measurements
evaluated and categorized into 5 levels as shown in Table 3, which are:
are below 2 %.
improbable, remote, possible, occasional, and frequently. The proba­
To comply with EN 50160, the minimum V5th must be 0.9 pu and the
bility assigned to each label is defined based on the user expert knowl­
maximum VUF95th must be 2 % as shown in (3)-(4), considering mea­
min edge of the problem and can be easily changed according to one‘s needs.
surements collected during 1-week. The index V5th is evaluated to each
max
one of the 200 consumers and VUF95th is evaluated to every 133 three-
4. Results and discussions
phase nodes in the grid.
min
V5th = 0.9pu (3) The proposed method is applied in a low-voltage system considering
two battery capacities (24 kWh and 75 kWh) and four charging power
VUF max
95th = 2% (4) levels (3.7 kW, 7.4 kW, 11 kW, and 17 kW). Five EVs penetration levels
(PL) are considered, evaluated as the ratio between the number of EVs

7
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 12. Number of vehicles charging simultaneously with 95 % of confidence interval. a) 3.7 kW, b) 7.4 kW, c) 11 kW, d) 17 kW.

Fig. 13. Charging section duration with 95 % of confidence interval. a) 3.7 kW, b) 7.4 kW, c) 11 kW, d) 17 kW.

Fig. 14. Residential demand and aggregated EVs peak demand for PL = 75 % with 95 % of confidence interval.

and the number of households (200 in this case): 0 %, 25 %, 50 %, 75 %, 4.1. Effects of battery capacity and charging power on EVs charging
and 100 %. Each case is defined by the combination of a penetration behavior
level, charging power, and battery capacity. In total, 33 cases are
analyzed. Fig. 12 and Fig. 13 show the number of vehicles charging simulta­
neously and the charging section duration as battery capacity and
charging power varies, respectively. Results show EVs with higher

8
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 15. Boxplot of transformer peak load for one week period (a) PL = 25 %, (b) PL = 50 %, (c) PL = 75 % and (d) PL = 100 %.

Fig. 16. Boxplot of transformer overload duration for one week period (a) PL = 25 %, (b) PL = 50 %, (c) PL = 75 % and (d) PL = 100 %.

battery capacity (75 kWh) decrease the number of vehicles charging the use of high charging power levels has a more significant impact on
simultaneously after 18 h because they charge less frequently and take transformer overloading due to higher peaks in power demand and the
longer to fully charge. Also, as the charging power increases, the number coincidence between EVs and residential demand. Then, the combina­
of vehicles charging simultaneously decreases because vehicles fully tion of long-range vehicles with high charging power levels, which is the
recharge faster. current trend, has a negative impact on the grid and can cause distri­
Vehicles with higher battery capacity tend to reduce the EV charging bution transformer overload. This effect is accentuated as the EV
peak demand and shift its occurrence to later hours because the number penetration level increases.
of vehicles charging simultaneously is lower and their battery takes Fig. 16 shows the boxplot of transformer overload duration during a
longer to fully charge, as shown in Fig. 14. However, as the charging week, considering different battery capacities, charging power, and EV
power increases, the aggregated EV peak demand also increases due to penetration levels. As the charging power increases, the overload
high power levels, and peak demand tends to occur earlier because the duration also increases. The same behavior is observed with the increase
charging duration decreases, also reducing the number of EVs charging of the EVs penetration level. It is observed that high-range electric ve­
simultaneously. hicles have a positive effect in reducing overload duration for lower
charging power levels (3.7 kW, 7.4 kW). However, for higher charging
4.2. Effects of battery capacity and charging power on transformer peak power levels (11 kW, 17 kW), long-range electric vehicles increase
and overload duration overload duration because EVs demand coincides with residential peak
demand.
Fig. 15 shows the boxplot of transformer weekly peak demand
considering different battery capacities, charging power, and EV pene­ 4.3. Effects of battery capacity on voltage quality
tration levels. Results show that EVs with higher battery capacity
slightly reduce the overload occurrence of the transformer. However, Fig. 17 shows the number of nodes with probability of violation on

9
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Fig. 17. Number of nodes with violation on Vmin max


5th and VUF95th . (a) PL = 25 %, (b) PL = 50 %, (c) PL = 75 % (d) PL = 100 %.

min
V5th max
and VUF95th to all cases analyzed. For a penetration level of 25 %, 4.4. Monte Carlo convergence
the probability of voltage and VUF violation is zero. As the penetration
level increases, the probability of violations also increases, especially In this paper, two commonly used metrics are adopted as stopping
when vehicles charge with 7.4 kW. EVs charging at high power levels criteria for Monte Carlo Simulation: coefficient of variation and confi­
represent a considerable increase in power demand, which may lead to dence interval [27]. The coefficient of variation is a statistical measure
possible voltage quality violations. However, these events are not veri­ of the uncertainty of estimation relative to the mean and can be evalu­
fied when high-power chargers of 11 kW and 17 kW are used, since these ated as shown in (5).
chargers have a three-phase connection and split demand between (√̅̅̅̅̅̅̅2 )
phases, relieving the power consumed in each phase. Then, high-power
σ̂μ
β= × 100 (5)
single-phase chargers worsen voltage unbalance conditions. μ
̂

where σ̂μ 2 is the variance of estimator and ̂


μ is the sample mean.

10
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

Table 4 higher power levels impact EVs behavior, transformer load, and voltage
Number of scenarios for each case. quality in a real low-voltage residential system. Monte Carlo Simulation
Charging Power Battery Capacity Penetration Level (%) was applied to consider uncertainties through a weekly Quasi-Static
(kW) (kWh) Time Series (QSTS) simulation with a 10-minute sampling interval
25 % 50 % 75 % 100 %
using OpenDSS software. EV charging behavior is modeled based on real
3.7 kW 24 kWh 1755 874 941 1210 data collected from the Electric Nation Project developed in the United
75 kWh 2000 2000 1652 2000
7.4 kW 24 kWh 1394 1467 1764 918
Kingdom, with real household demand curves provided by UK Data
75 kWh 2000 920 2000 975 Service. From the results, important conclusions can be addressed:
11 kW 24 kWh 793 323 329 328
75 kWh 2000 344 329 332 • EV battery capacity and charging power directly impacts vehicle
17 kW 24 kWh 548 323 329 326
charging frequency and duration, modifying aggregated EV demand
75 kWh 386 332 362 331
profile. Ignore this fact by considering an everyday charging
behavior can overestimate EV charging impact on the grid and the
To specify estimation accuracy, it is important to define a confidence availability of flexibility services such as congestion management,
interval, which is adopted in this paper as 95 % with 5 % of relative voltage regulation, and backup power for fault restoration.
width (wr), calculated by (6)-(7). • Vehicles with higher battery capacity shift aggregated EV peak de­
( ) mand to later hours of the day, avoiding coincidental between resi­
σ
̂
CI 95% = ±1.96 √̅̅̅̅ (6) dential and EVs peak demand. Consequently, transformer overload is
N slightly reduced. The use of higher charging power levels increases
[ ( )] transformer peak load more significantly due to higher peak power
σ
̂
wr = 2 × 1.96 √̅̅̅̅ × 100 (7) and coincidence between residential and EV demand.
N
• High-range electric vehicles reduce overload duration when low
where ̂ σ is the standard deviation and N is the number of samples. charging power levels are adopted (3.7 kW, 7.4 kW). For higher
The simulation stops when β ≤ 1 % and wr ≤ 5 % for all performance charging power levels (11 kW, 17 kW), high-range EVs increase
indices analysed, or when the number of simulations reaches 2,000, as a overload duration because EV peak demand overlaps residential
trade-off between accuracy and computation time. Table 4 shows the peak demand.
number of scenarios necessary to reach convergency in each case • The use of higher power levels to charge EVs may lead to voltage
simulated. As an example, Fig. 18 shows the convergence of the pro­ violations, especially when vehicles charge with single-phase char­
posed methodology considering battery capacity equal to 75 kWh, gers (7.4 kW) which worsen voltage unbalance conditions. Voltage
charging power of 3.7 kW, and penetration level of 75 %. As results violations are not verified when three-phase high-power chargers
show, 1,652 iterations were necessary to reach convergency. (11 kW, 17 kW) are used, because they split demand between phases,
relieving the power consumed in each phase.
5. Conclusions • As the penetration level increases, the probability of occurrence of
transformer overload and voltage violations also increases, as
This work proposed a probabilistic method to investigate how next expected.
generation of EVs models with higher battery capacity, charging at

Fig. 18. Monte Carlo Simulation convergence for VUF95th on node 53 for a battery capacity of 75 kWh and charging power of 3.7 kW, considering PL = 75 %. (a)
Confidence interval. (b) Coefficient of variation.

11
J. Henrique Angelim and C.M. Affonso International Journal of Electrical Power and Energy Systems 147 (2023) 108868

• The use of long-range vehicles combined with high charging power World. IEEE Power Energ Mag 2018;16:64–76. https://doi.org/10.1109/
MPE.2018.2863060.
levels in residential areas, which is the current trend, has a negative
[7] Barbosa T, Andrade J, Torquato R, Freitas W, Trindade FCL. Use of EV hosting
effect on distribution systems, increasing transformer overload rate capacity for management of low-voltage distribution systems. IET Gener Transm
and duration. Voltage quality violations can be avoided using Distrib 2020;14:2620–9. https://doi.org/10.1049/iet-gtd.2019.1791.
charging power with a three-phase connection since they split de­ [8] Betancur D, Duarte LF, Revollo J, Restrepo C, Díez AE, Isaac IA, et al. Methodology
to Evaluate the Impact of Electric Vehicles on Electrical Networks Using Monte
mand between phases. Carlo. Energies (Basel) 2021;14:1300. https://doi.org/10.3390/en14051300.
[9] Torres S, Durán I, Marulanda A, Pavas A, Quirós-Tortós J. Electric vehicles and
power quality in low voltage networks: Real data analysis and modeling. Appl
Energy 2022;305:117718. https://doi.org/10.1016/j.apenergy.2021.117718.
CRediT authorship contribution statement [10] Stiasny J, Zufferey T, Pareschi G, Toffanin D, Hug G, Boulouchos K. Sensitivity
analysis of electric vehicle impact on low-voltage distribution grids. Electr Pow
Jorge Henrique Angelim: Conceptualization, Investigation, Meth­ Syst Res 2021;191:106696. https://doi.org/10.1016/j.epsr.2020.106696.
[11] P. Rodríguez-Pajarón, A. Hernández, J. v. Milanović, Probabilistic assessment of
odology, Software, Data curation, Writing – original draft. Carolina M. the impact of electric vehicles and nonlinear loads on power quality in residential
Affonso: Visualization, Supervision, Writing – review & editing. networks, International Journal of Electrical Power and Energy Systems. 129
(2021). https://doi.org/10.1016/j.ijepes.2021.106807.
[12] Bollerslev J, Andersen PB, Jensen TV, Marinelli M, Thingvad A, Calearo L, et al.
Declaration of Competing Interest Coincidence Factors for Domestic EV Charging from Driving and Plug-In Behavior.
IEEE Trans Transp Electrif 2022;8:808–19. https://doi.org/10.1109/
The authors declare that they have no known competing financial TTE.2021.3088275.
[13] Dixon J, Bell K. Electric vehicles: Battery capacity, charger power, access to
interests or personal relationships that could have appeared to influence
charging and the impacts on distribution networks. ETransportation 2020;4:
the work reported in this paper. 100059. https://doi.org/10.1016/j.etran.2020.100059.
[14] Dixon J, Bukhsh W, Bell K, Brand C. Vehicle to grid: driver plug-in patterns, their
Data availability impact on the cost and carbon of charging, and implications for system flexibility.
ETransportation 2022;13. https://doi.org/10.1016/j.etran.2022.100180.
[15] Gonzalez Venegas F, Petit M, Perez Y. Plug-in behavior of electric vehicles users:
Data will be made available on request. Insights from a large-scale trial and impacts for grid integration studies.
ETransportation 2021;10:100131. https://doi.org/10.1016/j.etran.2021.100131.
[16] WPD, Smart Charging Project | Electric Nation : Electric Nation, Electric Nation
Acknowledgments Data. (2019). https://www.westernpower.co.uk/electric-nation-data (accessed
July 4, 2021).
This study was supported in part by National Council for Scientific [17] EPRI, OpenDSS, (2019). https://sourceforge.net/projects/electricdss/.
[18] Richardson I, Thomson M. One-Minute Resolution Domestic Electricity Use Data
and Technological Development (CNPq, Portuguese: Conselho Nacional 2008–2009, 2010.. https://doi.org/10.5255/UKDA-SN-6583-1.
de Desenvolvimento Científico e Tecnológico) and Coordination for the [19] da Silva AM, de Castro AM. Risk Assessment in Probabilistic Load Flow via Monte
Improvement of Higher Education Personnel (CAPES, Portuguese: Carlo Simulation and Cross-Entropy Method. IEEE Trans Power Syst 2019;34:
1193–202. https://doi.org/10.1109/TPWRS.2018.2869769.
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil) [20] A. Navarro-Espinosa, L.F. Ochoa, LVNS Network Models and LCT Profiles,
[Finance Code 001]. Manchester, 2015. www.enwl.co.uk/lvns.
[21] Met Office, Our seasons - Met Office, Our Seasons. (n.d.). https://www.metoffice.
gov.uk/weather/learn-about/met-office-for-schools/other-content/other-
References
resources/our-seasons (accessed October 12, 2020).
[22] N.J. Salkind, ed., Last Observation Carried Forward, in: Encyclopedia of Research
[1] IEA, Global energy-related CO2 emissions by sector, 2020 - Charts - Data & Design, SAGE Publications, Inc., 2455 Teller Road, Thousand Oaks California
Statistics - IEA, Global Energy-Related CO2 Emissions by Sector, 2020. (2021). 91320 United States, 2010. https://doi.org/10.4135/9781412961288.n211.
https://www.iea.org/data-and-statistics/charts/global-energy-related-co2- [23] Ross S. Introduction to Probability and Statistics for Engineers and Scientists.
emissions-by-sector-2020 (accessed July 1, 2022). Elsevier 2021. https://doi.org/10.1016/C2018-0-02166-0.
[2] IEA, Global EV Outlook 2021, OECD, 2021. https://doi.org/10.1787/d394399e- [24] E. Dudek, P. Barnfather, Electric Nation - Customer Trial Final Report, 2019.
en. https://www.westernpower.co.uk/downloads/64378 (accessed July 20, 2021).
[3] EV database, Compare electric vehicles - EV Database, (n.d.). https://ev-database. [25] Affonso CM, Kezunovic M. Technical and Economic Impact of PV-BESS Charging
org (accessed June 7, 2021). Station on Transformer Life: A Case Study. IEEE Trans Smart Grid 2019;10:
[4] Charging Smart: Analysis & Recommendation for Next Generation Home EV 4683–92. https://doi.org/10.1109/TSG.2018.2866938.
Charging, 2021. https://www.pecanstreet.org/. [26] CENELEC, Voltage Characteristics of Electricity Supplied by Public Distribution
[5] Dudek E. The Flexibility of Domestic Electric Vehicle Charging: The Electric Nation Systems, 2005.
Project. IEEE Power Energ Mag 2021;19:16–27. https://doi.org/10.1109/ [27] Rubinstein RY, Kroese DP. Simulation and the Monte Carlo Method, John Wiley &
MPE.2021.3072714. Sons Inc, Hoboken, NJ. USA 2016. https://doi.org/10.1002/9781118631980.
[6] Quiros-Tortos J, Ochoa L, Butler T. How Electric Vehicles and the Grid Work
Together: Lessons Learned from One of the Largest Electric Vehicle Trials in the

12

You might also like