Applied Energy 224 (2018) 330–339

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Effects of charging battery electric vehicles on local grid regarding T

standardized load profile in administration sector
⁎
Erik Blasius1, Zhenqi Wang1,
Chair of Energy Distribution and High Voltage Engineering, BTU Cottbus-Senftenberg, Siemens-Halske-Ring 13, Cottbus 03046, Germany

H I GH L IG H T S

• Analysis of pool pattern of university charging station with real use case and data.
• Model to represent the fleet and daily pool pattern is proposed and validated.
• Strategy with delayed charging start to avoid local grid bottleneck introduced.

A R T I C LE I N FO A B S T R A C T

Keywords: Integration of battery electric vehicles (BEV) as load could have an impact on the stability of local grids. With the
Work&charge experiment conducted in a research project, real-time metering data of the charging points was collected, with
Electric vehicle user behavior which the charging process, user behavior and furthermore modeling of default pattern on a working-place
Charging strategy charging station park are presented and further discussed in this paper. The simultaneity of load peak from
Peak load reduction
uncontrolled charging processes and from the local load profile could cause local grid bottleneck. A BEV pool
model to generate random charging sessions on a workday, which conforms the real behavior in this research
case, is proposed. Charging strategies were developed and considered as centralized or decentralized. The si-
mulation shows the influence of delay charging start time and demonstrate the potential of peak load reduction
from charging of several BEV and furthermore of a large-scale construction of charging infrastructure.

1. Introduction in low and medium voltage grids [2].

Energy storage capacities are very important in order to replace the
conventional power plants by wind and PV without jeopardizing the
grid stability, which means above all to balance generation and con-
sumption in all time frames. For the foreseeable future, the problem of
lacking flexibilities on the demand side will not be relieved.
Due to geography in Germany, installed generation capacity from
wind and solar has the largest share among all the renewable energy
sources, while the share of hydro-based generation and storage capa-
cities are relatively low [1]. As a result, the increment of market share
of installed generation capacity with high fluctuation in the German
power system will still last, because the power generation from wind
and solar are closely related to the real-time meteorological situation.
The parallel planning of traffic transition relying on renewable en-
ergy, the use of electric mobility (especially Battery Electric Vehicle-
BEV) and thus the increasing construction of charging infrastructure
will play an increasingly important role as integrated energy consumers
In this paper the main topic will be pinned on the analysis of typical
load behavior of a charging station park in the administration sector
(office building). For energy consumers in administration or industrial
sector, which have a high share of electricity consumption in Germany,
the power supply will be out of technical reason directly connected to
the grid through its own grid connection point, e.g. directly connected
via transformer to the medium-voltage grid, the load profile will either
be real-time measured or estimated with standard load profile (SLP)
depending on the total annual energy consumption [3]. However for
high annual energy consumption in the administration and industrial
sector an additional connection price calculated with the peak load
power will be charged, which is defined by the highest average load
power in a quarter hour timespan of the year. This surcharge can easily
exceed multiple thousand EUR [4].
With additional BEV charging in grids with simple structure (e.g. at
the end of single line grids), it could cause problem like increment of
peak load power, which can lead to significant voltage drops even at

⁎
Corresponding author.
E-mail addresses: Erik.Blasius@b-tu.de (E. Blasius), Zhenqi.Wang@b-tu.de (Z. Wang).
1
Both authors contributed equally to this work.

https://doi.org/10.1016/j.apenergy.2018.04.073
Received 29 January 2018; Received in revised form 29 March 2018; Accepted 26 April 2018
0306-2619/ © 2018 Elsevier Ltd. All rights reserved.
E. Blasius, Z. Wang Applied Energy 224 (2018) 330–339

Nomenclature Pt ,pool BEV Pool load power at t (kW)

Pt ,SLP SLP load power at t (kW)
Indices PoolFleet daily pool pattern for BEVs from Fleet
rnd random number in range [0, 1]
i index of BEV SoCsi starting state of charge of BEV i (%)
t time of day SoCti state of charge of BEV i at t (%)
SoCcci ,cv state of charge of BEV i at CCCV-phase conversion (%)
Variables SoCei charging end state of charge of BEV i (%)
i
Tsimp ,chrg charging duration of BEV i with simplified battery model
λ time constant of BEV in CV-Phase (Hrs)
BEVModeli BEV model of BEV i tai time of day for arrival of BEV i of (h)
i
ci boolean state for BEV i of Fleet appearance on the charging Tchrg ,cc charging duration of BEV i in CC-Phase (Hrs)
i
point on the day Tchrg ,cv charging duration of BEV i in CV-Phase (Hrs)
i
Ecap battery capacity of BEV i (kWh) i
Tchrg charging duration of BEV i (Hrs)
Edi energy demand of BEV i (kWh) tdi
time of day for departure of BEV i of (h)
Fleet designated BEV fleet consists of BEVs tei time of day for charging end of BEV i of (h)
Ncon number of connected BEV i on a day (kWh) Tpi parking duration (duration of stay) of BEV i (Hrs)
i
Pmax maximum charging power of BEV i (kW) tsi,central time of day for charging start of BEV i from centralized
Pti,chrg charging power for BEV i at t (kW) strategy (h)
Pti,max maximum charging power for BEV i at t (kW) tsi,decentral time of day for charging start of BEV i from decentralized
Pti,simp,chrg charging power with simplified battery model at t (Hrs) strategy (h)
Pmax ,load maximum load power appeared (kW) tsi,latest latest time of day for charging start of BEV i of (h)
Pmax ,SLP maximum load power of SLP (kW) tsi time of day for charging start of BEV i of (h)
Pt ,exce excessive load power at t (kW) Texce nominated duration of excessive load appearance in %
Pt ,load total load power at t (kW)

low penetration levels [5,6] and also extra energy losses [7].
The data to be researched are originated from a charging station
park consisting of 15 charging points and an aggregator, which un-
dertakes SCADA tasks. 15 BEVs (vehicle data in Table 1) were assigned
to employees of university for business and private use. The use as well
as charging in the charging station park were free of charge.
As a result, journeys took place primarily in sense of shuttle between
residence and the charging station park at BTU with additionally
journeys for private purposes. The users can therefore be characterized
as commuters in a work&charge use case. The charging processes were
conducted uncontrolled, which operated via a plug&charge logic
(‘dumb charging’). Such a user pattern at an office charging station park
can lead to problem like local overloading, as research in [8] indicates.
To avoid or minimize the increase of peak load power of load profile
caused by uncontrolled BEV charging, the avoidance of simultaneity of
load peaks is necessary. This can be optimized, e.g. by load shifting or
peak shaving, as discussed in [9]. Other options are tariff charging or
smart charging, as described in [10]. Different central managed char-
ging strategies were already researched in [11]. The possible commu-
nication possibilities between charging station and backend are ex-
plained in detail in [12]. The development of a charging scheduler can
be carried out by different means, e.g. Genetic Algorithm [12] and
Markov Decision Process [13]. For the special application case of
charging station in office building research works like [14,15,8,16]
have been done. The general pool pattern at a business charing station
was studied in [8]. In [14] a two-stage charging strategy for the com-
bination of charging and renewable generation in which the main op-
timization goal is to response to electricity market and pv-forecasting.
Momber et al. [16] studied the binding of plug-in hybrid EV into the
building energy management system to minimize the operation cost
with Distributed Energy Resources Customer Adoption Model (DER-
CAM). In [15] a real-time water-filling algorithm for local peak shaving
at non residential charging station was proposed and analyzed.
In this paper, a charging strategy is introduced, which essentially
based on the influence of the charging start time of connected BEVs.
The core of the strategy is to decouple the charging start time from the
arrival time (connection of BEVs to charging points), consequently to
start the charge at a later time. Unlike delayed charging strategy,
presented in [17], this work also takes into account the limitation of the
parking duration while working and the effects on the SLP G1 used for
the administration sector in Germany. The cases to be discussed are
limited on the charging processes which takes place on normal working
days, during while high peak load power is expected, and mainly from
those, who belongs to the employee group. Spontaneous charge, which
can be categorized as charging with very short charging times and thus
with less importance for flexibility purposes, will not be discussed. This
paper is organized as follows: Section 2 the pool charging load pattern
and the peak load problem caused by charging with additional SLP load
will be reviewed. In Section 3 the user behavior of BEV pool will be
modeled and generalized with statistical method for deeper investiga-
tion purposes. In Section 4 charging strategy is introduced and in Sec-
tion 5 testified and analyzed with the project fleet and a more re-
presentative fleet of BEV.
2. Problem description
Each of the 15 charging points is equipped with a measuring device,
which measures the active power in three phases over time. In order to
investigate the impact of BEVs on the power system, all connected BEVs
are considered as participants in a BEV pool and thus being analyzed.
There were totally 1075 charging sessions recorded including their real-
time active power and total charged energy over a seven months period.
2.1. Charged energy in a week
Due to the underlying behavior of vehicle users as commuters, BEVs
Table 1
German E-Cars CETOS technical data.
Item Value
Battery type Li-ion
Useable battery capacity Ecap 17.1 kWh
Nominal range 120 km
Max. charging power Pmax 9.5 kW(3 ∼ )
Phase 1- or 3-phase

331
E. Blasius, Z. Wang
are usually charged at the charging points only on normal workdays
(weekdays without national holidays) (Fig. 1).
The mean of charged energy per day on workdays is between 60 and
70 kWh. The box plot make it very clear that charged energy on
weekend is very rare. Therefore the analysis carried out in this paper
will be limited on the normal workdays.
2.2. Daily behavior of BEV pool on normal workdays
The power curve of BEV pool and the availabilities of BEV are re-
calculated into mean value in a quarter timespan and then represented
in Fig. 2 as box plot. The availabilities presented are nominated value
regarding to 15 BEVs in total. Since the BEVs are charged uncontrolled,
the charging process will always start with maximum permissible
power after plugging in. As a result, high peak loads occurred between
7:30 am and 10:00 am in the morning hours in comparison to the total
daily load (arrival of the majority of BEVs). The cumulation of in-
dividual charging processes lead to the high peak load power in the
morning, because of the simultaneity of arrival and the maximal
charging power at the starting time. In contrast, it turns out that the
vehicles are available over a relatively long period of time at the
charging points, so it is not necessary to start charging immediately
after the plugging. The charging processes are open for relatively long
time windows with the charging requirement still fulfilled. After about
14:30 o’clock the BEVs start to leave the charging station park one after
another. Between 9:15 am and 2:45 pm, however, the availability is
relatively stable (darkened) and in most cases higher than 30% (at least
5 out of 15 vehicles).
2.3. Load pattern analysis
The objective of the analysis of the charging profiles caused by the
BEV pool is to investigate its impact on standardized load profile in the
administration sector. Hereby the analysis of the mutual influence of
peak load power is of crucial importance, which are decisive for the
grids planing and the lifetime of grid equipment like cables, switches
and transformers (avoidance of grid bottlenecks, overheating) [9]. The
research will be conducted under following conditions:
• Grid structure: Simple unmeshed grid structure (radial grid).
• Grid connection point: all charging points are connected through
one connection point to the distribution grid.
• Standard load pattern: using SLP to simulate average load pattern in
administration sector.
• Pool pattern: using the average charging power of BEV pool on
normal workdays.
2.3.1. Load pattern with standard load profile (SLP)
A SLP is a representative load profile, which are used by the German
grid operators nationwide to predict the consumption behavior of end
consumers, who do not have a registered power measurement and
consume less than 100,000 kWh/a [3,18]. Different variants of SLP are
designed to categorize different types of end consumers (e.g. for
households and commercial). In this paper, the SLP G1 for general
commercial and administration end consumer will be used for analysis,
which e.g. represents the electricity consumption of office buildings.
The data originates from the GIPS project published by Stadtwerk Unna
[19]. The profile is normalized with its maximum load power as shown
in Fig. 3. Fig. 4 shows an example of load history on a winter day,
assuming that the maximum load power of the BEV pool is at 50%
penetration rate. In this case, the maximum load of the resulted load
profile is increased by 50% comparing to the original SLP G1. Con-
sidering that with different penetration levels of BEV pool, the incre-
ment of peak load power of resulted load profile is also varying. But
there is an obvious correlation between the peak power of the BEV pool
under uncontrolled charging and the peak power of the resulted load
332
Applied Energy 224 (2018) 330–339
profile, since the respective peaks of the two load profiles occur at the
same time of a day. This analysis proves that with increasing penetra-
tion rate of BEVs and the corresponding uncontrolled charging the local
grid connection point is additionally burdened, the critical limits for
load power and equipment temperature can be exceeded.
3. Modeling
In this section, the main goal is to discuss the technical realization of
creating a model, which can represent the scenarios to happen in a real-
life administration sector charging station park. The methodology and
definition of general mathematic model for pattern generations are
introduced. With the project data the model for project fleet will be
created and also validated. A possible model expansion for a real-life
fleet will also be the last focus in this section.
3.1. Methodology of modeling
Any charging processes will only take place when the vehicle is
connected to the charging point. The power curve during the charging
process are mainly decided by total energy demand, maximum charging
power, battery characteristics and if existed external controlling. For
the simulation of vehicle pool charging processes on a working place
charging station park, with which optimization strategy can be eval-
uated, a stochastic model without continuous daily correlation is pro-
posed. The model can generate stochastic daily cases, which could
possibly happen for a designated fleet. The main factors and flow dia-
gram are shown in Fig. 5. The input data of the modeling is the given
Fleet with registered BEVs consisting of one of more BEV models. These
definition can also be extended with other charging relevant parameters
e.g. temperature and λ (parameter for the CV-phase) for battery char-
acteristics, as defined as follows:
i i
Fleet = {〈i,BEVModeli,Ecap,Pmax ,⋯〉,⋯} (1)
On a working day the charging pattern Pooln,Fleet for the predefined
Fleet can be defined as:
PoolFleet = {〈BEV i,c i,tai,Tpi,Edi 〉,⋯
| ∀ BEV i ∈ Fleet ,
c i ∈ {False,True},
tai ∈ [0,24],Tpi ∈ [0,24],
Edi ∈ [0,Ecap
i
]} (2)
The factors shown in hexagons in Fig. 5 are the parts (c i,tai,Tpi,Edi )
to
be modeled or specified, which be done manually with historical data
or generated stochastically, in order to make the model work. A method
to generate variables ( fvariable ) will be proposed, with which the prop-
erties observed in the project can be represented. The general method is
to fit the observed parameter with empirical distributions e.g. with
normal distribution ( X ∼ N (μ, σ 2) ) similar to method in [20], and then
to sample stochastically (random value generation) from these dis-
tributions, with which wished number of days (daily pool patterns) can
Fig. 1. Charged energy in the pool during a week.
E. Blasius, Z. Wang
Fig. 2. (a) Charging power of BEV pool on normal workdays. (b) Relative BEV
availability of the pool on normal workdays.
Fig. 3. SLP G1 on normal workdays.
Fig. 4. Example load in winter.
be generated. In this paper the proposed method is realized with SciPy
Statistic Toolbox [21]. For the distribution fitting part the method
‘Maximum Likelihood Estimation’ is by default implemented [21,22].
The histogram and Kernel Density Estimation (KDE, a finer demon-
stration of histogram [22]) are applied for the visualization of the
original distribution of data and respectively the Probability Density
Function (PDF) defined with F (·) for the fitted distribution.
While sampling from given continuous distribution can be achieved
with ‘Inverse Transform Sampling’ method, the core of which is using
uniform distribution U (0,1) to generate random value defined as rnd,
which can be easily achieved on a computer, and then with reversed
PDF F −1 (·) expected sample x1,⋯x n can be calculated as x n = F −1 (rndn )
333
Applied Energy 224 (2018) 330–339
Fig. 5. Modeling process vehicle pool and simulator [10].
[23]. fvariable (rnd ) is defined as the function to generate variable from
rnd in this paper.
3.2. Pool pattern modeling
The fleet to be modeled are 15 project BEVs with the model data
defined in Table 1 denoted by Fleet proj with |Fleet proj| = 15. In this sec-
tion f c ∀ ,proj (rnd ),f t ∀ ,proj (rnd ),f T ∀ ,proj (rnd ),f E ∀ ,proj (rnd ) will be introduced.
a p d
3.2.1. Vehicle connections, arrival time, parking duration
The number of vehicle connections gives information about the
number of actual connected BEVs at the charging station on a working
day. In practice, the daily monitored vehicle connections fluctuate and
are much lower than expected. Users do not charge everyday, but al-
ways when they need to. For example, users do not have to charge if
they live only a few kilometers away from the working place while the
battery still has a relatively high State of Charge (SoC), by which sa-
tisfying travel range can still be provided. The possibility of vehicle
connection of each vehicle will then be recalculated with the vehicle
pool connected vehicles. The observed vehicle connections can be fitted
well with normal distribution (Gaussian distribution), as the PDF of
which is stated in Eq. (3) and the fitting result shown in Fig. 6a:
1 2 2
P (x ) = e−(x − μ) /2 − σ
σ 2π (3)
The total number of connected BEVs can be described as:
proj
Ncon ∼ N (7.35,2.112 ) (4)
E. Blasius, Z. Wang Applied Energy 224 (2018) 330–339

Table 2
The groups of users.
Group Distance (work-home) Number of BEVs Charging sessions

1 40–60 km 3 229
2 20–40 km 3 263
3 10–20 km 3 249
4 <10 km 6 334

The parking duration is fitted with normal distribution (Fig. 6c, Eq.
(8)). Since some employees of the BTU work under part-time employ-
ment contract (mainly 50% of a full-time job with 8–10 h per day), thus
a higher density around 4–5 h can be observed. To be able to fit the
general pattern, the fitting will be carried out in the range of 4–14 h.
The Tpi,proj can be calculated with Eq. (9).

Tpproj ∼ N (8.44,1.672) (8)

−1
f T ∀ ,proj (rnd ) = FN (rnd )
p (8.44,1.672) (9)

3.2.2. Energy demand

The energy demand of BEV depends highly on the traveled distance
while partially influenced by driving style. In the project, the BEV users
were classified into four groups considering their distances between
residence and workplace (Table 2).
The violin plot of session’s charged energy is shown in Fig. 7.
Therefore, it can be observed that there is a correlation between dis-
tances (work-home) and charged energy. Group 4 has comparatively
fewer sessions despite six BEVs, because apparently less charging were
required for the closer daily shuttle distances comparing to other three
groups. To model the BEV pool, the energy demand per charging ses-
sion can be approximately fitted with an equal distribution (purple
envelope). In order to reuse the model to other cases this distribution
can be free defined according to specific distance data. Thus a uniform
distribution will be used to describe Edproj :

Edproj ∼ U (0,Ecap
proj
) (10)
proj
f E ∀ ,proj (rnd ) = Ecap ·rnd (11)
d

3.3. Battery model

The typical charging curve of a Li-ion battery is determined by the

CCCV-method. During CC-phase the battery is charged with constant
current until the charging end voltage of the first battery cell is reached.
Thereafter, the battery is charged during CV-phase with constant
charging voltage and decreasing current until final rated voltage of the
battery is reached [24]. Both phases should be considered in modeling,

Fig. 6. Distribution fitting.

With this distribution, c i,proj can be calculated as follows:

−1
⎧True FN (7.35/| Fleet proj |,(2.11/| Fleet proj |)2) (rnd ) ⩾ 0.5
f c ∀ ,proj (rnd ) = −1
⎨ False FN (7.35/| Fleet proj |,(2.11/| Fleet proj |)2)
(rnd ) < 0.5
⎩ (5)
The arrival time of BEVs between 5 o’clock and 11 o’clock is fitted
with normal distribution (casual arrivals during other time points are
neglected since they are rare and have little influence on peak load
power), as this gives a good fitting to the observed values (Fig. 6b, Eq.
(6)) the calculation of tai,proj are stated in Eq. (7).
taproj ∼ N (8.32,1.012 ) (6)
−1
f t ∀ ,proj (rnd ) = FN (rnd )
a (8.32,1.012 ) (7) Fig. 7. Violin plot of charged energy in groups.

334
E. Blasius, Z. Wang Applied Energy 224 (2018) 330–339

in order to obtain realistic load behavior for the simulation. Eq. (12) 4.1. Time flexibility of charging BEV
shows the mathematical modeling of charging process of the battery.
The related parameters λ (the time constant during the CV-phase), In order to fully charge the vehicle in time, the time window of BEV i
SoCcc,cv (the SoC conversion point of the CCCV-phase) must be defined in which the charge can be delayed can be defined between tai and latest
beforehand. The impact of other factors on the charging process (tem- possible charging start time Tsi,latest . Tsi,latest can be calculated with time of
peratures, etc.) will not be considered. departure tdi and charging duration Tchrg i
and also limited by tai , as
i
⎧ Pmax 0 < SoCti < SoCcci ,cv
Pti,max = i
⎨ Pmax ·e−λt SoCcci ,cv < SoCti < 1 (12)
⎩
According to [24] the λ can be calculated with SoC cc,cv as follows:
i
Pmax 1
λi = i
·
Ecap 1−SoCcci ,cv (13)

For the BEVs in the project, the parameters are defined with the data
in project as follows:

SoCccCetos
,cv = 0.95 (14)

Thus Fleet model can be extended as:

i i
Fleet = {〈i,BEVModeli,Ecap,Pmax ,SoCcci ,cv 〉,⋯} (15)

3.4. Model validation

The model can be validated with comparison to original BEV pool

on main characteristics: the BEV availability and pool power during
daytime (Fig. 8a and b) and the energy demand of BEVs (Fig. 8c). As the
comparison implies, the model fits well to represent the real behavior,
with which further simulation of strategy can be sufficient accurately
carried out. As a result the project fleet can be defined as:
Fleet proj = {〈1,Cetos,17.1 kWh, 9.5 kW, 95%〉,⋯} with |Fleet proj| = 15
(16)
with the pool pattern generation functions defined in Eqs. (5), (7), (9),
(11).

4. Charging strategy

From Fig. 3, it can be assumed that the load power decreases since
lunchtime and remains at a lower level respectively to the peak load in
the morning. A simple charging strategy would therefore to start
charging after lunchtime and as late as possible, while the BEVs can still
be fully charged (no matter how high the starting SoC is). This does not
necessarily require communication to the backend since the start time
can be approximately generated decentralized.
In practice, due to the inaccessibility of data through vehicle and
charging point communication according to IEC 61851-1 [25], it is
necessary to use other communication channels. An essential require-
ment of smart charging strategies is a sufficient information exchange.
Entering relevant data can be easily realized from the user via smart
phone or through purchasing a charging ticket. The vehicle data e.g.
information about the maximum permissible charging power of the
battery and the battery capacity can be stored in the user profile, which
are transmitted in the authentication phase. Another possibility could
be automatic exchange between BEV and charging point through digital
communication, as already implemented while fast charging on a DC
charging point.
In this section, two charging strategies, which are designed for such
a system without and with centralized management, are introduced for
the local bottleneck prevention at working place charging station park,
both of which influence only the charging start time tsi , under the
condition that informations about the charging session is known to the
charging point or central management system. With BEVModel,ta,Tp,Ed
defined, the time flexibility of connected BEVs can also be calculated,
which are the essential factors for the charging strategy. Fig. 8. Model validation result.

335
E. Blasius, Z. Wang Applied Energy 224 (2018) 330–339

follows: as:
tdi = tai + Tpi (17) Pmax ,SLP = max ({Pt ,slp | ∀ t }) (27)

tsi,latest = max ({tdi−Tchrg

i
,tai}) Pt ,exce can be calculated as:
(18)
i Pt ,exce = max ({0,Pt ,load−Pmax ,SLP }) (28)
With vehicle model parameter and battery model introduced Tchrg
i
can be calculated with the charging duration in CC-phase (Tchrg,cc ) and with:
i
CV-phase (Tchrg ,cv ): Pt ,load = Pt ,SLP + Pt ,pool
i i i
Tchrg = Tchrg ,cc + Tchrg ,cv (19) Pt ,pool = ∑ Pti,chrg , ∀ i|c i = True (29)
With starting state of charge SoCsi ⩾ SoCcci ,cv,Tchrg
i
,cc equals 0, else For BEV i Pti,chrg can be calculated as:
i
Tchrg ,cc can be calculated as:
⎧0 t ∉ [tsi,tei]
i
Tchrg i i i i
,cc = (SoCcc,cv−SoCs )· Ecap/ Pmax (20) Pti,chrg =
⎨ Pti,max t ∈ [tsi,tei] (30)
⎩
The charging process in CV-phase from SoCt to SoCe with
As a result optimization goal is defined as minimizing Obj:
SoCcc,cv ⩽ SoCt ⩽ SoCe can be expressed as:
1 t
Obj = max ({Pt ,exce | ∀ t }) (31)
SoCe = SoCt + Ecap
∫
· 0 Pchrg,cv dt
t
Genetic algorithm will be implemented for the optimization, which
∫0 Pchrg,cv dt = (SoCe−SoCt )·Ecap is a search algorithm based on natural principle, with searching ran-
t
Pmax · ∫0 e−λt dt = (SoCe−SoCt )·Ecap domly and information exchange an optimal result can be expected
(21)
[26]. In this work, the algorithm is realized in python with DEAP-fra-
i
After evaluating the integral with t = Tchrg ,cv for Eq. (21), result can mework [27].
be calculated as: The main processes of genetic algorithm are shown in Fig. 9. The
i
−λTchrg process starts with definition of initial population, and then with se-
1−e ,cv
= (SoCe−SoCti )·Ecap
i i
/ Pmax lection, crossover and mutation simulating the law of nature, the al-
λi
ln (1 − (SoCe − SoCti)·λi·Ecap
i / Pi ) gorithm is searching for an efficiently optimized result in the defined
i max
Tchrg ,cv = − λi (22) space [28]. In this work, in order to improve the computational per-
formance, discrete time steps are used for optimization as 24 × 4 15-
The resulted flexibility time range to start charge is defined as: min time blocks, thus the BEV time parameter as input data will be
tsi ∈ [tai,tsi,latest ] (23) rounded to the nearest quarter. The initial population can be calculated
as:
The charging time window between [tsi,tei], with charging end time
limited by tdi can be defined as: Popinit = {〈rnd·(ts1,latest −ta1) + ta1,⋯rnd·(tsi,latest −tai) + tai 〉,⋯} (32)
tei = min ({tsi + i
Tchrg ,tdi}) (24) with i = {i| ∀ ∈BEV i Fleet ,c i
= True} . The size of initial population are
set as 50. The Fitness function to minimize is defined in Eq. (31) with
simplified battery model for discrete time steps as defined in Eq. (33).
4.2. Decentralized strategy
i i i
Tsimp ,chrg = Ed / Pmax
In case of decentralized strategy, the charging points will not be Pti,simp,chrg = Pmax
i
t ∈ [tsi,tsimp
i
,e ] (33)
coordinated with each other, thus according to the user defined flex-
ibility and the load pattern. For such a use case, in which a complex ICT The selection process will be realized with standard tournament
system is not required, it would be a feasible method to avoid local selection with a size of five and crossover process with standard two-
overloading, with Tsi set to the latest possible charging start time: point crossover at the crossover rate of 0.7, both of them are predefined
in DEAP-Framework [27]. The mutation will be realized with rando-
tsi,decentral = tsi,latest (25)
mizing at the encoding point within the allowed searching space at the
probability of 0.05. The resulting population with the best fitting value
4.3. Centralized strategy will be than be processed to continuous time line with randomizing the
charging start time to the quarter block ahead of tsi,central,best . This re-
In case of centralized strategy, with known flexibility of BEV i , the sulted as:
coordinator will search for the tsi,central within the allowed time windows
tsi,central = tsi,central,best −0.25·rnd (34)
for all connected BEVs defined as Eq. (26).
{tsi,central | ∀ BEV i ∈ Fleet ,c i = True,tsi ∈ [tai,tsi,latest ]} (26)
5. Case study
The key point is to minimize the positive excessive load Pt ,exce , in
which the base line is set with the maximum load power of SLP defined In this section, the impacts of the proposed charging strategies and

Fig. 9. Genetic algorithm processes.

336
E. Blasius, Z. Wang Applied Energy 224 (2018) 330–339

the uncontrolled case will be analyzed with the project fleet (Fleet proj ) Table 3
and a second more representative fleet (Fleet repr ), which consists of Definition of BEV classification [30].
different types of BEVs. A BEV combination will be proposed in this Vehicle class Energy consumption Typical Ecap Typical Pmax Percentage
section, with which a potential future scenario in Germany can be re- (kWh/km) (kWh) (kW)
presented. For both fleets the simulation will be conducted under two
charging levels, which represent a charging station park consisting of 1- Small 0.185 20 3.7/11.1 50
Midsize 0.226 24 3.7/11.1 40
phase or 3-phase charging points with different maximum charging
Large 0.243 28 3.7/11.1 10
power [29]. Higher charging rate, which is already realized with DC
charging points, will not be included in this study, since the cost of
construction of such a charging station park with a number of charging 5.1. Simulation results with project fleet
points is considerable. Test cases Fleet proj with 15 BEVs and Fleet repr
with 20, 50, 100 BEVs will be evaluated for the impacts on local load For Fleet proj the simulation was carried out in two charging levels:
pattern and the optimization effects from charging strategies. SLP will with 3.7 kW at Level 1 and 9.5 kW at Level 2 considering the charging
be scaled assuming the penetration rate of the fleet charging at the same point type whether as 1-phase wallbox or 3-phase charging point (ac-
time under each test case will be 50%, this is the condition for all fleet cordingly the BEV limits). The results of simulation are shown in Fig. 10
combinations at all charging levels. in form of average power curve (Fig. 10a, b), load duration curve (LDC)
(Fig. 10c, d) with x-axis as percentage of the simulation duration and y-
axis as penetration rate comparing to SLP load. And also charging start
time distribution as reference (Fig. 10e, f).

Fig. 10. Simulation result for project fleet over 200 Workdays.

337
E. Blasius, Z. Wang Applied Energy 224 (2018) 330–339

Fig. 11. LDC for representative fleet over 200 Workdays.

At both charging levels the both strategies delivered satisfactory with:

results, with which the peak load power Pmax ,load was reduced from
150% peak to less than 110% peak in both levels and duration with |Fleet small,repr | = 25,
excessive load power Texce was reduced from 15% to 5% at Level 1 and |Fleet midsize,repr | = 20,
10% to 2% at Level 2 . Both strategies delivered similar results for |Fleet large,repr | = 5 (36)
Fleet proj . At Level 2 Pmax ,load and Texce are even better optimized than at
The daily pool parameters can be approximately determined with
Level 1, since with higher charging power there is more time flexibility
the results in project as in Eq. (37).
for the pool.
Poolrepr = {〈BEV ,C i,tai,Tpi,Edi 〉| ∀ BEV ∈ Fleet repr } (37)
5.2. Analysis with a representative fleet
with:

In order to apply the model onto a scenario, which could possibly f c ∀ ,repr = f c ∀ ,proj
happen in next years, a fleet with BEVs from different categories de- f t ∀ ,repr = f ta∀ ,proj
a
fined with vehicle size can be differed from each other in energy re-
f T ∀ ,repr = f Tp∀ ,proj
levant parameters. Based on [30], a possible fleet combination is de- p

fined in Table 3. For example, assuming the size of the fleet is 50, fleet i,repr
f E i,repr = Emax ·rnd
d (38)
can be defined as:
Fleet with 3 different numbers of BEVs (20, 50, 100) at two charging
Fleet repr = ∪Fleet class, levels at charging power of 3.7 kW and 11.1 kW will be studied, since
∀ class ∈ {Small,Midsize,Large} (35) the number of BEVs may have impact on the local load pattern with or

338
E. Blasius, Z. Wang
without charging strategy.
As the enlarged LDC in Fig. 11 for the universal case implies, both of
the strategies are effective in reducing Pmax ,load and Texce . At Level 1
centralized strategy have slight advantage over decentralized strategy
at optimizing Pmax ,load , however at the increase of number of BEVs both
strategies have less impact on Texce while slightly reduced more of
Pmax ,load . At Level 2 both strategies are much more effective as at Level 1,
but with the increase of number of BEVs Texce are slightly larger than
before.
6. Conclusion and prospect
In this paper, with data from the project, analysis of BEV pool be-
havior with ‘dumb-charging’ and its impact on local load profile is
conducted, with which a model to represent the general BEV pool
characteristics is proposed and charging strategies for optimize the
local peak load power decentralized and centralized are proposed and
tested. Following results can be yielded:
• The characteristics of BEVs at charging points are mainly de-
termined by the user behavior.
• Uncontrolled charging of BEVs could lead to a considerable incre-
ment of the peak load power in administration sector.
• With delayed charge, peak load power can be effectively reduced,
which helps prevent local grid bottleneck.
The results show that even relative simple charging strategy can be
sufficient for the integration of large scale charging processes from
BEVs in existing power system. Based on such a strategy the construc-
tion of charging infrastructure at a large number but limited individual
maximum charging power can be carried out without further grid ex-
tension. Such a strategy can also be implemented in the charging
strategy toolbox because of its flexibility and low requirements for ICT-
support. Furthermore, with the model introduced in this paper, re-
search for the possibility of the stability contribution to the power
system and also relevant business models can be conducted.
Acknowledgment
The work was created as part of the project SMART Capital Region
2.0 in the joint project WindNODE (the showcase for intelligent energy
from the northeast Germany) from 2016 to 2020. WindNODE is one of
five selected German model regions of the SINTEG program (showcase
intelligent energy – digital agenda for the energy transition) funded by
the Federal Ministry of Economic Affairs and Energy.
References
[1] Kreyenberg D, Lischke A, Bergk F, Duennebeil F, Heidt C, Knoerr W, et al.
Erneuerbare Energien im Verkehr Potenziale und Entwicklungsperspektiven ver-
schiedener erneuerbarer Energieträger und Energieverbrauch der Verkehrsträger.
< http://elib.dlr.de/96208/ > .
[2] Schmidt P, Raksha T, Jöhrens J, Lambrecht U, Gerhardt N, Jentsch M. Analyse von
Herausforderungen und Synergiepotenzialen beim Zusammenspiel von Verkehrs-
und Stromsektor.
[3] Schumacher I, Würfel P, editors. Strategien zur Strombeschaffung in Unternehmen.
Wiesbaden: Springer Fachmedien Wiesbaden; 2015. http://dx.doi.org/10.1007/
978-3-658-07422-7.
[4] Bolay Sebastian Dr. Faktenpapier Strompreise in Deutschland 2017: Bestandteile,
Entwicklungen, Strategien.
[5] Clement-Nyns K, Haesen E, Driesen J. The impact of charging plug-in hybrid electric
vehicles on a residential distribution grid. IEEE Trans Power Syst
339
Applied Energy 224 (2018) 330–339
2010;25(1):371–80. http://dx.doi.org/10.1109/TPWRS.2009.2036481.
[6] Ying S. Auswirkungen auf die Niederspannungsnetze bei hoher Penetration von
innerstädtischen Photovoltaikanlagen und Elektrofahrzeugen, Doctoralthesis. BTU
Cottbus – Senftenberg; 2012. < https://opus4.kobv.de/opus4-btu/frontdoor/
index/index/docId/2363 > .
[7] Veldman E, Verzijlbergh RA. Distribution grid impacts of smart electric vehicle
charging from different perspectives. IEEE Trans Smart Grid 2015;6(1):333–42.
http://dx.doi.org/10.1109/TSG.2014.2355494.
[8] Islam MS, Mithulananthan N. Daily EV load profile of an EV charging station at
business premises: Melbourne, Australia, Nov 28-Dec 1, 2016; 2016. p. 787–92.
http://dx.doi.org/10.1109/ISGT-Asia.2016.7796485. < http://ieeexplore.ieee.
org/servlet/opac?punumber=7786610 > .
[9] Godina R, Rodrigues EM, Matias JC, Catalão JP. Smart electric vehicle charging
scheduler for overloading prevention of an industry client power distribution
transformer. Appl Energy 2016;178:29–42. http://dx.doi.org/10.1016/j.apenergy.
2016.06.019.
[10] Garcia-Valle R, editor. Electric vehicle integration into modern power networks,
power electronics and power systems. New York: Springer; 2013.
[11] Zhang L, Jabbari F, Brown T, Samuelsen S. Coordinating plug-in electric vehicle
charging with electric grid: valley filling and target load following. J Power Sources
2014;267:584–97. http://dx.doi.org/10.1016/j.jpowsour.2014.04.078.
[12] Detzler S. Lademanagement für Elektrofahrzeuge. Karlsruhe: KIT Scientific
Publishing; 2017.
[13] Walraven E, Spaan MTJ. Planning under uncertainty for aggregated electric vehicle
charging with renewable energy supply. FAIA 2016;285:904–12. http://dx.doi.org/
10.3233/978-1-61499-672-9-904.
[14] Wu Di, Zeng H, Lu C, Boulet B. Two-stage energy management for office buildings
with workplace EV charging and renewable energy. IEEE Trans Transport Electrif
2017;3(1):225–37. http://dx.doi.org/10.1109/TTE.2017.2659626.
[15] Zhang G, Tan T, Wang G. Real-time smart charging of electric vehicles for demand
charge reduction at non-residential sites. IEEE Trans Smart Grid 2017:1. http://dx.
doi.org/10.1109/TSG.2016.2647620.
[16] Momber I, Gómez T, Venkataramanan G, Stadler M, Beer S, Lai J, et al. Plug-in
electric vehicle interactions with a small office building: an economic analysis using
DER-CAM. In: IEEE PES General Meeting, Minneapolis, MN; 2010. p. 1–8. http://
dx.doi.org/10.1109/PES.2010.5589485.
[17] Götz A. Zukünftige Belastungen von Niederspannungsnetzen unter besonderer
Berücksichtigung der Elektromobilität: Dissertation. Professur Energie- und
Hochspannungstechnik, Fakultät für Elektrotechnik und Informationstechnik,
Technische Universität Chemnitz.
[18] Bundesregierung, Verordnung über den Zugang zu Elektrizitätsversorgungsnetzen:
(Stromnetzzugangsverordnung - StromNZV); 12/2017.
[19] Stadtwerke Unna GmbH. VDEW-Standardlastprofile [04.12.2017]. < http://www.
gipsprojekt.de/featureGips/Gips?SessionMandant=sw_unna&Anwendung=
EnWGKnotenAnzeigen&PrimaryId=132963&Mandantkuerzel=sw_unna&
Navigation=J > .
[20] Qian K, Zhou C, Allan M, Yuan Y. Modeling of load demand due to EV battery
charging in distribution systems. IEEE Trans Power Syst 2011;26(2):802–10. http://
dx.doi.org/10.1109/TPWRS.2010.2057456.
[21] The Scipy community. Statistics (scipy.stats) — SciPy v1.0.0 Reference Guide [25.
10.2017]. < https://docs.scipy.org/doc/scipy/reference/tutorial/stats.html > .
[22] NIST/SEMATECH. e-Handbook of Statistical Methods [10/30/2013]. < http://
www.itl.nist.gov/div898/handbook/ > .
[23] Lemieux C. Monte Carlo and quasi-Monte Carlo sampling, Springer series in sta-
tistics. New York NY: Springer Science + Business Media; 2009.
[24] Vagropoulos SI, Bakirtzis AG. Optimal bidding strategy for electric vehicle ag-
gregators in electricity markets. IEEE Trans Power Syst 2013;28(4):4031–41.
http://dx.doi.org/10.1109/TPWRS.2013.2274673.
[25] Leitinger C, Litzlbauer M. Netzintegration und Ladestrategien der Elektromobilität.
e & i Elektrotech Inform Tech 2011;128(1–2):10–5. http://dx.doi.org/10.1007/
s00502-011-0800-3.
[26] Walters DC, Sheble GB. Genetic algorithm solution of economic dispatch with valve
point loading. IEEE Trans Power Syst 1993;8(3):1325–32. http://dx.doi.org/10.
1109/59.260861.
[27] Fortin F-A, Rainville F-MD, Gardner M-A, Parizeau M, Gagné C. DEAP: evolutionary
algorithms made easy. J Mach Learn Res 2012;13:2171–5.
[28] Coley DA. An introduction to genetic algorithms for scientists and engineers. 3rd ed.
Singapore: WORLD SCIENTIFIC; 2005. http://dx.doi.org/10.1142/3904.
[29] Yilmaz M, Krein PT. Review of battery charger topologies, charging power levels,
and infrastructure for plug-in electric and hybrid vehicles. IEEE Trans Power
Electron 2013;28(5):2151–69. http://dx.doi.org/10.1109/TPEL.2012.2212917.
[30] Plötz P, Gnann T, Kühn A, Wietschel M. Markthochlaufszenarien für
Elektrofahrzeuge: Studie im Auftrag der acatech – Deutsche Akademie der
Technikwissenschaften und der Arbeitsgruppe 7 der Nationalen Plattform
Elektromobilität (NPE).