Content uploaded by Qiang Tang
Author content
All content in this area was uploaded by Qiang Tang on Nov 06, 2018
Content may be subject to copyright.
A Decision Function Based Smart Charging and Discharging Strategy
for Electric Vehicle in Smart Grid
Qiang Tang
1
&Mingzhong Xie
1
&Kun Yang
2
&Yuansheng Luo
1
&Dongdai Zhou
3
&Yun Song
1
#Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract
As the number of Electric Vehicle (EV) increases, the uncontrolled EV charging behaviors may cause the grid load fluctuations
and other passive effects. In order to balance the EV charging load in the smart grid, an Electric Vehicle Charging and
Discharging Strategy (EVCDS) which is based on a Charging Decision Function (CDF) as well as a Discharging Decision
Function (DDF) is proposed. In the CDF and DDF, there are three sub-functions related to the residual energy of battery, EV’s
charging habits, and the charging efficiency of charging station respectively. The residual energy of battery is used to calculate the
excepted probability that satisfies the user’s mileage requirement, which is an important factor. The charging habits are used to
calculate the second sub-function value, which stands for the user’s comfort. The charging efficiency and distance are combined
together to calculate the third sub-function. All the sub-functions are weighted and combined into the CDF and DDF to decide
whether to charge, discharge or do nothing. In the numerical results, we set up a scenario for the commercial vehicles and private
vehicles. After compared with other strategies, EVCDS performs well in terms of reducing the charging demand fluctuations and
improving the charging demand balance among charging stations.
Keywords Electric vehicle .Charging decision function .Discharging decision function .Charging demand balance .Smart grid
1 Introduction
In the past few years, the greenhouse effect has motivated the
development of the zero emission electric vehicles, which are
battery powered and will be the major participants in the trans-
port system [1]. In recent years, many countries have drawn
up some plans for the popularization of EV. For example, the
Chinese government has launched a Bten cities thousand
vehicles program^, and about two hundreds charging stations,
13 thousands charging piles are put into service [2]. The
International Energy Agency has set a target of having over
20 million EVs on roads by 2020 [3].
As the EV number increases, a lot of batteries are connect-
ed to the power grid, and the local power utilities will under-
take more charging burden. If there is no coordinated charging
strategy, the charging behaviors cannot be scheduled and the
power gird will suffer the overload and other passive effects.
For example, a 20% level of EV penetration would lead to a
35.8% increase in peak load demand in the uncontrolled
charging scenario [1]. Besides, the uncontrolled charging
and discharging, can also lead to the problems such as the
phase imbalance, voltage drop, harmonic wave and power
loss increment, etc.
In order to eliminate the passive effects, scheduling charg-
ing behaviors is inevitable. For a specific area, the electric
power utilization has peak stage and off-peak stage, and in
order to schedule the charging behaviors, the smart charging
strategy should be able to decide whether an EV should be
charged. If an EV needs to be charged, the smart charging
strategy should further decide where and when to charge. By
implementing the smart charging strategy, the charging load
can be arranged at the off-peak hour. Besides, in order to
*Qiang Tang
tangqiangcsust@163.com
Kun Yang
kunyang@essex.ac.uk
Dongdai Zhou
ddzhou@nenu.edu.cn
1
School of Computer and Communication Engineering, Changsha
University of Science and Technology, Changsha 410114, Hunan,
China
2
School of Computer Science and Electronic Engineering, University
of Essex, Colchester, Essex CO4 3SQ, UK
3
School of Information and Software, Northeast Normal University,
Changchun 130024, Jilin, China
Mobile Networks and Applications
https://doi.org/10.1007/s11036-018-1049-4
reduce the burden of charging station, the redundant energy
stored into some EVs can be discharged back to the charging
stations to satisfy other EVs’charging demand. The smart
charging strategy can be implemented in each EVand charg-
ing station, and actually the EVs and charging stations com-
pose a mobile social network as for the charging issue [4].
In this paper, we mainly focus on designing an EV charg-
ing and discharging strategy in the urban region, where there
are many EVs and the charging stations may be always
congested. The main contributions of this paper are proposing
a smart charging and discharging strategy that considers the
residual energy, user’s charging habit and the charging sta-
tion’s charging efficiency, and after combining the three fac-
tors together, the charging demand can be shifted efficiently
and the charging demand balance performance among differ-
ent charging stations is improved significantly. Compared to
our previous work accepted by the 1st EAI International
Conference on Smart Grid Inspired Future Technologies [5],
this paper has added the contents of discharge decision func-
tion, algorithms and flow chats for EVs and charging stations,
and the comparison scheme.
This paper is organized as follows. In Section 2the techni-
cal details of some related work is introduced. In Section 3,4
and 5, the charging and discharging decision functions are
introduced. In Section 6, the algorithms and flow chats for
EVs and charging stations are introduced. In Section 7,the
numerical results are presented.
2 Related work
In recent years, the EV charging and discharging strategies
have been studied in order to improve the peak-shaving per-
formance and reduce the congestion of charging stations. In
[6], Y. Cao et al. proposed an intelligent method to control
EV’s charging load by using the time-of-use (TOU) price.
At first, an optimization model is formulated to minimize the
charging cost, and then a heuristic algorithm is implemented.
In [7], Q. Tang et al. proposed a smart power model controlled
by a power market scheduling center (PMSC), which broad-
casts the price and gets the power capacities as well as the
power demands from the energy providers and users respec-
tively. Besides, a mechanism of identification and processing
(MIP) is integrated to process the malicious users and unstable
energy providers. In [8], J. Rivera et al. proposed an ADMM
(Alternating Direction Method of Multipliers) framework for
decentralized EV charging control for the large EV numbers.
The ADMM framework can deal with non-strictly convex
cost functions and applied to a range of EV charging control
problems, such as valley filling, price-based optimization. In
[9], K. Zhou et al. proposed a new decentralized random ac-
cess framework to schedule the PHEV charging. In this frame-
work, all the vehicles waiting for charging must have a certain
probability of the charging opportunity. In [10], L. Gan et al.
proposed a decentralized algorithm for scheduling the EV
charging. An optimal control problem is formulated at first,
and then by solving the problem in the iteration manner, the
final optimal charging profiles are obtained.
Apart from the charging behavior, the discharging is also
an important issue for the smart charging. If the discharging
scheduling is considered well, the power gird burden can be
reduced remarkably. In [11], an intelligent workplace parking
garage for plug-in hybrid electric vehicles (PHEVs) was pro-
posed based on a fuzzy logic power-flow controller. The
PHEVs were classified into five charging priorities, including
charging and discharging. In [12], S. Xie et al. proposed a fair
energy scheduling for EV charge and discharge, and a
contribution-based fairness is also proposed which gives dif-
ferent contribution values for different time and different ac-
tions. The scheduling problem is formulated as an infinite-
horizon Markov decision process and solved by a dynamic
programming to maximize the long-term fairness. In [13], R.
Yu et al. explored the EV mobility’s impaction to the DRM in
V2G system.After analysis the V2G mobile energy networks,
the EVs can be acting as energy transporters among different
districts, and the districts’DRM dynamics were formulated to
balance the power demand among different districts.
Besides the important issues described above, the EV
charging and discharging problem also contains other re-
lated issues, such as the renewable energy sources integra-
tion problem [14],andEVchargingwithenergystorage
technology [15].
3 Overview of EVCDS
The EVCDS is composed of two parts: CDF and DDF. Every
EV user has to calculate the values of CDF and DDF.
According to the values of CDF, DDF and the algorithm (will
be given in the later section), the EV user decides whether to
charge, discharge or do nothing. In this paper, we assume all
the EVs can communicate with all the charging stations, and
they can easily get the values of decision making factors, such
as the charging efficiencies of charging stations.
In the CDF and DDF, there are mainly three factors, includ-
ing residual energy of battery, user’s charging habit, and the
charging stations’charging efficiencies, needed to be consid-
ered. Residual energy of battery and the user’s charging habit
can be stored in the memory of EV. Only the charging effi-
ciencies of charging stations are received from all the charging
stations. If an EV determines to charge, it will select a charg-
ing station by considering both the charging efficiency and
distance factors. Besides if an EV determines to discharge, it
will also select a charging station to discharge its energy and
reduce the charging demand at that charging station. The
sketch map of EVCDS is shown in Fig. 1.
Mobile Netw Appl
4 Charging decision function
In this paper, we define three sub-functions: f
1
,f
2
and f
3
.Inf
1
,
the residual energy of battery is involved. In f
2
, the charging
habit is considered. In f
3
, the charging efficiency of the charg-
ing stations is contained. The CDF is defined as:
Cf
1;f2;f3
ðÞ¼
2arctan c1f1þc2f2þc3f3
ðÞ
πð1Þ
Where C(f
1
,f
2
,f
3
)∈(−1, +1), and c
i
is the weight of the f
i
.
After given a threshold θ, an EV determines to charge if C(f
1
,
f
2
,f
3
)>θ.
4.1 Residual energy of battery
The EV’s travel distances subject to a probability distribution.
Let p(x) denotes the probability density function of the next
trip’s travel distance, and xis the distance of next trip, which is
non-negative. Then we get the probability distribution func-
tion:
PmðÞ¼∫m
0pxðÞdx ð2Þ
Where P(m) represents a probability that the travel distance
is less than m. Although the residual energy of battery denoted
by E
r
can supply all the devices in the EV, such as the
computing and communication devices and it is very impor-
tant to consider the data and energy integrated communication
[16]inEVs,weassumetheE
r
are all consumed by the driv-
ing, and the travel distance mcan be expressed as:
m¼crErδð3Þ
Where δis the deviation and it satisfied 0 < δ<△,andwe
assume the travel distance mis uniformly distributed. In other
words, the distance m is uniformly distributed in the interval
[c
r
E
r
−Δ,c
r
E
r
+Δ], and the probability P(m) can be calculat-
ed according to (2). Since mis uniformly distributed, then we
get the probability density function for the travel distance:
φm;Er
ðÞ¼
1
2Δ;if m∈crEr−Δ;crErþΔ½
0;else
;
(ð4Þ
The excepted probability that the residual energy E
r
sat-
isfies the user’s mileage requirements can be calculated ac-
cording to:
EPE
r
ðÞ½¼∫þ∞
0φm;Er
ðÞ∫m
0pxðÞdx
hi
dm ð5Þ
According to (5), if we give E
r
a value, then the travel
distance can be fixed into a range, and each value m
i
in this
range has a probability P(m
i
) according to (2). Because the
travel distance is uniformly distributed in this range, excepted
probability can be rewritten as:
EPE
r
ðÞ½¼
1
2Δ∫crErþΔ
crEr−Δ∫m
0pxðÞdx
hi
dm ð6Þ
If E
r
increases, the travel distance mwill increase, and P(m)
will also increase, which causes the increasing of E[P(E
r
)]. So,
if the residual energy of battery is bigger, the charging will-
ingness is smaller and we define the sub-function f
1
as:
f1¼1−EPE
r
ðÞ½½
ρð7Þ
Where ρis an exponential, and f
1
is a nonlinear function.
In order to get the probability density function p(x), we
record the travel distances between two charging behaviors,
which indicates the travel distance for each charging. Assume
the recorded travel distances as x
j
(j=1,2,…,j
max
), and p(x)is
defined as:
pxðÞ¼ω1∑
j¼0
jmax
e−ω2x−xj
ðÞ
2
ð8Þ
Where ω
2
is a small constant. xis the travel distance of the
next trip and belongs to [0,M
max
], where M
max
is the
6HUYHU
:LUHOHVV&RPPXQLFDWLRQ
5RDG6LGH8QLW568
%URDGFDVWVLWVFKDUJLQJ
HIILFLHQF\WLPHE\WLPH
6FKHGXOHVWKH(9V
FKDUJLQJEHKDYLRU
7KHZLUHOHVVFRPPXQLFDWLRQRQO\
XVHGDPRQJWKH(9VFRPPXQLFDWLRQ
DVZHOODVWKH9,FRPPXQLFDWLRQ
&DOFXODWHVLWV&')DQG'')
'HFLGHVWRFKDUJHGLVFKDUJH
RUGRQRWKLQJ
7KLVXQLWLVUHVSRQVLEOHIRUWKH
FRPPXQLFDWLRQZLWKWKH(9V
5HVSRQVLEOHIRUWKHLQIRUPDWLRQ
SURFHVVLQJ
5HFHLYHVDOOWKHLQIRUPDWLRQDQG
VHQGVLWWRWKHGHVWLQDWLRQ
Fig. 1 The sketch map of EVCDS
Mobile Netw Appl
maximum travel distance.. According to (8) and the property
of that the sum of all the probabilities is 1, ω
1
is obtained as:
ω1¼1
∑Mmax
x¼0∑jmax
j¼0e−ω2x−xj
ðÞ
2ð9Þ
4.2 The charging habit of electric vehicle
The EV’s charging habit is also an important factor that
influences the charging decision. By learning the charging
habits, the strategy will know when and where to charge
is comfortable for an EV user. We assume that the vehicle
will not be charged or discharged during the traveling. We
set three statuses for each EV, which are charging,
discharging or waiting at a charging station (this status
is called as charging status for abbreviation), driving on
the road (driving status for abbreviation) and parking at
company or home (parking status for abbreviation) re-
spectively. For the charging habits, we only consider
two factors, which are charging time t, and charging place
p=(p
x
,p
y
). In each specific time, the EV user gets a
record R=(t,b,p), and saves it into a sequence. In the
record, the charging behavior bhas three values 1, 0 and
−1, which stands for charging status, parking status and
driving status respectively.
The system learns the charging habits according to the
records saved in the sequence. It at first learns in the time
dimension:
CttðÞ¼ctarctan ∑
N
i¼1
bi½
t−ti½
ðÞ
2ð10Þ
Where Nis the number of training samples, and the latest N
records will be selected as the training sample. c
t
is a constant.
tis the current time. We use arctan function to map the value
into a controllable range (−π/2, +π/2).
In (10),thetimecycleisaday,i.e.24hours.Wedefine
thatthetimedifferencebetweenoneday’s 23:00 pm and
another day’s 23:00 pm is zero. We also bound the time
difference into 12 hours:
Δt¼Δt;if Δt<12
12−Δt;else case ;
ð11Þ
Where △t=|t-t
[i]
|. According to (11), the time difference
between 23:00 pm and 2:00 am is 3 hours, not 21 hours.
Similarly, the system also learns the charging habits in the
spatial dimension:
CppðÞ¼cparctan ∑
N
i¼1
bi½
p−pi½
kk
2ð12Þ
Where c
p
is a constant, and pis the current position. The
sub-function f
2
is defined by:
f2¼CttðÞþCppðÞ−Uð13Þ
Where Uis a constant.
4.3 The charging efficiency of Charging Station
In order to evaluate the charging efficiency, we define an
efficiency variable:
e¼
σ
∑K
i¼1σi
;if σ<∑K
i¼1σi
1;else
;
(ð14Þ
Where σis the total power capacity of a charging station
and σ
i
is the charging power of the EV
i
which is connected
with charging plot and waiting for charging or staying in
charging. Kis the total number of EVs that want to charge at
this charging station.
We assume there are Lcharging stations. For a specific EV,
there are different distances between it and different charging
stations. By considering both the distance and charging effi-
ciency, the following charging decision sub-function is de-
fined:
f3¼α∑
L
j¼1
ej
arctandj
=∑
L
i¼1
1
arctandi
!
γ
−βð15Þ
Where α,βand γare parameters. According to (15), we
know that if the charging efficiency e
j
increases, the value of f
3
becomes larger. In order to consider the straight-line distance
influence on f
3
, every charging station’se
j
is divided by an
arctan function with straight-line distance d
j
as an input pa-
rameter. In this paper, we set β≥α, which means the function
f
3
is less than 0.
4.4 Charging station selection
If an EV has calculated all the sub-functions, the charging
decision function in (1) can be easily calculated out. By com-
paring the value of (1)withθ, the EV can decide whether to
charge or not at the current time. If one EV decides to charge,
it will select a charging station. We propose the following
selection function:
CSw
i¼weeiþwd
1
arctandi
;i∈1;2;…;L½ ð16Þ
WhereCSw
iis a weight of the i-th charging station and cal-
culated out by each EV. The w
e
is a weight of charging
Mobile Netw Appl
efficiency, and w
d
is a weight of distance. Each EV will select
the charging station with the maximum weight CSw
i.
5 Discharging decision function
In order to relieve the charging demand of the charging sta-
tions, we further propose a discharging strategy based on the
three sub-functions. According to the value of CDF, three
factors are considered: residual energy, charging efficiency,
and charging habit. We also use these factors to decide wheth-
er an EV wants to discharge or not. The discharging decision
function is:
Df
1;f2;f3
ðÞ¼
2arctan c1f1−c2f2þc3f3
ðÞ
πð17Þ
Where the c
1
f
1
is the weighted first sub function, and if its
value is smaller, the residual energy is more, and the possibil-
ity of discharging is bigger. The -(c
2
f
2
) is the charging habit
comfort value. If its value is small, the c
2
f
2
is big, and the
possibility of discharging is big. The c
3
f
3
is the weighted
charging efficiency, and if its value is small, the charging
efficiency of charging stations is small, which means the
charging stations are busy now, and the possibility of
discharging is big.
All the parameters of DDF are the same as that of
CDF. If an EV has calculated out the DDF value which
is smaller than -θ, the EV should decide to discharge. If
an EV decides to discharge, the next problem is how to
choose a charging station. In order to balance the charg-
ing demand among the charging stations, the EV will
select the charging station with the lowest charging ef-
ficiency and the nearest distance. Then we define the
charging station selection function for the discharging
action:
DSw
i¼−weeiþwd
1
arctandi
;i∈1;2;…;L½ ð18Þ
The DSw
iis a weight of the i-th charging station and calcu-
lated by each EV. The w
e
is a weight of charging efficiency,
and the w
d
is a weight of distance. Each EV will select the
charging station with the maximalDSw
ito discharge.
6 Smart charging and discharging algorithm
EVCDS
Every EV calculates the CDF and DDF value at specific time.
According to the values, the EV decides the next decision
based on its current status. If an EV is not staying at a charging
station, it will execute the Algorithm 1.
Where CS means Charging Station. If an EV is already
staying at a charging station, it will execute the Algorithm 2.
The charging stations just supply the charging and
discharging service. We assume the charging power equals
to the discharging power, and there is no queue for the
discharging. If an EV has finished its discharging, it can charge
Mobile Netw Appl
the energy back without any waiting in the queue in the next
hours as an award for its help of relieving the charging pres-
sure. The flow chat of the charging station is shown in Fig. 2.
7Numericalresults
7.1 Environment and parameters settings
In this paper, we set up a scenario to evaluate the performance
of EVCDS based on the MATLAB. Firstly, we define a square
area with the size as 15 km × 15 km. Then, we mainly simu-
late two types of EV, which are private EVs and commercial
EVs such as bus and taxi. The two types of EV have different
driving pattern. As for the private cars, it is driven by its owner
mainly in the morning and evening, and it is immobile at most
of the time, and we assume the private EV stays at the charg-
ing station when it has arrived at home or work place. The
commercial EVs move at most of the time in a day, and we
assume their working time interval is 24 hours like the taxis in
China. The commercial EVs have to charge or discharge in
their working time.
In order to describe the strategies of these two types EV, we
further propose two flow chats for them respectively. The flow
chat executed by the private EVs is shown in Fig. 3.
The commercial EVs will execute the flow chat shown in
Fig. 4.
According to the flow chats of private EV and commercial
EV, the two types of vehicles make decision at different time.
For the private vehicles, they make decisions at the time when
theywillgotowork,gohome,andeachhourwhentheyare
staying at a charging station. For the commercial vehicles, they
make decisions at the time when they have finished a passenger
travel or each hour when they stay at a charging station.
Each private vehicle has its own home coordinate and com-
pany coordinate. The time going to work is generated random-
ly, which belongs to the interval [7:00 am, 9:00 am]. The time
going back home is also generated randomly, which belongs
to the interval [16:00 pm, 18:00 pm].
The coordinate of the private vehicle’shomeisgenerated
according to the following probability distribution
Kh
pe−x−15ðÞ
2þy−15ðÞ
2
10 .The company coordinate is generated accord-
ing to the following probability distribution Kc
pe−x2þy2
10 . All the
EVs are distributed in the 15 km × 15 km area, which means
the integrals of the above two distribution formulations on the
range [0, 15] for both xand yare the same and equal to 1.
Then, we further get the Kh
p:
Kh
p¼1
∫15
0∫15
0e−x−15ðÞ
2þy−15ðÞ
2
10 dxdy
ð19Þ
And the Kc
p:
Kc
p¼1
∫15
0∫15
0e−x2þy2
10 dxdy
ð20Þ
We set 2500 private EVs and 2000 commercial EVs in the
area. In order to meet the charging demand of these vehicles,
we set 5 charging stations in the area. Each station can charge
200 vehicles and provides 200 waiting places. The five charg-
ing stations are located at the four vertexes as well as the
6WDUW
(QG
'LVFKDUJHGDW
ODVWKRXU"
$UUDQJHLWWR
FKDUJH
LPPHGLDWHO\
$UUDQJHLWLQWRWKH
TXHXHWRZDLWIRU
FKDUJLQJ
(YHQW$Q(9
FRPHVWRWKH
FKDUJLQJ
VWDWLRQ
&KDUJLQJRU
GLVFKDUJLQJ"
$UUDQJHLWWR
GLVFKDUJH
LPPHGLDWHO\
%URDGFDVWVLW
FKDUJLQJHIILFLHQF\
DWLQWHUYDOV
/LVWHQLQJWKH(9
FKDUJLQJRU
GLVFKDUJLQJHYHQWV
(YHQW$Q(9
KDVEHHQIXOO\
FKDUJHG
(YHQW$Q(9
KDVEHHQIXOO\
GLVFKDUJHG
$UUDQJHWKH
FKDUJLQJ
FRQQHFWRUWRWKH
QH[W(9
/HDYHRUQRW"(QG /LVWHQHGHYHQW
W\SH"
1
<
1
'LVFKDUJLQJ
&KDUJLQJ
Fig. 2 The flow chart of the charging station
6WDUW
&DOO$OJRULWKPWR
JHWVXIILFLHQW
HQHUJ\
:DQWWRJRWR
ZRUNLQJSODFH"
$UULYHDWWKH
ZRUNLQJSODFH" &DOO$OJRULWKP
:DQWWRJR
EDFNKRPH"
&DOO$OJRULWKPWR
JHWVXIILFLHQW
HQHUJ\
$UULYHDWWKH
KRPH" &DOO$OJRULWKP
1RSODQDQG
VWD\DWD
FKDUJLQJ
VWDWLRQ"
,VWKLVDQHZ
WLPHVORW"
&DOO$OJRULWKPWR
FKDUJHRU
GLVFKDUJH
11
<
<
<
<
<(QG
Fig. 3 The flow chat of private EV
6WDUW
)LQLVKHGD
WUDYHO" &DOO$OJRULWKP
$OUHDG\VWD\V
DWDFKDUJLQJ
VWDWLRQ"
$QHZWLPH
VORW" &DOO$OJRULWKP
1
(QG
Fig. 4 The flow chat of commercial EV
Mobile Netw Appl
center of this square area respectively. The charging and
discharging power are set as 10KW.
When calculate the practical travel distance, we multi-
plied the straight-line distance between the EV and its desti-
nation by a random multiplier uniformly distributed in [1.4,
1.7]. The average energy consumption is set as 5.05 km/(kw·
h) according to [17]. At the beginning, SOC (State of
Charge), which represents the residual energy of battery, is
set as a uniformly distributed number belonging to [0.3, 0.7].
The battery capacities have four kinds of specifications:
40kwh, 44kwh, 47kwh, 53kwh. All the EVs are randomly
assigned a capacity. The other parameters are set in Table 1:
7.2 Compared strategies
We compare EVCDS with a basic charging strategy (BCS),
a random access strategy (RAS) which is similar to the idea
in [9], and a fuzzy logic charging control (FLCC) which is
similar to the strategy in [11].
In the BCS, an EV will decideto charge only if its SOC is
lower than a lower limit. The EV will select the nearest
charging station for charging.
In RAS, a probability function is proposed, and each
vehicle can be independently charged in accordance with
the probability function. In order to be adjusted to this sim-
ulation scenario, we set the access probability function as:
PaSOCðÞ¼e−3⋅SOC ð21Þ
If an EV has calculated out the P
a
,itcomparesP
a
with a
randomly generated number r
a
and if P
a
>r
a
, charging event
variable C
a
is 1. In order to delay charge at low charging
efficiency, we set another probability function related with
charging efficiency:
PeWe
ðÞ¼e−3⋅We ð22Þ
Where W
e
is a weighted charging efficiency which can be
calculated by:
We¼∑
L
j¼1
ej
arctandj
=∑
L
i¼1
1
arctandi
ð23Þ
If an EV has calculated out the P
e
,itcomparesP
e
with a
randomly generated number r
e
,andifP
e
<r
e
, charging event
variable C
e
is 1.
The final charging decision variable C
d
is:
Cd¼Ca&Ceð24Þ
If C
d
is 1, the EV decides to charge. There is no discharging
decision in RAS.
As for the FLCC, according to the main idea of [11], we
give the membership function of residual energy classified
into five states which are “very sufficient”,“sufficient”,“stan-
dard”,“lack”,“very lack”. We also give the membership func-
tion of charging efficiency which is also classified into five
standards: “very high”,“high”,“medium”,“low”,“very low”.
The fuzzy charging logic function value F
V
varies from −1to
1. EVs will choose to “charge”,“discharge”,and“do nothing”
Table 1 The parameters settings
Parameter c
1
c
2
c
3
ργ c
t
Value 1 0.2 0.7 5 5 0.5
Parameter c
p
Uαβθ w
e
/w
d
Value 0.5 1 1 1 0.1 3
20 30 40 50 60 70 80
0
0.5
1
Residual Energy (%)
very lack lack standard sufficient very sufficient
(a) Membership Function of Residual Energy
50 55 60 65 70 75 80 85 90 95 100
0
0.5
1
Charging Efficiency (%)
very low low medium high very high
(b) Membership Function of Residual Energy
-1 -0.5 0 0.5 1
0
0.5
1
Fuzzy Logic Function Value
Discharge Do nothing Charge
(c) Membership Function of Charging Decision
Fig. 5 Membership function of fuzzy charging logic function value
Table 2 Fuzzy charging logic function value
Charging efficiency
Very low Low Medium High Very high
Residual energy
Very lack 0.5 0.6 0.8 0.9 1
Lack 0.2 0.4 0.5 0.6 0.8
Standard −0.2 0.2 0.3 0.5 0.6
Sufficient −0.5 0 0.1 0.3 0.4
Very sufficient −1−0.5 −0.2 0 0.1
Mobile Netw Appl
according to the probability P
c
,P
d
and P
n
respectively. For
example, if the logic function value is −1, the probability of
discharging P
d
is 100%, and if the logic function value is 1,
the probability of charging P
c
is 100%. The following equa-
tions are used to calculate the probability:
Pc¼FV;if FV>0
0;if FV≥0
ð25Þ
Pd¼−FV;if FV<0
0;if FV≥0;
ð26Þ
Pn¼1−FV;if FV>0
1þFV;if FV≤0
ð27Þ
Where P
n
represents the probability of “do nothing”.The
logic function value F
V
is determined according to the residual
energy level and the charging efficiency.
Figure 5(a) shows the membership function of residual
energy. If an EV has the residual energy level as 35%, both
the probabilities of “very lack”and “lack”are “50%”.Thena
random number ris generated to determine whether the EV
belongs to the “very lack”class or not. The principle is also
suitable for the charging efficiency classification. Figure 5(b)
shows the membership function of charging efficiency.
According to the classifications of residual energy and
charging efficiency, we give the fuzzy charging logic function
value F
V
in Table 2and Fig. 5(c). After getting the value of F
V
,
the probabilities of P
c
,P
d
and P
n
can be easily get according to
(25)~(27). After comparing P
c
,P
d
and P
n
with a randomly
generated number which belongs to [0, 1], the EV can easily
determine the next action such as charging, discharging and
do nothing.
7.3 Load demand fluctuations
In this subsection, we will simulate the load curves of pri-
vate vehicles, commercial vehicles and total demand. The
simulation time period is 72 h. In the simulation process,
the BCS, RAS and FLCC only decide an EV whether to
charge or discharge, and we simply assume the EVs which
using these three strategies select the nearest charging sta-
tion to charge or discharge. The results of different strate-
gies are shown in Fig. 6.
0 10203040506070
0
2000
4000
6000
8000
10000
12000
time
charging demand (Kw.h)
Private EV Demand Commercial EV Demand Total Demand
(a) Charging Demand of BCS
0 10203040506070
0
2000
4000
6000
8000
10000
12000
time
charging demand (Kw.h
Private EV Demand Commercial EV Demand Total Demand
(b) Charging Demand of RAS
010203040506070
-2000
0
2000
4000
6000
8000
10000
time
charging demand (Kw.h)
Private EV Demand Commercial EV Demand Total Demand
(c) Charging Demand of FLCC
010203040506070
-2000
0
2000
4000
6000
8000
10000
time
charging demand (Kw.h)
Private EV Demand Commercial EV Demand Total Demand
(d) Charging Demand of EVCDS
Fig. 6 Charging demand curves
BCS RAS FLCC EVCDS
0
1
2
3
4x 10
6
Fig. 7 Total charging demand variance
Mobile Netw Appl
According to the Fig. 6, we find that the BCS strategy
cannot shift the charging demand efficiently, because it
only considers the SOC statues. In RAS strategy, it
considers both the SOC and charging efficiency of charg-
ing station, which leads to a result that when the demand
of private EVs reach the peak, the commercial EVs will
reduce their demand in this time period. As for the total
demand curve, the peak-shaving performance is not good
enough. The FLCC strategy and EVCDS has the strongest
ability to shift the charging demands. In these strategies,
when the private EVs have a charging peak, the commer-
cial EVs will charge at another time slot. Besides, the
FLCC and EVCDS have the ability to discharge when
the peak hour appeared, thus the two strategies have the
best performance in terms of peak-shaving.
In order to introduce the fluctuations of the demand curves,
we calculate the varianceof the total charging demands,which
is shown in Fig. 7.
AccordingtoFig.7, we find that our strategy EVCDS
and FLCC have small variances, which means that the
demand curve has smaller fluctuations compared with oth-
er two strategies BCS and RAS. Besides, the EVCDS’s
variance is almost the same as that of FLCC, which illus-
trates that the peak-shaving abilities of these two strategies
are almost the same.
7.4 Charging demand balance among different
charging stations
In order to introduce the load balance among different charging
stations, we calculate the charging loads of different EV types
for each charging station. Because the EVs in BCS, RAS and
FLCC strategies will select the nearest charging stations for
charging or discharging, they did not consider the load balance
among charging stations at all. The total charging demand for
each charging station is presented in Fig. 8:
According to the Fig. 8, we find that the charging demand
balance of EVCDS is the best of all the strategies, which is
determined by the charging station selection method of CDF
and DDF. The standard deviation of charging demand, which
can further illustrate the charging demand balance perfor-
mance, is shown in Fig. 9.
Station1 Station2 Station3 Station4 Station5
0
1
2
3
4
5x 105
Charging Demand(Kw.h)
Private EV Demand
Commercial EV Demand
Total Charging Demand
(a) Charging Demand Balance of BCS
Station1 Station2 Station3 Station4 Station5
0
1
2
3
4
5
6
7x 105
Charging Demand (Kw.h)
Private EV Demand
Commercial EV Demand
Total Charging Demand
(b) Charging Demand Balance of RAS
Station1 Station2 Station3 Station4 Station5
0
0.5
1
1.5
2
2.5
3
3.5
4x 10
5
Charging Demand(Kw.h)
Private EV Demand
Commercial EV Demand
Total Charging Demand
(c) Charging Demand Balance of FLCC
Station1 Station2 Station3 Station4 Station5
0
0.5
1
1.5
2
2.5
3
3.5
4x 105
Charging Demand (Kw.h)
Private EV Demand
Commercial EV Demand
Total Charging Demand
(d) Charging Demand Balance of EVCDS
Fig. 8 Load balance comparison
BCS RAS FLCC EVCDS
0
2
4
6
8
10
12 x 104
Fig. 9 Standard deviation of charging demand
Mobile Netw Appl
8 Conclusion
In this paper, we propose a distributed charging and
discharging strategy EVCDS for electric vehicle, which is
supported by the Charging Decision Function (CDF) and
Discharging Decision Function (DDF). After considering the
residual energy of battery, charging habits, charging efficien-
cy, the values of CDF and DDF can be calculated out, which
are used to decide whether to charge, discharge or do nothing.
Compared with other strategies such as FLCC and RAS,
EVCDS performs well in terms of peak shaving and balancing
thechargingdemandamongdifferentchargingstations.
In this paper, the EVCDS is a mechanism to balance the
spatial charging load among different charging stations, but
the utilities of the EV users and charging stations are not con-
sidered, which will be further studied in the future.
Acknowledgments The work presented in this paper is in part funded by
a project supported by the Scientific Research Fund of Hunan Provincial
Education Department (No.13C1023), and a project supported by the
Natural Science Foundation of Hunan Province, China (Grant No.
13JJ4052). It is also partly funded by a project supported by the
National Natural Science Foundation of China (Grant No. 61303043,
61772087).
References
1. Qian K, Zhou C, Allan M, Yuan Y (2011) Modeling of load de-
mand due to EV battery charging in distribution systems. IEEE
Trans Power Syst 26(2):802–810
2. Yang H, Yang S, Xu Y, Cao E, Lai M, Dong Z (2015) Electric
vehicle route optimization considering time-of-use electricity price
by learnable partheno-genetic algorithm. IEEE Trans Smart Grid
6(2):657–666
3. Rigas ES, Ramchurn SD, Bassiliades N (2015) Managing electric
vehicles in the smart grid using artificial intelligence: a survey.
IEEE Trans Intell Transp Syst 1(4):1619–1635
4. Vastardis N, Yang K (2013) Mobile social networks: architectures,
social properties, and key research challenges. IEEE Commun Surv
Tutorials 15(3):1355–1371
5. Tang Q, Xie M-Z, Wang L, Luo Y-S, Yang K, Load Balanced A
(2017) Charging strategy for electric vehicle in smart grid. In:
Proceedings of the first international conference on smart grid in-
spired future technologies, pp 155–164
6. Cao Y, Tang S, Li C, Zhang P, Tan Y, Zhang Z, Li J (2012) An
optimized EV charging model considering TOU price and SOC
curve. IEEE Trans Smart Grid 3(1):388–393
7. Tang Q, Yang K, Zhou D, Luo Y, Yu F (2016) A real-time dynamic
pricing algorithm for smart grid with unstable energy providers and
malicious users. IEEE Internet Things J 3(4):554–562
8. Rivera J, Wolfrum P, Hirche S, Goebel C, Jacobsen H-A (2013)
Alternating direction method of multipliers for decentralized elec-
tric vehicle charging control. In: Proceedings of the 52nd IEEE
Annual Conference on Decision and Control (CDC), p 6960–6965
9. Zhou K, Cai L (2014) Randomized PHEV charging under distribu-
tion grid constraints. IEEE Trans Smart Grid 5(2):879–887
10. Gan L, Topcu U, Low SH (2012) Optimal decentralized protocol
for electric vehicle charging. IEEE Trans Power Syst 28(2):940–
951
11. Ma T, Mohammed OA (2014) Optimal charging of plug-in electric
vehicles for a Car-Park infrastructure. IEEE Trans Ind Appl 50(4):
2323–2330
12. Xie S, Zhong W, Xie K, Yu R, Zhang Y (2016) Fair energy sched-
uling for vehicle-to-grid networks using adaptive dynamic pro-
gramming. IEEE Trans Neural Netw Learn Syst 27(8):1697–1707
13. Yu R, Zhong W, Xie S, Yuen C, Gjessing S, Zhang Y (2016)
Balancing power demand through EV mobility in vehicle-to-grid
mobile energy networks. IEEE Trans Ind Inf 12(1):79–90
14. Mwasilu F, Justo JJ, Kim E-K, Do TD, Jung J-W (2014) Electric
vehicles and smart grid interaction: a review on vehicle to grid and
renewable energy sources integration. Renew Sust Energ Rev 34:
501–516
15. Sbordone D, Bertini I, Pietra BD, Falvo MC, Genovese A,
Martirano L (2015) EV fast charging stations and energy storage
technologies: a real implementation in the smart microgrid para-
digm. Electr Power Syst Res 120:96–108
16. Yang K, Yu Q, Leng S, Fan B, Wu F (2016) Data and energy
integrated communication networks for wireless big data. IEEE
Access 4:713–723
17. Frieske B, Kloetzke M, Mauser F (2013) Trends in vehicle concept
and key technology development for hybrid and battery electric
vehicles. In: Proceedings of the 2013 world electric vehicle sympo-
sium and exhibition (EVS27), p 1–12
Mobile Netw Appl
A preview of this full-text is provided by Springer Nature.
Content available from Mobile Networks and Applications
This content is subject to copyright. Terms and conditions apply.
