Control

IFAC andofCIGRE/CIRED
TransmissionWorkshop
and Distribution
on Smart Grids
IFAC and
October
Control ofCIGRE/CIRED
11-13, Workshop
2016. Prague,
Transmission Czechon
Republic
and Distribution Smart Grids
IFAC and
Control
October ofCIGRE/CIRED
Transmission
11-13, Workshop
Czechon
and Distribution
2016. Prague, Smartonline
Republic
Available Grids at www.sciencedirect.com
Control
Octoberof Transmission
11-13, and Distribution
2016. Prague, Smart Grids
Czech Republic
October 11-13, 2016. Prague, Czech Republic
ScienceDirect
Controlled Charging of Electric
IFAC-PapersOnLine Vehicles
49-27 to Minimize Energy Losses
(2016) 324–329
Controlled Charging ofinElectric
DistributionVehicles to Minimize Energy Losses
Systems
Controlled Charging ofinElectric
DistributionVehicles to Minimize Energy Losses
Systems
Controlled Charging of Electric Vehicles to Minimize Energy Losses
inS. Distribution
M. Mousavi *, Systems
Damian Flynn **
inS. Distribution Systems
M. Mousavi *, Damian Flynn **
* School of Electrical and Electronic S. M. Mousavi *, Damian
Engineering, FlynnCollege
University ** Dublin, Dublin, Ireland
S. M. Mousavi *, Damian Flynn **
* School of Electrical and Electronic
(e-mail: Engineering, University
mohammad.mousaviagah @ College
ucd.ie). Dublin, Dublin, Ireland
* School
** Schoolof ofElectrical
Electricaland Electronic
(e-mail:
and Engineering,
Engineering,University
mohammad.mousaviagah
Electronic University @ College
CollegeDublin,
ucd.ie). Dublin,Dublin,
Dublin,Ireland Ireland
* School of Electrical and Electronic
(e-mail: Engineering,
mohammad.mousaviagah University @ College
ucd.ie). Dublin, Dublin, Ireland
** School of Electrical and Electronic Engineering, University
(e-mail:damian.flynn @ ucd.ie) College Dublin, Dublin, Ireland
** School of Electrical and (e-mail: mohammad.mousaviagah
Electronic Engineering, University @ ucd.ie).
(e-mail:damian.flynn @ ucd.ie)College Dublin, Dublin, Ireland
** School of Electrical and Electronic (e-mail:damian.flynn @ ucd.ie)College Dublin, Dublin, Ireland
Engineering, University
Abstract: Controlled electric vehicle (e-mail:damian.flynn (EV) charging scenarios @ ucd.ie)
are proposed, each characterized with an
Abstract:
algorithm Controlled
and associated electric vehicle (EV)and
computational charging scenarios are
communication proposed, each
requirements, to becharacterized
adopted by with an EV an
Abstract:
algorithm Controlled
aggregator and
or system electric
associated vehicle
computational
operator. The (EV)and
proposed charging scenarios
communication
scenarios include are proposed,
requirements,
uniform, each
to becharacterized
random, adopted by with
conditional-random, an EV an
Abstract:
algorithm Controlled
and electric
associated vehicle (EV)and
computational charging scenarios are
communication proposed, each to becharacterized with an
aggregator
and valley-fillingor system operator.
charging The proposed
scenarios. Different scenarios includerequirements,
from previous uniform,this
studies, random,
paper adopted on by easy-to-
conditional-random,
focuses an EV
algorithm
aggregator and associated
or system computational
operator. The proposed and communication
scenarios requirements,
includeisstudies,
uniform, to
random, be adopted by
conditional-random, an EV
and valley-filling
implement charging charging
scenarios. scenarios.
Further, Different
a modeling from previous
framework presented this topaper focusestheonimpact
investigate easy-to-of
aggregator
and or system
valley-filling operator.
charging The proposed
scenarios. Different scenarios
from include studies,
previous uniform, random,
this conditional-random,
implement
the proposed charging
charging scenarios.
scenarios Further,
on energy a modeling
losses in framework
distribution is presented
systems, as topaper
compared focuses
investigate to an the onimpact
easy-to-
uncontrolled of
and valley-filling
implement charging charging
scenarios. scenarios.
Further, Different
a modeling from previous
framework studies,
is presented this paper focuses
to investigate on easy-to-
the proposed
charging charging
scenario and scenarios
a on energy
reference losses with
scenario in distribution
no EVs. systems,
The as compared
modeling to anthe
framework impact
uncontrolled
considers of
implement
the proposed charging
charging scenarios.
scenarios Further,
on energy a modeling
losses framework is
in distribution presented
systems, to investigate
as compared toItanisthe impact of
uncontrolled
charging
uncertainties scenario
involved andin a reference
the behavior ofscenario
low voltagewith no EVs.and
customers The EVmodeling
charging framework
loads. considers
applied to a
the proposed
charging charging
scenario scenarios
andinvarious
a referenceon energy losses in distribution systems, as compared to an uncontrolled
uncertainties
distribution involved
system for the behavior
case ofscenario
low voltage
studies, withcustomers
including no EVs.and
different The EVmodeling
charging
penetrations and framework
loads. considers
It is applied
combinations of to a
EVs
charging
uncertainties scenario
involved and the a reference ofscenario withcustomers
no EVs.and The modeling framework considers
distribution
with various system forinvarious
characteristics. behavior
As case
seen lowthe
studies,
by voltage
including
results, different
although EVvalley-filling
charging
penetrations
the andloads. It is applied
combinations
charging to a
of EVs
algorithm
uncertainties
distribution involved
system forinsolution
the behavior
various of low voltage customers and EV charging andloads. It is applied to a
with various
represents the characteristics.
optimal Ascaseseen
from studies,
the theincluding
results,ofdifferent
by perspective energy penetrations
although the valley-filling
losses, the uniform combinations
charging
chargingalgorithmof EVs
scenario
distribution
with various system for various case studies, including different penetrations and combinations of EVs
represents
emerges
with variousasthea characteristics.
optimal solution
quasi-optimal
characteristics.
Asfromseenthe
solution,
Asfromseenthe
by perspective
havingthe lower
results,ofalthough
energy losses,
the results,computational
by perspective although
the andvalley-filling
the uniformcharging
communication
the valley-filling
charging
charging
algorithm
scenario
requirements,
algorithm
represents
emerges
which makes asthea optimal
quasi-optimal
itoptimal solution
easier tosolution solution,
be adopted having lower of energy
computational losses,
and the uniform
communication charging scenario
requirements,
represents
emerges astheaitquasi-optimal frombythe
solution,
EVperspective
aggregatorsoforenergy systemlosses,operators. Further, with
the uniform chargingappropriately
scenario
which makes
selected easier tothebe conditional-random
coefficients, adopted by having lower
EV aggregators
charging computational
or algorithm
system operators.and
can communication
also Further,
exhibit with a requirements,
appropriately
performance
emerges
which as
makes a itquasi-optimal
easier to be solution,
adopted by having
EV lower
aggregators computational
or system and
operators. communication
Further, with requirements,
appropriately
selected coefficients,
comparable toitthat of the theuniform
conditional-random
charging charging algorithm can also exhibit a performance
which
selected makes easier
coefficients, tothebe adopted by EValgorithm.
conditional-random aggregators
chargingor algorithm
system operators. can also Further,
exhibit with a appropriately
performance
comparable to that of the uniform charging algorithm.
selected
© coefficients,
Distribution
2016, IFACto(International
Keywords:
comparable the
that of the uniform conditional-random
System, Electric
Federation
charging Vehicle,
of Automatic charging
algorithm. Controlled algorithm
Charging, canPoweralso
Control) Hosting by Elsevier Ltd. All rights exhibit
System a performance
Operation, Smart
reserved.
Keywords: Distribution
comparable
Grids to that of theSystem, uniformElectric
chargingVehicle,
algorithm. Controlled Charging, Power System Operation, Smart
Grids
Keywords: Distribution System, Electric Vehicle, Controlled Charging, Power System Operation, Smart
Keywords: Distribution System, Electric Vehicle, Controlled Charging, Power System Operation, Smart
Grids
Grids  of EVs, new load peaks were created that might exceed the
1. INTRODUCTION  of EVs, new
available load peakscapacity.
transformer were created Several that charging
might exceed the
profiles
 of EVs, new load peaks were created that charging
might exceed the
1. INTRODUCTION available
were analysedtransformer
to prevent capacity.
EVs Several
from causing harmful profiles
new
Electric vehicle (EV)1. technology
INTRODUCTION is seen as a key component of EVs, new
available
were analysed
load peakscapacity.
transformer were created Several that charging
might exceed the
profiles
Electric vehicle (EV) 1. INTRODUCTION
technology is seen as a key component daily peaks.
available A to prevent EVs
large-scale
transformer
from causing
distribution
capacity. Several planning harmful
charging model new
was
profiles
to reduce carbon emissions within the transport sector. As a were analysed to prevent
Electric
to reduce vehicle
carbon (EV) technology
emissions withinis the seentransport
as a key As a developed
component
sector.
daily peaks.
were analysed in A large-scale
Fernandez
to prevent et EVs
al.
EVs
from to
distribution
(2011) causing
from causing
planning
evaluate harmful
model
the impact
harmful
new
was
new
result,
Electric many
vehicleautomotive
(EV) manufacturers
technology is seen have
as a begun
key to place
component daily peaks.
developed in A large-scale
Fernandez distribution
(2011) to planning
evaluate model
the was
impact
to reduce
result, many carbon emissions
automotive within
manufacturers the transport
have begunsector. of various penetration levels
et al. of
As a daily peaks. A large-scale distribution planning model was
to place EVs on system investments.
an increased
to reduce emphasis
carbon on
emissions the development
within the transport of EVs, including
sector. developed
of various in Fernandez
penetration levels
As a These studies suggested that distribution systems can
et al. of(2011)
EVs to
on evaluate
system the impact
investments.
result, many
an increased
pure battery automotive
emphasis onmanufacturers
electric vehicles (BEVs), have
the development andof begun
EVs,
plug-in tohybrid
place
including developed in Fernandezlevels et al.of(2011) to system
evaluateinvestments.
the impact
result, many emphasis
automotive manufacturers have begun to place These of various penetration
studies suggested that EVs ofon
distribution if systems EVs can
an increased
pure battery
electric vehicleselectric
(PHEVs). on the
vehiclesThe development
(BEVs), of
batteries andof
bothEVs,
plug-in hybrid accommodate
including
technologies of various
These studies
higher
penetration
suggested
penetrations
levels of
that EVs on EVs,
system
distribution
the
investments.
systems
are
can
an
pureincreased
battery emphasis
electric on the
vehicles development
(BEVs), andof EVs,
plug-in including
hybrid accommodate
intelligently higher
charged penetrations
during off-peak of EVs,
hours. if
Thisthe EVs was
result are
electric
can vehiclesfrom
bebattery
charged (PHEVs).
domestic The electrical
batteries of both technologies
sockets (Richardson These studies higher
accommodate suggested that distribution
penetrations ofstudies
EVs, if systems
theresult
EVs wascan
are
pure
electric vehicleselectric
(PHEVs). vehiclesThe (BEVs),
batteries amount and
of plug-in
both technologies hybrid intelligently
further analysedcharged in during
a number off-peak of hours. This
that aimed at
can
et be 2012),
al. charged from
demanding domestic a electrical
significant sockets (Richardson
of electrical accommodate
intelligently higher during
charged penetrations
off-peak of hours.
EVs, if Thistheresult
EVs was are
electric
can vehicles
be 2012),
charged (PHEVs).
from The batteries of both technologies further
proposing analysed
methodologiesin a number
for of
controlled studies that
charging aimed
of EVs at
to
energy,
et al.
can be charged may domestic
whichdemanding
from domestic
electrical
a significant
adversely impactamount
electrical
sockets
power of
sockets
(Richardson
electrical intelligently
distribution
(Richardson further
proposing analysedcharged
methodologies
during
inobjectives,
a number
for
off-peak hours. This
ofminimizing
controlled studies that
charging
result
aimed
of EVs
was at
to
al. 2012),
energy,
et which demanding
mayet adversely a 2010).
significant
impact amount
power ofdistribution
electrical optimize
further utility-side
analysed in a number e.g. of studies load
that of variance
aimed at
systems (Clement al. Investigation into these
electrical optimize proposing methodologies
utility-side for
objectives,ete.g.controlled charging EVs to
et al. 2012),
energy,
systems which demanding
(Clement may a significant
adversely2010). impact amount
power
Investigation
ofdistribution
into these (Gan et al.
proposing 2013; Karfopoulos
methodologies for al.minimizing
controlled Tangload
2015;charging al.variance
et of 2014),
EVs to
impacts
energy, iswhich
one ofmay the main
et al.
adverselypriorities
impact for system
power operators
distribution to optimize
(Gan etlossesutility-side
2013; objectives,
Karfopoulos e.g.al.minimizing
2015; Tangload al.variance
2014),
to energy (Sortomme et al.et2011), and voltage
systems is (Clement al. 2010). Investigation intosystem
these al. et deviations
impacts
develop one of the
controlled et main
charging priorities
algorithms forthat
system
minimize operators optimize
(Gan utility-side
2013; objectives,
Karfopoulos e.g. minimizing
2015; Tang load al.variance
2014),
systems is(Clement
impacts one of the al. 2010).
et main priorities Investigation
for system into these
operators to energy
(Wu et
et losses
al.
al. (Sortomme
2011), EV aggregator
et al. 2011),
et al.
and and
EV voltage
owner et deviations
objectives,
develop controlled
operational costsofinthecharging
themain
presencealgorithms
of high that minimizeofsystem
penetrations EVs. (Gan etlosses
energy al. 2013; Karfopoulos
(Sortomme al. 2015;
et2011), and Tang et deviations
voltage al. 2014),
impacts is one priorities for system operators to (Wu
e.g. maximizing 2011), EV
aggregator aggregator
et al. and EV owner objectives,
et profit (Geng al. 2013;deviations
Jin et al.
develop controlled et al.
operational costs in charging
the presence algorithms
of high that minimizeofsystem
penetrations EVs. energy
(Wu losses
al. 2011),(Sortomme
EV al. 2011),
aggregator and
et voltage
and
EV owner objectives,
develop
Much
operationalcontrolled
effort has been
costs in charging
thedevoted
presencealgorithms
in of
thehigh that
literature minimize ofsystem
to investigating
penetrations EVs. e.g.
2013;maximizing
etMomber aggregator
et al. 2015), profitand (Geng
minimizing
et al. 2013;
EV Jin et al.
charging
(Wumaximizing
et al. 2011), EV aggregator and EV owner objectives,
Much effortcosts
operational
the adverse has been
impacts in the devoted
of presence
EVs on in power
thehigh
of literature
systems. While some e.g.
to investigating
penetrations of EVs. 2013;
cost
e.g. (HeMomber
et
maximizing al. aggregator
2013),
et 2015), profit
al.respectively.
aggregator profit
and(Geng
Aminimizing
(Geng
et al.
general 2013;
conclusion
2013;
Jin et
EV charging
Jin fromal.
Much
the
studies,effort
adverse has
suchimpacts been
as those devoted
of EVs on in
presented in the
power literature
Denholm systems.
andto investigating
While(2006)
Short some cost 2013;
the Momber
(He et al. 2013),
reviewed et
studiesal. 2015),
respectively.
is that and and
fromAminimizing
et
minimizing
general
the
al. EV
conclusionoffrom
perspectives
et
charging al.
the
Much
the effort
adverse has been
impacts devoted
of EVs in
on inthe
power literature
systems. to investigating
While some 2013; Momber et al. 2015), EV charging
studies,
and
the Ramossuchet
adverse
asal.
those
impacts
presented
(2008)
of EVs assessed
on in
Denholm
powerthe effects and
systems. ofShort
EVs (2006)
While on
somethe cost the (Hethe
reviewed
utility,
cost
et al.
(He et al.
2013),
studiesrespectively.
aggregator,
2013),
isandthatEV
respectively.
from A the
owner,general
A the
generalthe conclusion
perspectives
optimal charge
conclusion
offrom
the
studies,
and Ramos
generation such as
and
et those
al. presented
(2008)
transmission assessed Denholm
the
sectors, effects
mostand of Short
EVs
works (2006)
on the
have the reviewed
utility,
scheduling the studies
aggregator,
algorithm is that
and
minimizes EV from owner,
the load perspectives
the optimal
variance, offrom
the
charge
tracks the
studies,
and Ramossuchetasal.those presented
(2008) assessed in Denholm
the effects and of Short
EVs (2006)
on the the reviewed
utility, the studies isand
aggregator, thatEV from the perspectives
owner, the optimal of the
charge
generation
focused on and transmission
distribution systems. sectors,
In most
Richardson works
et al. have
(2010), scheduling
required algorithm
electricity minimizes
profile in the
the load variance,
day-ahead tracks
market, the
and
and Ramos et (2008) assessed the effects
mostofet EVs onhave the scheduling
(2010), utility, the algorithm
aggregator, and EV the owner, the optimal charge
generation and al. transmission sectors, works minimizes load variance, tracks the
focused
an on distribution
analysis is performed systems.
on the In Richardson
voltage profiles and
al. loading required
charges electricity
the vehicle profile
at times in
with the
the day-ahead
lowest market,
electricity and
price,
generation
focused on and transmission
distribution systems. sectors,
In Richardson most et works have scheduling
(2010), required algorithmprofile
electricity minimizes in the load
the variance,
day-ahead tracks and
market, the
an analysis
levels of is performed
distribution system on the voltage
components profiles
in the and
al. loading
presence of charges the vehicle at times with the lowest electricity price,
focused
an on distribution
analysis is performed systems.
on the In Richardson
voltage profiles etand (2010), respectively.
al. loading required electricity profile in the day-ahead
charges the vehicle at times with the lowest electricity price, market, and
levels
EVs. Itofwas
distribution system components innumbers
the presence of respectively.
an analysis
levels is shown
performed that even
on thefor modest
voltage profiles and of EVs,
loading charges
Despite the thevehicle
high level at times of with the lowest
attention electricity
that has focused price,
on
EVs.
both Itofwas
factorsdistribution
shown
may system
that
exceed evencomponents
safe for modest
operating innumbers
limits. theThe presence
of EVs,
impact of
of respectively.
levels
EVs. ofwas
It distribution
shown system
that even components
for modest innumbers
the presence of EVs, Despite theand
of identifying
respectively. high level of the
minimizing attention
adverse that has focused
impacts of charging on
both factors
charging EVsmay exceed safetransformers
on distribution operating limits. was The examined
also impact of Despite the high level of attention that has focused on
EVs.
both It was
factors shown that even for modest numbers of EVs, identifying
EVs, the and
authors minimizing
believe the
substantialadverse studyimpacts
is stillof charging
needed to
charging
in Shao
both factors al.may
etEVs (2009),
may
exceed
on distribution
exceedwhere safeittransformers
safe
operating
was shownlimits.
operating
wasin
that
limits.
The
also
The
impact
theexamined
presence
impact
of Despite the high level of attention that has focused on
of identifying
EVs, the and minimizing
authors believe the adverse
substantial studyimpacts
is stillofneeded
charging to
charging
Shao etEVs on distribution
where ittransformers wasin alsotheexamined
presence identifying and minimizing the adverse impacts
in
charging
al. (2009),
EVs on distribution
was shown that
transformers was also examined EVs, the authors believe substantial study is stillofneeded
charging to
in Shao et©al.
Copyright 2016
(2009),
IFAC
where it was shown that in the presence324 EVs, the authors believe substantial study is still needed to
in Shao et al. (2009), where it was shown that in the presence
Copyright
2405-8963 © © 2016,
2016 IFAC
IFAC (International Federation of Automatic Control)
324Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2016 responsibility
IFAC 324Control.
of International Federation of Automatic
Copyright © 2016 IFAC
10.1016/j.ifacol.2016.10.712 324
2016 IFAC CTDSG

October 11-13, 2016. Prague, Czech RepublicS.M. Mousavi et al. / IFAC-PapersOnLine 49-27 (2016) 324–329 325

compare easy-to-implement controlled charging algorithms, assumed to be charged from their arrival state of charge, ���� ,
and to investigate the impacts that such algorithms may have towards the departure state of charge, ���� , under level 2 of
on distribution system operation, especially from the the SAE J1772 standard with a maximum rate of 6.6 kW
perspective of reducing energy losses. In this context, this during consecutive 15-min intervals, based on the proposed
paper brings two main contributions to the literature. First, a charging algorithms. The energy required for charging an EV
modeling approach is proposed to investigate the impact of is calculated by Eq. (1) [see Ota et al. (2012) for details].
charging EVs on energy losses. The proposed approach Table 1. Characteristics of Battery Models
considers the uncertainty in the characteristics of EV models
based on data from their manufacturers. It simulates EV Model/ Characteristics BEV1 BEV2 PHEV1 PHEV2
charging using the characteristics of battery packages used in
Voltage (VNom) [V] 325.6 364.8 345.6 334.7
the vehicles. Further, regardless of the previously used
assumption of fully discharged EVs at plug-in time, EV Capacity (CNom) [Ah] 50 66.2 15 47.8
decharging is modelled based on daily travelled distance for Energy Capacity [kWh] 16.28 24.15 5.2 16
the vehicles. Secondly, this paper proposes and compares
several charging scenarios, including uniform, random, Sensitivity Parameter (α) 15 15 15 15
conditional-random and valley-filling scenarios, each Electrical Range (km) 130 160 23 64
characterized with an algorithm and its associated
computational and communication requirements to be
adopted by an EV aggregator or system operator. The results ������� � ���� ����� − ���� )
are analysed and compared with those from uncontrolled ���� − ����
charging and a reference scenario with no EVs. The ������ �� � �
���� − ����
remainder of this paper is organized as follows: Section 2 ���� (1)
presents our modeling approach. Section 3 discusses the �������������������������� �� � �
���� − ����
proposed charging algorithms and associated methods to ����
calculate energy losses in the presence of EVs. Simulation −����� �� � �
���� − ����
results are presented in Section 4, and conclusions are given
in Section 5. where,��� ���� and ���� are the characteristics of each EV,
as defined in Table 1. It is also assumed that the EVs are to
2. EV CHARGING AND CUSTOMER LOAD MODELING be fully charged when the charging period ends. The arrival
A distribution system with N customers is considered. The state of charge is obtained for each EV using Eq. (2).
electrical load of the customers was modelled based on ���
���� � ���� − � ���� (2)
typical load data containing 15-min time-series electricity ��
demand for high, medium and low use customers over where, �� and ��� are the electrical range and daily travel
various seasons across a one year period (Real Market Design distance of the EV, respectively. While the electrical range is
Service 2015). Different demand profiles were randomly considered based on Table 1, the daily travel distance is
assigned to each of the customers of the distribution system modelled based on data provided in Table 2, which is an
based on the method presented in Richardson et al. (2010). extract of historical data from Commuting in Ireland (2015).
Assuming a maximum of one EV per house, each customer Although the data from Table 2 is valid for vehicles with
may have an EV or not. EV penetration is defined as the internal combustion engines, the driving habits of the EV
number of customers with an EV relative to the total number owners are assumed to be similar.
of customers. Thus, the maximum number of EVs, Table 2. Statistics of Daily Travel Distance
corresponding to 100% EV penetration, is equal to N.
However, in real cases, the EV penetration may be anywhere Distance Probability
from 0 to 100%. For any given EV penetration, the EV < 5 km 0.10
owners were assigned by a random sequence of k integers,
representing the customer number with an EV. The EV 5 – 10 km 0.19
characteristics were considered based on four EV models, 10 – 20 km 0.19
including two BEVs: Mitsubishi i-MiEV (Hosokawa et al., 20 – 30 km 0.14
2008) (BEV1) and Nissan Leaf (Yoshioka, 2011) (BEV2), as
well as two PHEVs: Toyota Prius (Yoda, 2010) (PHEV1) and 30 – 50 km 0.17
Chevrolet Volt (Chevrolet Volt Battery, 2015) (PHEV2). 50 – 70 km 0.06
Table 1 shows the characteristics of the vehicles. In this table,
70 –100 km 0.08
VNom and CNom are the nominal voltage and capacity of the
battery, respectively, and α is a sensitivity parameter between > 100 km 0.07
the state of charge and open circuit voltage of the battery.
Three cases were studied, each representing a certain share of
It was assumed that the EVs are charged during the night- BEVs and PHEVs among the EV fleet. The first case
time charging period, e.g. from 22:00 to 06:00, and this is considers 50% share for both BEVs and PHEVs, while the
regardless of the EV arrival time. This assumption requires other cases assume only the presence of BEVs or PHEVs. In
that an adjustable start delay should be applied for EVs that each case, the battery characteristic and the daily travel
arrive before the charging period starts. The EVs were

325
2016 IFAC CTDSG
326
October S.M. Mousavi et al. / IFAC-PapersOnLine 49-27 (2016) 324–329
11-13, 2016. Prague, Czech Republic
distance of each EV are randomly selected respectively from
Tables 1 and 2. The daily travel distance is multiplied by 1.2
in winter to account for the air conditioning load. The arrival
state of charge and the charging energy required by each EV
is calculated by Eqs. (1) and (2). Next, the required charging
energy is allocated to each EV for particular 15-min intervals
based on the proposed charging scenarios of Section 3. Load
flow is carried out in DIgSILENT Power Factory and the EV
charging load (if any) is added to each customer load during
15-min intervals, if all the voltage and line limits are
respected. The voltage limits ensure that the bus voltages are
within the normal operating range of 0.95 to 1.05 p.u. and the
line limits prevent overloading line sections and transformers.
Finally, the energy loss is calculated by writing a script in
DIgSILENT Power Factory to sum up the power losses of all
line sections over the entire 15-min intervals of the examined
day.
3. ENERGY LOSSES IN THE PRESENCE OF PROPOSED
EV CHARGING SCENARIOS
Five scenarios are studied, each characterized with a charging
algorithm, along with a reference scenario with no EVs. The
proposed scenarios consider a centralized charging structure
and are based on uncontrolled, random, conditional random,
uniform and valley-filling charging algorithms. Of these, the
uncontrolled scenario corresponds to a situation where there
are no smart meters available to control EV charging, while
the other scenarios are based on the assumption that each
household is equipped with a smart meter with load control
capability, utilized remotely by an EV aggregator, or system
operator, to control EV charging over 15-min intervals.
Scenarios 1 and 2 have no communication burden, while
scenarios 3-6 require a communication link between the EVs
and an aggregator or system operator. The scenarios are
summarised as follows.
Scenario 1- Reference scenario
For each 15-min interval of a typical day, customer loads are
obtained from the database, a load flow calculation is
performed, and the energy losses are calculated. No network
violations are considered.
Scenario 2- Uncontrolled charging scenario
Uncontrolled charging of EVs is applied during the charging
period. It is assumed that EVs are charged at their maximum
charging rate starting at 10 p.m. until they are fully charged.
No voltage and line limits are considered. The energy loss is
calculated based on the following procedure.
Step 1) For each charging interval �, retrieve customer loads
across the typical days.
Step 2) Obtain arrival state of charge, ���� , for each EV, and
calculate the required energy, ��������� , to charge each EV.
Step 3) Calculate the charging time, ��������� = ��������� /
����� , where ����� is the maximum charging rate;
Step 4) Allocate charging energy ��������� = 0��� � �����
to the EV during each 15-min interval �, if � � � � ��������� ;
Step 5) Repeat steps 2-4 for all EVs to obtain the charging
loads;
326
Step 6) For each charging interval �, perform load flow
calculation, and obtain the energy loss.
Scenario 3- Random charging scenario
EVs are charged at their maximum charging rate across
random 15-min charging intervals selected by a central
controller. The logic behind the random selection of the
charging intervals is to reduce computational burden when
the algorithm is adopted by an EV aggregator or system
operator in real time. Steps 1-2 and 5-6 remain unchanged
with respect to the uncontrolled charging scenario. Steps 3
and 4 are modified as follows.
Step 3) Initialise the charged energy, �������� = 0 for each
EV.
Step 4) While (�������� � ��������� ), generate an integer �,
� � ��� � � ���� �; to represent the interval when the EV is
charged. ���� is the total number of 15-min intervals of the
charging period. Allocate charging energy 0��� � ����� to
the EV during interval �. Perform load flow calculation.
Update �������� if the voltage and line limits are respected,
otherwise, remove interval � from the list of possible charging
intervals and repeat this step.
Scenario 4- Conditional random charging scenario
The random charging algorithm may charge EVs during peak
hours, resulting in an increase in energy losses, if no voltage
and line limits are violated. This is avoided by modifying
Step 4 of the random charging scenario as follows.
Step 3) While (�������� � ��������� ), generate an integer �,
� � ��� � � ���� �. Allocate charging energy 0��� � ����� to
the EV during interval �, if � is within the peak load period
and �������� � �� � ��������� , or if � is within the medium
load period and �������� � �� � ��������� , or if � is within
the valley load period. Perform load flow calculation. Update
�������� if the voltage and line limits are respected, otherwise,
repeat this step.
�� and �� are predefined coefficients in the interval [0, 1],
and are initially set to 0.33 and 0.67, respectively. The logic
behind these values is to divide the required energy of each
EV into three equal portions, to charge the EV with different
priorities when the charged energy, �������� , is within each
portion. The peak, medium and valley load periods are also
defined based on the system operator’s experience from the
load shape on the distribution system.
Scenario 5- Uniform charging scenario
In order to reduce the computational burden again, it is
assumed that the charging rate can be adjusted from zero to a
maximum, and that each EV is charged constantly at an
adjusted rate during the entire charging period. The step-by-
step procedure is the same as for the random charging
algorithm, with the following modification to Step 4.
Step 4) Allocate charging energy ��������� = ��������� /����
to the EV during each interval �, � � ��� � � ���� �. Perform
load flow calculation. If any voltage or line limits are violated,
remove interval � from the set of 15-min charging intervals,
and repeat this step. Otherwise, add the EV charging load to
the customer loads in each interval �.
2016 IFAC CTDSG

October S.M. Mousavi et al. / IFAC-PapersOnLine 49-27 (2016) 324–329
11-13, 2016. Prague, Czech Republic 327

Scenario 6- Valley-filling charging scenario the seasonal energy losses in winter tend to exceed the annual
In order to minimize the incremental energy losses in the averages regardless of the charging scenario. Further, due to
presence of EVs, this algorithm first finds the predicted lower home/work activities of customers, the energy losses
valley periods in the daily load profile, and then charges EVs tend to be marginally lower at weekends than weekdays
in the valley times. This scenario has the highest under the studied scenarios. Further, as can be seen, the
computational burden among the examined charging energy losses increase under the proposed EV charging
scenarios. Again, the procedure is the same as the random scenarios, as compared to the reference scenario with no EVs.
charging algorithm with Step 4 modified as follows. Such an increase may exceed 60% of the reference value, and
tends to be higher under uncontrolled charging than any other
Step 4) While (�������� � ��������� ), allocate charging scenario. This is due to the lack of coordinated control, where
energy ���� � ����� to the EV during each interval �, EVs are independently charged during the first few hours of
� � ��� � � ���� �. Perform load flow calculation based on the the charging period, which tends to coincide with the higher
present load profile and calculate the corresponding energy system load. On the contrary, adopting a valley-filling
losses for all charge allocations. Find the allocation with charging algorithm, which tends to charge EVs in the early
minimum energy losses, and name it as ��. The 15-min morning, may limit the incremental energy losses to 21% of
interval �� can be regarded as the valley time of the load the reference values, even at 50% EV penetration.
profile. Allocate the charging energy to the EV during
Considering the other scenarios, uniform charging seems
charging interval ��, and modify the load profile during
more effective than random charging in reducing the energy
interval ��. loss in the presence of EVs. This is due to the quadratic
relationship between the energy loss and the supplied load. In
4. SIMULATION RESULTS fact, the uniform scenario, although not optimal, is not far
The proposed algorithms of Section 3 are applied to a LV away from the optimal result, and accordingly, a uniform
residential distribution system in a suburban area in Dublin, charging can be regarded as a quasi-optimal charging
Ireland (McKenna and Keane, 2015). The system, as shown solution, which has lower computational and communication
in Fig. 1, includes a 10/0.4 kV transformer supplying 134 requirements, and therefore is easier to implement by the
residential customers through 1.2 km of three-phase mains system operator or EV aggregators. Further, it can be seen
cables and 980 m of single-phase service cables. The that including the conditional statements in the random
technical specifications for the system components were charging introduces a slight improvement (≈3%) in the
supplied by Electricity Supply Board (ESB) Networks, who results. This is because the load demand usually drops after
are the distribution system operator in the Republic of midnight, and a conditional random scenario charges EVs at
Ireland. a higher rate after midnight.
Table 3. Daily Energy Loss
Scenario Sum- Sum- Win- Win- Annual
WD WE WD WE Ave.
(kWh) (kWh) (kWh) (kWh) (kWh)
Reference 153 117 350 319 267
Uncontrolled 240 188 588 523 443
Random 213 165 493 450 375
Fig. 1. Residential suburban distribution system Conditional Random 206 162 477 441 363
First, an analysis is performed to assess the impact of Uniform 193 148 450 408 342
charging EVs on energy losses under the analysed scenarios. Valley-filling 181 142 428 386 324
The EV penetration is assumed to be 50%, which means that
half of the customers own an EV. The EV population is When analyzing the results of the conditional random
considered to consist equally of BEVs and PHEVs, with the scenario, care should be paid to selection of the coefficients
characteristics shown in Table 1. In order to compare the �� and �� . For more investigation, they are changed as
proposed scenarios under different loading profiles, a Monte denoted in Table 4, and the results are monitored.
Carlo simulation is performed for typical days, corresponding Table 4. Coefficients of Conditional Random Charging
to different seasons of the year, and the daily energy losses
are calculated for each simulation. The arrival SOC of each Cases �� ��
EV battery is calculated randomly as explained in Section 2. Case 1 0.33 0.67
The departure SOC is obtained based on the scenarios of
Section 3. Table 3 shows the results averaged for summer Case 2 0.50 0.75
weekdays and weekends (Sum-WD and Sum-WE), winter Case 3 0.25 0.50
weekdays and weekends (Win-WD and Win-WE).
As indicated by the results (not shown here), when the
In Table 3, the first result that claims attention is that due to coefficients are changed from case 1 to case 2, the annual
the higher customer loads, driven by heating loads in winter, energy loss decreases from 1.36 p.u. to 1.32 p.u. (considering

327
2016 IFAC CTDSG
328 11-13, 2016. Prague, Czech RepublicS.M. Mousavi et al. / IFAC-PapersOnLine 49-27 (2016) 324–329
October

the daily energy loss of the reference scenario as the base case for different start times (8 p.m., 10 p.m., 12 midnight),
quantity), whereas in case 3, it increases to 1.39 p.u. This and end times (4 a.m., 6 a.m., 8 a.m.), respectively, in per
means that decreasing the coefficients results in the unit of the annual energy loss against the reference scenario.
conditional random scenario shifting towards that of the Table 6. Daily Energy Loss for EV Penetrations
random scenario (1.40 p.u.), because with low coefficient
values, EV charging may occur at peak load hours as likely Scenario/Penetration 30% 50% 80%
as the valley load periods. On the contrary, by increasing the
Uncontrolled 337 kWh 443 kWh 675 kWh
coefficients, most EVs are charged during the valley load
hours, and therefore, the annual energy loss is shifted towards Random 326 kWh 375 kWh 511 kWh
the valley-filling scenario (1.21 p.u.). It is noted that based on Conditional Random 316 kWh 363 kWh 458 kWh
the typical shape of the aggregated load at distribution
substations, the peak load period is from 16:00 to 20:00, and Uniform 307 kWh 342 kWh 412 kWh
the valley load period is from 02:00 to 07:00. Valley-filling 293 kWh 324 kWh 394 kWh
In order to investigate the impact of EV characteristics on the 1.45
results, two additional cases are studied, first when the EV

Annual energy losses (in per unit of reference scenario)

Conditional Random
population is only composed of BEVs, and then, when it only Uniform
Optimal
1.4
contains PHEVs. Table 5 shows the results of these extreme
cases. By comparing Tables 3 and 5, it can be seen that the
energy losses represent their maximum and minimum values
1.35

under the BEV and PHEV cases, respectively, which

correspond to 1.058 p.u. and 0.971 p.u. (on average), 1.3

considering the daily energy loss of the BEV+PHEV case as

the base quantity. This can be explained by the characteristics 1.25

of BEVs, especially their higher electrical range and energy

capacity over PHEVs, resulting in a higher charging demand 1.2

upon their arrival to make the vehicles ready for next day use.
1.15
Table 5. Daily Energy Loss for EV Technologies 8 p.m. 10 p.m. 12 p.m.
Start of charging period (hour)

Scenario/EV Mix BEV+PHEV BEV PHEV Fig. 2. Variation of the per unit energy loss as a function of
Uncontrolled 443 kWh 470 kWh 432 kWh the start time of the charging period
Random 375 kWh 392 kWh 358 kWh 1.45
Annual energy losses (in per unit of reference scenario)

Conditional Random
Conditional Random 363 kWh 383 kWh 349 kWh Uniform
1.4 Optimal
Uniform 342 kWh 361 kWh 336 kWh
Valley-filling 324 kWh 349 kWh 318 kWh 1.35

In order to investigate the impact of EV penetration on the 1.3

results, the penetration level is first decreased to 30%, and
then increased to 80%. Table 6 shows the energy losses 1.25
calculated under the BEV+PHEV case.
From Table 6, it can be observed that, as expected, the energy 1.2
losses increase as the EV penetration rises. However, such an
increase caused by going from 30% to 80% EV penetration 1.15
under the uncontrolled charging scenario may exceed 100% 4 a.m. 6 a.m.
End of charging period (hour)
8 a.m.

of the preliminary daily energy loss (267 kWh), while the

increased energy losses remain below 40% for valley-filling
and uniform charging, and below 55% for the conditional
random charging scenario. Thus, it can be concluded that the
valley-filling, uniform and conditional-random algorithms are
the most efficient algorithms among the studied scenarios to
make distribution systems capable of accommodating a wide
range of EV penetrations.
The results of Tables 5 and 6 are based on the definition of
the charging period, which begins at 10 p.m., and ends at 6
a.m. on the next day. In order to investigate the impact of the
start/end time of the charging period on the results, sensitivity
analyses are carried out under the three best performing
charging algorithms from the perspective of reducing energy
losses. Figs. 2 and 3 show the results under the BEV+PHEV
Fig. 3. Variation of the per unit energy loss as a function of
the end time of the charging period
From Fig. 2, it can be seen that the efficiency of the uniform
charging algorithm improves with a later start of the charging
period. This is because the algorithm distributes EV charging
loads more evenly across the charging period, and a later start
time imposes the charging loads after the peak load hours.
However, the annual loss remains almost the same under both
the conditional random and valley-filling (optimal) charging
scenarios, because the adopted algorithms of these scenarios
inherently charge most EVs during off-peak hours in the
early morning. Contrary to these results, as can be seen from
Fig. 3, the annual energy loss tends to decrease with a later
end charging period, regardless to the adopted charging

328
2016 IFAC CTDSG
October S.M. Mousavi et al. / IFAC-PapersOnLine 49-27 (2016) 324–329
11-13, 2016. Prague, Czech Republic
algorithm. This is due to the low customer loads in the early
morning and charging more EVs in this period may decrease
the energy loss. Although Figs. 2 and 3 depict the results of
the charging algorithms for various definitions of the
charging period, more realistic results can be obtained by
considering stochastic plug-in and departure times of the EVs
and dynamic electricity tariffs, which are the subject of our
future work.
6. CONCLUSIONS
A modeling framework has been presented to analyse the
impact of various proposed charging scenarios, characterized
by uniform, random, conditional-random, and valley-filling
charging algorithms, on energy loss in the presence of EVs in
distribution systems. The modeling framework and the
proposed algorithms have been applied to a distribution
system. As revealed by the results, the proposed valley-filling
algorithm emerged as the best solution for controlled
charging of EVs from the perspective of reducing the energy
loss, but it has the cost of computation. Further, the uniform
and conditional-random algorithms can be regarded as quasi-
optimal solutions with easier application and less
computational requirements. Further, high EV penetrations
might increase energy losses, especially with uncontrolled
and random algorithms. The sensitivity of the results has
been studied against the definition of the charging period.
Based on the results obtained from the test network, later start
and later end to the charging period tends to improve the
performance of the charging algorithms, as customer loads
are lower during morning hours compared to the evening.
REFERENCES
Chevrolet Volt Battery. (2015). [Online]. http://media.gm.
com/content/dam/Media/microsites/product/volt/docs/bat
tery_101.pdf
Clement, K., Haesen, E., and Driesen, J. (2010). The impact
of charging plug-in hybrid electric vehicles on a
residential distribution grid. IEEE Trans. Power Syst.,
vol. 25, pp. 371–380.
Commuting in Ireland. Census 2011 Profile Door to Door.
(2015) [Online]. http://www.cso.ie/en/census/census/.
Denholm, P. and Short, W. (2006). An Evaluation of Utility
System Impacts and Benefits of Optimally Dispatched
Plug-in Hybrid Electric Vehicles, NREL National
Renewable Energy Lab.
Fernandez, L., San Roman, T., Cossent, R., Domingo, C., and
Frias, P., (2011). Assessment of the impact of plug-in
electric vehicles on distribution networks, IEEE
Trans.Power Syst., vol. 26, pp. 206–213.
Gan, L., Topcu, U., and Low, S. H. (2013). Optimal
decentralized protocol for electric vehicle charging.
IEEE Trans. Power Systems, vol. 28, pp. 940–951.
Geng, B., Mills, J., and Sun, D. (2013). Two-stage charging
strategy for plug-in electric vehicles at the residential
transformer level. IEEE Trans. Smart Grid, vol. 4, pp.
1442–1452.
He, Y., Venkatesh, B., and Guan, L. (2012). Optimal
scheduling for charging and discharging of electric
329
329
vehicles. IEEE Trans. Smart Grid, vol. 3, pp. 1095–
1105.
Hosokawa, T., Tanihata, K., and Miyamoto, H. (2008).
Development of i-MiEV next-generation electric vehicle
(second report). Mitsubishi Motors Tech. Rev., vol. 20,
pp. 52–59.
Jin, C., Tang, J., and Ghosh, P. (2013). Optimizing electric
vehicle charging with energy storage in the electricity
market. IEEE Trans. Smart Grid, vol. 4, pp. 311–320.
Karfopoulos, E. and Hatziargyriou, N. (2015). Distributed
coordination of electric vehicles providing V2G services.
IEEE Trans. Power Systems, vol. 31, pp. 329-338.
McKenna, K., and Keane, A. (2015). Residential load
modeling of price-based demand response for network
impact studies. IEEE Trans. Smart Grid, vol. 7, pp.
2285-2294.
Momber, I., Wogrin, S., and San Roman, T. (2015). Retail
pricing: a bi-level program for PEV aggregator decisions
using indirect load control. IEEE Trans. Power Systems,
vol. 31, pp. 464-473.
Ota, Y., Taniguchi, H., Nakajima, T., Liyanage K. M., Baba,
J., and Yokoyama A. (2012). Autonomous distributed
V2G (vehicle-to-grid) satisfying scheduled charging.
IEEE Trans. Smart Grid, vol. 3, pp. 559-564.
Ramos, A., Olmos, L., Latorre, J. and Perez-Arriaga, I.
(2008). Modeling medium term hydroelectric system
operation with large-scale penetration of intermittent
generation, in Proc. XIV Latin and Iberian Conf.
Operations Research, Colombia.
Real Market Design Service. Standard Load Profiles, (2015)
[Online]. http://rmdservice.com/standard-load-profiles/.
Richardson, P., Flynn, D., and Keane, A. (2012). Optimal
charging of electric vehicles in low-voltage distribution
systems. IEEE Trans. Power Syst., vol. 27, pp. 268–279.
Richardson, P., Flynn, D., and Keane, A. (2010). Impact
assessment of varying penetrations of electric vehicles
on low voltage distribution systems. In Proc. IEEE
Power and Energy Soc. General Meeting, Minneapolis,
MN.
Shao, S., Pipattanasomporn, M., and Rahman, S. (2009).
Challenges of PHEV penetration to the residential
distribution network. In Proc. IEEE Power and Energy
Soc. General Meeting, Calgary, Canada.
Sortomme, E., Hindi, M., MacPherson, S., and Venkata, S.
(2011). Coordinated charging of plug-in hybrid electric
vehicles to minimize distribution system losses. IEEE
Trans. Smart Grid, vol. 2, pp. 198–205.
Tang, W., Bi, S., and Zhang, Y. (2014). Online coordinated
charging decision algorithm for electric vehicles without
future information. IEEE Trans. Smart Grid, vol. 5, pp.
2810–2824.
Wu, D., Aliprantis, D., and Ying, L. (2011). On the choice
between uncontrolled and controlled charging by owners
of PHEVs. IEEE Trans. Power Del., vol. 26, pp. 2882–
2884.
Yoda, T. (2010). Development of battery pack & system for
plug-in hybrid. In Proc. Adv. Automotive Battery Conf.
Yoshioka, N., (2011). New era of mass-produced BEVs. In
Proc. Adv. Automotive Battery Conf.