Content uploaded by Karin Anna Hummel
Author content
All content in this area was uploaded by Karin Anna Hummel on Dec 03, 2014
Content may be subject to copyright.
Optimal Deployment of Charging Stations for
Electric Vehicular Networks
Andrea Hess
University of Vienna, Austria
andrea.hess@univie.ac.at
Francesco Malandrino
Politecnico di Torino, Italy
malandrino@tlc.polito.it
Moritz B. Reinhardt
University of Vienna, Austria
moritz.reinhardt@univie.ac.at
Claudio Casetti
Politecnico di Torino, Italy
casetti@polito.it
Karin Anna Hummel
ETH Zurich, Switzerland
karin.hummel@tik.ee.ethz.ch
Jose M. Barceló-Ordinas
UPC Barcelona, Spain
joseb@ac.upc.edu
ABSTRACT
In a smart city environment, we look at a new, upcoming generation
of vehicles running on electric power supplied by on-board batter-
ies. Best recharging options include charging at home, as well as
charging at public areas. In this setting, electric vehicles will be
informed about public charging stations using wireless communi-
cations. As the charging stations are shared resources, cooperat-
ing electric vehicles have the potential to avoid unbalanced use of
recharging stations and lengthy waiting times.
We present a model for electric vehicles and their battery de-
pletion, vehicle mobility, charging stations, and give a solution for
optimal placement of charging stations in a smart city. Our place-
ment approach is based on genetic programming and simulation
of electric vehicles which move on a real map of a European city.
We show that after a few evolution steps, an optimal solution of
the charging infrastructure is derived based on mean trip times of
electric vehicles.
Categories and Subject Descriptors
C.2.1 [Computer Communication Networks]: Network Archi-
tecture and Design—Wireless communication; I.2 [Artificial In-
telligence]: Problem Solving, Control Methods, and Search
Keywords
Electric Vehicles, Charging Infrastructure Deployment, VANETs,
Urban Mobility Modeling
1. INTRODUCTION
Despite being at a relatively embryonic stage in terms of mar-
ket penetration, electric cars represent the most environmentally
friendly vehicle as they have absolutely no emissions (unless one
accounts for the emission of their primary energy provider, i.e., the
power stations). Electric vehicles (e-vehicles) will for sure con-
tribute to overcome the predictable shortage of fossil fuel in the
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
UrbaNE’12, December 10, 2012, Nice, France.
Copyright 2012 ACM 978-1-4503-1781-8/12/12 ...$15.00.
near future. To support e-vehicles, driving ranges have to be en-
larged and electricity suppliers will need to redesign the power
grid around new charging stations and schedule their provisioning
around the clock in order to offer a high-quality service in terms of
usability, low delay, and security.
Vehicle manufacturers are already equipping new vehicles with
navigation and infotainment devices as well as with wireless tech-
nology to support vehicle to vehicle (V2V) and vehicle to infras-
tructure (V2I) communication provided by, e.g., 3G and future 4G
cellular networks or IEEE 802.11p. Conceivably, this trend will
likely continue for e-vehicles. Up-to-date on-board information
about the location of a charging station and its status in terms of
number of free plugs and average waiting time at the moment is
likely to be provided in future vehicles. This information may help
individual users to dynamically optimize their travel in terms of re-
duced search time for charging stations, fares, and overall, reduced
trip time. Even governments may leverage this Intelligent Trans-
port Infrastructure (ITS) to foster a decrease of pollution via proper
policies (e.g., adapting fares). Additionally, ITS services can in-
form vehicles about the current road traffic conditions to allow a
better choice during peak times. Our final vision is a network of
electric vehicles which cooperate to efficiently use the charging in-
frastructure.
In this paper, we assume the situation where networked electric
vehicles are cruising until they detect a low battery charge and, as
they are connected via, e.g., a vehicular ad-hoc or cellular network,
they receive advertisements of charging stations and navigate to
these stations. With today’s technology, vehicle charging is still
a time-consuming task although fast charging systems capable to
charge a battery from 0 to 50% in a few minutes are being devel-
oped1. Hence, in a future setting, on demand charging will be one
option of a plethora of charging systems, e.g., installed at parking
lots, shopping malls, homes, campuses, etc. (We review related
work on scenarios of e-vehicles in Section 2.)
In this setting, the main contributions of our paper are:
•We present a model for e-vehicles including state-of-the-art
ranges and battery depletion figures and discuss the infras-
tructure necessary to operate them in an urban environment,
i.e., charging stations (Section 3). Further, we introduce a
mobility model for on demand charging including a state
change (navigation change) for e-vehicles in case of low bat-
tery power (Section 4).
•We present a novel approach to optimizing the deployment
1http://www.betterplace.com
1
of charging stations using genetic programming. As the de-
ployment of charging stations depends on the mobility of e-
vehicles, which is not observable in real urban areas at higher
scales, the real-world behavior is not yet clear. Thus, we se-
lected the genetic programming approach which is capable of
adapting to mobility. We describe our approach in Section 5.
•We give first results of a simulation-based study of synthetic
vehicular mobility traces on a real city map, i.e., the city map
of Vienna, and show that indeed after a few evolution steps,
the genetic algorithm produces a solution which is, with sta-
tistical certainty, optimal (Section 6).
2. RELATED WORK
The integration of electric vehicles into the existing power grid,
road traffic, and an ITS infrastructure raises challenging issues that
will be discussed in this section.
Recent literature on smart grid systems started paying attention
to the integration of vehicles as participants in the electricity mar-
ket. In [11], e-vehicles are assumed to be consumers and recharge
their batteries through the grid, but also conserve capacity. They
are expected to react to pricing information when being parked (to
delay, reduce, or switch-off charging), and to sell power to the elec-
tric grid when feasible (Vehicle to Grid (V2G) power). Although
one single vehicle is not providing enough power to sell, vehicle
aggregation can save costs as shown in [11]. The V2G concept de-
pends on the power required for moving the vehicles themselves,
whether hybrid vehicles are used (carbon and electric power), and
on the standing/parking times. In our work, vehicles are integrated
as consumers only.
Further, approaches to balance the load on the power grid by
regulating charging demand have been investigated. In [4], a con-
gestion pricing model represents driver preferences by the willing-
ness to pay (charging costs and rates). Other concepts for charg-
ing scheduling include load shifting dependent on predicted de-
mand [1] or measurement-based grid congestion control [2]. In
our approach, we take the perspective of the driver and aim at min-
imizing the trip time.
Since future urban mobility behavior will be significantly af-
fected by the properties of electric vehicles, mobility models taking
the charging demands into account are required. Early simulation
studies of e-vehicle mobility account for the energy consumption
and charging demands of vehicles during every-day trips. In [23], a
mobility model that stochastically determines transitions between
daily activities associated with parking locations (workplace, shop-
ping center, etc.), is used to investigate the impact of charging
strategies on the energy supply system. When modeling the driving
behavior for an information system about nearest charging stations
and energy prices, user classes can be differentiated. In [5], users
are characterized based on personal information about age, gender,
energy market behavior (e.g., minimum accepted charge level), trip
type (workday or weekend), and car properties (e.g., battery type).
In [19], mobility of e-vehicles is described as being dependent on
their need for recharging as well as on the waiting time for charg-
ing. The network of charging stations is modeled as a graph and
vehicles are attracted by a station and, finally, queue (modeled by
queuing networks). Similarly, our approach uses a basic vehicular
mobility model and a mode change depending on a battery charge
threshold. In our overall concept, the mobility behavior will be
further influenced by coooperative decisions about navigation.
To facilitate e-vehicle cooperation, technologies for (ad-hoc) com-
munication between vehicles and with smart grid infrastructures are
needed. Recently proposed systems utilize, e.g., cellular network-
based text messaging [5] or WMNs [3] for exchanging position,
charging state, and available stations nearby, to coordinate charging
load in a power grid. In [8], the authors discuss the advantages and
disadvantages of different communication technologies available
for smart grids, such as wireless mesh networks (WMNs), cellular
networks, powerline communication, and digital subscriber lines.
Finally, the improved placement of charging stations has already
been targeted in a few studies. For instance, charging stations are
placed according to actual energy demands at highway sections as
portable storage units [15] or based on day- and nighttime demand
estimations for residence and workplace areas of a city [7]. In [21],
station topologies are compared in terms of spatial coverage by
modeling driver’s daily activities as well as e-vehicle adoption (in-
fluenced by driver inconvenience in the surrounding). A placement
algorithm for charging stations of a door-to-door electric bus ser-
vice is evaluated in [12] in terms of customer acceptance. The po-
sitioning is based on customer distributions extracted from Tokyo
taxi traces. In [18], the optimal placement of V2G stations for dis-
charging vehicle batteries onto the grid, to compensate for variabil-
ities of energy generation by renewable sources, is studied.
To get an idea of present charging facilities, we refer the reader
to the alternative fueling station locator2. This database holds about
4000 electric stations in the US, most of them providing one to four
slow and standard charging plugs.
3. BATTERY CHARGING CHARACTERIS-
TICS OF E-VEHICLES
Users of e-vehicles expect high QoS (Quality of Service) in the
delivery of energy given by electricity utility companies. In this
context, QoS is defined in terms of how fast a battery can be charged
and discharges, energy price, charging delay, and the impact of the
charging method on the battery performance. There are several
strategies to deliver batteries to e-vehicles [20]: (i) the battery is
included in the vehicle price and thus belongs to the vehicle owner,
(ii) the battery is owned by a third party that leases them, (iii) the
battery is owned by utilities that receive a fixed rate added to the
distribution rate. Batteries further differ in terms of charging capa-
bility as defined in the SAE J1772 standard3for electrical connec-
tors for e-vehicles:
1. Level 1 (slow charging) implies AC energy to the on-board
charger of the vehicle, e.g., 120V/16Afor 1.92kW charging;
charging time tc:∼10 h.
2. Level 2 (standard charging) implies AC energy to the on-
board charger of the vehicle, e.g., 208-240VAC, single phase,
12A-80Afor 2.5-19.2kW;tc:∼6-8 h.
3. Level 3 (fast charging) implies DC energy from an off-board
charger; there is no minimum energy requirement but the
maximum current specified is 400Aand 240kW continuous
power supplied; tc:∼30 min.
The energy can be supplied by electricity utilities that charge bat-
teries at specific places (home, work, shopping centers, posts in the
street, etc.). The daily traffic routine of people plays an important
role in the analysis of how batteries are charged and discharged. For
example, Zhao et al. [23] show that the locations of parking cars are
predominant at work and at home, while shopping only takes place
in business hours. It seems clear that Level 1 charging is ideal for
2http://www.afdc.energy.gov/locator/stations/
3published versions of the standard can be purchased here:
http://standards.sae.org
2
charging the vehicle at home (e.g., at night) at low prices. Level 2
charging seems a good solution for charging while the user resides
at a place for a long time (e.g., at work), whereas Level 3 seems the
solution for fast charging at higher prices when the battery is get-
ting empty and the user does not stop for long. An alternative for
recharging batteries is swapping them for charged ones. This solu-
tion implies that a charging station is capable of storing a sufficient
number of batteries for different vehicle models and of charging
them fast. First deployments have recently taken place4, however
the logistics to implement battery swapping require battery stan-
dard agreements between e-vehicle manufacturers [20].
The battery consumption of e-vehicles depends on the intrinsic
characteristics of the battery and how the vehicle consumes energy
while traveling. For example, let us look at the Nissan Leaf5. The
Nissan Leaf is powered by a 24kWh lithium ion battery pack. Us-
ing an on-board 3.3kW charger, the Leaf can be fully recharged
from empty in 8 hours from a 220V/30ALevel 2 supply. The bat-
tery pack is expected to retain 70% to 80% of its capacity after
10 years assuming Level 1/2 charging. On the other hand, 80%
of its capacity can be charged in 30 minutes in a Level 3 charger.
However, using fast charging as the primary way of recharging, the
normal and gradual battery capacity loss is about 10% higher than
regular 220Vcharging over a 10-year period.
For the Nissan Leaf, the battery lasts for around 160 km on the
EPA (Environmental Protection Agency’s) city driving cycle – at
40 km/h and with air condition off. However, on highways – at 90
km/h and with air condition on – the battery lasts for around 110
km. Considering stop and go traffic in winter – at 25 km/h and
heater on – the battery lasts for around 100 km and in heavy stop
and go traffic at 10 km/h and with air condition on, the battery lasts
for around 70 km.
Communications too have an impact on the battery. For exam-
ple, a DCMA-86P2 using 5.9GHz DSRC protocol supporting com-
munications to vehicle (V2V) or to roadside (V2R) consumes 250
mW at 6 Mb/s. If a lightweight communication protocol sends 10
packets/s of size 1500 Bytes at 6 Mb/s, the power consumption is
approximately in the order of 0.018 kWh. For a 24 kW battery, that
consumption represents around 0.075% of the battery power.
Depletion Model: We use a simple linear depletion model
which assumes an average depletion based on the maximum range
known from data sheets of manufacturers, e.g., a range of 70–
160 km in an urban area for the Nissan Leaf (100% capacity, i.e.,
fully charged battery). The depletion follows the equation
Ct2=Ct1−d·u, (1)
where Ct1,C
t2are the capacities of the battery in % at the two
subsequent timesteps t1and t2,dis the distance traveled in km
between t1and t2, and uis the average battery use given in % per
km (where Ct1≥d·u). For example, when assuming a range of
100 km (100% capacity), then u=1%per km.
Charging Model: Similar to the depletion model, we use a lin-
ear fast charging model (Level 3 charging) of about 30 min charg-
ing time (from 0 to 100% capacity). The charging follows the equa-
tion
Ct2=Ct1+(t2−t1)·r, (2)
where Ct1,C
t2are the battery capacities in % at two subsequent
timesteps t1and t2, and ris the charging value in % per min (where
4see, e.g., http://www.betterplace.com/
5http://en.wikipedia.org/wiki/Nissan_Leaf
Ct1+(t2−t1)·r≤100). For example, when assuming a charging
time of 30 min (from 0% to 100%), then r∼3.33% per min.
4. E-VEHICLE MOBILITY
Electric vehicular networks show a new type of mobility behav-
ior which is determined by the purpose of a trip, but also by the
need to recharge the battery more frequently in a more resource-
constrained environment than by using traditional fuel. Mobility
behavior of e-vehicles has been already described in the scope of
vehicular ad-hoc networks (VANETs). Vehicular mobility models
address different aspects as reflected by the classification provided
in [6]: (i)stochastic models select topologies and road character-
istics randomly; (ii)traffic stream models look at vehicle flows at
a macroscopic level; (iii)car following models account for the re-
lationship between cars and surrounding vehicles in terms of dis-
tance, speed, acceleration, etc.; (iv)queueing network models fo-
cus on car densities on roads, and (v)behavioral models address
driving characteristics derived from social rules. Among the most
popular vehicular models are the Nagel-Schreckenberg model [17],
the Intelligent Driver Model (IDM) [22], and the Krauss model [14].
The Krauss model provides a safe distance, introduction of a target
speed, a jitter to the response to a stimulus, and a stochastic ap-
proach to model human irrational behavior (see, e.g., the summary
of vehicular mobility models provided in [10]).
4.1 Extending the SUMO mobility simulator
A widely-known vehicular traffic simulator is SUMO6(for fur-
ther available traffic simulators see [9, 16]). The SUMO simulator
generates mobility based on road networks where movements be-
tween source and destination roads are determined by, e.g., a short-
est path algorithm. We use and extend the SUMO simulator to
generate feasible mobility traces for e-vehicles.
To simulate e-vehicular traffic in a realistic environment, street
maps including details about street type, number of lanes, speed
limitations, etc. can be imported to SUMO from geodata sources.
Figure 1 shows a topology extracted from OpenStreetMap7data
of Vienna including traditional fuel gas stations. We re-use actual
locations of existing fuel gas stations as an approach to initially
position charging stations in an urban environment. However, one
caveat is that charging station placement will look different in ten
years from now and that different time scales apply for recharging
and refueling. The rationale behind using current station positions
is their placement in locations preferable w.r.t. demand, reachabil-
ity, maintaining a minimum distance to the next station, etc.
Figure 1: Extract of Vienna’s street network imported to
SUMO (gas station locations highlighted red).
In addition to positions of charging stations, we extend SUMO
by introducing the behavior of e-vehicle charging and use the de-
pletion and charging model for batteries introduced in Section 3.
6http://sumo.sourceforge.net/
7http://www.openstreetmap.org/
3
Furthermore, the charging model has to assume a possible number
of charging plugs provided by a station.
Finally, we assume that the vehicles are equipped with commu-
nication as well as positioning and navigation support. While, a
priori, vehicles choose a destination based on their trip purpose and
use a traditional car following model, we add a route adaptation
decision logic which changes the navigation of e-vehicles when
recharging is required.
4.2 Navigating to charging stations on demand
Whenever charging is needed, e-vehicles change their behavior
and try to reach a charging station. This behavior is triggered by a
lower threshold termed Tl. Whenever the State of Charge (SOC) is
less than or equal to Tl, the vehicle changes to follow the attraction
model. In the current implementation, the attraction model forces
navigating to the charging station closest in distance to avoid run-
ning out of battery on the way to the charging station, queuing if
necessary, and recharging the battery until an upper threshold Tu
is reached. Alternatively, an attraction model may minimize over-
all trip length or trip time. In these cases, care has to be taken
to avoid vehicles running out of battery while approaching the se-
lected charging station. After charging, the e-vehicle resumes its
purpose-driven navigation (e.g., following the shortest path to a
destination). Figure 2 illustrates the state transitions between the
basis mobility model and the attraction mobility model.
Basis
model
Attraction
model
SOC level Tl
SOC level Tu
Figure 2: Transition between mobility models triggered by each
e-vehicle’s level of charging.
Energy supply is another aspect to model. Providing energy sup-
ply for each charging station can be either realized by a smart grid
or by primary power stations serving a set of nearby stations. To
model the sharing of resources between different secondary power
stations in the latter case, location and capacity of primary stations
have to be taken into account. For our first experiments, we model
only the supply distribution between the secondary stations assum-
ing that they are all served by the same primary station.
5. CHARGING STATION DEPLOYMENT
As discussed in Section 3, electric vehicles will not have a tremen-
dous range even in the best of conditions, and charging will be a
time-consuming task, definitely lengthier than current gas pump
stops. It is therefore crucial that the deployment of charging sta-
tions be as much as possible in tune with the expected mobility of
electric vehicles. It is also important that new variables, specific for
electric charging stations, be cast in the problem to be addressed.
We study the problem of charging station deployment through
the lenses of optimization and, given a generic city map, we assume
that:
•there are pprimary power stations each capable of serving
up to XkWh;
•charging stations can be deployed in any of mcandidate lo-
cations;
•each station could be equipped with up to kpower plugs.
Notice that, with the above assumptions, each station could host
kvehicles in parallel. The amount of power served to such vehi-
cles is limited by the capacity Xof its primary station, divided by
the number of power plugs currently in use at any of the charging
stations served by the same primary station.
Beside the above constraint, it is conceivable that additional con-
straints can be applied, such as a monetary budget: each candidate
location, if selected, drains the budget by some amount. It follows
that we can have a limit on the number of both charging stations
and the plugs that can be deployed.
As far as the optimization objective is concerned, we select a
metric linked to the user satisfaction, namely, the average trip time
of electric vehicles. Such a trip time will account for the whole
route of the vehicle: from its origin to the selected charging station
and from it to the final destination, plus the queuing and recharge
times.
5.1 Deployment Optimization
Feasible charging station deployments are represented by asso-
ciating discrete quantities (the number of power plugs to deploy)
to the (discrete) elements of a set (the candidate locations). Mod-
eling the deployment through a standard MILP model would result
in an overwhelming number of discrete variables, and be computa-
tionally infeasible even for small instances of the problem. Indeed,
a MILP modelling would not even be feasible unless we derive
a closed-form expression for the vehicle trip time (as opposed to
simulating the possible solutions). Therefore, we resort to a ge-
netic programming approach [13]. The fitness function, i.e., the
objective to minimize, corresponds to the delta of trip time defined
earlier. Individual solutions correspond to feasible deployments,
i.e., mapping of station IDs in [0,m]to number of plugs in [0,k].
Our algorithm works as follows:
1. the solution pool is initialized with Prandom solutions;
2. for each solution in the pool, compute its fitness value;
3. use such fitness values to sort the solutions in the pool;
4. discard the bottom fraction f·Pof the solution pool, which
then has size (1 −f)·P;
5. pair-wise combine the solutions still in the pool (e.g., select-
ing all the stations present in both solutions) and obtain f·P
new ones;
6. add the newly-constructed solutions to the pool;
7. go to Step 2.
The whole process is repeated for the target number Gof gener-
ations. The outcome of the process is a set of feasible solutions,
along with their fitness values. The best among such solution is,
with statistical certainty [13], very close to the actual optimal solu-
tion.
Along with its simplicity, a major advantage of genetic program-
ming is that we can tune its behavior by setting the parameters P,
G,f. For example, a larger solution pool (i.e., a higher P) puts
more emphasis on the exploration phase, i.e., trying many different
potential solutions. On the other hand, increasing the number Gof
generations puts more emphasis on the exploitation phase, i.e., im-
proving the existing solutions as much as possible. Given the spe-
cific scenario that we take into account, we need to take a balanced
approach, and therefore set the parameters as shown in Table 1.
4
2400
2500
2600
2700
2800
2900
3000
0 20 40 60 80 100 120 140
Average trip time [s]
Consecutively generated solutions
(a)
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100 120
CDF
Trip time [min]
solution w. t(min)
solution w. t(max)
(b)
0
50
100
150
200
0 20 40 60 80 100 120
Num. of queuing vehicles
Simulation time [min]
solution w. t(min)
solution w. t(max)
(c)
Figure 3: (a) Average trip time tfor the deployment solutions generated by the optimizer. (b-c) Comparison of the solutions with
minimum and maximum tin terms of (b) cumulative density functions (CDFs) for trip time and (c) number of vehicles in charging
station queues.
Traffic model
Simulation time 7200 s
Area size ∼36 km2
Number of e-vehicles 2160
Car following model Krauss model
Depletion model
Battery capacity 24 kWh
E-vehicle driving range 100 km
Energy consumption per km u1%
Lower SOC threshold Tl5%
Upper SOC threshold Tu20%
Charging model
Number of charging stations m6
Plugs per station k0, 3, 6, 9, or 12
Number of primary stations p1
Maximum total capacity 30 plugs
Charging speed r3.33% per min
Waiting queue length infinite
Genetic algorithm
Solution pool size P40
Discard fraction f0.5
Number of generations G5
Table 1: Simulation parameters.
6. EXPERIMENTS
During each experiment, the optimizer iteratively inputs possible
charging station configurations to the traffic simulation to compute
the corresponding fitness value. In order to account for both power
capacity limits and budget constraints, the total number of charging
plugs that can be served is set to 30. The vehicular traffic is simu-
lated on the city center topology of Vienna (Figure 1), where 720
vehicles are initially placed and additional 720 are inserted every
hour (one every 5 s of the remaining simulation time) to keep the
traffic density on a constant level. Source/destination locations are
randomly selected (uniform distribution). An upper battery level
threshold of 20% is used since 20 km range can be considered as
adequate for the area size of 6×6 km. Besides, this reduces the
impact of charging duration on trip time as the optimization should
depend on navigation and queuing time, and additionally, allows
emulating more charging services in shorter simulation time. As
car following model we selected the Krauss model since it provides
basic vehicular behavior as described in Section 4 and a robust im-
plementation of the Krauss model exists already in SUMO. Table 1
summarizes the parameter setting used for traffic, depletion, charg-
ing, and optimization model setup.
We now discuss the results of the optimization process. Fig-
ure 3(a) shows the average trip time, i.e., the fitness value, for the
solutions generated by the optimizer, as described in Section 5.1.
The solution pool is initialized with P=40randomly generated
solutions. Then, for each of the G=5subsequent generations,
the worst P·f=20solutions are discarded and replaced with
new ones, coming from the combination of the others. The whole
process takes P+G·f·P= 140 possible solutions into ac-
count. The optimal solution (corresponding to an average trip time
of 2438 s, identified by a circle in the plot) is found in the last gen-
eration. Here, three charging stations are active with 9 to 12 plugs,
whereas in the solution with the largest trip time (t= 2981 s) all
the available plugs are activated on one station only. Notice that,
unless we test all possible solutions, we cannot exclude that by let-
ting the algorithm run longer, i.e., for more generations, we could
find a better one.
To show how the impact varies according to different placement
and configuration of charging stations, we compare two metrics for
the best and the worst solutions. First, Figure 3(b) shows the cu-
mulative density functions (CDFs) of the observed trip times for
the solutions achieving best and worst fitness. While the same frac-
tion of trips have a duration shorter than 10 min (i.e., recharging
might not be necessary), the solution minimizing the trip time rises
more steeply between about 10 and 50 min. Further, 3% less vehi-
cles have here a trip time larger than 120 min. Second, Figure 3(c)
shows the number of vehicles in queues for the same example solu-
tions over the simulation time. The queue length values show that
the capacity limit of 30 plugs even causes queuing times in the op-
timal solution. However, the charging demands of our simulation
setting can be handled in this case on a more acceptable level (26
vehicles are waiting on average), while the worst solution exhibits
an increasing congestion (198 vehicles are waiting after 120 min).
7. CONCLUSION
We reviewed some critical issues in electric vehicular networks
due to battery depletion and charging. Hereby, we identified mobil-
ity modeling and charging station deployment as two of the most
critical tasks for the widespread adoption of electric vehicles. We
assumed the likely situation, that electric vehicle drivers resort to
information about charging stations and, in future, to cooperative
5
navigation (based on message exchange) to decrease waiting- and
trip times.
We characterized the frequency and duration of the battery charg-
ing process and proposed a charging model. Further, we introduced
a mobility model that includes a state change between general ve-
hicular mobility and navigation to a charging station on demand.
To show the feasibility of our approach, we integrated our models
into a widely-used traffic simulator. Finally, a genetic program-
ming approach has been employed, which finds a virtually-optimal
charging station deployment so as to achieve the minimum possible
trip times.
Acknowledgment
The work described in this paper was partially supported by the
EC’s FP7 Network of Excellence Euro-NF SJRP e-VENETs. Fur-
ther, this research was supported by a Marie Curie IEF within the
EC’s FP7 (contract PIEF-GA-2010-276336 MOVE-R).
8. REFERENCES
[1] C. Ahn, C.-T. Li, and H. Peng. Optimal Decentralized
Charging Control Algorithm for Electrified Vehicles
Connected to Smart Grid. Journal of Power Sources,
196(23):10369 – 10379, 2011.
[2] O. Ardakanian, C. Rosenberg, and S. Keshav. Real-time
Distributed Congestion Control for Electrical Vehicle
Charging. In ACM SIGMETRICS Greenmetrics Workshop,
2012.
[3] M. Erol-Kantarci, J. Sarker, and H. Mouftah.
Communication-based Plug-In Hybrid Electrical Vehicle
Load Management in the Smart Grid. In IEEE Symposium on
Computers and Communications, 2011.
[4] Z. Fan. Distributed Charging of PHEVs in a Smart Grid. In
IEEE International Conference on Smart Grid
Communications (SmartGridComm), 2011.
[5] J. Ferreira, P. Pereira, P. Filipe, and J. Afonso. Recommender
System for Drivers of Electric Vehicles. In International
Conference on Electronics Computer Technology (ICECT),
2011.
[6] M. Fiore. Vehicular Mobility and Network Simulation.
Handbook on Vehicular Networks, S. Olariu and M. C.
Weigle Eds., Taylor and Francis, 2008.
[7] I. Frade, A. Ribeiro, G. Goncalves, and A. Antunes. Optimal
Location of Charging Stations for Electric Vehicles in a
Neighborhood in Lisbon, Portugal. Transportation Research
Record, 2252:91–98, 2011.
[8] V. Gungor, D. Sahin, T. Kocak, S. Ergut, C. Buccella,
C. Cecati, and G. Hancke. Smart Grid Technologies:
Communication Technologies and Standards. IEEE
Transactions on Industrial Informatics, 7(4):529–539, 2011.
[9] J. Härri, F. Filali, and C. Bonnet. Mobility Models for
Vehicular Ad Hoc Networks: A Survey and Taxonomy. IEEE
Communications Surveys & Tutorials, 11(4):19–41, 2009.
[10] H. Hartenstein and K. Laberteaux, editors. VANET –
Vehicular Applications and Inter-Networking Technologies.
John Wiley and Sons, 2010.
[11] S. Kamboj, W. Kempton, and K. S. Decker. Deploying
Power Grid-integrated Electric Vehicles as a Multi-agent
System. In International Conference on Autonomous Agents
and Multiagent Systems (AAMAS), 2011.
[12] H. Kameda and N. Mukai. Optimization of Charging Station
Placement by Using Taxi Probe Data for On-Demand
Electrical Bus System. In International Conference on
Knowledge-Based and Intelligent Information and
Engineering Systems (KES), 2011.
[13] J. Koza and R. Poli. Genetic Programming. In E. K. Burke
and G. Kendall, editors, Search Methodologies, pages
127–164. Springer US, 2005.
[14] S. Krauß. Microscopic Modeling of Traffic Flow:
Investigation of Collision Free Vehicle Dynamics. PhD
thesis, University of Cologne, 1998.
http://e-archive.informatik.uni-koeln.de/319/.
[15] Z. Li, Z. Sahinoglu, Z. Tao, and K. H. Teo. Electric Vehicles
Network with Nomadic Portable Charging Stations. In IEEE
Vehicular Technology Conference Fall (VTC-Fall), 2010.
[16] F. J. Martinez, C. K. Toh, J.-C. Cano, C. T. Calafate, and
P. Manzoni. A Survey and Comparative Study of Simulators
for Vehicular Ad Hoc Networks. Wireless Communications
and Mobile Computing, 11:813–828, 2011.
[17] K. Nagel and M. Schreckenberg. A Cellular Automaton
Model for Freeway Traffic. Journal de Physic,
I(2):2221–2229, 1992.
[18] F. Pan, R. Bent, A. Berscheid, and D. Izraelevitz. Locating
PHEV Exchange Stations in V2G. In IEEE International
Conference on Smart Grid Communications
(SmartGridComm), 2010.
[19] H. Qin and W. Zhang. Charging Scheduling with Minimal
Waiting in a Network of Electric Vehicles and Charging
Stations. In ACM International Workshop on Vehicular
Inter-networking (VANET’11), 2011.
[20] A. Senart, S. Kurth, and G. Le Roux. Assessment Framework
of Plug-in Electric Vehicles Strategies. In IEEE International
Conference on Smart Grid Communications
(SmartGridComm),2010.
[21] T. Sweda and D. Klabjan. An Agent-based Decision Support
System for Electric Vehicle Charging Infrastructure
Deployment. In IEEE Vehicle Power and Propulsion
Conference (VPPC), 2011.
[22] M. Trieber, A. Hennecke, and D. Helbing. Congested Traffic
States in Empirical Observations and Microscopic
Simulations. Physical Review E, 62(2):1805–1824, 2000.
[23] L. Zhao, P. Awater, A. Schafer, C. Breuer, and A. Moser.
Scenario-based Evaluation on the Impacts of Electric Vehicle
on the Municipal Energy Supply Systems. In IEEE Power
and Energy Society General Meeting, 2011.
6