This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
1

Smart Charging Strategy for Electric Vehicle

Charging Stations
Zeinab Moghaddam, Student Member, IEEE, Iftekhar Ahmad, Member, IEEE,Daryoushr Habibi, Member, IEEE,
and Quoc Viet Phung, Member, IEEE

NOMENCLATURE
Abstract—While the concept of transportation electrification
holds enormous prospects in addressing the global environmental t Current time.
pollution problem, in reality the market penetration of plug-in M Total number of PEVs.
electric vehicles (PEVs) has been very low. Consumer concerns N Total number of charging stations.
over the limited availability of charging facilities and unaccept-
k Type of charge at charging option (k = 1: swap battery,
ably long charging periods are major factors behind this low k = 2 : DC fast charging and k = 3: AC level 2 charging).
penetration rate. From the perspective of the electricity grid,
a longer PEV peak-load period can potentially overlap with
T drive (s, j) Driving time from current location of P EVi to the charg-
the residential peak-load period, making energy management
ing station j.
more challenging. A suitably designed charging strategy can help
to address these concerns, which motivated us to conduct this T drive (j, d) Driving time from the charging station j to the destination
research. In this paper, we present a smart charging strategy for a d.
PEV network that offers multiple charging options, including AC Tjwait (k) Waiting time for charging option k at the charging station
level 2 charging, DC fast charging and battery swapping facilities j.
at charging stations. For a PEV requiring charging facilities, Dmax (t) Maximum driving distance with the remaining energy.
we model the issue of finding the optimal charging station as a E ca Battery capacity of PEV.
multi-objective optimization problem where the goal is to find a Ejca Maximum power capacity of each charging station j.
station that ensures the minimum charging time, travel time and
charging cost. We extend the model to a meta-heuristic solution in Egmax Maximum capacity of grid g.
the form of an ant colony optimization. Simulation results show E T ravel Energy consumption per unit driving distance for each
that the proposed solution significantly reduces waiting time and PEV.
charging cost. SoC(t) State of the charge of the battery for a PEV.
λj Arrival rate at the charging station j.
Index Terms—Plug-in electric vehicles, charging stations,
charging strategies. µkj Service time at the charging station j for charging option
k.
Pi (t) Power consumption of PEVs.
I. I NTRODUCTION Lkj (t) Queue length at time t for charging option k at charging
station j.

P lug-in electric vehicles (PEVs) powered by electricity

from low carbon emission grids can provide significant
benefits in terms of reducing the climate impact from trans-
πjR

Wjk
Probability transition matrix for R possible state at charg-
ing station j.
Waiting time at charging station j for charging option k.
portation and minimizing transport grid’s reliance on oil-based di Driving destination for PEV.
fuels. PEVs provide a cleaner and quieter environment, and v(s, j) Average speed between source to a charging station.
reduce operating costs at the same time [1]. Due to their usage P rk Price for charging option. k
pattern in urban areas, PEVs can potentially operate as flexible Cjrecharge (k) Recharging cost for charging option k at charging station
electric loads to support the operation of power systems and j.
the integration of renewable energy sources [2]. The vision of Tj
max− wait
(k) Maximum waiting time for charging option k at charging
using parked PEVs as storage devices for renewable energy station j.
has also attracted increasing interest in recent years [3], [4],
[5]. Although PEVs are considered to be an important part
of the next generation smart grid system [6], their market loads [8]. Currently, charging infrastructures are not widely
penetration is still relatively low and faces a number of available in all major cities, and because of long waiting
challenges. Firstly, drivers’ range anxiety is a key issue which times at charging stations, the recharging process can cause
must be managed by organizing better communication with significant delays [9], [10].
the smart grid interface, to facilitate timely and fast recharging The relevant industries consider this delay to be a major
at public charging stations. Secondly, uncoordinated charging challenge and are exploring all available options to reduce
strategies in a limited charging infrastructure can increase the the waiting time at public charging stations [11], commonly
average recharge time [7] and contribute to an increase in peak known as PEV networks. A smart charging strategy can make
a major contribution to the efficient management of available
The authors are with the School of Engineering, Edith Cowan
University, 270 Joondalup Drive, Joondalup, WA, 6027, Australia. E- resources in PEV networks. Recognizing the significance of
mail:[email protected], {i.ahmad,d.habibi,q.phung}@ecu.edu.au smart charging in the context of PEV networks, researchers

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
2

have investigated and presented various smart charging strate-

7 3600
gies in recent years, targeting the reduction of range anxiety Traffic Counts
and charging time [12]. However, a gap still exists in the 6
Residential Load 3400
literature since none of the existing works considers multiple 3200

Traffic Counts (%)

charging and dynamic pricing options at the charging stations 5
3000
in their PEV networks [13], [14]. In practice, a charging

Load (MW)
station, like a traditional gas station (selling petrol, diesel, 4 2800
LPG) can have multiple charging options with dynamic price 2600
information. A smart grid can collect important information 3
about the current status (e.g., available number of sockets, 2400
2
queue status, price etc.) of every charging station in a PEV 2200
network. The grid can then provide this information in real-
1 2000
time to the individual PEV user [15], [16]. This information 12 14 16 18 20 22
can be taken into account to calculate a path to the destination Time of Day
which would reduce the time and cost of charging. This
Fig. 1. Traffic counts and residential loads in NSW, Australia
provided the motivation for this research.
In this paper, we present a smart charging strategy for a PEV
network that supports multiple charging options at its charging
stations. We model multiple charging options as a multi-server residential load profile in an urban area in New South Wales
queuing system in order to estimate the waiting time for each (NSW), Australia [17], [18]. It is evident that the period
charging option at a charging station. The queuing system between 2 p.m. and 6 p.m. is the busiest, when most of the
is presented in detail in Section III.A. We then model the vehicles are on the streets. The load profile for a PEV charging
optimal charging station-finding problem as a multi-objective network is expected to follow a similar trend, as most of the
optimization problem where the objective is to minimize the PEVs are also expected to be on the roads during this period
travel time, waiting time and charging cost. The optimization and would need to recharge. This is analogous to traditional
problem and related constraints are presented in Section III.B. gas stations, where the load is generally high in the late
Considering the industry demand for a robust solution, we afternoon. Figure 1 also demonstrates that there is an overlap
extend and model the problem as a meta-heuristic optimization between the daily residential peak load and the PEV load. This
problem. A detailed description of the meta-heuristic solution overlap would be a major challenge for the power industry
is presented in Section IV. The main contributions of this work since an extended period of overlap can put significant stress
can be summarized as follows: on both the generation and distribution sides of the energy
industry. An uncoordinated charging strategy can result in a
• We introduce a smart charging strategy that considers
long queue at hotspot areas during busy period, which would
multiple charging options and relevant price information
increase the charging time (i.e., waiting time at a station plus
at each charging station in a PEV network. We model
time to recharge) and the overlap period between the PEV
the research question as a multi-objective optimization
and the residential load. This challenge can be mitigated by
problem, and reflecting the need for a real-time solution,
introducing a smart charging strategy where the smart grid can
we also present a meta-heuristic solution.
collect real-time information about loads at various charging
• We show that dynamic price variation at charging stations
stations and pass this information to individual PEVs, enabling
can be a useful mechanism to control the average charg-
PEVs to come up with their charging plans (i.e., the optimum
ing time, which ultimately can prove pivotal in reducing
charging station along the route to the destination). Most of the
the overlap extension between the PEV and residential
research [19], [20], [21], [22] on PEV charging strategies has
peak load periods.
focused on controlling residential charging patterns to avoid
• We verify the significance of the proposed solution by
the potential overloads, stresses, voltage deviations and power
using a computer simulation on a Washington City PEV
losses that may occur in distribution systems from domestic
network.
PEV charging activities. Some researchers [23], [24] have also
The remainder of this paper is organized as follows: Sec- investigated the simultaneous utilization of distributed renew-
tion II introduces previous work on PEV charging in urban able resources and PEVs to improve the performance of smart
areas. Section III describes a smart charging strategy for PEVs power distribution networks. Most recently, researchers have
and presents a system model and the proposed method for started to investigate charging strategies for public charging
solving the problem of the minimization of total travel time stations.
and recharging cost. Section IV presents the simulation results
and discusses the introduced case study. Finally, SectionVI In [25], Amini et al. proposed a framework for interde-
concludes with a summary of the study. pendent power and electrified transportation networks that
utilized the communication of PEVs with competing charging
stations to exchange information such as electricity price,
II. R ELATED W ORK energy demand and time of arrival. While the framework
Figure 1 shows the average afternoon traffic distribution solves an important problem in the area of optimal power
as a percentage of the total vehicle counts, and a typical flow and vehicle routing, Amini et al. do not consider multiple

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
3

B D AC Level 2 :
queue

DC Fast Charging EVs depart the

EVs arrival :
queue station
A F

Battery Swapping
queue :

Charging Sockets
C E

Fig. 3. The M/M/s/C queuing model for a charging station

Fig. 2. A graph representation of a PEV network

infrastructure. Using the coordination between the charging

charging options and associated queuing models to address the stations and PEVs, their approach focused on minimizing
problem of longer waiting times at charging stations. In [26], the waiting time at charging stations and the overall travel
Shun-Neng et al. investigated the PEV charging problem and time. Considering the complexity of the optimization problem,
proposed two types of charging station selection algorithms: they adopted a meta-heuristic (A* search) approach to find
the first solution utilized only local information (e.g., SoC, a solution. However, this work did not consider multiple
geographical position etc.) relating a PEV and the second charging options and price variation among charging stations
solution utilized the global information obtained through the in their model.
interaction between the PEVs and the charging station server As discussed above, while researchers have introduced a
using a mobile telecommunications network. Their work number of smart charging strategies for PEVs, none of the
showed that the performance of the charging algorithm that existing work presents an integrated solution that consid-
used global information was better than the performance of ers multiple charging options, waiting time, travel time and
the charging algorithm that used local information. However, recharging cost. In the present study, we introduce a smart
their work focused on waiting time and did not consider the charging strategy, focusing on total travel time reduction, with
multiple charging option, cost and travel time. different charging options at charging stations and also taking
into account various prices for each charging option.
In [27], Pourazarm et al. solved a path-finding problem
within a graph of charging station nodes using a dynamic
III. S MART C HARGING S TRATEGY
programming solution. They applied a grouping technique
based on traffic flows with multi-vehicle routing to achieve A. System Model
the shortest path. This work, however, did not consider the A PEV network can be considered as a weighted undirected
waiting time and recharging cost at charging stations. Similar graph G = (V, E), where V is the set of charging stations
to the work presented in [27], Sweda et al. [28] introduced which is denoted as j : (j = 1...N ) and E is the set of
a recharging plan for PEVs to find a charging station with connecting paths between the nodes as shown in Fig. 2. Each
the shortest path. Their model was designed for an urban charging station provides three charging options: i) DC fast
environment where the number of routes can be very large charging, ii) AC level 2 charging and iii) battery swapping
and the number of charging stations is limited. They proposed facilities [32]. Each charging option has a queue, and each
a pre-processing approach to save computations in an urban queue has a specific service rate, waiting time and price. The
environment. Their work, however, focused on finding the queue length is influenced by factors such as PEV arrival rates,
shortest path based on the minimum travel time only. In [29], and service times (i.e., time to fully recharge). The queue
a distributed charging scheduling protocol was proposed to length is an important parameter as it determines the waiting
minimize waiting times in the charging stations. The authors time before the actual service is offered. The PEV arrival
used a theoretical approximation model which was based rate is also partially influenced by price information. In our
on the arrival rate of PEVs at each charging station and system model, each PEV indexed by i at a time instant t can
achieved a high performance in terms of waiting time. How- be attributed by its current state of charge SoCi (t), current
ever, they did not consider a multi-server queuing system for location Si and intended destination di . All charging stations
different charging options. In addition, they did not consider and PEVs are connected to the smart grid and can exchange
the minimization of total travel time and recharging cost in information in real-time. If a PEV plans to go from a source
their objective function. In [30], Gusrialdi et al. proposed to a destination node and its current SoC suggests that it will
an optimized charging strategy using a stochastic model for not have enough stored energy, it will have to recharge at a
controlling the PEV arrival rate at charging stations in order charging station. A PEV driver may also prefer to charge at
to minimize the demand flow and the waiting time. However, a faster rate from a fast charging station instead of using the
they did not consider the impact of variable price and multiple slow charging facility at his/her accommodation. The research
charging options at a charging station. question then translates into finding the optimum charging
Razo et al. in [31] proposed a smart charging approach station that does not significantly increase the travel time to
to plan charging stops on highways with limited charging the destination and offers the best price for recharging.

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
4

• s: the number of servers.

• C: the system capacity ( the number of PEVs that can
be parked at the station).
Every PEV that arrives at a station can immediately recharge
if there is an available socket in the charging station. If all
Fig. 4. The state transition diagram of the M/M/s/C queuing system the sockets are busy, the PEV needs to wait in a queue until
a socket becomes available. Each charging station has three
queues, one for each charging option. PEVs in each queue are
For a PEV, the total travel time depends on the time required served by s servers. Each server operates with an exponential
to reach the charging station, the time spent waiting in the service rate µkj facilitated by a socket. The PEV arrival process
queue for the preferred charging option, service/charging time at a charging station follows a Poisson distribution with a mean
and time required to reach the destination. As such, in the arrival rate λj . The service rates are constrained by the infras-
following sub-sections we show how to model and calculate tructure (i.e., sockets and charging rates) whereas the mean
the total travel time takings the queuing delay and charging PEV arrival rate during various hours of the day can be derived
cost for each charging option into account. from historical data. A smart charging strategy that influences
1) Driving Time from Source to Destination: For a PEV the route selection decision can have an impact on arrival rates
at a time instant t, if T drive (s, d) indicates the driving time at a charging station. However, the traffic/load analysis based
from its current location s to the destination d and j indicates on the data collected over a period of time is expected to
the charging station along the way, then T drive (s, d) can be capture this impact. Based on the principle of Markov chains
obtained by: [34], the state transition diagram for the M/M/s/C queuing
process can be derived and depicted as shown in Fig. 4, where
D(s, j) D(j, d) each state of the chain corresponds to the number of PEVs in
T drive (s, d) = + (1)
v(s, j) v(j, d) the queue. When new PEVs arrive or a PEV recharges and
where the first term of the above equation is the time departs the station, the queing process moves to a different
required for the PEV to travel from its current location to the state. This state transition is essentially a stochastic process
charging station and the second term states the travel time from with Xr being the random variable that represents the value
the charging station to the specific destination. Here, D(s, j) of the chain at step r. This stochastic process with state space
and D(j, d) indicate the route distance from the source to the ξ = {1, 2, 3, ...} exhibits the property of a Markov chain
charging station and the charging station to the destination, because of the P (Xr+1 = j|Xr = i, Xr−1 = xr−1 , ..., X0 =
respectively. v(s, j) and v(j, d) indicate the average moving x0 ) = P (Xr+1 = j|Xr = i) attribute.
speed of the PEV from its source to the charging station and The state transition probability matrix P for the above
the charging station to the destination, respectively. To make mentioned stochastic process can be given as Eq. 3 [35], [36]:
sure that the PEV does not run out of charge before reaching
the charging station, D(s, j) must be less than Dmax , the 
−λ λ 0 ... 0 0

maximum distance that the PEV can travel based on its current  µ −(λ + µ) λ ... 0  0
SoC. The maximum distance that the PEV can travel based 
 0 2µ −(λ + 2µ) . . . 0

 0
on its SoC at a current time t, can be given by: P = 

 ... 

E ca SoC(t)
 0 0 ... λ 0 
Dmax (t) = (2) 0 0 ... −(λ + cµ) λ
E T ravel (3)
where E ca is the total capacity of batteries in PEV, and Definition 1:
E T ravel represents its energy consumption per unit of trav- In the probability matrix P, P T .π T = 0T , where π is the
eling distance.In addition to the net driving time as indicated row vector that contains the stationary distributions.Assuming
by Eq. (1), the PEV would have to wait in the queue at a the occupancy rate with ρ = λ/cµ, we can obtain the rth
charging station before it obtains access to the desired facility. stationary distribution:
The waiting time in the queue at a charging station can be
estimated using a queuing system which is presented in the λr (cρ)r
πr = π0 = π0 (4)
next section. n!µr r!
2) Queuing Model for a Charging Station: Figure 3 shows
the queuing system used for charging stations with multiple and reaching the system capacity C, we obtain the capacity
charging options in our model. To estimate the waiting time for each station which is shown by c:
for each charging option at a station, we use the M/M/s/C
model [33], [7], where the letters have the following meaning: (cρ)c
πc+r = ρr πc = ρr π0, (5)
• the first M (Markov): Markovian (exponential) PEVs
c!
r=0,1,2, ..., c+r=C
arrival time distribution. P∞
• the second M (Markov): Markovian (exponential) charg- Now if we consider p=0 πp = 1, we can obtain the π0
ing time distribution. stationary distribution:

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
5

peak-hours a preference for full charging would lead to longer

1 waiting times. Longer waiting times can stretch the PEV peak
π0 =  i  i (6) period and make it overlap with the residential peak period.
Pc 1 λ
PR 1 1 λ
i−0 ( i! µ + j=c+1 C j−c )( c! ) µ Partial charging (i.e., σ ¡ 0.9) can encourage users to postpone
Using this formula, we can also obtain the mean queue their full charging process and depart the queue early. Partial
lengths for each queue: charging can be influenced by a suitable dynamic pricing
model, and together, these measures can provide an effective
R
X solution to the problem of longer waiting times during peak
E(Lq ) = (p − c).πp (7) hours.
k=c

where πp is the pth stationary distribution. After substituting

B. Proposed Model
the proper values and simplification of the equation, queue
length can be given as: The objective of the proposed charging strategy is to find a
charging station along the path such that the total travel time
ρ(cρ)c  - including driving time from current location to destination,
1 − ρR−c − (R − c)ρR−c (1 − ρ)

E(Lq ) = π0 . waiting and charging time at a charging station - and the charg-
c!(1 − ρ)2
(8) ing cost are minimized. Mathematically, considering equations
Using the Little law at each charging station for three types (1), (9) and (13), the objective of our charging strategy can be
of charging options, the mean waiting time at each queue can modeled as: for ∀i ∈ I find a charging station j that minimizes
be given as[37], [38], [39]: total travel time and recharging cost, as below:

∀k:(k=1, 2, 3)
E(Lkj ) min[x(T drive (s, d)) + yTjwait (k) + zCjrecharge (k)] (15)
E(Wjk ) = (9)
λj (1 − πjR )
where x, y and z are the positive coefficients of the objective
Therefore, for the queue corresponding to the battery swap- function.
ping facility at a charging station j, the mean waiting time is s.t.
calculated as follows:

µ1j T drive (s, d) ≤ ((E ca SoC(t))/E T ravel ) (16)

E(Wj1 ) = (10)
λj (µ1j − λj )(1 − πjR )
for the DC-fast charging queue, the mean waiting time is max− wait
Tjwait (k) <= Tj (k) (17)
calculated by:

µ2j
E(Wj2 ) = (11) SoCmin ≤ SoC(t) (18)
λj (µ2j − λj )(1 − πjR )
M
and for the AC-normal charging queue, the mean waiting X
time is calculated by: ∀j ∈ J E ca ≤ Ejca , (19)
i=1
µ3j
E(Wj3 ) = (12) Ejca < Eca,
max
(20)
λj (µ3j − λj )(1 − πjR ) grid

3) Charging Cost for a PEV with a Specific Charging Op- In the optimization problem, the summation of driving time
tion at the Charging Station: The charging cost of each PEV from source to destination, waiting time at the station for a
depends on its associated charging option and the associated specific queue, and charging cost for each P EVi at a charging
charging rate at the station j. For a PEV, the charging cost is station j should be minimized. Constraint (16) shows the
calculated through the formula below: constraint for the driving time from the current source to a
charging station, which is explained in Eq. (2). In constraint
Cjrecharge (k) = ((dcharging P rk ) j ∈ J (13) (17) the maximum waiting time for each queue at charging
stations is presented. This should be less than the maximum
where dcharging is the charging demand of each PEV, which
waiting time at each charging station for different charging
depends on the remaining and target SoC for each PEV:
options. Constraint (18) indicates that initial amount of SoC
for a PEV at charging station j should be greater than SoCmin .
dcharging = σE − SoC (14)
An additional constraint is defined in Eq. (19), which states
Here, σ is the coefficient for partial charging, which can that the summation of the charging power capacity for all
vary from the current SoC to the maximum SoC (e.g., 90% PEVs at a charging station should be less than the maximum
battery capacity). Under normal circumstances, most users capacity of that charging station, and constraint (20) considers
would prefer full charging (i.e.,σ = 0.9). However, during the maximum grid capacity for each charging station.

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
6

IV. S MART C HARGING S TRATEGY U SING function, the routing algorithm should consider the lowest
M ETAHEURISTIC A LGORITHM travel time, the lowest recharge cost and the lowest waiting
In the previous section, we have shown that the proposed time at each station, as the heuristic information.
smart charging strategy can be formulated as a classical ACO is not only sensitive to the number of variables but
optimization problem. However, the optimization solution is also runs faster, which is an important ability in relation to
NP-hard due to the path discovery mechanisms [37]. This solve dynamic vehicle routing and refuelling problems. ACO
motivated us to further investigate the research problem and is a probabilistic search algorithm which has specific charac-
find a heuristic solution. teristics, considering the problems in terms of the heuristic
information in the probabilistic decision, and the strategy of
updating the pheromone trail in the path is based on the search
A. Ant Colony Optimization (ACO) objectives [39], [43].
ACO is a category of swarm intelligence that analyzes the
survival behavior pattern of the insects in solving complex
B. Proposed Smart Charging Strategy Using Ant Colony Op-
optimization problems [38]. ACO is a widely recognized meta-
timization
heuristic approach and has been successfully used across
diverse domains, including vehicle routing and scheduling The main reason for choosing ACO in our proposed sce-
[40], [41], [39], [42]. The main characteristic of ACO is that nario is that using the heuristic information in the ACO
every single ant in a colony can construct a possible solution algorithm our model can estimates the minimum travel time
by considering both heuristic and stochastic information and and charging cost at every node of the graph to accelerate the
exchanging that information with the ants and the environment computation. With the arrival of a new PEV every neighbor
[38]. The collective intelligence gathered from a group of ants exploration allows us to test for the defined constraints and
traveling along alternate routes leads to the identification of an exclude non-feasible alternatives at the first stage. However,
optimal route [28], [43]. As a description of the ACO algo- the conventional ACO algorithm keeps track of the variable
rithm, consider a colony with m ants that iteratively exploits it aims to minimize along the entire search. In the proposed
the graph and searches for feasible solutions to the problem. model, the smart-strategy in Algorithm. 1 is iteratively called
At each iteration u, an ant k moves stochastically, based on up to provide up-to-date information to the PEV driver. As
a constructive decision policy which uses the information of information is shared between the smart grid and a PEV driver,
pheromone trails and attractiveness to obtain the probability the following processes occur:
for choosing the next node, as below: • Loading initial information: the modified ACO algorithm
receives as input a graph describing all possible paths,
τij (u)α η the current SoC of the PEVs battery, the maximum
Pijk (u) = P αη
ifj ∈ Nik (21)
l∈N k τ il (u) energy consumption of the PEV, the coordination of the
i
source and destination, the colony number, the number
Here Nik is the set of feasible neighborhood nodes in the
of ants in each colony, the initial pheromone level and
graph. For a node i, a set of feasible neighborhood nodes
the coefficient of the creation and evaporation of the
indicates that the list of nodes that are directly accessible from
pheromone.
node i and candidates for searching based on the heuristic
• Selecting specific heuristic information to obtain the
information. Nodes that are in the feasible neighborhood set
probability distribution function: In a conventional ACO
offers a better chance of finding the optimum solution [41],
algorithm, for calculating the probability distribution in
[44], τij (u) is the sum of pheromones deposited between
Eq. 21 only the pheromone and one heuristic value
nodes i and j which denotes the desirability of the move
(distance) are considered. In the smart charging strategy,
between the nodes; ηij is heuristic information which specifies
since PEVs look for an optimum path in terms of mini-
the attractiveness of that move, and α, β are parameters
mum travel time and charging cost, we introduced a mod-
which control the relative weight of pheromone and heuristic
ification where we maintained the driving time, waiting
information. During the completion of a tour, each ant deposits
time and charging cost as a set of heuristic information.
the pheromone information on the respective edges of its path.
Considering the three objectives in the aforementioned
There is a rule for updating pheromone information, as below:
objective function, the ACO algorithm for each j ∈ Nik
m
X can be implemented with the Eq. 23:
k
τij (u + 1) = ρτij (u) + ∆τij (u) ∀(i, j) (22)
k=1
τij (u)α [(T drive (s, d))Tjwait (k)Cjrecharge (k)]β
where 0 ≤ ρ ≤ 1 is the pheromone evaporation rate, Pijk (u) = P
which causes the pheromone value to decrease over time to l∈Nik τij (u)α [(T drive (s, d))Tjwait (k)Cjrecharge (k)]β
k
prevent a local optimum, and ∆τij (u) refers to the inputs (23)
of ants between nodes i, j. Consequently, each ant moves
through those nodes similar to its partial solution. Although where the variables τij (u), T drive (s, d)), Tjwait (k) and
recharge
the convergence properties of the ACO algorithm have been C (j, k) are the pheromone intensity, driving time,
proven, the probabilistic decisions depend on the problem- waiting time and charging cost of a path. The parameters α
definition and user preferences [38]. For the proposed objective and β are constants, which determine the relative influences of

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
7

Algorithm 1 ACO algorithm for the smart charging strategy

Require: Graph Of Charging Stations, Charging Station Specifica-
tions, PEVs Specifications, SoC , Destination Coordination;
Ensure: optimum path, P
1: SetAdjustableParameters (G, P EV , t, α, ρ, τ , β, SoC, η, CS);
2: Initialize (t,counter,iteration);
3: // t: time slot;
4: Set t ← 0;
5: while maximum iteration is not met do
6: t ← t + 1;
7: Initialize τglobal ;
8: for each node j of graph G do
9: if j is in graph G’s selected path then
10: Call function of driving time;
11: Call function of waiting time;
12: Call function of charging cost;
13: Pg ← 0;
14: else
15: for each ant n do
16: Initialize τlocal ;
17: for P ← 1 to G do
18: while constraints are not met do
19: g ←ηP g (t);
20: xnP g ← link(P g);
21: end while
22: Update τlocal ;
23: end for
24: end for
25: Sort (RouteSolution(N )) ascendingly;
p
26: Ibest (t)= RouteSolution(1);
27: Update τglobal ;
28: end if
29: end for
30: end while
31: Update Information (P athInf o, P);
Fig. 5. Flowchart of the proposed smart charging strategy

obtained for calculating the probability function of the

the pheromone and heuristic parameters on the PEV’s decision. modified ACO. As illustrated in Fig. 5, the first step
For the probability distribution, there is a trade-off between the is exchanging information between PEVs and charging
three objectives. The process of finding an optimum path is stations using the smart grid panel in real-time. For
described in the following: planning and managing the charging strategy, the smart
grid needs PEV components and the charging station’s
• Pheromone initialization: Collect the possible solution at
specifications. The next step uses our proposed algorithm
each iteration and update the pheromone values using the
to implement a charging strategy for PEVs along their
general Eq. (22), τij (u + 1) symbolizes the pheromones
trip. During this process, the algorithm needs to consider
of a vehicle moving from the current location s to a
the capacity of each charging station and the lengths of
destination d or stopping at node j for charging during
the queues which are updated for all the iterations in
time period (u + 1) with the objective of minimizing
the simulation. The analytical and numerical results are
travel time. Here, the updated value of the pheromone
explained in Section V.
is a function of time and is controled by the number of
• Local pheromone update: While constructing the path
PEVs on each route.
decision solution, a local pheromone update is executed
• Generate the best possible path: This step uses the mod-
so that the visited path becomes less attractive, allowing
ified ACO algorithm which is described in algorithm 1
the next PEV to explore other paths. This local update
for the PEV network which receives as input a graph de-
can be defined as:
scribing all possible paths including charging stations, the
source and destination nodes, and the SoC of the PEVs.
Each time a neighbouring node is explored, the function l
τij (u + 1) = (1 − ρ)τ (u) + τij ∀(i, j) ∈ iterl (24)
Feasible nodes is responsible for selecting feasible nodes
from the graph with a calculation of the distance from where based on the ACO algorithm, for the local best solu-
each current location to all charging stations, then using tion at each local iteration, only the PEV which constructed the
the output of this function (which is a set of feasible best solution is allowed to deposit the pheromone information.
l
nodes, with function get nodes and using sub functions τij is the incremental amount of the local updating phase for
w-time and cost-to-charge) the heuristic values can be the local iteration.

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
8

• Evaluation of the tour: When the PEV arrives at the desti-

nation, for each PEV tour the optimization function value
(total travel time), denoted as drive-time in Algorithm 1 is
calculated. The PEV tour that utilized the optimum path
with the minimum travel time and charging cost among
all the PEV tours that are found in the previous iterations
is selected as the best tour.
• Global pheromone update: After a few iterations, each
node is then able to estimate the potential path, regardless
of the varying conditions of the model topology and
the PEV traffic. In fact, the global pheromone update
rule is only employed by a PEV that has constructed
the best global solution so far and this update gives the
PEVs more opportunities to explore the search space, thus
balancing the need for exploration and PEV exploitation.
The global pheromone update at each global best solution
can be defined as:
g
τij (u + 1) = (1 − ρ)τ (u) + τij ∀(i, j) ∈ iterg (25)
g
where τij is the incremental amount of global updating
phase for the global iteration.

V. S IMULATION AND R ESULTS

A. Simulation Model
In the simulation, we implemented a PEV network for
the Washington City road network along the driving route
from Oregon to Vancouver, as shown in Fig. 6. The PEV
network consisted of 28 charging stations, each equipped with
two charging options (CHAdeMO Fast charger and AC level
2 charger) and a battery swap facility. We considered up
to 1000 PEVs comprising Nissan Leaf (30kWh), BMW i3
(22kWh) and Smart Ed (16 kWh), with their SoC modeled
as a uniform distribution in the 10%˜90% range.The PEV
arrival at each charging station was modeled as a Poisson
distribution. The number of sockets for a charging option at
Fig. 6. The Washington green highway PEV network[45]
a station was distributed uniformly in the 1˜10 range. For the
fast DC charging option, each socket was assumed to supply
20 kW power [46], [47], [48]. For the AC level II option, We compared the proposed solution with the active scheduling,
each socket supplied 7 kW power [49] and the average time known as the AS model in the literature. In the AS model [31],
to swap a battery was considered as 3 minutes [50]). PEVs the smart charging strategy is modeled as a meta-heuristic
were assumed to travel at a maximum speed of 60 miles/hour, optimization problem (A∗ search algorithm) where the goal
consuming 0.12kWh/mile [51]. The time of use (i.e., dynamic) was to find charging stations that reduced the travel time.
tariff was modeled using an exponential function [52] with the The AS model, however, does not consider multiple charging
minimum and maximum rate set as $0.24 and $0.46 [53], [54], options (i.e., does not include a queuing model) and the costs
[55], respectively. Price variation at different charging stations associated with recharging at a charging station. To the best of
was implemented using a uniform distribution within the -10% our knowledge, for a PEV network, the AS model is the most
to +10% range of the standard time of use tariff rate. We used relevant and best performing smart charging solution available
MATLAB to perform the simulation. in the literature.
The discussion of the results begins with Fig. 7, which
B. Results and discussion shows the normalized average waiting time for PEVs at
In this study, we analyzed the average waiting time, travel charging stations. We normalized the average waiting time
time, charging cost and charging station queue occupancy. The over the travel time, i.e., the period of time a PEV has to wait
waiting time results from the time spent waiting in the queues at a charging station on average expressed as a fraction of the
plus charging time, whereas the travel time is made up of travel time. As the figure suggests, the average waiting time
driving time plus waiting time. The queue occupancy param- increases with the increasing volume of traffic. This is because
eter indicates the current occupied queue length expressed as for a bigger fleet of PEVs, more PEVs are required to share
a fraction of the maximum queue length at a charging station. the limited charging facilities. The result, however, shows that

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
9

0.75
AS model AS model
70 Proposed model
Proposed model
0.7

Charging Station Queue Occupancy (%)

0.65 60
Waiting Time / Travel Time (s)

0.6
50

0.55
40
0.5

0.45 30

0.4
20

0.35
100 300 500 700 900 1000 12 13 14 15 16 17 18 19 20 21 22
Number of EVs Time of Day (Afternoon hours)

Fig. 7. Normalized waiting time for different number of PEVs Fig. 9. Charging station queue occupancy in hotspot areas

4200
AS model model, the waiting time corresponding to each charging option
Proposed model
4000 at a charging station is calculated based on the queuing model,
which when communicated to the PEV users, can help them
to make a better decision.
Average Travel Time (s)

In Fig. 9, we present the comparison of average queue

3500
occupancy at charging stations in hotspot areas (i.e., nodes
with greater number of edges) for both models. It is apparent
that the AS model results in higher average queue occupancy
because the model does not use queue information specific to
an individual charging type. In our proposed model, PEVs tend
3000 to distribute themselves more uniformly among the available
charging stations, thereby reducing the pressure on charging
stations in hotspot areas.
Just as gas/petrol prices vary from place to place, prices
2500 per charging option can vary between charging stations. In
100 300 500 700 900 1000 the proposed model, we therefore considered the price per
Number of EVs
charging option at various charging stations, which ultimately
Fig. 8. Average travel time for different number of PEVs generated a lower average charging cost. In Fig. 10, we present
the average charging cost during various hours of the day.
It is evident that because the price information is not taken
the waiting time is significantly lower in the proposed solution into account while calculating the route in the AS model, the
compared to the AS model. This is because in our model, we average charging cost in the proposed model is significantly
implemented multiple charging options as a queuing model lower than the average cost in the AS model.
and considered the queuing delay for each charging option Figure 11 illustrates the average waiting time during the
while selecting the best charging station for a PEV, which peak PEV load period in response to dynamic charging prices.
was missing in the AS model. In our model, information in As shown in the figure, the higher price option encourages
relation to the queuing delay for each charging option at every many PEV users to go for either partial charging or to avoid
charging station was communicated to a PEV, which allowed charging during the peak load period, making the average
the PEV to make an informed decision while choosing the waiting time shorter. Since the ultimate decision in relation
best charging station. Figure 8 presents a comparison of the to the choice of time to recharge PEVs belongs to owners, a
average travel time in the proposed and AS models. It is service provider cannot enforce a policy to not to serve/charge
evident that compared to the AS model, the proposed model a PEV, even when it creates a problem for the grid. But
reduces the average travel time. This is because in the AS the behaviour of PEV users can be significantly influenced
model, many PEVs show up at charging stations (e.g., hotspot by setting a suitable price so that both the users and grid
areas) where the queue lengths corresponding to their preferred can benefit. In our model, through the introduction of the
charging options are too long, and PEVs are required to wait recharge cost in the objective function, we capitalized on this
for a longer period before they can be served. In our proposed potential. An extended overlap period between the PEV and

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
10

8.8
AS model 8.5 High price effect
Proposed model Low price effect
8.5

8
8

Recharge Cost ($)

Recharge Cost ($)

7.5
7.5

7
7

6.5 6.5
12 13 14 15 16 17 18 19 20 21 22 12 13 14 15 16 17 18 19 20 21 22
Time of Day Time of Day (Afternoon hours)

Fig. 10. Average charging cost in AS and proposed model Fig. 12. Average charging cost for the proposed model with price variation

0.6 140
Low price effect Proposed model
High price effect AS model
125

0.5
Waiting Time / Travel Time (s)

Load (kW)

100
0.4

75
0.3

0.2 50
100 300 500 700 900 1000 12 13 14 15 16 17 18 19 20 21
Hour of Day
Number of EVs

Fig. 11. Improvement in waiting time with price variation Fig. 13. PEV load with AS model and proposed model

residential peak periods can be avoided by setting a higher coefficient value for a PEV was assigned a value in the range
recharge price in our model, which would then encourage of its current SoC to the maximum SoC using a normal
PEV users to perform partial charging at charging stations distribution. The mean value of this distribution was assumed
during busy hours and/or complete the charging process at to decrease exponentially with increasing charging price. For
home/charging stations during off-peak hours. It should be partial charging, there is a trade-off between the recharge price
noted that an optimum dynamic pricing model, which would and the delay at charging stations. As shown in Fig. 12, the
have the greatest benefit for both the grid and PEV users, average peak hour recharging cost with dynamic pricing is
can be modeled as an optimization problem. This, however, is relatively higher.
not within the scope of this work. We intend to investigate We also investigated the impact of reduced average waiting
and develop an optimum dynamic pricing model for PEV time achieved by the proposed model on the charging load
charging stations in our next work. In this work, in order for the PEV network that we implemented in our simulation.
to show the impact of pricing on partial charging and the As evident in Fig. 13, in the proposed model, the average
average waiting time during the peak load period, we used PEV load per charging station is higher during the busiest
two indicative maximum price levels, (i.e., US$ 0.46 vs 0.56 hour compared to the AS model. This is because the proposed
per kWh as the maximum price). In response to the higher model efficiently uses the available PEV charging facilities by
charging price during the peak load period, the partial charging providing information in relation to the queue length and price

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
11

6 TABLE I
S UMMARIZED S IMULATION R ESULTS

Evaluation parameters AS Proposed Improvement

5 model model
Average waiting time (min) 48.5 36.2 25.3%
Average travel time (min) 68.3 58.2 15%
Charging cost (US$) 8.9 7.5 16%
Objective Function

Table I summarizes the performance improvements of the

3 proposed model over the AS model for a fleet size of 1000
PEVs. The Table clarifies the reduction in the average waiting
time and the average travel time as well as the recharging cost.
2
VI. C ONCLUSION
In this paper, we proposed a smart charging strategy for a
1
0 200 400 600 800 1000 PEV network that offers multiple charging options at charging
Number of Iteration stations. Just as traditional gas stations have different capaci-
ties and pricing options, charging stations can have different
Fig. 14. Convergence graph with ACO for the optimal paths
capacities and pricing options, and the recharge price for each
option can vary from one station to another. In a scenario
150
like this, it is important to adopt a charging strategy that
identifies the most suitable charging station for a PEV user,
so that the user can recharge at the minimum cost and reach
his/her destination without a significant delay. We modeled the
research challenge as a multi-objective optimization problem
125
where the goal was to reduce the charging time, travel time
Solving Time (s)

and charging cost. We used a queuing model to estimate the

delay at various charging stations. To mitigate the challenge
of longer waiting times and the potential overlap between the
peak PEV and residential load periods, we also introduced
100 the concept of partial charging. We showed that pricing could
be used as a useful tool to encourage PEV drivers to choose
the partial charging option during peak load hours. In light of
the significant time complexity of the optimization solution,
we solved the research problem by introducing an ACO-based
75 meta-heuristic solution. The simulation results confirm that the
100 300 500 700 900 1000
Number of EVs proposed solution significantly reduces the average charging
delay (up to 25%) and cost (up to 15%). In our future work,
Fig. 15. Average computation time for PEVs we will investigate the optimum dynamic pricing model to
minimize the overlap between the peak PEV and residential.

for the preferred charging options to PEV users. The reduced

average waiting time achieved by the proposed model also R EFERENCES
causes the PEV load curve to decline at a faster rate, making [1] N. Daina and J. W. Polak, “Hazard based modelling of electric vehicles
energy management for the residential load less challenging charging patterns,” in Transportation Electrification Asia-Pacific (ITEC
for the grid. Asia-Pacific), 2016 IEEE Conference and Expo. IEEE, 2016, pp. 479–
484.
Since the smart charging strategy is intended largely for [2] A. Emadi, “Transportation 2.0,” IEEE Power and Energy Magazine,
PEV users, it is important to have a solution that can be vol. 9, no. 4, pp. 18–29, 2011.
solved in real-time. This motivated us to investigate the [3] U. K. Debnath, I. Ahmad, D. Habibi, and A. Y. Saber, “Improving
battery lifetime of gridable vehicles and system reliability in the smart
time complexity of the proposed ACO-based smart charging grid,” IEEE Systems Journal, vol. 9, no. 3, pp. 989–999, 2015.
solution. Figure. 14 presents the convergence test result for the [4] T. Moore, “Electrification and global sustainability,” EPRI Journal,
proposed solution. This suggests that the ACO-based approach vol. 23, no. 1, pp. 42–50, 1998.
[5] E. Karden, S. Ploumen, B. Fricke, T. Miller, and K. Snyder, “Energy
converges to the optimal solution after 440 iterations for a storage devices for future hybrid electric vehicles,” Journal of Power
fleet of 500 PEVs. In real-time, ACO-based approach requires Sources, vol. 168, no. 1, pp. 2–11, 2007.
around 127s to find the best route for a PEV (Fig. 15). The [6] J. Shen, S. Dusmez, and A. Khaligh, “Optimization of sizing and battery
cycle life in battery/ultracapacitor hybrid energy storage systems for
computation time for each PEV remains relatively constant, electric vehicle applications,” IEEE Transactions on Industrial Infor-
even for a large fleet of PEVs. matics, vol. 10, no. 4, pp. 2112–2121, 2014.

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
12

[7] R. Garcia-Valle and J. G. Vlachogiannis, “Letter to the editor: Electric [28] T. M. Sweda and D. Klabjan, “Finding minimum-cost paths for electric
vehicle demand model for load flow studies,” 2009. [Online]. Avail- vehicles,” in Electric vehicle conference (IEVC), IEEE international.
able: http://www.tandfonline.com/doi/abs/10.1080/15325000802599411 IEEE, 2012, pp. 1–4.
Last accessed :july 2017 [29] H. Qin and W. Zhang, “Charging scheduling with minimal waiting in
[8] S. Deilami, A. S. Masoum, P. S. Moses, and M. A. Masoum, “Real- a network of electric vehicles and charging stations,” in Proceedings of
time coordination of plug-in electric vehicle charging in smart grids to the Eighth ACM international workshop on Vehicular inter-networking.
minimize power losses and improve voltage profile,” IEEE Transactions ACM, 2011, pp. 51–60.
on Smart Grid, vol. 2, no. 3, pp. 456–467, 2011. [30] A. Gusrialdi, Z. Qu, and M. A. Simaan, “Scheduling and cooperative
[9] A. Abdulaal, M. H. Cintuglu, S. Asfour, and O. Mohammed, “Solving control of electric vehicles’ charging at highway service stations,” in
the multi-variant ev routing problem incorporating v2g and g2v options,” Decision and Control (CDC), 2014 IEEE 53rd Annual Conference on.
IEEE Transactions on Transportation Electrification, 2016. IEEE, 2014, pp. 6465–6471.
[10] R. Abousleiman and R. Scholer, “Smart charging: System design and [31] V. del Razo and H.-A. Jacobsen, “Smart charging schedules for highway
implementation for interaction between plug-in electric vehicles and the travel with electric vehicles,” IEEE Transactions on Transportation
power grid,” IEEE Transactions on Transportation Electrification, vol. 1, Electrification, vol. 2, no. 2, pp. 160–173, 2016.
no. 1, pp. 18–25, 2015. [32] H.-Y. Mak, Y. Rong, and Z.-J. M. Shen, “Infrastructure planning for
[11] L. HANSEN, An electric vehicle fill-up of the future to last electric vehicles with battery swapping,” Management Science, vol. 59,
minutes, not hours, Bay Area News Group, Jan. 2017. [Online]. no. 7, pp. 1557–1575, 2013.
Available: http://www.eastbaytimes.com/2017/01/05/an-electric-vehicle- [33] H. Yudai and K. Osamu, “A safety stock problem in battery switch
fill-up-of-the-future-to-last-minutes-not-hours/ Last accessed :july 2017 stations for electric vehicles,” in The 8th International Symposium on
[12] N. Rauh, T. Franke, and J. F. Krems, “Understanding the impact of Operations Research and Its Applications (ISORA), 2009, pp. 332–339.
electric vehicle driving experience on range anxiety,” Human factors, [34] M. J. Neely, C. E. Rohrs, and E. Modiano, “Equivalent models for
vol. 57, no. 1, pp. 177–187, 2015. queueing analysis of deterministic service time tree networks,” IEEE
[13] Y.-T. Liao and C.-N. Lu, “Dispatch of ev charging station energy re- Transactions on Information Theory, vol. 51, no. 10, pp. 3576–3584,
sources for sustainable mobility,” IEEE Transactions on Transportation 2005.
Electrification, vol. 1, no. 1, pp. 86–93, 2015. [35] D. Said, S. Cherkaoui, and L. Khoukhi, “Multi-priority queuing for
[14] C. Le Floch, F. Belletti, and S. Moura, “Optimal charging of electric electric vehicles charging at public supply stations with price variation,”
vehicles for load shaping: A dual-splitting framework with explicit con- Wireless Communications and Mobile Computing, vol. 15, no. 6, pp.
vergence bounds,” IEEE Transactions on Transportation Electrification, 1049–1065, 2015.
vol. 2, no. 2, pp. 190–199, 2016. [36] Y. Xia and D. Tse, “On the large deviation of resequencing queue size: 2-
[15] P. D. Lund, J. Mikkola, and J. Ypyä, “Smart energy system design m/m/1 case,” in INFOCOM 2004. Twenty-third AnnualJoint Conference
for large clean power schemes in urban areas,” Journal of Cleaner of the IEEE Computer and Communications Societies, vol. 4. IEEE,
Production, vol. 103, pp. 437–445, 2015. 2004, pp. 2374–2382.
[16] L. Gan, U. Topcu, and S. H. Low, “Optimal decentralized protocol for [37] P. V. Krishna, V. Saritha, G. Vedha, A. Bhiwal, and A. S. Chawla,
electric vehicle charging,” IEEE Transactions on Power Systems, vol. 28, “Quality-of-service-enabled ant colony-based multipath routing for mo-
no. 2, pp. 940–951, 2013. bile ad hoc networks,” IET communications, vol. 6, no. 1, pp. 76–83,
[17] K. Bhavnagri, “How do i use electricity throughout the day, the 2012.
load curve,” Solar Energy Consultant, Tech. Rep., 2010. [Online].
[38] K. M. Sim and W. H. Sun, “Ant colony optimization for routing
Available: https://www.solarchoice.net.au/blog/how-do-i-use-electricity-
and load-balancing: survey and new directions,” IEEE Transactions on
throughout-the-day-the-load-curve/ Last accessed :july 2017
Systems, Man, and Cybernetics-Part A: Systems and Humans, vol. 33,
[18] S. N. I. Guide, “Daily trafffic counts,” Tech. Rep., no. 5, pp. 560–572, 2003.
2017. [Online]. Available: https://www.service.nsw.gov.au/service-
[39] J. E. Bell and P. R. McMullen, “Ant colony optimization techniques for
centre/roads-maritime-services Last accessed :july 2017
the vehicle routing problem,” Advanced engineering informatics, vol. 18,
[19] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of charging
no. 1, pp. 41–48, 2004.
plug-in hybrid electric vehicles on a residential distribution grid,” IEEE
[40] M. Dorigo, M. Birattari, C. Blum, M. Clerc, T. Stützle, and A. Winfield,
Transactions on Power Systems, vol. 25, no. 1, pp. 371–380, 2010.
Ant Colony Optimization and Swarm Intelligence: 6th International
[20] A. S. Masoum, S. Deilami, P. Moses, M. Masoum, and A. Abu-Siada,
Conference, ANTS 2008, Brussels, Belgium, September 22-24, 2008,
“Smart load management of plug-in electric vehicles in distribution and
Proceedings. Springer, 2008, vol. 5217.
residential networks with charging stations for peak shaving and loss
minimisation considering voltage regulation,” IET Generation, Trans- [41] M. Dorigo and L. M. Gambardella, “Ant colony system: a cooperative
mission & Distribution, vol. 5, no. 8, pp. 877–888, 2011. learning approach to the traveling salesman problem,” IEEE Transac-
tions on evolutionary computation, vol. 1, no. 1, pp. 53–66, 1997.
[21] A. Khaligh and S. Dusmez, “Comprehensive topological analysis of con-
ductive and inductive charging solutions for plug-in electric vehicles,” [42] B. Yu, Z.-Z. Yang, and B. Yao, “An improved ant colony optimization
IEEE Transactions on Vehicular Technology, vol. 61, no. 8, pp. 3475– for vehicle routing problem,” European journal of operational research,
3489, 2012. vol. 196, no. 1, pp. 171–176, 2009.
[22] B. Koushki, A. Safaee, P. Jain, and A. Bakhshai, “A bi-directional single- [43] Z. Li, K. Dey, M. Chowdhury, and P. Bhavsar, “Connectivity supported
stage isolated ac-dc converter for ev charging and v2g,” in Electrical dynamic routing of electric vehicles in an inductively coupled power
Power and Energy Conference (EPEC), 2015 IEEE. IEEE, 2015, pp. transfer environment,” IET Intelligent Transport Systems, vol. 10, no. 5,
36–44. pp. 370–377, 2016.
[23] M. Amini and A. Islam, “Allocation of electric vehicles’ parking lots in [44] M. Gruber, J. van Hemert, and G. R. Raidl, “Neighbourhood searches
distribution network,” in Innovative Smart Grid Technologies Conference for the bounded diameter minimum spanning tree problem embedded
(ISGT), 2014 IEEE PES. IEEE, 2014, pp. 1–5. in a vns, ea, and aco,” in Proceedings of the 8th annual conference on
[24] M. H. Amini, M. P. Moghaddam, and O. Karabasoglu, “Simultaneous Genetic and evolutionary computation. ACM, 2006, pp. 1187–1194.
allocation of electric vehicles parking lots and distributed renewable [45] Alternative Fueling Station Locator. [Online]. Available:
resources in smart power distribution networks,” Sustainable Cities and http://www.westcoastgreenhighway.com/ Last accessed :july 2017
Society, vol. 28, pp. 332–342, 2017. [46] S. Bakker, P. Leguijt, and H. van Lente, “Niche accumulation and
[25] M. H. Amini and O. Karabasoglu, “Optimal operation of interdependent standardization–the case of electric vehicle recharging plugs,” Journal
power systems and electrified transportation networks,” arXiv of Cleaner Production, vol. 94, pp. 155–164, 2015.
preprint arXiv:1701.03487, 2017. [Online]. Available: https://arxiv.org/ [47] ABB, “Abb launches cost-effective 20kw dc fast charger
Last accessed :july 2017 for the americas,” Tech. Rep., 2017. [Online]. Available:
[26] S.-N. Yang, W.-S. Cheng, Y.-C. Hsu, C.-H. Gan, and Y.-B. Lin, http://www.abbaustralia.com.au, Last accessed :july 2017
“Charge scheduling of electric vehicles in highways,” Mathematical and [48] CHAdeMO User Manual, July. [Online]. Avail-
Computer Modelling, vol. 57, no. 11, pp. 2873–2882, 2013. able: http://store.evtv.me/proddetail.php?prod=20kwchademo&cat=23,
[27] S. Pourazarm, C. G. Cassandras, and A. Malikopoulos, “Optimal routing Last accessed :july 2017
of electric vehicles in networks with charging nodes: A dynamic [49] P. Morrissey, P. Weldon, and M. OMahony, “Future standard and fast
programming approach,” in Electric Vehicle Conference (IEVC), 2014 charging infrastructure planning: An analysis of electric vehicle charging
IEEE International. IEEE, 2014, pp. 1–7. behaviour,” Energy Policy, vol. 89, pp. 257–270, 2016.

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TTE.2017.2753403, IEEE
Transactions on Transportation Electrification
13

[50] A. G. UNDA, P. Papadopoulos, S. SKARVELIS-KAZAKOS, L. CIP- Daryoush Habibi (M’95- SM’99) graduated with a
CIGAN, and N. JENKINS, “Electric vehicle battery swapping stations, Bachelor of Engineering (Electrical) with First Class
calculating batteries and chargers to satisfy demand.” Honours from the University of Tasmania in 1989
[51] F. Badin, F. Le Berr, H. Briki, J. Dabadie, M. Petit, S. Magand, and a PhD from the same University in 1994. His
and E. Condemine, “Evaluation of evs energy consumption influencing employment history includes Telstra Research Lab-
factors, driving conditions, auxiliaries use, driver’s aggressiveness,” in oratories, Flinders University, Intelligent Pixels Inc.,
Electric Vehicle Symposium and Exhibition (EVS27), 2013 World. IEEE, and Edith Cowan University, where he is currently
2013, pp. 1–12. a Professor and the Executive Dean of Engineering.
[52] M. Bouroche, “Meeting qos requirements in a dynamically priced His research interests include engineering design for
commercial cellular network,” 2003. sustainable development, reliability and quality of
[53] SDGE, “Ev rates can help you save,” Tech. Rep., 2017. [Online]. service in communication systems and networks,
Available: https://www.sdge.com Last accessed :july 2017 smart energy systems, and environmental monitoring technologies. He is a
[54] T. Covert, M. Greenstone, and C. R. Knittel, “Will we ever stop using Fellow of Engineers Australia, and Editor-in-Chief of the Australian Journal
fossil fuels?” The Journal of Economic Perspectives, vol. 30, no. 1, pp. of Electrical and Electronic Engineering
117–137, 2016.
[55] M. Kathryn, “How much does it cost to charge an electric
car?” Tech. Rep., 2016. [Online]. Available: https://pluginamerica.org,
Last accessed :july 2017

Zeinab Moghaddam received the M.Sc. degree

in Embedded System Design from the University
of Lugano, Switzerland, in 2012. She is currently
pursuing the Ph.D. degree with the School of En-
gineering, Edith Cowan University, Australia. Her
research interests include management of smart grid
and renewable energy systems with focus on Electric
Vehicles applications.

Iftekhar Ahmad is currently working as a Senior Quoc Viet Phung is currently working as a post doc-
Lecturer with the School of Engineering, Edith toral research fellow with the School of Engineering,
Cowan University, Australia. He received the Ph.D. Edith Cowan University, Australia. He received the
degree in communication networks from Monash Ph.D. degree in communication networks from Edith
University, Australia, in 2007. His research interests Cowan University, Australia, in 2010. His research
include 5G technologies, green communications, interests include 5G technologies, green communi-
QoS in communication networks, software-defined cations, wireless sensor networks and computational
radio, wireless sensor networks and computational intelligence.
intelligence.

2332-7782 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.