226 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO.

1, MARCH 2019

A Coordinated Dynamic Pricing Model for

Electric Vehicle Charging Stations
Zeinab Moghaddam , Student Member, IEEE, Iftekhar Ahmad, Member, IEEE,
Daryoush Habibi , Senior Member, IEEE, and Mohammad A. S. Masoum , Senior Member, IEEE

Abstract— The charging loads of plug-in electric vehi- μkj Service time at CS j for charging option.
cles (PEVs) within a network of charging stations (CSs) are not −−→
uniformly distributed. The load distribution is skewed toward the β j (t) Price coefficient vector of CS j .
−−→
stations located in the hotspot areas, instigating longer queues g j (t) Gain vector of CS j .
and waiting times, particularly during afternoon peak traffic −−→
hours. This can lead to a major challenge for the utilities in the Q(t) Inverse correlation matrix of the price vec-
form of an extended PEV load period, which could overlap with tor in the RLS algorithm.
the residential evening peak load hours, increase peak demand,
j (t)
φ min Minimum value for profit at CS j
and cause serious issues, such as network instability and power [$/time slot].
outages. This paper presents a new coordinated dynamic pricing
model to reduce the overlaps between residential and CS loads by π jR Probability transition matrix for the R pos-
inspiring the temporal PEV load shifting during evening peak sible states at CS j .
load hours. The new idea is to dynamically adjust the price σ Charging price coefficient.
incentives to drift PEVs toward less popular/underutilized CSs. c j (t) Electricity purchase price [cents/kWh].
We formulate a constraint optimization problem and introduce a recharge
heuristic solution to minimize the overlap between the PEV and Cj (k) Charging cost for charging option k at CS j
residential peak load periods. Our extensive simulation results [cents/kWh] (using time of use price).
indicate that the proposed model significantly reduces the overlap d j (t) Charging demand at CS j .
and the PEV load during evening peak hours. dst Destination of PEV.
Index Terms— Charging station (CS), electric vehicle, pricing e j (t) Prediction error of CS j .
model. E caj Maximum power capacity of CS j [kW].
N OMENCLATURE I j (t) Amount of incentive from the grid
A. Abbreviations [cents/kWh].
ACO Ant colony optimization. j Index of a CS.
BSSs Battery storage systems. L kj (t) Queue length for option k in CS j (number
CS Charging Station. of PEVs in a queue).
NP Nondeterministic polynomial-time hardness. p j (t) Charging price at CS j [cents/kWh].
PEV Plug-in electric vehicle. j (t)
pmax Upper bound of the charging price at CS j
RLS Recursive least square. [cents/kWh].
SoC State of Charge [%]. j (t)
pmin Lower bound of the charging price at CS j
UP Uncoordinated pricing. [cents/kWh].
r (t) Residential load [MW].
B. Parameters
χm Correlation coefficient of iteration m for src Current source of PEV.
the updated price vector of CS j . t Time consisting of 24 time slots; t = t,
t Time slot (time interval) equal to 1 h. 2t, 3t, 24 h.
γ j (t) Uniform queue occupancy of CS j . T drive ( j, dst) Driving time from CS j to destination.
κ j (t) Actual queue occupancy of CS j . T drive (src, j ) Driving time from current location of PEV
κ max to CS j .
j (t) Maximum queue occupancy of CS j .
T jwait (k) Waiting time for charging option k at CS j .
κ min
j (t) Minimum queue occupancy of CS j .
xj Price coefficient for price vector at CS j
λj PEV Arrival rate at CS j .
based on the actual and uniform queue
Manuscript received October 18, 2018; revised December 23, 2018; occupancy at CS j .
accepted January 24, 2019. Date of publication February 1, 2019; date of
current version March 19, 2019. (Corresponding author: Zeinab Moghaddam.) C. Constants
Z. Moghaddam, I. Ahmad, and D. Habibi are with the School of Engi-
neering, Edith Cowan University, Joondalup, WA 6027, Australia (e-mail: E ca Capacity of the EV network [kW].
[email protected]; [email protected]; [email protected]). E gca Maximum capacity of grid assigned to a CS [kW].
M. A. S. Masoum is with the Department of Engineering, Utah Valley
University, Orem, UT 84058 USA (e-mail: [email protected]). k Charging option (k = 1: swap battery; k = 2: dc
Digital Object Identifier 10.1109/TTE.2019.2897087 charging; and k = 3: ac charging.
2332-7782 © 2019 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.
MOGHADDAM et al.: COORDINATED DYNAMIC PRICING MODEL FOR ELECTRIC VEHICLE CSs 227

M Total number of PEVs.

N Total number of CSs.

I. I NTRODUCTION

T HE PEVs are rapidly becoming a major component of

cities around the world, and their market penetration is
expected to significantly grow in the upcoming years. As the
expectations for future PEV sales increase, there is a greater
research focus on the development of charging infrastructure.
Charging infrastructure is broadly divided into three categories
based on the PEV charging speed. Levels 1 and 2 are in the cat-
egory of ac charging, while level 3 represents dc charging [1].
Many countries around the world have already established
Fig. 1. Daily traffic counts and daily residential loads in NSW [5], [6].
a network of electric vehicle CSs, commonly known as an
EV network [2]. Such infrastructures are eventually expected
to be as ubiquitous as the traditional gas stations. An EV
network provides the opportunity for quick recharging on develop a coordinated dynamic pricing model for the temporal
demand, thereby significantly addressing the range anxiety PEV load shifting to reduce the PEV load during the evening
problem among consumers. In addition, EV networks can play hours.
a critical role in peak load shaving by offering the charging In the literature, various smart charging strategies have been
facilities during off-peak hours. However, utilities have serious proposed to address the concerns raised by the PEV owners
concerns over the negative impacts of such stations on the and utilities. Most of these works aimed at reducing the
stability of the power grid [3]. For instance, simultaneous average waiting period and charging cost [10]–[12]. In our
charging of too many vehicles at CSs can substantially increase previous work [7], we presented a solution to reduce the
the power demand at that station and impose detrimental waiting time and charging cost. This provides guidelines for
impacts on the grid components. However, intelligent routing the PEV owners to select the best available CS based on the
of PEVs can turn this challenge into an opportunity by viewing minimization of distance, charging price, and waiting time at
the vehicles as mobile storage devices with charge/discharge the CS. The scope of the work presented in [7], however,
capabilities [4]. does not include the PEV load shifting and price coordination
As more PEVs join the grid, the waiting time at CSs among CSs in an EV network. In this paper, we show that
in combination with the actual road traffic will constitute a the price coordination among CSs can be an effective tool in
major challenge. During peak hours, CSs may end up with balancing the PEV load, which ultimately reduces the overlap
long queues that can directly impact the comfort of PEV between the PEV and residential loads.
owners. To resolve these challenges, we need a systematic Lack of price coordination within a CS network leads to a
PEV-scheduling technique that not only takes into account the nonuniform distribution of charging load across the CSs (e.g.,
distribution of the electricity load but also reduces the traffic long queues at hotspot areas, whereas other CSs are not
congestion and waiting times at the CSs. utilized to their maximum capacities). This results in underuti-
Fig. 1 presents the average afternoon traffic pattern as a lization of the maximum capacity of the CS network, therefore
percentage of the total vehicle counts and a typical residential increasing the charging times and the overlap period between
load profile in an urban area in New South Wales (NSW), the PEV and residential loads. As a result, the PEV load
Australia [5]–[7]. Fig. 1 demonstrates that the busiest time during the evening hours can be high enough to cause network
slots are between 2 P. M . and 6 P. M . when most of the PEVs are instability problem due to an increase in total load (i.e.,
expected to be on the streets and would need recharging. The residential peak load plus PEV load).
temporal PEV load profile of an EV network is also expected In a consumer-driven market, utilities cannot directly deny
to follow a similar pattern [7]. Utilities have already expressed services to PEV owners even when a grid stability problem
their concerns over the possibility of facing an overlap between looms. However, PEV owners’ behavior can be influenced
the PEV load and daily residential load periods. An extended by adjusting the charging prices. Therefore, price incen-
overlap period is likely to put significant stress on both the tives/signals can be used as the important management tools
generation and distribution sides of the energy industry [7]. in an EV network. Consequently, dynamic pricing has been
The PEV load combined with the increasing residential load an active area of research, and researchers have made some
during the evening hours, commonly known as the “duck important contributions in the literature, such as [13]–[16].
curve” problem, offers a major challenge (i.e., high peak-to- However, none of the existing solutions addresses the chal-
average demand ratio) for the energy industry. Researchers lenge of reducing the overlap between the PEV and residential
have been working to shift some loads from the evening hours load periods.
to address the duck curve challenges [8], [9]. This motivated In this paper, we present a new coordinated dynamic pricing
us to investigate the charging strategies in an EV network and model to reduce the overlap between the PEV and residential
228 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO. 1, MARCH 2019

loads during the evening hours. The coordinated pricing policy to establish the interaction between PEVs and the power grid
provides a price vector that encourages a uniform distribution in order to ensure a safe and semistable load on the grid and
of PEV loads across all CSs so that the EV network can be minimize the cost. However, they focused on residential PEV
utilized to its maximum capacity. This leads to a reduction charging without considering EV CSs.
in PEV load during the evening hours. However, this is not a The electrification of transportation brings both opportuni-
trivial approach since charging cost is not the only influencing ties and challenges to existing critical infrastructures [21]. For
factor. EV owners are influenced by other factors as well, example, Wu et al. [22] addressed the challenges of energy
including driving distance and waiting time, which we also scheduling in office buildings integrated with photovoltaic
consider in our approach. In summary, the major contributions systems and workplace PEV charging for public users. They
are as follows. proposed to leverage day-ahead power market and time-of-
1) We introduce a new coordinated dynamic pricing model use electricity and used stochastic programming to address
for the temporal PEV load shifting to reduce the over- the uncertainties in PEV charging. They proposed a model to
lap between loads associated with CS and residential estimate the demands of charging a PEV at the workplace.
networks. To the best of our knowledge, this is the first However, there is a limited flexibility for selecting the CSs
paper of its kind, which deals with temporal PEV load since they did not consider real-time pricing in their model
shifting using price incentives/signals in a bid to reduce and overlooked various charging options.
the evening peak demand. Wong and Alizadeh [23] presented a framework of PEV
2) We formulate the research challenge as a constrained CSs using the queueing model. They proposed a price strategy
optimization problem. We propose solutions to other rel- based on social optimal congestion that enforces the CSs prices
evant subproblems, including the estimation of demand to the customer in order to minimize the total latency of PEV
in response to charging price variation for each station users and total electricity costs of CSs. However, in their
during various time slots. Considering the time complex- queueing model for CSs, they did not mention the charging
ity of the optimum solution, we introduce a rule-based rates, and also they did not consider any load management
model to derive the appropriate price information. of PEVs at peak hours. There are also a number of industry
3) The proposed model significantly reduces the overlap projects [24] and research papers [25], [26] on fast CSs with
between the PEV peak load period and the residential BSSs to store the purchased electricity from the grid at the
peak load period, which ultimately leads to a lower cheapest hours of a day and sell it to PEVs at peak hours.
evening peak demand. In order to quantify and bench- Bai et al. [25] studied the power demand of the CS and
mark the benefits, we implement our proposed coordi- proposed an optimum design of a fast-CS equipped with BSSs.
nated dynamic pricing model on the Washington green Similarly, Nargestani et al. [26] have investigated the CSs
highway EV network [17] and compare the results with with BSSs but also estimated the optimum sizing of BSSs
the smart charging strategy of [7] that uses an ACO. for controlling the charging demand. However, none of the
It should be noted that a pricing model has an impact on existing literature discusses changing the price at CS and
multiple parties, including the electricity utilities, PEV, and CS controlling the PEVs demand at peak hours. Another important
owners. In this paper, we consider the potential grid stability factor in managing a CS is the power market. Similar to petrol
issue caused by an increase in peak demand during the evening stations, multiple CSs in the same area may belong to different
peak hours as the major challenge. As such, direct profit owners; therefore, competition between different CSs should
maximization of CS owners is not within the scope of this be considered [27]. Lee et al. [28], Etesami et al. [29], and
paper. Yuan et al. [30] have employed approaches based on game
We assume that since the utilities will enjoy the maximum theory to model an interplay among multiple PEVs or between
benefit from the reduction of the peak load, they will need to PEVs and power grid. They have established a competition
provide necessary incentives (e.g., adjustment of selling price system based on game theory for CSs in terms of increasing
to CSs, reward for contribution to the reduction of peak load, CS profits; however, the impacts of traffic load of PEVs
and so on [18], [19]) to the CSs to promote a dynamic pricing on the power grid at peak hours are not considered. Erol-
model. One can argue that increasing the capacity of the utility Kantarci and Hussein [31] proposed a prediction-based charg-
during the evening peak hours can solve the problem, but this ing scheme that receives dynamic pricing information via
would require capital investment and the high evening peak wireless communication, predicts the market prices during the
demand (i.e., the duck curve problem) would not help with charging period, and determines an appropriate time of day
the economics of this decision (i.e., a low return on investment to charge the vehicle at low cost. However, their prediction-
caused by the high peak-to-average demand ratio). based charging scheme was based on a simple, lightweight
classification technique that is suitable for implementation on
II. R ELATED W ORK a vehicle or a CS. Therefore, they have not considered the
In this section, we present previous studies related to the impact of driving and charging patterns, such as the increase
estimation of PEVs charging load, pricing models, and PEV in the demand for charging at peak hours in their charging
charging strategies with emphasizes on reducing the overlap scheme.
between PEVs charging and residential loads during peak Junjie et al. [32] presented two indirect methods for the PEV
hours. Abousleiman and Scholer [20] presented a novel hard- management system in the power grid. They used the market-
ware design and implementation logic for a smart grid system and price-based controls to minimize the communication cost
MOGHADDAM et al.: COORDINATED DYNAMIC PRICING MODEL FOR ELECTRIC VEHICLE CSs 229

and computational complexity. However, both control strate-

gies have some limitations, such as the uncertainty of the
price-based control strategy. Moreover, they did not present
real-time verifications of the proposed control strategies.
Kim et al. [33] developed an algorithm by controlling
charging price and the number of vehicles to be scheduled
as well as the charging/discharging of BSSs so that the CS
could achieve profit maximization. However, their work did
not consider PEV load shifting. In [13], a price strategy for
the economic operation of CSs equipped with the renewable
resources has been proposed. In their proposed pricing model,
they suggested a stochastic approach for dealing with the
uncertainty of the renewable energy sources to make an opti-
mal price decision in order to maintain the operational cost at
the minimum level. Cherikad et al. [16] proposed a distributed
dynamic pricing model for PEVs charging and discharging
scheduling and building energy management in a microgrid.
Their model was based on the decentralized communication
Fig. 2. Architecture of the PEV CS.
architecture, and they used a linear optimization approach to
achieve the efficient price decision to maximize the PEVs
utility and maintains the microgrid stability.
the waiting time parameter is a key concern from PEV owners’
Chen et al. [14] using a dynamic linear program for PEVs
perspective, this parameter is not a direct concern for the
charging process proposed a real-time price strategy for the
utility. Instead, the PEV load, which is a direct concern for the
CSs to minimize the cost of electricity purchased from the
utility, is estimated based on a RLS approach [see (10)–(15)
grid. However, temporal PEV load shifting was not the focus
and Fig. 3] and included in the problem formulation.
of this paper.
As discussed earlier, although researchers have made impor-
tant contributions in the areas of charging strategy, dynamic A. Charging Station System Model
pricing, and load management, none of the existing solu- We consider a city with a set of CSs under different
tions presents a coordinated pricing model to control the
ownership structures that buy electricity from the power grid,
PEVs demand at CSs and mitigate the extra stress on the as shown in Fig. 2. At each CS, there are multiple charging
power grid at peak hours, which is the main focus of this options, including ac level 2 charging, dc fast charging, and
paper.
battery-swapping facilities. In smart charging, both PEVs and
the electricity utility are directly involved. In the existing
III. F ORMULATION OF P ROPOSED
works [7], [12], smart charging strategies that have been
DYNAMIC P RICING M ODEL
designed for PEV owners attempt to find the optimum CS for
This section presents the proposed coordinated dynamic a PEV by considering multiple factors, including driving time,
pricing model for EV CSs in terms of (1)–(20), which is used waiting time, and the price to charge at a CS. Mathematically,
in Section IV to present a new EV charging strategy to reduce the objective of such smart charging strategies can be given
the overlaps between residential and CS loads by inspiring the as [7]
temporal PEV load shifting during evening peak load hours.  
recharge
In EV smart charging, two entities, namely, the PEVs min x(T drive (src, dst)) + yT jwait (k) + zC j (k) (1)
and the electricity utility, are directly involved. The opti-
mization model included in (1)–(5) is intended for PEVs. where x, y, and z are the positive coefficients of the objective
This model finds the optimum CS for an electric vehicle function. For a PEV, T drive (src, dst) shows the total driving
so that the travel time, waiting time, and charging cost can time from its current location src to the final destination dst,
recharge
be minimized. In the proposed model, we highlight that while j shows the CS along the way and C j (k) indicates
charging price is an influential factor in the decision-making the charging cost. In addition to the driving time, a PEV
process of PEV owners, and the utility can influence their needs to stay in the queue at a CS to access an available
behaviors (e.g., which CS to choose and when, and thereby charging socket. As a result, the total waiting time [T jwait (k)]
time-shift the load) by controlling the charging price at at a CS j depends on the waiting time in the queue [E(W kj )]
various CSs. and the time to recharge. Average waiting times for the three
Equations (6)–(9) constitute the main problem formulation types of charging options at time t can be estimated using the
in this paper. This formulation is intended for the utility, where M/M/s/C and the Little’s law models [34]
the goal is to find the price vector for various time slots, which  
 k E L kj
will ultimately reduce the amount of overlap between the PEV E Wj =   ∀k : (k = 1, 2, 3). (2)
and residential loads during evening hours. As a result, while λ j 1 − π jR
230 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO. 1, MARCH 2019

Therefore, for the queue of the battery-swapping facility at optimization model can be given as follows:
CS j , the mean waiting time can be obtained [7]

N

  μ1j min Z = r (t) + dj (6)

E W 1j =   . (3) j =1
λ j μ1j − λ j 1 − π jR
s.t. pmin
j ≤ p j (t) ≤ pmax
j ∀ j = 1, . . . , N, ∀t (7)
Similarly, the mean waiting times for dc fast and ac charging 
N
queues are d j (t) ≤ E ca ∀t (8)
j =1
  μ2j d j (t)[ p j (t)+ I j (t)−c j (t)] ≥ φ min
E W 2j =    (4) j (t) ∀ j = 1, . . . , N, ∀t.
λ j μ2j − λ j 1 − π jR (9)
  μ3j
E W 3j =   . (5) Constraint (7) defines the upper ( pmax j ) and lower
λ j μ3j − λ j 1 − π jR min
bounds ( p j ) for the charging price at a CS. Constraint
In addition to the parameters, such as driving time and (8)
 indicates that the aggregate PEV loads across all CSs
waiting time, the other parameter that influences the decision [ Nj=1 d j (t)] should be less than the maximum capacity of
of a PEV owner is the charging cost. Charging cost depends on the EV network (E ca ). Constraint (9) ensures that the price
the charging prices at CSs. Existing works [7], [12] assume vector does not lead to revenue loss for CSs. I j (t) represents
that all CSs use their own pricing models, which offer no the incentive [36], [37] provided to CS j by the utility for the
price coordination among CSs and no price incentive to PEVs time slot t in appreciation of the contribution of the CSs in
for selecting lightly loaded CSs during various hours of the reducing the peak load. c j (t) indicates the electricity cost for
day. While this assumption causes no major problem for PEV the time slot t at the CS j .
owners, such an approach introduces a major concern for the The PEV and residential load profiles can vary from day
electricity utility. to day (e.g., weekdays versus weekends). The optimization
One risk associated with such charging strategies is that model, however, can still find the optimum price vector, which
even when there are several other lightly loaded CSs available would lead to a reduction of overlap and peak load during the
nearby while a PEV waits at a heavily loaded CS located evening hours. In cases when the overlap is not a significant
at a hotspot area, the additional driving time and energy concern for the utility (e.g., low PEV load in the afternoon
consumption would discourage many PEV owners to use hours during a public holiday), the period of interest can be
the lightly loaded CSs. The other risk associated with such adjusted to reflect this in the optimization model.
charging strategies is that many PEV owners may prefer partial The model is independent of the number of CSs and
charging at heavily loaded CSs [7], [35], meaning that these distances between the stations. The optimization problem can
owners are likely to plug into the grid straightaway after they be solved given that the impact of price vector on PEV loads
reach home, which will increase the peak residential load. [i.e., f j p j (t)] at various CSs [see (10)] can be quantified.
The overall risk is that due to the nonuniform distribution of This is not a trivial task since the demand is influenced by
charging load, the maximum capacity of the charging network multiple factors, including the locations of CSs, their distances
will not be fully utilized during the peak PEV load period, from PEVs, and the relative price differences in their offered
meaning that a fraction of this PEV load will shift and overlap charging price
with the residential peak load. This is a major concern for the ⎡ ⎤ ⎡ ⎤
d1 (t) f 1 ( p(t))
utility and a coordinated dynamic pricing model can be useful ⎢ d2 (t) ⎥ ⎢ f 2 ( p(t)) ⎥
here to influence PEV owners to select lightly loaded CSs. ⎢ ⎥ ⎢ ⎥
⎢ .. ⎥ = ⎢ .. ⎥. (10)
In Section III-B, we show how a coordinated time-varying ⎣ . ⎦ ⎣ . ⎦
pricing model can be represented as an optimization problem. d N (t) f N ( p(t))

In Section III-C, we show how a demand vector can be

B. Proposed Optimization Model for Temporal estimated in response to a change in the price vector in an EV
PEV Load Shifting network.
Let us consider a set of N CSs in an EV network. The price
−−→
vector P(t) ∈ R N indicates the charging prices offered by the
C. Temporal Demand Estimation at Charging Stations
CSs in the network, where p j (t) stands for the charging price
in Response to a Price Vector
offered by the CS j at t. We consider a day, which is divided
into t time slots. The main objective of the proposed model In this section, we consider the estimation of the PEV
−−→
is to obtain an appropriate price vector P(t) (i.e., decision charging demands at the CSs. Using the general demand
variable) for the time slots during the busy hours to minimize function [38], we can characterize the PEVs demand based
the overlap between the PEV load [d j (t)] and residential load on the price fluctuations. Therefore, charging demand function
[r (t)]. Mathematically, the objective function of the proposed can be characterized by price variations during different time
MOGHADDAM et al.: COORDINATED DYNAMIC PRICING MODEL FOR ELECTRIC VEHICLE CSs 231

all CSs at different time slots. The rule-based model attempts

to find a price vector that encourages a uniform distribution
of PEV loads across all CSs so that the EV network can be
utilized to its maximum capacity during the peak PEV load
period.
The rule-based model starts with input parameters, which
include the current time slot t, current demand κ j (t) at each
−−→
Fig. 3. Dynamic state presentation of the price vector. CS j , and the target demand vector γ (t) in the EV network
−−→
obtained under the uniform distribution condition. γ (t) at
slots in a day, which can be given as the time slot t can be derived from a charging strategy by
⎧ uniformly allocating the PEVs among all CSs according to
⎡ ⎤ ⎪ d (t) = α1 (t) − β1,1 p1,k (t) + · · · + β N,1 p N,k (t)
⎪ 1
⎪ their capacities (i.e., sockets). It should be noted that the
d1 (t) ⎪
⎨d2 (t) = α2 (t) − β2,1 p2,k (t) + · · · + β N,1 p N,k (t)
⎢ .. ⎥ demand parameters in the rule-based solution are expressed in
⎣ . ⎦= .
⎪
⎪..
terms of queue occupancy in contrast to the demand parameter
d N (t) ⎪⎪
⎩ in the optimization model, which is expressed in terms of
d N (t) = α N (t)+β1,N p1,k (t) + · · · − β N,N p N,k (t) energy demand.
(11) The rule-based model adjusts the price vector at the start
where [α1, . . . α N ] are factors other than price (e.g., distance, of each time slot so that the corresponding demand vector,
location, and hourly aggregate load) which affect charging which is derived from the RLS algorithm, moves closer to the
demand, and [β1 , . . . β N ] are the price coefficients that influ- solution under the uniform distribution condition. In order to
ence the demand. For demand estimation, we need to obtain benchmark the proposed dynamic pricing model, we include
the price coefficients during various hours of the day. the proposed solution in a smart charging strategy modeled
Equation (11) defines a linear system of PEVs demand as an ACO, which is presented in Section IV. The flowchart
function, which is affected by the price coefficients of all the of the charging strategy incorporating the proposed dynamic
CSs in the EV network. The price elasticity coefficients of such pricing model is illustrated in Fig. 4.
a linear system can be estimated by the RLS algorithm (see
IV. C HARGING S TRATEGY U SING THE P ROPOSED
Fig. 3) [39]. RLS is an adaptive filter algorithm that recursively
DYNAMIC P RICING M ODEL
finds the price coefficients at different time slots relating to the
input parameters. For applying the RLS algorithm, we need As discussed in Section III, in order to find an optimum
the following equations: CS for a PEV, a charging strategy considers multiple factors,
−−→−−→ including the total driving time, total waiting time (queue
e j (t) = d j (t) − β j (t) p j (t) (12) and charging time), and the price to charge. In this section,
−−→−−−−−−→
−−→ p j (t) Q j (t − 1) we describe the implementation of a charging strategy using
g j (t) = −−−−−−→−−→ (13) ACO [7] considering the impact of the proposed pricing
 + Q j (t − 1) p j (t)
−−→ −−→ −−→ model on the grid and PEV owners. We selected ACO as a
β j (t) ← β j (t) + e j (t)g j (t) (14) metaheuristic solution to provide a robust solution in real time.
−−−→ 1 −−−−−−→ 1 −−→−−→ Using this approach, we can include real-time information
Q j (t) = Q j (t − 1) − g j (t) p j (t) (15)
  about the distance to a CS, waiting times at the CS, and
where e j (t) is the prediction error, while 0 <  ≤ 1 is the updated pricing information for each time slot. This enables
forgetting factor that helps us to capture the last charging us to estimate the optimum path in the CS network using the
demands and ignore the old information and track charging proposed pricing model based on the following steps.
−−→
demand variation over time. g j (t) is the gain vector and Q(t) 1) Initialize the parameters, including the CSs number and
is the inverse correlation matrix of the input vector. In the their attributes, number of PEVs and their attributes and
RLS algorithm, for a given time slot, the price coefficients are pheromone level, as well as the evaporation rate of the
derived, and the demand at a CS can be estimated for a given pheromone for the ACO approach in the EV network.
price vector. This demand information is required to find the 2) Calculate Pikj (u) in each iteration with respect to the
optimum solution of the optimization problem. heuristic values, such as travel time, waiting time, and
The model described in Section III is a nonconvex NP-hard updated price at each CS
constrained optimization problem. Since the optimal solution Pikj (u)
is untraceable and computationally expensive, we look for  θ
a suboptimal solution. Considering the need for a robust τi j (u)α (T drive (src,dst))T jwait (k) p j (t)
=    .
solution, we solve the problem by introducing an iterative rule- α drive (src,dst))T wait (k) p (t) θ
l∈N k τi j (u) (T j j
based pricing model, which is described in Section IV. i
(21)
D. Proposed Rule-Based Pricing Model In (21) at each iteration u, using the updated informa-
In this section, we introduce the rule-based instructions, tion of pheromones and attractiveness of the heuristic
as shown in Table I, to obtain the spatial price vector for information in the model, a PEV moves stochastically
232 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO. 1, MARCH 2019

TABLE I
O PERATION RULES OF THE P ROPOSED DYNAMIC P RICING M ODEL

Fig. 4. Smart charging strategy using the proposed dynamic pricing model.

vector [see (16) and (17)] as follows [40]:

  
τi j (u + 1) = σ (1 − ρ)τ (u) + τilj ∀(i, j )] ∈ iterl
(22)
  g 
τi j (u + 1) = σ (1 − ρ)τ (u) + τi j ∀(i, j ) ∈ iterg
(23)
g
where τilj and τi j are the incremental amounts of the local and
global updates at each iteration, respectively.
Fig. 4 shows the flowchart of the proposed dynamic pricing
approach to minimize the overlap between PEVs and residen-
tial loads.

V. S IMULATION R ESULTS AND D ISCUSSION

A. Simulation Model
We tested the proposed solution using a simulation based
on the Washington City road EV network, as shown in Fig. 5.
Each CS is equipped with two charging options (dc fast
charger and ac level 2 charger) and a battery swap facility.
We considered a mix of up to 500 PEVs comprising of Nissan
Leafs (30 kWh), BMWs i3 (22 kWh), and Smart Eds (16 kWh)
with uniform SoC distribution in the range of 10%∼90%. The
PEV arrival at each CS is modeled as a Poisson distribution.
to obtain the probability function of choosing the next The number of sockets for a charging option at a station is
node in a graph of CS. also distributed uniformly in the range of 1∼10. The fast dc
3) For the path constructions, the local and global is 50 kW [43] and the ac level II charging is 22 kW [44],
pheromones are updated using the new price coefficient while the average time to swap a battery is considered to be
MOGHADDAM et al.: COORDINATED DYNAMIC PRICING MODEL FOR ELECTRIC VEHICLE CSs 233

TABLE II
S IMULATION PARAMETERS

Fig. 6. Hourly electricity price of the CSs.

Fig. 5. Washington green highway EV network [17].

3 min [45]. Simulation parameters are presented in Table II. Fig. 7. Hourly queue occupancy of the CSs in Fig. 5 with (a) UP model
The charging prices for all CSs at each time slot are generated and (b) proposed model.
based on the proposed pricing model of Section III. It should
be noted that the peak charging price at CSs is maintained
below the peak residential price (i.e., 55 cents/kWH). This B. Results and Discussion
is to encourage PEV owners to use CSs instead of charging In this paper, we analyze the hourly electricity price,
sockets at homes during evening hours since the load from too the queue occupancy, and the average waiting times at CSs as
many PEVs charging at the same time is likely to cause stress well as the probability of overlap between PEV and residential
on local substations. Simulation results using the MATLAB loads with the proposed dynamic pricing model described in
software package are presented in Figs. 6–12 and summarized Section III. We compare the proposed solution with existing
in Table III. The price vector in the EV network [see (14)] is studies that use an UP model for all CSs [7], [12]. In the UP
updated hourly using the proposed pricing model, as discussed model, CSs independently choose their own charging price.
in Section III. The discussion of the results begins with Fig. 6 that shows
234 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO. 1, MARCH 2019

Fig. 9(c) shows that the average price at CSs with the UP
model is higher than the proposed model during the day time
and lower than the proposed model during the afternoon hours.
This is because, in the proposed model, considering the PEV
demands at each time slot, price will change to encourage a
uniform distribution of PEV loads across all CSs.
In Fig. 9(d), we have normalized the average waiting time
over the travel time for both models. However, the result shows
that at first, the waiting time in the proposed model is higher
than UP model, but it decreases significantly during peak
hours. This is because, in the proposed model, information
in relation to the queuing delay for each charging option at
Fig. 8. Impact of the proposed coordinated price model on the PEVs load. every CS is considered to update the price of the next time slot,
which attracts the PEV owners to start charging early due to
the low charging price. Fig. 10(a) shows the PEV and residen-
the hourly electricity prices provided by the proposed model tial loads during a typical weekend in NSW [46], [47]. Note
for the 15 CSs in the EV network of Fig. 5. As Fig. 5 suggests, that the overlap between the PEV and residential loads during
charging prices at CSs are: 1) moderate before 3 P. M . (to peak hours (see Fig. 1) is lower in the weekend compared
motivate early PEV charging that could be supplied from BSSs to weekdays. Fig. 10(b) illustrates that the proposed model
and/or renewable energy resources; 2) high from 3 P. M . to maintains its superior performance in terms of the reduced
4 P. M . due to the increase in demand; and 3) low during the load overlapping during the weekend as well. Considering
late evening hours due to a drop in demand at CSs. The price the popularity of Tesla’s PEVs and dc fast CSs, we also
trend corresponds to the PEV load profile where the PEV peak implemented the proposed coordinated dynamic pricing model
load is observed at 4 P. M . If there is a change in the PEV in a scenario where PEVs were considered to have the capacity
load profile (e.g., PEV peak load shifts left or right depending equivalent to Tesla Model 3 (80.5-kWh battery capacity [38])
on days or there is an increase in demand at CSs during the PEVs, and the ac charging sockets were replaced with dc fast
evening hours), the proposed dynamic pricing model would charging sockets in all CSs. Relevant results shown in Fig. 11
follow the load trend and adjust the price at various hours indicate a similar trend observed in Fig. 9 where the proposed
accordingly. We investigate the average queue occupancy at model was found to outperform the existing model in terms of
all CSs during different times of the day in Fig. 7. Fig. 7(a) peak demand during the evening peak hours and the overlap
presents the queue occupancy of the UP model. This shows period.
that when there is no price coordination among CSs, PEV Since a smart charging strategy needs to be robust, a solu-
demands remain divergent during different time of a day, while tion is desired in real time. We have investigated the real-
Fig. 7(b) shows that changing the prices at CSs using the time computation time (in seconds) required by the proposed
proposed model influences the queue occupancy (i.e., load at dynamic pricing model to provide a solution. We have also
a CS). As shown in Fig. 7, the proposed model distributes the investigated the computation time required for each PEV to
load among CSs more uniformly so that the EV network can find the optimum CS. As evident in Fig. 12(a), the computation
be utilized to its maximum capacity during the peak traffic time of the rule-based model increases at a moderate rate with
period. Fig. 8 shows the impact of the proposed coordinated the number of CSs. Fig. 12(b) demonstrates the computation
price strategy on the PEVs load during different time slots. time for PEVs to find a CS. As it shows, the average
As it indicates, the PEV charging load during the peak period computation time remains relatively constant as the number
decreases, resulting in uniform distribution of PEV loads of PEVs increases.
across all CSs. Fig. 9 highlights the potential advantages of Although the proposed solution relies on the estimated PEV
the proposed approach compared with the UP model in terms and residential loads, the uncertainty in loads will have impacts
of hourly PEV load demands, overlap of PEV and residential on the outcomes. This uncertainty, however, is captured in our
loads, average charging prices, and normalized waiting times. simulation results, where the PEV arrival process at source
As evident in Fig. 9(a), the net PEV load per CS is significantly nodes is implemented using the Poisson distributions with
lower during the peak residential load period (e.g., 6 P. M . to arrival rate λ [see (2)–(5) and Table II]; hence, the actual
8 P. M .) in the proposed model compared to the PEV load in the PEV load at various stations varies from the estimated load
existing UP model. This is because, the proposed model makes during various time slots. As such, the results presented in
the best use of the maximum capacity of the EV network Section V captured this uncertainty in PEV load (see Figs.
during the peak PEV load period by encouraging uniform 6–11). For the residential load, we have relied on the data
load distribution through the implementation of a coordinated supplied by the utility and assumed that their estimation error
pricing model. Fig. 9(b) presents the probability of an overlap is not significantly high.
between PEVs and residential loads during different time slots. Table III summarizes the performances improvements of the
As the PEVs loads decrease during the peak hours with the proposed dynamic pricing strategy over the existing UP model,
proposed model, it is obvious that the probability of an overlap given a fleet size of 500 PEVs. Table III clarifies that the
between PEVs and residential loads will decrease significantly. proposed model provides a significant reduction of the overlap
MOGHADDAM et al.: COORDINATED DYNAMIC PRICING MODEL FOR ELECTRIC VEHICLE CSs 235

Fig. 9. Performance comparison of the UP model (see [7], [12]) and the proposed model. (a) PEV charging loads. (b) Probability of overlap between PEV
and residential loads. (c) Average PEV charging price. (d) Normalized waiting time in the queues.

Fig. 10. Performance comparison of the UP model [7], [12] and the proposed model for a typical weekend in NSW [46], [47]. (a) Weekend traffic counts
and residential load. (b) PEV charging load of weekend.

and the peak load during the evening hours. It should be Similarly, the proposed model would work for different PEV
noted that the duck curve challenge motivated us to investigate and residential load profiles (e.g., weekend versus weekdays).
the overlap between the PEV and residential loads in this However, the period of interest can be different depending
paper. However, the proposed model can be applied to reduce on the load profiles, and the utility would have the option
the overlap in other scenarios as well, for example, PEV to adjust this period of interest. The proposed model works
and commercial loads by changing the period of interest and by providing price incentives to EV owners to use less
load profile in the optimization and the rule-based model. popular/underutilized CSs. As such, the impact of the proposed
236 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO. 1, MARCH 2019

Fig. 11. Performance comparison of the UP model (see [7], [12]) and the proposed model with high capacity batteries for PEVs and dc charging in CSs.
(a) PEV charging loads. (b) Probability of overlap between PEV and residential loads.

Fig. 12. Computation time for the proposed charging strategy. (a) Computation time for the proposed rule-based model. (b) Average computation time of
ACO. Computer specification includes Intel i5 2.4-GHz CPU, 8-GB RAM.

TABLE III the same level of attention compared to other factors, such
S UMMARIZED S IMULATION R ESULTS as driving time and energy consumption. In cases where the
less popular/underutilized stations are located far from other
stations, the price incentive may not significantly influence
EV owners’ behavior because of the additional driving time
and energy consumption. As such, the impact of the proposed
solution can be limited in such networks.
The work presented in this paper is intended for the elec-
tricity utility. However, since smart charging involves both
the utility and the PEV owners, a real-time communication
platform, such as mobile applications, would be useful in
maintaining real-time communication between the two entities.
The mobile application can capture any change in the dynamic
price information, which would influence the PEV owners’
decisions. It can also analyze historical data and provide useful
information to the PEV owners, such as the period of a
day or the day of a week, when the charging price is low.

VI. C ONCLUSION
solution would be more significant for a large EV network We have introduced, implemented, and evaluated a new
since the disparity in terms of load distribution is expected coordinated dynamic pricing model for vehicle charging in
to be high in a large network. The proposed solution would CS networks that will encourage temporal PEV load shifting
work well for a scenario, where the price consideration enjoys to reduce their overlaps with the residential peak load periods.
MOGHADDAM et al.: COORDINATED DYNAMIC PRICING MODEL FOR ELECTRIC VEHICLE CSs 237

The dynamic pricing model was formulated as a constrained [13] Y. Guo, J. Xiong, S. Xu, and W. Su, “Two-stage economic operation of
optimization problem. We showed how the PEVs demand can microgrid-like electric vehicle parking deck,” IEEE Trans. Smart Grid,
vol. 7, no. 3, pp. 1703–1712, May 2016.
be estimated in response to charging prices at various CSs. [14] Q. Chen et al., “Dynamic price vector formation model-based automatic
We also presented a rule-based heuristic solution to address demand response strategy for PV-assisted EV charging stations,” IEEE
the dynamic pricing challenge in real time. The proposed Trans. Smart Grid, vol. 8, no. 6, pp. 2903–2915, Nov. 2017.
[15] C. D. Korkas, S. Baldi, S. Yuan, and E. B. Kosmatopoulos, “An adap-
solution was included in a charging strategy, which was then tive learning-based approach for nearly optimal dynamic charging of
implemented using an ACO model for bench marking. The electric vehicle fleets,” IEEE Trans. Intell. Transp. Syst., vol. 19, no. 7,
findings of this paper revealed significant benefits to both the pp. 2066–2075, Jul. 2018.
consumers by minimizing the waiting times at the CSs (by up [16] D. A. Chekired, L. Khoukhi, and H. T. Mouftah, “Decentralized cloud-
SDN architecture in smart grid: A dynamic pricing model,” IEEE Trans.
to 22.15%) and the utilities by shifting the charging demands Ind. Informat., vol. 14, no. 3, pp. 1220–1231, Mar. 2018.
to reduce their overlaps (by up to 46.7%) and average PEV [17] (Jan. 2018). Alternative Fueling Station Locator. Accessed: Jan. 2018.
loads (by up to 67.2%) with peak residential loads and mitigate [Online]. Available: http://www.westcoastgreenhighway.com
[18] H. Aalami, G. R. Yousefi, and M. P. Moghadam, “Demand response
grid congestion. model considering EDRP and TOU programs,” in Proc. IEEE/PES
In this paper, we considered that all CSs would cooperate Transmiss. Distrib. Conf. Expo., Apr. 2008, pp. 1–6.
with the utility, which would have a direct influence in [19] E. Shahryari, H. Shayeghi, B. Mohammadi-Ivatloo, and M. Moradzadeh,
“An improved incentive-based demand response program in day-ahead
changing/controlling the charging prices at various CSs in the and intra-day electricity markets,” Energy, vol. 155, pp. 205–214,
network. In many cases, CSs can be owned by independent Jul. 2018.
entities, and the utility may not have a direct control over how [20] R. Abousleiman and R. Scholer, “Smart charging: System design and
these independent entities choose/change their prices. In such implementation for interaction between plug-in electric vehicles and the
power grid,” IEEE Trans. Transport. Electrific., vol. 1, no. 1, pp. 18–25,
cases, a reward-based model will be required, where the utility Jun. 2015.
will provide rewards to CSs for their contributions in temporal [21] W. Su, J. Wang, and Z. Hu, “Planning, control, and management
PEV load shifting so that a potential grid stability problem strategies for parking lots for PEVs,” in Plug in Electric Vehicles in
Smart Grids. Singapore: Springer, 2015.
can be avoided. In our future work, we will investigate a [22] D. Wu, H. Zeng, C. Lu, and B. Boulet, “Two-stage energy management
reward-based model for the profit management of CSs in an for office buildings with workplace ev charging and renewable energy,”
EV network. IEEE Trans. Transport. Electrific., vol. 3, no. 1, pp. 225–237, Mar. 2017.
[23] P. Wong and M. Alizadeh, “Congestion control and pricing in a network
of electric vehicle public charging stations,” in Proc. 55th Annu. Allerton
Conf. Commun., Control, Comput. (Allerton), Oct. 2017, pp. 762–769.
R EFERENCES
[24] (Jun. 2017). 34 New Fast-Charging Stations With Energy Storage
for EVs to be Deployed Along Trans-Canada Highway. Accessed:
[1] S. F. Tie and C. W. Tan, “A review of energy sources and energy
May 2018. [Online]. Available: https://electrek.co/2017/07/21/fast-
management system in electric vehicles,” Renew. Sustain. Energy Rev.,
charging-stations-trans-canada-highway
vol. 20, pp. 82–102, Apr. 2013.
[25] S. Bai, D. Yu, and S. Lukic, “Optimum design of an EV/PHEV charging
[2] T. Franke and J. F. Krems, “What drives range preferences in electric
station with DC bus and storage system,” in Proc. IEEE Energy Convers.
vehicle users?” Transport Policy, vol. 30, pp. 56–62, Nov. 2013.
Congr. Expo. (ECCE), Sep. 2010, pp. 1178–1184.
[3] S. Deilami, A. S. Masoum, P. S. Moses, and M. A. S. Masoum, “Real-
time coordination of plug-in electric vehicle charging in smart grids to [26] S. Negarestani, M. Fotuhi-Firuzabad, M. Rastegar, and
minimize power losses and improve voltage profile,” IEEE Trans. Smart A. Rajabi-Ghahnavieh, “Optimal sizing of storage system in a fast
Grid, vol. 2, no. 3, pp. 456–467, Sep. 2011. charging station for plug-in hybrid electric vehicles,” IEEE Trans.
Transport. Electrific., vol. 2, no. 4, pp. 443–453, Dec. 2016.
[4] U. K. Debnath, I. Ahmad, D. Habibi, and A. Y. Saber, “Improving
battery lifetime of gridable vehicles and system reliability in the smart [27] J. Tan and L. Wang, “Real-time charging navigation of electric vehicles
grid,” IEEE Syst. J., vol. 9, no. 3, pp. 989–999, Sep. 2015. to fast charging stations: A hierarchical game approach,” IEEE Trans.
[5] K. Bhavnagri, Solar Energy Consultant. (2010). How do I Use Electricity Smart Grid, vol. 8, no. 2, pp. 846–856, Mar. 2015.
Throughout the Day—The Load Curve. Accessed: May 2018. [Online]. [28] W. Lee, L. Xiang, R. Schober, and V. W. S. Wong, “Electric vehicle
Available: https://www.solarchoice.net.au/blog/how-do-i-use-electricity- charging stations with renewable power generators: A game theoretical
throughout-the-day-the-load-curve analysis,” IEEE Trans. Smart Grid, vol. 6, no. 2, pp. 608–617, Mar. 2015.
[6] SNI Guide. (2017). Daily Trafffic Counts. Accessed: May 2018. [29] S. R. Etesami, W. Saad, N. Mandayam, and H. V. Poor. (2017). “Smart
[Online]. Available: https://www.service.nsw.gov.au/service-centre/ routing of electric vehicles for load balancing in smart grids.” [Online].
roads-maritime-services Available: https://arxiv.org/abs/1705.03805
[7] Z. Moghaddam, I. Ahmad, D. Habibi, and Q. V. Phung, “Smart charging [30] W. Yuan, J. Huang, and Y. J. A. Zhang, “Competitive charging station
strategy for electric vehicle charging stations,” IEEE Trans. Transport. pricing for plug-in electric vehicles,” IEEE Trans. Smart Grid, vol. 8,
Electrific., vol. 4, no. 1, pp. 76–88, Mar. 2018. no. 2, pp. 627–639, Mar. 2017.
[8] C. Le Floch, F. Belletti, and S. Moura, “Optimal charging of electric [31] M. Erol-Kantarci and T. M. Hussein, “Prediction-based charging of
vehicles for load shaping: A dual-splitting framework with explicit PHEVs from the smart grid with dynamic pricing,” in Proc. IEEE Local
convergence bounds,” IEEE Trans. Transport. Electrific., vol. 2, no. 2, Comput. Netw. Conf. (LCN), Oct. 2010, pp. 1032–1039.
pp. 190–199, Jun. 2016. [32] J. Hu, C. Si, M. Lind, and R. Yu, “Preventing distribution grid conges-
[9] A. Baniasadi, D. Habibi, O. Bass, and M. A. S. Masoum, “Optimal real- tion by integrating indirect control in a hierarchical electric vehicles’
time residential thermal energy management for peak-load shifting with management system,” IEEE Trans. Transport. Electrific., vol. 2, no. 3,
experimental verification,” IEEE Trans. Smart Grid, to be published. pp. 290–299, Sep. 2016.
doi: 10.1109/TSG.2018.2887232. [33] Y. Kim, J. Kwak, and S. Chong, “Dynamic pricing, scheduling, and
[10] C. Luo, Y.-F. Huang, and V. Gupta, “Placement of EV charging energy management for profit maximization in PHEV charging sta-
stations—Balancing benefits among multiple entities,” IEEE Trans. tions,” IEEE Trans. Veh. Technol., vol. 66, no. 2, pp. 1011–1026,
Smart Grid, vol. 8, no. 2, pp. 759–768, Mar. 2017. Feb. 2017.
[11] Z. Yi and P. H. Bauer, “Spatiotemporal energy demand models [34] M. J. Neely, C. E. Rohrs, and E. Modiano, “Equivalent models for
for electric vehicles,” IEEE Trans. Veh. Technol., vol. 65, no. 3, queueing analysis of deterministic service time tree networks,” IEEE
pp. 1030–1042, Mar. 2016. Trans. Inf. Theory, vol. 51, no. 10, pp. 3576–3584, Oct. 2005.
[12] V. del Razo and H.-A. Jacobsen, “Smart charging schedules for highway [35] M. Keskin and B. Çatay, “Partial recharge strategies for the electric
travel with electric vehicles,” IEEE Trans. Transport. Electrific., vol. 2, vehicle routing problem with time windows,” Transp. Res. C, Emerg.
no. 2, pp. 160–173, Jun. 2016. Technol., vol. 65, pp. 111–127, Apr. 2016.
238 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 5, NO. 1, MARCH 2019

[36] F. Wang, H. Xu, T. Xu, K. Li, M. Shafie-Khah, and J. P. Catalão, Iftekhar Ahmad (M’07) received the Ph.D. degree
“The values of market-based demand response on improving power in communication networks from Monash Univer-
system reliability under extreme circumstances,” Appl. Energy, vol. 193, sity, Clayton, VIC, Australia, in 2007.
pp. 220–231, May 2017. He is currently an Associate Professor with the
[37] A. S. B. Humayd and K. Bhattacharya, “Design of optimal incentives School of Engineering, Edith Cowan University,
for smart charging considering utility-customer interactions and dis- Joondalup, WA, Australia. His current research inter-
tribution systems impact,” IEEE Trans. Smart Grid, to be published. ests include 5G technologies, green communica-
doi: 10.1109/TSG.2017.2771508. tions, quality of service in communication networks,
[38] R. Christensen, Plane Answers to Complex Questions: The Theory of software-defined radio, wireless sensor networks,
Linear Models. Albuquerque, NM, USA: Springer, 2011. and computational intelligence.
[39] J. G. Proakis and D. G. Manolakis, Digital Signal Processing: Prin-
ciples, Algorithms, and Applications. Parsippany, NJ, USA: Simon &
Schuster Company, 1992. Daryoush Habibi (M’95–SM’99) received the B.E.
[40] Z. Li, K. Dey, M. Chowdhury, and P. Bhavsar, “Connectivity sup- degree (Hons.) in electrical engineering and the
ported dynamic routing of electric vehicles in an inductively coupled Ph.D. degree from the University of Tasmania,
power transfer environment,” IET Intell. Transport Syst., vol. 10, no. 5, Hobart, TAS, Australia, in 1989 and 1994, respec-
pp. 370–377, Jun. 2016. tively.
[41] SDGE. (2017). Ev Rates can Help You Save. Accessed: Feb. 2018. His employment history includes Telstra Research
[Online]. Available: https://www.sdge.com Laboratories, Melbourne, VIC, Australia, Flinders
[42] M. Kathryn. (2016). How Much Does it Cost to Charge an Electric Car?. University, Adelaide, SA, Australia, Intelligent Pix-
Accessed: Oct. 2018. [Online]. Available: https://pluginamerica.org els, Inc., and Edith Cowan University, Joondalup,
[43] (Jul. 2017). CHAdeMO User Manual. Accessed: May 2018. [Online]. WA, Australia. He is currently a Professor, the Pro
Available: http://store.evtv.me/proddetail.php?prod=20kwchademo& Vice-Chancellor, and the Executive Dean of Engi-
cat=23 neering with Edith Cowan University. His current research interests include
[44] P. Morrissey, P. Weldon, and M. O’Mahony, “Future standard and fast engineering design for sustainable development, smart energy systems, envi-
charging infrastructure planning: An analysis of electric vehicle charging ronmental monitoring technologies, and reliability and quality of service in
behaviour,” Energy Policy, vol. 89, pp. 257–270, Feb. 2016. communication systems and networks.
[45] M. Mahoor, Z. S. Hosseini, A. Khodaei, and D. Kushner. (2017). “Elec- Dr. Habibi is a fellow of Engineers Australia and the Institute of Marine
tric vehicle battery swapping station.” [Online]. Available: https://arxiv. Engineering, Science and Technology.
org/abs/1710.06895
[46] (Nov. 2018). Electricity Price and Demand. Accessed: Nov. 2018.
[Online]. Available: https://www.aemo.com.au/Electricity/National- Mohammad A. S. Masoum (S’91–M’91–SM’05)
Electricity-Market-NEM/Data-dashboard received the B.S. and M.S. degrees from the Uni-
[47] (Nov. 2018). Transport Roads Maritime Services. Accessed: Nov. 2018. versity of Colorado, Denver, CO, USA, in 1983 and
[Online]. Available: https://www.rms.nsw.gov.au/about/corporate- 1985, respectively, and the Ph.D. degree from the
publications/statistics/traffic-volumes/index.html University of Colorado, Boulder, CO, USA, in 1991,
all in electrical and computer engineering.
He was with the Iran University of Science
and Technology, Tehran, Iran, from 1993 to 2003,
Zeinab Moghaddam (S’15) received the M.Sc. and Curtin University, Perth, WA, Australia, from
degree in embedded system design from the Univer- 2004 to 2018. He is currently an Associate Professor
sity of Lugano, Lugano, Switzerland, in 2012. She is with the Department of Engineering, Utah Valley
currently pursuing the Ph.D. degree with the Smart University, Orem, UT, USA. He has published over 300 papers, including
Energy Research Group, School of Engineering, over 120 journal articles. He has co-authored Power Quality in Power Systems
Edith Cowan University, Joondalup, WA, Australia. and Electrical Machines (Elsevier, 2008 and 2015) and Power Conversion of
Her current research interests include the manage- Renewable Energy Systems (Springer, 2011 and 2012).
ment of smart grid and renewable energy systems Dr. Masoum is an Editor of the IEEE T RANSACTIONS ON S MART G RID
with a focus on electric vehicles applications. and IEEE P OWER E NGINEERING L ETTERS and the Guest Editor of the IEEE
T RANSACTIONS ON I NDUSTRIAL I NFORMATICS and IET Renewable Power
Generation.