3386 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO.

9, SEPTEMBER 2019

Optimal Charging Scheduling by Pricing for EV

Charging Station With Dual Charging Modes
Yongmin Zhang , Member, IEEE, Pengcheng You , Student Member, IEEE,
and Lin Cai , Senior Member, IEEE

Abstract— With the increasing penetration of electric vehicles However, due to limited cruising range, EVs may require
(EVs) and various user preferences, charging stations often frequent recharging when they travel to a faraway destination,
provide several different charging modes to satisfy the various and hence charging convenience is one of the most impor-
requirements of EVs. How to effectively utilize the charging
capacity to minimize the service dropping rate is a pressing tant concerns for EV owners. In addition, due to the high
and open issue for charging stations. Given that EV owners cost of battery replacement, battery lifetime-related cost is
are price-sensitive to the charging modes, we intend to design another important concern for EV owners. Given the slow
an optimal pricing scheme to minimize the service dropping evolution of battery technologies, the key issue to is how
rate of the charging station. First, we formulate the operation to address these challenges without waiting for new battery
of a dual-mode charging station as a queuing network with
multiple servers and heterogeneous service rates, and analyze the technologies.
relationship between the service dropping rate of the charging Charging stations play an important role in providing
station and the selections of EVs. Then, we formulate a customer charging services to EVs. Typically, there exist two dif-
attrition minimization problem to minimize the number of EVs ferent charging modes for conventional public chargers:
that leave the charging station without being charged and propose i) AC Level II (charging power typically is 10-22 kW); and
an optimal pricing approach to guide and coordinate the charging
processes of EVs in the charging station. The simulation has been ii) Direct Current Quick Charging (DCQC) (charging power
conducted to evaluate the performance of the proposed charging typically is 50-120 kW). Given the EV charging requirement,
scheduling scheme and show the efficiency of the proposed pricing AC Level II has a longer charging duration while DCQC
scheme. has a shorter charging duration, which may further reduce
Index Terms— Charging stations, charging modes, queuing battery lifetime. Because of deployment cost concerns, both
theory, charging scheduling, pricing. the number of charging stations and the number of chargers
in charging stations are limited. How to use limited charging
station resources to satisfy the various EV charging require-
I. I NTRODUCTION ments has attracted attention, such as developing intelligent
charging station architectures [3]–[6], and optimizing the loca-
E LECTRIC vehicles (EVs) have been considered to be a
key technology to cut down the massive greenhouse gas
emissions from the transportation sector, and they are also
tion and sizing of charging stations [7]–[11]. Nevertheless,
the non-cooperative charging behaviors of EVs will lead to
expected to mitigate the fossil fuels scarcity problem [1]. congestion at charging stations and reduce their operational
Thanks to the policies and plans for promoting EVs from efficiency.
the regions and countries worldwide (e.g., the sales of EVs To guide and coordinate the charging behaviors of EVs,
including PHEVs in US will reach 50% of total sales of mobile there have been extensive researches on developing charging
vehicles by 2030, and Europe has the similar targets [2]), scheduling schemes, for EV charging stations [12]–[17], for
the amount of EVs is expected to reach a sizable market share battery swapping stations [18]–[20], and for charging sta-
in the next decade. tions with renewable energy [21]–[24]. However, these works
assumed that all the chargers in the charging station are using
Manuscript received October 19, 2017; revised August 13, 2018; accepted the similar charging mode, such as AC Level II or DCQC.
October 5, 2018. Date of publication November 2, 2018; date of current
version August 27, 2019. This work was supported in part by the Natural Since different EVs may have different charging behaviors
Sciences and Engineering Research Council of Canada (NSERC) and in part and charging service requirements, i.e., short charging dura-
by NSFC under Grant 61702450 and Grant 61629302. The Associate Editor tion or long battery lifetime or both of them, the charging
for this paper was E. Kosmatopoulos. (Corresponding author: Lin Cai.)
Y. Zhang is with the State Key Laboratory of Industrial Control Technology station with single charging mode has a low flexibility and
and Innovation Joint Research Center for Industrial Cyber Physical Systems, adaptability to satisfy the requirements of multi-class cus-
Zhejiang University, Hangzhou 310027, China, and also with the Department tomers, and thus limits the service quality.
of Electrical and Computer Engineering, University of Victoria, Victoria,
BC V8P 5C2, Canada (e-mail: [email protected]). To deal with the various EV charging requirements,
P. You is with the Department of Mechanical Engineering, Johns Hopkins the charging station can install two types of chargers, such
University, Baltimore, MD 21218 USA (e-mail: [email protected]). as part of them with the AC Level II mode and part of
L. Cai is with the Department of Electrical and Computer Engineering,
University of Victoria, Victoria, BC V8P 5C2, Canada (e-mail: [email protected]). them with the DCQC mode. Consequently, the charging station
Digital Object Identifier 10.1109/TITS.2018.2876287 can provide different charging services to different classes of
1524-9050 © 2018 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.

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL CHARGING SCHEDULING BY PRICING FOR EV CHARGING STATION WITH DUAL CHARGING MODES 3387

customers based on their behaviors. Furthermore, guiding the

EV owners to select adequate charging modes can reduce the
congestion and improve the service quality of the charging
station. This motives us to design an optimal scheduling
scheme for the charging station with dual charging modes
to minimize the service dropping rate. To the best of our
knowledge, this is the first paper addressing the charging
scheduling problem for the charging station with dual charging
modes by designing an optimal pricing scheme.
Generally, the selections of EV owners depend on several
factors, i.e., service fee, charging duration, waiting time,
etc [12]. Given that EV owners are price sensitive, we analyze
the relationship between the service dropping rate and the
service rate of the charging station and then design an optimal
pricing scheme to guide and coordinate the charging processes
of EVs, such that the number of EVs that leave the charging
station without being charged can be minimized and the
operation efficiency of the charging station can be improved.
The contributions of this papers can be summarized as follows:
• We model the operation processes of the charging station
with dual charging modes as a queuing network with
multiple servers and heterogeneous service rates.
• Based on the queuing theory, we analyze the relationship
between the selections of EVs and the service dropping
rate of the charging station, and prove that the service Fig. 1. Operation model and queuing model for the charging station. (a)
dropping rate is a convex function of the service rates. Operation model. (b) Queuing model.
• We formulate a customer attrition minimization problem
for the charging station and propose an optimal pricing
scheme to guide and coordinate the charging processes of EVs are service fee, waiting time, battery lifetime-related
of EVs to minimize the service dropping rate. cost, etc. In this paper, we assume that the service fee and
• Simulation results show that the proposed pricing scheme the battery lifetime-related cost will affect the EVs’ selections
can reduce the service dropping rate of the charging while the waiting time will affect the service dropping rate
station with dual charging modes significantly comparing of the charging station. When the corresponding queue length
to the charging station with a single charging mode. is too long, EV will leave the charging station without being
The rest of the paper is organized as follows: Section II charged. Otherwise, EV will be charged by one charger with
presents the operation models for the charging station, and the selected charging mode immediately or late.
formulates the charging scheduling problem as a customer For each EV, it needs to select one charging mode when it
attrition minimization problem based on queuing theory. The arrives at the charging station. Then, it will be charged when
relationship between the selections of EVs and the service there is at least one available charger with the selected charg-
dropping rate of the charging station is analyzed, and an ing mode; otherwise, it needs to wait in the corresponding
optimal pricing scheme is designed by utilizing the price sensi- queue until one charger with the selected charging mode is
tiveness of EV owners in Section III. Section IV demonstrates available, or leaves the charging station directly without being
the operational performance analysis based on simulation charged if the corresponding queue length is too long. Here,
results. Finally, Section V concludes our work. the percentage of EVs that leave the charging station directly
denotes the service dropping rate, which is one of the main
factors related to the Quality of Service (QoS) of the charging
II. S YSTEM M ODEL AND P ROBLEM F ORMULATION station. Our objective is to design an optimal pricing scheme
Considering a charging station, there are N1 AC chargers based on EV owners’ preferences to guide them to select
and N2 DC chargers to provide charging services for connected adequate charging modes, such that the service dropping rate
EVs.1 Given the charging demand of one EV, both the service of the charging station can be minimized. The operation model
fee and the battery lifetime-related cost of the AC chargers and the queuing model for the charging station are shown
are much lower than the DC chargers. But, the charging in Fig. 1. A summary of notations is given in Table I.
duration for the AC chargers is much longer. As shown in Note that, given the ubiquitous communication networks,
[12] and [25], the dominant factors that affect the selections in the future, EVs and charging stations can exchange pricing
and waiting time information remotely. Such that, an EV can
1 For simplicity, we use the AC charger and the DC charger to denote the
decide which charging mode it prefers and whether go to
charger with the AC Level II charging mode and the charger with the DCQC
charging mode, and the AC mode and the DC mode instead of the AC Level the charging station for charging service or not based on its
II charging mode and the DCQC charging mode in this paper, respectively. QoS. In this scenario, the queuing model and the proposed

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
3388 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 9, SEPTEMBER 2019

TABLE I Let λtA and λtD denote the expected number of EVs that select
N OTATION D EFINITIONS the AC mode and the DC mode during time slot t, respectively.
We have
λtA = λt p̄tA and λtD = λt p̄tD . (3)
Generally, based on the existing data analysis in [27],
the EV charging requirements follow a lognormal distribution
with a mean value E and a standard deviation of E d . Thus,
we assume that the expected charging requirement of each EV
is E. Let T A and T D denote the expected service time for one
AC charger and one DC charger to fully charge a battery with
the capacity of B C , respectively. T A > T D always holds. The
mean service times for one EV selecting the AC mode and
the DC mode are E T A /B C and E T D /B C , respectively.
Based on the selections of EVs, they enter two separate
queues: 1) one queue for EVs that select the AC mode, denoted
by Q tA ; 2) another queue for EVs that select the DC mode,
denoted by Q tD . The queue lengths of Q tA and Q tD at time
slot t can be given by the following equations:
Q tA = max{0, Q tA−1 + λtA − VtA − OtA − L tA }, (4)
Q tD = max{0, Q tD−1 + λtD D
− Vt − OtD − L tD }, (5)
where Vt A and VtD respectively denote the total number of
the available AC chargers and DC chargers at slot t, OtA and
OtD respectively denote the number of EVs that have been
fully recharged by the AC chargers and the DC chargers and
leave the charging station during time slot t, and L tA and L tD
respectively denote the number of EVs that select the AC mode
and the DC mode and leave the charging station without being
charged due to the long corresponding queue length.
In this paper, the maximum queue length is used to denote
the maximal tolerable queue length of the EV owners, denoted
pricing algorithm in this paper can still be applicable, using
by Q̄ tA and Q̄ tD respectively. If Q tA ≥ Q̄ tA (Q tD ≥ Q̄ tD ),
the concept of virtual queues.
the coming EVs that select the AC (DC) mode will leave the
charging station without being charged; otherwise, they will
A. Operation Model of Charging Stations be charged immediately or wait in the corresponding queue.
Considering the realistic scenarios that the average EV Let L t denote the service dropping rate of the charging station
arrival rate during peak vs. off-peak hours may be different, during time slot t. The expected value of L t can be given by
we divide the whole day into T time slots (i.e., an hour is
one time slot), and assume that the average EV arrival rate L t = L tA + L tD , (6)
within the time slot remains the same, while that for different where
time slots may change. Let t, t = 1, 2, . . . , T , denote t-th time
slot. The arrival of EVs at each time slot follows a Poisson L tA = λtA Pr {Q tA |Q tA ≥ Q̄ tA }, (7)
process with the average rate of λt during time slot t [26]. The L tD = λtD Pr {Q tD |Q tD ≥ Q̄ tD }. (8)
probability for n EVs arriving at the charging station during
time slot t is given by Obviously, the service dropping rate L t depends on the
selections of EVs λtA and λtD , and the queue lengths Q tA and
eλt (λt )n Q tD , as well as the maximum queue lengths Q̄ tA and Q̄ tD . Since
P{n} = , n = 0, 1, 2, · · · .
n! the values of Q̄ tA and Q̄ tD are determined by EV owners, it is
When EVs arrive at the charging station, they will select difficult for the charging station to change them. According to
their charging modes based on their preferences and the (4) and (5), Q tA and Q tD mainly depend on the selections of
corresponding service fee. Let p̄tA and p̄tD respectively denote EVs λtA and λtD , respectively, which can be tuned to reduce
the probability for one EV selecting the AC mode and the DC the service dropping rate L t .
mode during time slot t, respectively. The values of p̄tA and
p̄tD satisfy
B. Selection Model of EVs
p̄tA + p̄tD = 1, (1) Generally, due to the high construction cost of the DC mode,
0 ≤ p̄tA , p̄tD ≤ 1. (2) the service fee of the DC mode is much higher than that for

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL CHARGING SCHEDULING BY PRICING FOR EV CHARGING STATION WITH DUAL CHARGING MODES 3389

the AC mode. Furthermore, the battery lifetime-related cost for minimization problem. Since EV arrives at different time slot
the AC mode is much lower than that for the DC mode [28]. follow a Poisson process, which has independent increments,
Different EVs may have the different preferences, but are price we just need to minimize the service dropping rate of the
sensitive [16]. Since different charging stations always have charging station during each time slot, such that the total
the similar service fee of the AC mode, we set the service fee service dropping rate can be minimized. By now, the customer
of the AC mode as a constant and adjust the service fee of attrition minimization problem can be formulated as follows:
the DC mode.
Let CtA and CtD denote the service fee per kWh during time P0: min L t = λt p̄tL ,
C tD
slot t, and C BA and C BD denote the battery lifetime-related cost D
per kWh, with the AC mode and the DC mode, respectively. s.t. C D ≤ CtD ≤ C ,
Here, CtA < CtD and C BA < C BD always hold. Generally, p̄tD = −aCtD + b, (14)
the service fee of the DC mode can only be adjusted in a + p̄tD =
p̄tA 1, (15)
D
given range. Let C D and C denote the lower bound and upper
1 ≤ p̄tA , p̄tD
≤ 0, (16)
bound on the service fee of the DC charger. Thus, the service
fee CtD should always satisfy (4), (5), and (12). (17)
D The objective is to minimize the service dropping rate L t by
C D ≤ CtD ≤ C . (9)
adjusting the service fee CtD . The first constraint defines the
As mentioned above, the selections of EVs mainly depend available range of CtD . The constraints (14)-(16) describe the
on the service fees CtA and CtD and the battery lifetime-related relationship between the service fee CtD and the selections of
costs C BA and C BD . Specifically, in this paper, we define p̄tA and EVs, p̄tA and p̄tD . The other constraints show relationships
p̄tD as follows: among p̄tL , Q tA , Q tD , p̄tA , and p̄tD .
p̄tD = −aCtD + b, (10)
III. O PTIMAL EV C HARGING S CHEDULING S CHEME
p̄tA = 1− p̄tD , (11)
To solve the customer attrition minimization problem,
D
where a = A 1
, b = aC , and β denotes the we first analyze the performance measure of the queue system
C t +β−(C BD −C BA )
service time reduction cost of the DC mode comparing with to explore the possible range of optimal price and the rela-
the AC mode. It can be seen that a is a price-dependent term tionship between the service dropping rate L t and the service
and b is a price-independent term. fee CtD . Then, we transform the primal problem into a convex
D optimization problem. At last, we propose an optimal pricing
To ensure 0 ≤ p̄tA , p̄tD ≤ 1, we set C D = C − (CtA + β −
(C BD − C BA )), which usually is larger than 0. Such that, when scheme to guide the EVs to select adequate charging modes,
CtD equals its lower bound C D , all the coming EVs will select such that the total service dropping rate of the charging station
D can be minimized.
the DC mode, i.e., p̄tD = 1 and p̄tA = 0; when CtD = C ,
According to the definitions of p̄tA and p̄tD , it can be found
all the coming EVs will select the AC mode, i.e., p̄t = 0 D
that the relationships among λtA , λtD and CtD are linear. For
and p̄tA = 1. Note that, our proposed algorithm also applies
any given service fee CtD and arrival rate λt , λtA and λtD can
to other probability models, such as linear or concave ones.
be calculated according to (3), (10) and (11). Also, CtD and
λtA can be obtained when the value of λtD is given. Hence,
C. Service Dropping Rate of Charging Station in the following sections, we can replace the variable CtD by
Let p̄tL denote the probability for EVs leaving the charging λtD for easy computation.
station without being charged during time slot t. The expected
values of p̄tL can be given by A. Performance Measures Based on Queue Theory
p̄tL = Pr {Q tA |Q tA ≥ Q̄ tA } p̄tA + Pr {Q tD |Q tD ≥ Q̄ tD } p̄tD . (12) Let μ A and μ D denote the average service rate for one AC
charger and that for one DC charger, respectively. Based on the
Thus, the expected value of the service dropping rate L t can system model, the values of μ A and μ D , and their relationship
be rewritten as can be given by
L t = λt p̄tL , (13) BC
μA = , (18)
due to λtA = λt p̄tA and λtD = λt p̄tD . It can be found that ET A
the service dropping rate L t can be reduced by adjusting the BC
μD = , (19)
service fee CtD , which affects the values of p̄tA and p̄tD . ET D
TD
μA = A μD. (20)
D. Customer Attrition Minimization Problem T
The service dropping rate is not only an important parameter Obviously, μ A < μ D since T D < T A .
for the loss of benefit, but also one of the major factors for the The charging processes of EVs at the charging station
satisfaction of EV owners. To minimize the service dropping can be formulated as two independent queuing networks,
rate of the charging station, we formulate a customer attrition i.e., M/M/N1 with ρtA = λtA /(N1 μ A ) and M/M/N2 with

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
3390 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 9, SEPTEMBER 2019

ρtD = λtD /(N2 μ D ), respectively. Here, ρtA and ρtD are the CASE II: If λt < N1 μ A + N2 μ D , it means that there exists
corresponding utilization factors for different charging modes. an optimal service fee ĈtD that can satisfy ρtA < 1 and ρtD < 1
For the optimal service fee ĈtD , we have the following simultaneously. If ρtA < 1 and ρtD < 1, it means that λtA <
Theorem: N1 μ A and λtD < N2 μd always hold, and the service dropping
Theorem 1: For any given arrival rate λt , the available rate L t can be calculated by
range for the optimal service fee ĈtD can be given by
⎧ L t = λtA Pr {Q tA |Q tA ≥ Q̄ tA } + λtD Pr {Q tD |Q tD ≥ Q̄ tD }
⎪
⎪ N1 μ A + bλt − λt = λt − λtA Pr {Q tA |Q tA < Q̄ tA } − λtD Pr {Q tD |Q tD < Q̄ tD }.
⎪
⎪
⎪
⎪ aλt
⎪
⎪ (24)
⎪
⎪ bλt − N2 μ D
⎨ ≤ ĈtD ≤ , if λt ≥ N1 μ A + N2 μ D ;
aλt It can be found that (24)<(21) and (24)<(23) always hold.
⎪ bλt − N2 μ
⎪ D
⎪ Hence, the optimal service fee ĈtD should satisfy
⎪
⎪
⎪
⎪ aλt
⎪
⎪ N1 μ A + bλt − λt
⎪ < Ĉ D < N1 μ + bλt − λt , otherwise.
A
⎩ t λt p̄tA < N1 μ A ⇒ ĈtD < ,
aλt aλt
Proof: According to the queuing theory, if with infinite bλt − N2 μ D
λt p̄tD < N2 μ D ⇒ ĈtD > .
buffer, the necessary conditions for these two queue systems aλt
to be stable are ρtA < 1 and ρtD < 1. If the mean arrival
The available range for the optimal service fee ĈtD is obtained.
rate is greater than the mean service rate, the necessary
conditions for the stable queue systems cannot be satisfied,
According to the available range for the optimal service
which makes the server keeping busy and the queue growing
fee ĈtD given by Theorem 1, if the arrival rate λt satisfies
without bound2 [29]. Thus, based on the arrival rate λt and
λt ≥ N1 μ A + N2 μ D , we have the following Lemma for the
the service rate, we classify the available range of ĈtD into the
optimal service fee ĈtD :
following two cases:
Lemma 1: The optimal service fee ĈtD can be any value
CASE I: If λt ≥ N1 μ A + N2 μ D , there is no available
in the available range [ N1 μ aλ
A +bλ −λ
t bλt −N2 μ
D
λ p̄ A λ p̄ D t
, ] when λt ≥
service fee satisfying Nt μt A < 1 and Nt μt D < 1 simultaneously. t aλt
1 2 N1 μ A + N2 μ D .
If ρtA ≥ 1, the chargers with the AC mode will keep
Proof: For the service dropping rate L t given by (22),
busy since their queuing system becomes overloaded and
it can be found that the minimal value of the service dropping
the service dropping rate will increase with respect to λtA ,
rate L t is a constant when λt ≥ N1 μ A + N2 μ D , and the only
and L tA ≈ λtA − N1 μ A ; otherwise, the service dropping rate
condition for the optimal service fee ĈtD is to make sure of
L tA ≈ λtA Pr {Q tA |Q tA ≥ Q̄ tA } = λtA (1 − Pr {Q tA |Q tA < Q̄ tA }).
that ρtA ≥ 1 and ρtD ≥ 1 are satisfied simultaneously. Thus,
The parameters for L tD are similar to those for L tA . Hence,
the optimal service fee ĈtD can be any value in the available
the service dropping rate L t can be given by the following
range [ N1 μ aλ
A +bλ −λ
t bλt −N2 μ
D
cases: t
t
, aλt ].
Therefore, we only need to design the optimal pricing scheme
• If λtA > N1 μ A and λtD < N2 μ D , we have
for the problem when λt < N1 μ A + N2 μ D .
L t ≈ λt − N1 μ A − λtD Pr {Q tD |Q tD < Q̄ tD }; (21) Let n 1 and n 2 denote the steady state of the charging
processes of the AC mode and the DC mode, in which there are
• If λtA ≥ N1 μ A and λtD ≥ N2 μ D , we have n 1 and n 2 customers in their corresponding systems, include
the customers in service, respectively. Let ptA (n 1 ) denote the
L t ≈ λt − N1 μ A − N2 μ D ; (22)
probability that in steady state the number of customers present
• If λtA < N1 μ A and λtD > N2 μ D , we have in the AC mode is n 1 during time slot t and ptD (n 2 ) denote
the probability that in steady state the number of customers
L t ≈ λt − N2 μ D − λtA Pr {Q tA |Q tA < Q̄ tA }. (23) present in the DC mode is n 2 during time slot t. Then, we have
⎧
Since Pr {Q tA |Q tA < Q̄ tA } < 1 and Pr {Q tD |Q tD < Q̄ tD } < 1, ⎪ (N1 ρtA )n
⎪
⎨ ptA (0) , if n ≤ N1 ;
(22)<(21) and (22)<(23) always hold. To minimize the service
ptA (n) = n!n (25)
dropping rate L t , ρtA ≥ 1 and ρtD ≥ 1 should be satisfied ⎪ (ρ A ) N1 1N
⎪
⎩ ptA (0) t , if n ≥ N1 ;
simultaneously. Thus, the available range for the optimal
⎧ N1 !
service fee ĈtD is
⎪
⎪ (N 2 ρt )
D n
⎨ ptD (0) , if n ≤ N2 ;
N1 μ A + bλt − λt n!n
λt p̄tA ≥ N1 μ A ⇒ ĈtD ≥ , ptD (n) = N2 (26)
aλt ⎪
⎪ (ρ ) N2
D
⎩ ptD (0) t , if n ≥ N2 ;
bλt − N2 μ D N2 !
λt p̄tD ≥ N2 μ D ⇒ ĈtD ≤ .
aλt where
1 −1
N
2 In a realistic queuing system, the buffer size is limited, so the queue length
(N1 ρtA )n (N1 ρtA ) N1 1
will be limited, while a portion of the arrivals will be blocked when the buffer ptA (0) = [ + ]−1 , (27)
is full. n! N1 ! 1 − ρtA
n=0

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL CHARGING SCHEDULING BY PRICING FOR EV CHARGING STATION WITH DUAL CHARGING MODES 3391

2 −1
N
(N2 ρtD )n (N2 ρtD ) N2 1 where
ptD (0) = [ + ]−1 . (28)
n! N2 ! 1 − ρtD 1
n=0 L tA = N1 μ A ρtA ptA (n 1 )
In this paper, we use the steady-state distribution of EVs to 1 − ρtA

model the service processes of the charging station and solve A A N1N1 (ρtA )n1 +1 1
= N1 μ pt (0) , (33)
the customer attrition minimization problem. N1 ! 1 − ρtA
1
B. Service Dropping Rate of Charging Station L tD = N2 μ D ρtD ptD (n 2 )
1 − ρtD
Since the maximal queue lengths Q̄ tA and Q̄ tD for EVs 
N2N2 (ρtD )n2 +1 1
during time slot t are constant, the service dropping rate L t = N2 μ D ptD (0) . (34)
can be given by N2 ! 1 − ρtD
∞
 ∞
 It can be found that L tA and L tD have the similar structure.
Lt = λtA ptA (n 1 ) + λtD ptD (n 2 ) (29) Thus, we first prove that the second part L tD is an increasing
n 1 =N1 + Q̄ tA n 2 =N2 + Q̄ tD and convex function of λtD , and then we can derive that the
∞ A first part L tA is a decreasing and convex function of λtD due
λtA ( ptA (N1 + Q̄ tA ))(N1 + Q̄ t )
where λ A p A (n )
n 1 =N1 + Q̄ t t t
A 1 = A to λtA = λt − λtD . Because ρtA is a linear function of λtA and
1−ρt
denotes the blocking traffic for the queue of the AC mode such a variable substitution will not change the convexity of
 D
λ D ( p D (N + Q̄ D ))(N2 + Q̄ t ) the objective function, we take L tD as the objective function
and ∞ λ D p D (n 2 ) = t t 2 tD denotes
n 2 =N2 + Q̄ tD t t 1−ρt and ρtD as the variable in the following part. For the second
the blocking traffic for the queue of the DC mode. According part L tD , we have the following theorem:
to the definitions of ptA (n) and ptD (n) given by (25) and (26), Theorem 2: The service dropping rate L tD is an increasing
it can be found that ptA (n) and ptD (n) also depend on the and convex function of the service rate ρtD .
values of λtA and λtD . Hence, the service dropping rate L t Proof: The proof can be found in Appendix A.
mainly depends on the values of λtA and λtD . Since L tA and L tD have the similar structure and λtA = λt −
λtD , we have the following Lemma for the relationship between
C. Problem Transformation L tA and ρtD :
According to Theorem 1 and the relationship among λtA , Lemma 2: The service dropping rate L tA is a decreasing and
λtD , p̄tA , p̄tD and CtD
given by (3), (10) and (11), the optimal convex function of the service rate ρtD .
λtD should satisfy Proof: The proof can be found in Appendix B.
Theorem 3: The service dropping rate L t is a convex func-
λt − N1 μ A < λtD < N2 μ D . (30)
tion of λtD .
By now, the primal Problem P0 can be transformed to the ∂ 2 L tA ∂ 2 L tD
Proof: Since L t = L tA + L tD , D 2 > 0 and D 2 > 0,
∂(ρt ) ∂(ρt )
following problem P1, in which the variable is λtD , i.e., ∂λtD
∂2 Lt
we have > 0. Since λtD = ρtD N2 μ D , = N2 μ D > 0
P1: min L t , (31) ∂(ρtD )2 ∂ρtD
always holds. Hence, ∂ DL t2 = (N2 μ D )2 ∂ DL t2 > 0. Thus,
2 2
λtD
∂(λt ) ∂(ρt )
s.t. λt − N1 μ A < λtD < N2 μ D . (32) the service dropping rate L t is a convex function of λtD .
In this problem, the goal is to find the optimal λtD to minimize Because the relationship between λtD and CtD is linear, the ser-
the service dropping rate L t , and the available range for λtD vice dropping rate L t also is a convex function of CtD .
is given by constraint (32). If the objective function L t is
a convex function of λtD , Problem P1 can be transformed E. Optimal Pricing Scheme
into a typical convex optimization problem. Thus, we first Since the objective function L t is a convex function of
establish that, if the problem is a convex optimization problem, λtD and the constraint (32) is a linear constraint for λtD ,
the solution of this problem is indeed toward the global the transformed Problem P1 is a convex optimization problem.
optimum [30]. To solve this problem, we first prove that Since this optimization problem will be solved by the charging
convexity of the service dropping rate L t with respect to λtD , station, it can be solved using existing centralized tools, such
and then propose an optimal pricing scheme to obtain the as fmincon function [31] or CVX toolbox [32] in Matlab.
optimal price ĈtD . We omit the details of how to solve this problem.
By solving Problem P1, the optimal λtD can be obtained,
D. Proof of Convexity then the optimal pricing ĈtD can be calculated by
According to (29), the service dropping rate L t is
b λD
1 1 ĈtD = − t . (35)
L t = λtA ptA (n 1 ) + λtD ptD (n 2 ) , a aλt
1 − ρtA 1 − ρtD
The process for obtaining the optimal price ĈtD can be
where n 1 = N1 + Q̄ tA and n 2 = N2 + Q̄ tD . Due to λtA =
sketched as Algorithm 1.
N1 μ A ρtA and λtD = N2 μ D ρtD , we have
Theorem 4: The minimal L t can be achieved by the optimal
L t = L tA + L tD , pricing scheme shown in Algorithm 1.

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
3392 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 9, SEPTEMBER 2019

Algorithm 1 Optimal Pricing Scheme for the Charging Station

Initialization λt , N1 , μ A , N2 , μ D , Q̄ tA , Q̄ tD
• If λt ≥ N1 μ A + N2 μ D
Chooses λtD randomly in [N2 μ D , λt − N1 μ A ];
• Else
Obtains the optimal λtD by solving Problem P1;
• end
Calculates the optimal price ĈtD by (35);
return ĈtD Fig. 3. Optimal price ĈtD where CtA = $0.15/kW h and selections of EVs
at different time. (a) Optimal price ĈtD . (b) Selection of EVs.

Fig. 2. The EV arrivals of the charging station.

Fig. 4. The numbers of EVs that leave the charging station without being
charged and the corresponding expected queue lengths at different times.
Proof: According to Theorem 1 and Lemma 1, if λt ≥ (a) L tA and L tD . (b) Q tA and Q tD .
N1 μ A + N2 μ D , the minimal service dropping rate L t ≈ λt −
N1 μ A − N2 μ D can be obtained by any value in the available
range, i.e., ĈtD ∈ [ N1 μ aλ
A +bλ −λ
t bλt −N2 μ
D

t
t
, aλt ]. If λt < N1 μ A + EVs can only select the AC mode, i.e., N̄1 = 31 and Q̄ tA = 10;
N2 μ , since the customer attrition minimization problem is a
D
ii) charging station with the DC mode, in which all the AC
convex optimization problem according to Theorem 3, there chargers are replaced by the DC chargers, i.e., N̄2 = 23
exists an unique optimal solution, which can be obtained by and Q̄ tD = 18, while the service fee of the DC mode is
solving Problem P1 [30]. Thus, the proposed optimal pricing $0.63/kW h.
scheme in Algorithm 1 can minimize the service dropping The optimal price ĈtD of the charging station with dual
rate L t . charging modes and the expected selections of EVs can be
found in Fig. 3. It can be found that the available range for the
IV. P ERFORMANCE E VALUATION optimal price changes and the optimal price of the DC mode
In order to demonstrate the performance of the proposed is time-varying due to time-varying EV arrival rate. With the
algorithm, we take one charging station with dual charging increase of the arrival rate λt , the available range for optimal
modes and time-varying arrivals of EVs as a case study, price will be narrowed, which means that the selections for
and then analyze the effects of the number of chargers, the optimal price become less, and the optimal price ĈtD for
the maximal waiting queue length, and the arrival rate on the the DC mode will be decreased, such that more EVs will
optimal pricing of the charging station. select the DC mode. However, since the service fees in the
charging stations with single mode are constants, all the EVs
in i) charging station with the AC mode can only select the
A. Case Study
AC mode while only part of EVs in ii) charging station with
Consider a charging station with N1 = 15 AC chargers and the DC mode will select the DC mode due to its high battery
N2 = 8 DC chargers. The average service rates for each AC lifetime-related cost.
charger and DC charger are μ A = 25 per hour and μ D = 12 5 The total service dropping rates for different charging
per hour, respectively. According to [33], the cost for replacing stations and the expected queue lengths can be found in Fig. 4.
a 24kW h battery is about $5500, and the lifetime with the It can be found that the optimal pricing scheme can minimize
AC mode and the DC mode are about 650 and 500 cycles, the service dropping rate since it can guide the EVs selecting
respectively. Thus, we can derived that C BA ≈ $0.35/kW h and the suitable charging mode to improve the utilization of
C BD ≈ $0.46/kW h. The maximal queue lengths are Q̄ tA = 10 chargers in the charging station. For i) the charging station
D
and Q̄ tD = 8. We set CtA = $0.15/kW h, C = $0.8/kW h, with the AC mode, part of EVs leaves due to its limited service
β = $0.2/kW h, and E = 16kW h. The arrivals of EVs during ability and the long waiting queue, while for ii) the charging
different time slot can be found in Fig. 2. Assume that there station with the DC mode, part of EVs will not select this
exists another two construction plan: i) charging station with charging station due to its high battery lifetime-related cost and
the AC mode, in which all the DC chargers (including the high charging service fee. For the charging station with dual
waiting space) are replaced by the AC chargers, and all the charging modes under optimal pricing scheme, by adjusting

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL CHARGING SCHEDULING BY PRICING FOR EV CHARGING STATION WITH DUAL CHARGING MODES 3393

Fig. 5. Optimal solution with the increase of arrival rate λtA : a) the optimal price ĈtD ; b) the selections of EVs λtA and λtD ; c) the service dropping rates
L tA and L tD ; and d) the expected queue lengths Q tA and Q tD .

TABLE II B. Relationship Between Arrival Rate and System

T HE S ERVICE D ROPPING R ATE (U NIT: EV S ) Performance
The maximal service ability of the charging station with
dual charging modes is 25.2 EVs/hour. In order to explore the
performance of our proposed pricing scheme, we set the arrival
rate from [1, 35] and respectively show the optimal price ĈtD
and its available range, the minimal service dropping rates L tA
and L tD , the selections of EVs λtA and λtD , and the expected
queue lengths for the AC mode and the DC mode in Fig. 5.
From the simulation results in Fig. 5(a), it can be found
that, when the arrival rate of EVs is smaller than the maximal
the service fee of the DC mode, both the number of EVs service ability of the charging station, with the increase of
that leave the charging station without being charged and the the arrival rate λt , the available range for the optimal price
expected queue lengths can be minimized, especially when the becomes narrow to ensure that both of the utilization factors
arrival rate is high. ρtA and ρtD are smaller than 1, and the optimal price ĈtD
With same arrival rates and construction space, i) charging decreases; otherwise, the available range for the optimal price
station with the AC mode has the smallest service ability and becomes wider to ensure that both of the utilization factors ρtA
longest waiting queue, ii) charging station with the DC mode and ρtD are larger than 1, and the optimal price ĈtD can be any
has the largest service ability and the smallest waiting queue, value in the available range. Specially, when the arrival rate of
and charging station with dual charging modes under optimal EVs is smaller than the maximal service ability of the charging
pricing scheme has a middle service ability and waiting queue. station, the charging station needs to select an optimal pricing
In addition, i) charging station with the AC mode has the by solving Problem P1 to minimize the total service dropping
largest service dropping rate due to limited service ability, and rate for both the AC mode and the DC mode; and when the
ii) charging station with the DC mode has a higher service arrival rate of EVs is larger than the maximal service ability of
dropping rate due to its high battery lifetime-related cost, and the charging station, the charging station just needs to select
the charging station with dual charging modes under optimal one available value in the available range since any value in
pricing scheme can minimize the service dropping rate, which the available range will obtain the same service dropping rate.
has a higher flexibility and adaptability to deal with the time- Figs. 5(b)-5(d) show the selections of EVs, the service
varying arrival rate of EVs. dropping rates and the expected queue lengths, respectively.
The expected service dropping rate for different charging With the increase of EV arrival rate, the selections of EVs
stations can be found in Table II. It can be found that the for both the AC mode and the DC mode, the service dropping
charging station with dual charging modes under the proposed rate and the expected queue lengths will be increased. In order
pricing scheme can minimize the total service dropping rate of to minimize the service dropping rate, it can be found that
the charging station. The charging station with the AC mode the selections of EVs for both the AC mode and the DC
has the highest service dropping rate due to its limited charging mode increase smoothly. Also, the service dropping rate grows
service rate even when it has 31 chargers. The charging station when the arrival rate of EVs is larger than a certain threshold
with the DC mode has a middle service dropping rate due to its and the expected queue lengths reach their upper bounds of
high service fee and high battery lifetime-related cost. As we the charging station. Since the service rate of the AC mode
know, since the charging station with the DC mode has enough is much lower than the service rate of the DC mode, more
charging service ability to service more EVs, reduce its service EVs are assigned to select the DC mode to minimize the total
fee can cut down the service dropping rate. However, we found service dropping rate.
that only when the charging station with the DC mode reduce
its the service fee to $0.59/kW h, which is much lower than
the service fee in charging station with dual charging modes, C. The Effects of N1 and N2 on Service Dropping Rate L t
it has the similar service dropping rate of the charging station To explore the effect of the charging facilities in the
with dual charging modes. charging station on the service dropping rate L t , we fix the

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
3394 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 9, SEPTEMBER 2019

the multiple servers and heterogeneous service rates, and

analyzed the relationship between the service dropping rate
and the selections of EVs. Then, by making use of price
sensitiveness of EV owners, we designed an optimal pricing
scheme to guide and coordinate the charging processes of EVs
to minimize the service dropping rate of the charging station.
Simulation results are provided to demonstrate the efficiency
of the proposed charging scheduling scheme.
In our future work, we will consider the optimal pricing
Fig. 6. The minimal service dropping rate L t affected by the values of N1 scheme for the charging station, where the arrival rate of the
and N2 : (a) N1 = 15 and N2 ∈ [1, 20]; (b) N2 = 8 and N1 ∈ [1, 20].
charging station depends on the service fee and the EVs can
change their selections when the selected queue length is too
long. Also, we intend to design an algorithm to determines the
optimal number of chargers with dual charging modes based
on the distribution of EVs, which can maximize the total profit
of the charging stations and improve the service quality of the
charging station.

A PPENDIX A
Fig. 7. The minimal service dropping rate L t affected by the values of Q̄ tA Proof: For the special case when Q̄ tD = 0, according
and Q̄ tD : (a) Q̄ tA = 10 and Q̄ tD ∈ [1, 20]; (b) Q̄ tD = 8 and Q̄ tA ∈ [1, 20].
to the Erlang’s C formula [34], the loss probability for
M/M/N/N can be given by
EV arrival by λt = 22, and perform the following simula-
N2N2 (ρtD ) N2 1
tions: 1) fixing N1 = 15 while adjusting N2 from 1 to 20, B(N2 , N2 ρtD ) = ptD (0)
the minimal service dropping rate L t is shown in Fig. 6(a); N2 ! 1 − ρtD
2) fixing N2 = 8 while adjusting N1 from 1 to 20, the minimal (N2 ρtD ) N2
service dropping rate L t is shown in Fig. 6(b). It can be N2 !(1−ρtD )
= 
found that the minimal service dropping rate L t in the first N2 −1 (N2 ρtD )n (N2 ρtD ) N2
+
simulation is decreasing much faster than that in the second n=0 n! N2 !(1−ρtD )
one. That is because the service rate of each DC charger is  N2 −1
N2 !(1 − ρtD ) −1
much higher than that of each AC charger. For a charging = +1 . (36)
station with limited space, more DC chargers can reduce the n=0
n!(N2 ρtD ) N2 −n
service dropping rate, but may decrease the number of EVs
due to its high battery lifetime-related cost. The existing works [35], [36] have proved that B(N2 , N2 ρtD )
is an increasing and convex function of ρtD when
D. The Effects of Q̄ tA and Q̄ tD on Service Dropping Rate L t N2 is given. Thus, both of the first and the sec-
ond derivatives B(N2 , N2 ρtD ) with respect to ρtD are
Since the maximal queue length affects the minimal service
larger than zero, i.e., ∂ B(N2 , N2 ρtD )/∂ρtD > 0 and
dropping rate L t , we conduct several simulations to demon-
∂ 2 B(N2 , N2 ρtD )/∂(ρtD )2 > 0.
strate the effects of Q̄ tA and Q̄ tD on the minimal service
According to (34), L tD can be rewritten as
dropping rate L t . First, we fix Q̄ tA = 10 and adjust Q̄ tD
from 1 to 20, whose minimal service dropping rate L t is D +1
shown in Fig. 7(a). Then, we fix Q̄ tD = 8 and adjust Q̄ tA L tD = N2 μ D (ρtD ) Q̄ t B(N2 , N2 ρtD ). (37)
from 1 to 20, whose minimal service dropping rate L t is
illustrated in Fig. 7(b). Obviously, the service dropping rate The first derivative L tD with respect to ρtD is
L t in the first simulation will decrease more quickly than that  D +1
in the second simulation. That is because the service rate for ∂ L tD D ∂(ρt )
D Q̄ t
= N2 μ B(N2 , N2 ρtD )
the DC chargers is much higher than that of the AC charger. ∂ρtD ∂ρtD
However, increasing the waiting space of the charging station D +1 ∂ B(N2 , N2 ρtD )
may not improve the service dropping rate significantly, since + (ρtD ) Q̄ t .
∂ρtD
the service dropping rate mainly depends on the service ability
of the charging station. D
∂(ρtD ) Q̄ t +1 D
Since N2 μ D > 0, = ( Q̄ tD + 1)(ρtD ) Q̄ t > 0,
∂ρtD
V. C ONCLUSION ∂ B(N2 ,N2 ρtD )
(ρt ) Q̄ t +1
D
B(N2 , N2 ρtD ) > 0, D > 0, and ∂ρtD
> 0,
In this paper, we modeled the operation of the charging ∂ L tD
station with dual charging modes as a queuing network with we have ∂ρtD
> 0.

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL CHARGING SCHEDULING BY PRICING FOR EV CHARGING STATION WITH DUAL CHARGING MODES 3395

The second derivative L tD with respect to ρtD is [10] X. Wang, C. Yuen, N. U. Hassan, N. An, and W. Wu, “Electric vehicle
charging station placement for urban public bus systems,” IEEE Trans.
∂ 2 L tD  2 D Q̄ tD +1
D ∂ (ρt )
Intell. Transp. Syst., vol. 18, no. 1, pp. 128–139, Jan. 2017.
= N2 μ B(N2 , N2 ρtD ) [11] C. Luo, Y.-F. Huang, and V. Gupta, “Placement of EV charging
∂(ρtD )2 ∂(ρtD )2 stations—Balancing benefits among multiple entities,” IEEE Trans.
D +1 Smart Grid, vol. 8, no. 2, pp. 759–768, Mar. 2017.
∂(ρtD ) Q̄ t B(N2 , N2 ρtD )
+2 [12] E. Azadfar, V. Sreeram, and D. Harries, “The investigation of
∂ρtD ∂ρtD the major factors influencing plug-in electric vehicle driving pat-
terns and charging behaviour,” Renew. Sustain. Energy Rev., vol. 42,
∂ 2 B(N2 , N2 ρtD )
+ (ρtD ) Q̄ t +1
D pp. 1065–1076, Feb. 2015.
. (38)
∂(ρtD )2 [13] J. Timpner and L. Wolf, “Design and evaluation of charging station
scheduling strategies for electric vehicles,” IEEE Trans. Intell. Transp.
D D +1
∂ 2 (ρtD ) Q̄ t +1 D −1 ∂(ρtD ) Q̄ t Syst., vol. 15, no. 2, pp. 579–588, Apr. 2014.
Since = Q̄ tD ( Q̄ tD +1)(ρtD ) Q̄ t > 0, > [14] X. Cheng et al., “Electrified vehicles and the smart grid: The ITS
∂(ρtD )2 ∂ρtD
∂ B(N2 ,N2 ρtD ) ∂2 B(N2 ,N2 ρtD ) ∂2 L tD perspective,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 4,
0, > 0, and > 0, > 0 always pp. 1388–1404, Aug. 2014.
∂ρtD ∂(ρtD )2 ∂(ρt )2
D
holds. [15] Z. Tian et al., “Real-time charging station recommendation system for
∂ L tD ∂ 2 L tD electric-vehicle taxis,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 11,
Since > 0 and > 0 always hold, the service pp. 3098–3109, Nov. 2016.
∂ρtD ∂(ρtD )2
dropping rate L tD is an increasing and convex function of the [16] W. Yuan, J. Huang, and Y. J. A. Zhang, “Competitive charging station
pricing for plug-in electric vehicles,” IEEE Trans. Smart Grid, vol. 8,
service rate ρtD . no. 2, pp. 627–639, Nov. 2017.
[17] Y.-C. Hung and G. Michailidis, “Optimal routing for electric vehicle
A PPENDIX B service systems,” Eur. J. Oper. Res., vol. 247, no. 2, pp. 515–524, 2015.
[18] M. R. Sarker, H. Pandzic, and M. A. Ortega-Vazquez, “Optimal opera-
Proof: Since λtA = L t − λtD , ρtA = λtA /N1 μ A and ρtD = tion and services scheduling for an electric vehicle battery swapping sta-
λtD /N2 μ D , we have tion,” IEEE Trans. Power Syst., vol. 30, no. 2, pp. 901–910, Mar. 2015.
[19] X. Tan, B. Sun, and D. H. Tsang, “Queueing network models for
L t − N2 μ D ρtD electric vehicle charging station with battery swapping,” in Proc. IEEE
ρtA = , (39) SmartGridComm, Nov. 2014, pp. 1–6.
N1 μ A [20] P. You, Z. Yang, Y. Zhang, S. H. Low, and Y. Sun, “Optimal charging
schedule for a battery switching station serving electric buses,” IEEE
and the first derivation of ρtA with respect to ρtD is Trans. Power Syst., vol. 31, no. 5, pp. 3473–3483, Sep. 2016.
[21] W. Lee, L. Xiang, R. Schober, and V. W. S. Wong, “Electric vehicle
∂ρtA N2 μ D
= − . (40) charging stations with renewable power generators: A game theoretical
∂ρtD N1 μ A analysis,” IEEE Trans. Smart Grid, vol. 6, no. 2, pp. 608–617, Mar. 2015.
[22] C. Luo, Y.-F. Huang, and V. Gupta, “Stochastic dynamic pricing for EV
Since L tD has the same structure with L tA , it can be easily charging stations with renewable integration and energy storage,” IEEE
∂LD ∂2 L D ∂ρ D Trans. Smart Grid, vol. 9, no. 2, pp. 1494–1505, Mar. 2017.
proved that ∂ρ Dt > 0 and ∂(ρ D )t 2 > 0. Due to ∂ρtA < 0, [23] W. Tushar, C. Yuen, S. Huang, D. B. Smith, and H. V. Poor, “Cost
t t t
∂ L tD ∂ 2 L tD minimization of charging stations with photovoltaics: An approach with
we have < 0 and > 0. Thus, the service dropping EV classification,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 1,
∂ρtA ∂(ρtA )2
pp. 156–169, Jan. 2016.
rate L tD is a decreasing and convex function of ρtA . [24] D. Wu, H. Zeng, C. Lu, and B. Boulet, “Two-stage energy manage-
ment for office buildings with workplace EV charging and renewable
R EFERENCES energy,” IEEE Trans. Transport. Electrific., vol. 3, no. 1, pp. 225–237,
Mar. 2017.
[1] S. Habib, M. Kamran, and U. Rashid, “Impact analysis of vehicle-to-grid [25] G. Li and X.-P. Zhang, “Modeling of plug-in hybrid electric vehicle
technology and charging strategies of electric vehicles on distribution charging demand in probabilistic power flow calculations,” IEEE Trans.
networks—A review,” J. Power Sour., vol. 277, pp. 205–214, Mar. 2015. Smart Grid, vol. 3, no. 1, pp. 492–499, Mar. 2012.
[2] I. S. Bayram, A. Tajer, M. Abdallah, and K. Qaraqe, “Capacity plan- [26] M. Alizadeh, A. Scaglione, J. Davies, and K. S. Kurani, “A scalable
ning frameworks for electric vehicle charging stations with multiclass stochastic model for the electricity demand of electric and plug-in
customers,” IEEE Trans. Smart Grid, vol. 6, no. 4, pp. 1934–1943, hybrid vehicles,” IEEE Trans. Smart Grid, vol. 5, no. 2, pp. 848–860,
Jul. 2015. Mar. 2014.
[3] M. Wang, H. Liang, R. Zhang, R. Deng, and X. Shen, “Mobility-aware [27] T. Ma and A. O. Mohammed, “Optimal charging of plug-in electric
coordinated charging for electric vehicles in VANET-enhanced smart vehicles for a car-park infrastructure,” IEEE Trans. Ind. Appl., vol. 50,
grid,” IEEE J. Sel. Areas Commun., vol. 32, no. 7, pp. 1344–1360, no. 4, pp. 2323–2330, Jan. 2014.
Jul. 2014.
[28] M. Shirk and J. Wishart, “Effects of electric vehicle fast charging on
[4] E. Yudovina and G. Michailidis, “Socially optimal charging strategies
battery life and vehicle performance,” SAE Tech. Paper 2015-01-1190,
for electric vehicles,” IEEE Trans. Autom. Control, vol. 60, no. 3,
2015.
pp. 837–842, Mar. 2015.
[5] Y. Liu, R. Deng, and H. Liang, “Game-theoretic control of PHEV [29] A. O. Allen, Probability, Statistics, and Queueing Theory. New York,
charging with power flow analysis,” AIMS Energy, vol. 4, no. 2, NY, USA: Academic, 2014.
pp. 379–396, 2016. [30] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
[6] Z. Xu, W. Su, Z. Hu, Y. Song, and H. Zhang, “A hierarchical framework Cambridge Univ. Press, 2004.
for coordinated charging of plug-in electric vehicles in China,” IEEE [31] D. M. Etter and D. C. Kuncicky, Introduction to MATLAB.
Trans. Smart Grid, vol. 7, no. 1, pp. 428–438, Jan. 2016. Upper Saddle River, NJ, USA: Prentice-Hall, 2011.
[7] Z. Zhu, Z. Gao, J. Zheng, and H. Du, “Charging station location problem [32] M. Grant, S. Boyd, and Y. Ye. (2015). CVX: MATLAB Software for Dis-
of plug-in electric vehicles,” J. Transp. Geogr., vol. 52, pp. 11–22, ciplined Convex Programming. 2009. [Online]. Available: http://stanford.
Apr. 2016. edu/~boyd/cvx
[8] S. H. Chung and C. Kwon, “Multi-period planning for electric car [33] J. Voelcker. (2014). Nissan Leaf $5,500 Battery Replacement
charging station locations: A case of Korean expressways,” Eur. J. Oper. Loses Money, Company Admits. [Online]. Available: http://www.
Res., vol. 242, no. 2, pp. 677–687, 2015. greencarreports.com/news/1093463_nissan-leaf-5500-battery-
[9] A. Y. S. Lam, Y.-W. Leung, and X. Chu, “Electric vehicle charging replacement-loses-money-company-admits
station placement: Formulation, complexity, and solutions,” IEEE Trans. [34] R. B. Cooper, “Queueing theory,” in Proc. ACM Conf., 1981,
Smart Grid, vol. 5, no. 6, pp. 2846–2856, Nov. 2014. pp. 119–122.

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.
3396 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 20, NO. 9, SEPTEMBER 2019

[35] H. L. Lee and M. A. Cohen, “A note on the convexity of performance Pengcheng You (S’14) received the B.S. degree
measures of M/M/c queueing systems,” Appl. Probab., vol. 20, no. 4, (Hons.) in electrical engineering and the Ph.D.
pp. 920–923, 1983. degree in control science and engineering from
[36] W. Grassmann, “The convexity of the mean queue size of the M/M/c Zhejiang University, China, in 2013 and 2018,
queue with respect to the traffic intensity,” Appl. Probab., vol. 20, no. 4, respectively. He was a Visiting Student with the
pp. 916–919, 1983. Singapore University of Technology and Design
and the California Institute of Technology and a
Research Intern with the Pacific Northwest National
Laboratory. He is currently a Post-Doctoral Fellow
with the Department of Mechanical Engineering,
Johns Hopkins University. His research focuses on
smart grid, especially electric vehicle, and power market.

Lin Cai (S’00–M’06–SM’10) received the M.A.Sc.

and Ph.D. degrees in electrical and computer engi-
neering from the University of Waterloo, Waterloo,
Canada, in 2002 and 2005, respectively. Since 2005,
she has been with the Department of Electrical
and Computer Engineering, University of Victoria,
where she is currently a Professor. Her research
interests span several areas in communications and
Yongmin Zhang (S’12–M’15) received the Ph.D. networking, with a focus on network protocol and
degree in control science and engineering from architecture design supporting emerging multimedia
Zhejiang University, Hangzhou, China, in 2015. He traffic over wireless, mobile, ad hoc, and sensor net-
was a Visiting Student with the California Insti- works. She was a recipient of the NSERC Discovery Accelerator Supplement
tute of Technology. He is currently a Post-Doctoral Grants in 2010 and 2015 and the Best Paper Award at the IEEE ICC 2008 and
Research Fellow with the Department of Electrical the IEEE WCNC 2011. She has served as a TPC Symposium Co-Chair
and Computer Engineering, University of Victoria, for the IEEE Globecom 2010 and Globecom 2013, as an Associate Editor
BC, Canada. His research interests include resource for the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS, the IEEE
management and optimization in wireless networks, T RANSACTIONS ON V EHICULAR T ECHNOLOGY, the EURASIP Journal on
smart grid, and mobile computing. He received the Wireless Communications and Networking, the International Journal of Sensor
Best Paper Award from the IEEE PIMRC 2012 and Networks, and the Journal of Communications and Networks, and as a
the IEEE Asia–Pacific Outstanding Paper Award in 2018. Distinguished Lecturer of the IEEE VTS Society.

Authorized licensed use limited to: National Physical Laboratory. Downloaded on August 08,2024 at 06:37:44 UTC from IEEE Xplore. Restrictions apply.