You are on page 1of 10

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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2660584, IEEE
Transactions on Smart Grid


Performance Evaluation of a Multi-standard Fast
Charging Station for Electric Vehicles
Ioannis Zenginis, John Vardakas, Member, IEEE, Nizar Zorba, Member, IEEE,
and Christos Verikoukis, Senior Member, IEEE

Abstract—Electric Vehicles (EVs) have been considered as and Volkswagen are equipped with the Combo inlet [5]. Both
a feasible solution to deal with the high fuel consumption CHAdeMO and Combo inlets operate with DC power. On the
and greenhouse gas emissions caused by conventional vehicles. other hand, other EV models, such as Renault Zoe, use AC
However, long charging times and drivers’ range anxiety are power for fast charging [6].
the main disadvantages of EVs. A key factor that is expected
to mitigate these problems and facilitate the wide adoption of The projected high penetration levels of EVs and the ne-
EVs will be the effective operation of fast charging stations cessity for wide deployment of fast charging facilities will
(FCSs). In this paper, the operation of a FCS is evaluated, in introduce large loads on the power grid [7]. Furthermore, the
terms of operator’s profits and customers’ waiting time in the EVs’ fast charging duration is long when compared to the
queue. The FCS contains both DC and AC outlets that provide duration of refilling conventional vehicles. This may lead to
high power levels, while the various EV models are classified by the formation of queues and long delays [8]. It is therefore
their battery size and the fast charging option they use (DC or crucial to develop realistic and accurate models that estimate
AC). The operator’s daily profits and the queue waiting time are the charging load of EVs and evaluate the level of services
initially computed by considering that the EVs recharge under
a Flat-rate Pricing Policy (FPP). In order to avoid a long queue
provided by FCSs. Equally important is the determination of
build-up at the FCS, a new pricing policy is then proposed. The charge scheduling policies in order to mitigate the negative
intuition behind the Scheduled Pricing Policy (SPP) is that users impacts of the additional EV load on the grid, such as dete-
are deterred to charge more than an arranged energy threshold, rioration of power quality, transformer overloads, peak load
thus reducing the load and the waiting time at the FCS. increase, etc. Charge scheduling policies are also important
for FCS operators, the main concerns of which are to achieve
considerable profits, while providing high Quality of Service
I. I NTRODUCTION (QoS) to their customers.
Electric Vehicles (EVs) are an emerging technology that will There are several studies in the literature that analyze the
bring major changes in the transportation and energy sectors. operation of FCSs based on queuing theory models. Li et al.
Due to their high energy efficiency and zero tailpipe emissions, [9] derive the overall charging load of Plug-in Hybrid EVs
EVs represent a promising pathway to increase energy security (PHEVs) in a FCS and examine the impact of uncontrolled EV
and reduce air pollution, especially when they are coupled charging on the distribution grid by performing Probabilistic
with a low-carbon power generation mix [1]. In 2015, the Power Flow Calculations (PPF). Gusrialdi et al. [8] propose
global EV stock exceeded one million, while the target for a higher level distributed scheduling algorithm together with
2020 has been set to 20 million EVs [2]. The role of Electric a lower level cooperative control policy for individual EVs
Vehicle Supply Equipment (EVSE) in achieving these targets in order to optimize the operation of a network of FCSs on
will be decisive. EVSE facilities are classified into three types a highway. The proposed methodology aims at adjusting the
depending on the power level they provide. Types 1 and 2 are percentage of the EVs to be charged at individual FCSs so that
appropriate for workplaces, parking lots and residencies, since all the FCSs are uniformly utilized and the total waiting time
the charging duration under these power levels ranges between is minimized. The works in [10]-[12] consider the operation of
1-36 hours. On the contrary, the charging duration using type networks of FCSs where each station draws a certain amount of
3 fast charging equipment ranges between 0.2-1 hour [3]. power form the distribution grid. This configuration assures the
The wide deployment of Fast Charging Stations (FCSs) reliable operation of the grid, however when the EVs’ charging
is crucial, especially in densely populated areas, where the demands are higher than the available power, an amount of
majority of the EV users may not have access to night-time customers is blocked. In order to mitigate this problem, authors
charging at home. Furthermore, FCSs are considered as an in [10] introduce a decentralized control mechanism where
effective solution for mitigating drivers’ range anxiety [4]. For the network operator offers price incentives, so that customers
this reason, the majority of EV models contain fast charging accept being routed to stations less busy than others. In this
inlets; Nissan Leaf and Mitsubishi i-MiEV are equipped with way more customers are served with the same amount of
the CHArge de MOve (CHAdeMO) inlet, while BMW i3 grid resources and the revenue of the operator is maximized.
A pricing based control mechanism is also proposed in [11]
I. Zenginis and J. Vardakas are with Iquadrat, Barcelona, Spain e-mail: aiming at ensuring that only a small percentage of EVs is
({izenginis, jvardakas}
N. Zorba is with Qatar University, Doha, Qatar, (e-mail:
blocked. This is achieved by incentivizing EV drivers to shift
C. Verikoukis is with the Telecommunications Technological Centre of their charging requests from peak to less congested periods.
Catalonia (CTTC), Barcelona, Spain, (e-mail: The charging management scheme proposed in [12] targets

1949-3053 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.

In what derivation of the EVs’ recharging patterns compared to [15]. 2. Afterwards. Instead. FCS is high and the queue waiting time rises to unacceptably Nevertheless. batteries by 50%. rather than actual needs. while the FCS is modeled simultaneously. in the queue and the tail of the queue waiting time under the subject to a maximum queue waiting time value. See http://www.1109/TSG. The flat fee per obtained kWh pricing scheme is assumed to be an exponentially distributed random variable is used in this paper as a baseline policy. Finally. only CHAdeMO and Combo outlets of the same CS cannot operate one EV class is considered in [15]. which traffic periods. this may common feature of the aforementioned studies is the simplified lead to long waiting times in the queue. the amount of energy obtained by the EV drivers who are blocked by their preferred station to EVs is likely to be high because it is mainly determined by the visit a nearby station which provides lower power levels. a Combo and an AC outlet in the same cabinet The EVs served by the DC outlets are divided into CDC [17]. although there are multi-standard Charging Spots (CSs) which contain A. drivers of conventional vehicles who do not always full their II.html for more information. as well as with an AC outlet that provides PAC .. the EVs’ mean waiting time in each queue and the tail of the queue waiting time. . the simultaneous operation of as an M/M/s queue. which is a generally distributed thresholds and increases the price per obtained kWh for those random variable. CDC ). This issue is taken into is located in a densely populated area. we formulate the SPP and present the aforementioned studies consider that EV drivers’ intention is to analysis of the FCS when the FPP or the SPP are activated recharge their batteries up to the maximum possible departure during different parts of the day. the EVs’ mean data regarding the amount of energy obtained during a fast waiting time in the queue and the tail of the queue waiting charging session. especially during peak approach adopted for modeling the EVs’ charging time. the EVs that request service by either a CHAdeMO or a Combo outlet form a DC queue. we model a FCS classes. Citation information: DOI 10. which implies that the SoC increase is an the AC outlet with one of the DC outlets is feasible. It also time are calculated.[21]. AC). we derive as a multi-class M/G/s queue.. customers that request amounts of energy greater than the Specifically. based on their battery size. Based on the aforementioned novelties. while it also indicates the in [13] and [14]. since this amount can last for the one-way According to the technical specifications of the CSs [17]. operator’s daily profit margin are determined given that the This consideration enables the utilization of real-case statistical EVs’ arrival rates vary during a day. follows. EVs’ recharging pattern and charging time distribution a CHAdeMO. authors in [13] model a FCS as an M/G/s/K queue arranged thresholds. . We conclude in Section V. A different pricing only on the power level. as well as the customers’ (SPP) is activated during parts of the day where the load of the mean waiting time in the queue and blocking probability. Each CS is equipped account in [15]. the EVs’ mean waiting time calculation of the maximum arrival rate capacity of a FCS. the most prevalent pricing policies in the existing market of On the contrary. the EV classes and their recharging patterns under the The increase of the batteries’ SoC is considered to be a random FPP are initially this in not a realistic approach. A drivers’ preferences. where the FCS operator sets energy (SoC) of the EVs’ batteries. FCS O PERATION UNDER THE FPP tanks. the aforementioned analysis considers that only high levels. and on the departure SoC of the batteries. but has not been fully edited. IEEE Transactions on Smart Grid 2 IEEE TRANSACTIONS ON SMART GRID the reduction of customers’ blocking probability by motivating day. such as the data presented in [16]. In Section II. a more holistic approach is adopted for the service by an AC outlet form a separate AC queue. the and [14]. but also on the arrival State of Charge policy is also proposed. where a FCS is modeled This paper is organized as follows. Both of the FPP. but republication/redistribution requires IEEE permission. Under the FPP. where the EVs’ charging time depends not upper bound of the queue waiting time. The proposed Scheduled Pricing Policy and derive the operator’s profits.2017. while the EVs that request In this paper. This issue is neglected in the existing literature. due to the fact that necessarily assumed to fully recharge their batteries as in [13] the analysis of the AC system follows a similar process. In Section IV we evaluate SoC level. it is reasonable to assume that EV users will obtain the amount of energy required so as to satisfy their We consider a FCS that contains s multi-standard CSs and preferences or/and estimated needs. reason. since the operation of the FCS and the effectiveness of the SPP by EV drivers may not always fully recharge their batteries. On the contrary. where the EVs served by a FCS located on with a CHAdeMO and a Combo outlet that provide the same a highway are assumed to increase the mean SoC of their power PDC . Bc with c = (1. as two multi-class M/G/s queues (one for the AC and one for The amount of energy Ec that a c-class EV obtains is: the DC charging queue). trip to the final destination. since it is one of depending only on the charging power level.. However. Personal use is permitted. This model is used for the the FCS operator’s daily profit. like providing numerical examples. EVs are classified not only by their derivation of the AC performance metrics is described in a battery size. In the rest of Section II the operation of leads to a more flexible analysis because the EVs are not the DC system is thoroughly analyzed.2660584. The accuracy of the proposed analysis is evaluated one EV class is served by the FCS (all EVs have the same through extensive simulations. Despite this realistic approach. the system’s load and the variable that may follow any possible distribution function. which show a close agreement battery size). The concept of classifying EVs by their different between analytical and simulation results battery sizes is incorporated in [14]. Under the FPP. a more sophisticated approach is adopted fast charging services [18] . In turn. in order to calculate the operator’s Ec = ∆SoCc Bc (1) overall daily profit margin. In Section III.This article has been accepted for publication in a future issue of this journal. Those metrics are where ∆SoCc is the increase in its battery SoC during the initially derived under the assumption that a Flat-rate Pricing charging session. Content may change prior to final publication. but also by their fast charging option (DC or more synoptic manner. For this exponentially distributed random variable. the amount of energy EVs ob- Policy (FPP) is implemented during the entire duration of the tain depends only on the drivers’ preferences and/or estimated 1949-3053 (c) 2016 IEEE.

δ (10) Considering a constant power level PDC during EVs’ fast Under the FPP.This article has been accepted for publication in a future issue of this journal.2017. . the process followed for the derivation of c=1 the mean waiting time and the tail of the queue waiting time For the same reason (Poisson arrivals). Therefore. Tc is also a random variable. Personal use is permitted. Relations mc = tgc (t)dt (5) (11) and (12) compute the operator’s revenue and expenses during a time interval Iδ . Furthermore. ∆SoCc ranges between a minimum and a maximum value while the mean energy provided to the EVs is obtained by: ∆SoCc. EVs’ waiting time in the queue λδ = λc.min = ∆SoCc. respectively.δ (12) d PDC h=1 gc (t) = Gc (t) = fc (xc (t)). The EVs’ mean waiting time C X in the queue in a multi-class M/G/s system is determined based aδ = ac. the expected arrival rates may be higher ∆ during afternoon when people return from their jobs and the X traffic is heavy. The aggregated arrival rate at time interval Iδ Γ= .δ represents the mean number of CSs occupied Gδ (t) = P (Tδ ≤ t) = kc.δ . Therefore. as well as the power rate of the load of c-class EVs during the same interval is defined as: AC outlets. the mean arrival rate of c-class EVs during a Given the battery sizes. the total load of the in the DC system is described in detail. Therefore.2660584. the operator’s overall profit margin is In a multi-class queuing system where the arrival process of expressed as the normalized profit: each class is Poisson. (8) on the analysis presented in [14]. hence. Furthermore.δ . depending mainly on the δ=1 traffic. For example. while the arrival rates of the RVNDC = RVNδ . while the arrival rates of the AC classes. . In practice. 2. since the price does not alternate with the amount of charged to the EVs during an interval Iδ is calculated by the obtained energy. Citation information: DOI 10. its revenue during the interval Iδ is: Bc RVNδ = R EV.δ = τδ PV.min and ∆SoCc. the superposed arrival process is also (RVNDC + RVNAC ) − (EXPDC + EXPAC ) Poisson [23]. (14) the majority of people are in their homes and the traffic is δ=1 lighter. (6) ((1)-(14)).org/publications_standards/publications/rights/index. (15) RVNDC + RVNAC equals the sum of the arrival rates of each EV class: C X C. ∆SoCc is considered to be a multiplying the mean number of occupied CSs aδ by the power random variable that follows a general Cumulative Distribution output of each CS: Function (CDF) Fc (x) = P (∆SoCc ≤ x).δ = aδ PDC (9) ing Probability Distribution Function (PDF) fc (x). (7) In this Section.δ Gc (t) (16) by c-class EVs at time interval Iδ .html for more information. the operator charges the served EVs with a charging [22]. the mean power c=1 1949-3053 (c) 2016 IEEE. C X Accordingly. The first and foremost step c=1 is to derive the CDF of the superposed charging time of the The total load of the system aδ represents the mean number system Tδ : of busy CSs in the steady state condition of the system [24]. fc (x) is truncated in the interval [∆SoCc. we make the following value Tc. but has not been fully edited. but republication/redistribution requires IEEE permission. compared to the arrival rates at night when EXPDC = EXPδ .δ (11) Tc = ∆SoCc .max = ∆SoCc. (2) PDC In order to compute the operator’s expenses (cost of supplied Based on (2). while the derivation of system equals the sum of the loads of each EV class [23]: the corresponding AC metrics is described in a more synoptic manner in the end of the Section. (4) dt Bc which expresses the sum of products of the mean charging Z energy during each time slot by the energy price. with a minimum energy) during the same interval. By assuming that a day consists of ∆ time intervals. Hδ } and τδ= h=1 dh . Content may change prior to final publication. with a correspond- PV. the recharging patterns and the time interval Iδ of duration τδ is denoted as λc. the price Rh that the FCS operator Gc (t) = P (Tc ≤ t) = P (∆SoCc ≤ PDC buys energy is constant. (13) EV classes may differ during a day.max (Bc /PDC ).: PERFORMANCE EVALUATION OF A MULTI-STANDARD FAST CHARGING STATION FOR ELECTRIC VEHICLES 3 needs. ∆SoCc. FCS load and operator’s daily profit margin the DC system are respectively obtained as follows: We consider that the EVs of each class arrive at the FCS ∆ X following a Poisson process. the PDF gc (t) and the Hδ mean mc of Tc are respectively obtained as follows: X EXPδ = Rh dh PV.min .max .1109/TSG. the charging time Tc of a c-class EV is: constant price R. where h= {1. ac.min (Bc /PDC ) and a maximum value considerations: i) the interval Iδ consists of Hδ time-slots Zh PHδ Tc.δ . the daily revenue RVNAC and expenses EXPAC for the AC system are calculated by following the same process ac.δ = λ ]. The CDF of Tc is derived by of duration dh . ii) the following relation: During these time-slots. The operator’s expenses are then Bc t) = Fc (xc (t)) (3) determined by the following relation: where xc (t)=(PDC /Bc )t. EV. IEEE Transactions on Smart Grid ZENGINIS et al. . Hence. the operator’s daily revenue and expenses for B. See http://www. .δ mc .

org/publications_standards/publications/rights/index.2017.1. Over these time intervals the s!(1 − ρδ )2 r=0 r! s! s operator activates the SPP. ρ) and OG.M/M/s= + 1− (21) for QoS satisfaction limit WQ . ρδ )Lδ.. The aforementioned selection reflects longer than a predefined time period TL . The the mean mδ and the coefficient of single-class analysis of [25]: variation vδ of the superposed charging time distribution are (  ) then derived as follows: s(1 − ρδ TL P (TQ. SPP dictates that the 1 − ρδ EVs are allowed to obtain up to Ethr. the activated energy Finally.     The main feature of the SPP is that the FCS operator sets 1 θ various energy thresholds Ethr. In order to derive the 1949-3053 (c) 2016 IEEE. they are charged with a higher price Rj >R. EVs’ mean waiting time in the multi-class queue by using an approximation method for a single-class M/G/s system III. ρδ and aδ into the enters an arbitrary CS.M/G/s The consideration of multiple thresholds minimizes the effect Wδ = . in case they request to 0 get more r than Ethr. (26) of an abrupt energy reduction that could result in significant λδ decrease in customers convenience and comfort. (20) approximates the mean The queue waiting time depends on the EVs’ arrival rates number Lδ.δ is charging events [27]. ρδ )= 1+Φ(θ)ζ(ρδ ) 1−exp − (23) 2 Φ(θ)ζ(ρδ ) which correspond to thresholds in the change of SoC of c- class EVs ∆SoCc. J) for all EVs.M/M/s s The mean waiting time and the tail of the queue waiting time Note that for a queuing system to have a finite queue in steady for the AC system are calculated through (16)-(29) by applying state.j < Wω ≤ Wl. when the waiting time lies on the range Wl. the mean waiting time Wδ of customers in the queue threshold is Ethr. Lδ. energy obtained during a charging session.M/M/s (22) rates and recharging patterns.δ > TL ) = Q(s. However.. which denotes the the EVs’ mean covered distance between two consecutive probability that customers’ waiting time in the queue TQ. .M/D/s system. which aims at reducing the system’s load and queue waiting time by affecting both the EVs’ arrival Lδ.This article has been accepted for publication in a future issue of this journal.thr.. Content may change prior to final publication. IEEE Transactions on Smart Grid 4 IEEE TRANSACTIONS ON SMART GRID where kc. . especially during peak traffic hours. but republication/redistribution requires IEEE permission.j values depending on the level Φ(θ) = ( − 2).j .δ = (29) ρδ = .j . Taking into account that the FCS is M/G/s system by using: i) the mean number of customers located in a densely populated area.M/M/s Lδ.M/D/s is ap. (18) (ρδ s)s X αr δ αδs  αδ −1 m2δ Q(s. actual needs. ψ(s. rather than where Lδ. (25) 8(1 + θ) 1−θ s+1 of the queue waiting time. See http://www. ρδ ) exp − (27) Z d mδ OG. Besides the mean waiting time in the queue. the EVs’ arrival rates waiting in the queue in an M/M/s Lδ. Specifically.M/D/s ≈ ψ(s. The determination of the battery sizes. Moreover. the recharging patterns and the arrival rates the aforementioned parameters enables the computation of the of the AC classes to (1)-(8). Lδ.δ /λδ denotes the probability that a c-class EV the superposed multi-class metrics mδ .M/M/s is computed by (21) and Lδ. In this section we consider that there "s−1 #−1 is a subset {Ω} ∈ {∆} of time intervals Iω .J criterion for assessing the effective operation of a FCS would can be determined by taking into account statistical data for be the tail of the queue waiting time. higher waiting time during the interval Iδ is computed using Little’s law [24]: levels correspond to the activation of lower energy thresholds.html for more information.M/M/s and an M/D/s are expected to be high.j /Bc . This probability is the operator’s intention to promote a more sensible recharging approximated through the following relation. Citation information: DOI 10.δ = λc. ρδ < 1 is a necessary condition. gδ (t) = Gδ (t) (17) dt where Q(s. 2. varδ = t2 gδ (t)dt − m2δ . Furthermore. by incorporating pattern during peak traffic periods. ii) the square of the coefficient of variation Furthermore.1109/TSG. Specifically.j (j = 1. under the FPP.2660584. Researchers in [26] report that the EV drivers usually overestimate their energy 1 + vδ2 needs.j kWh at the same price ζ(ρδ ) = (24) ρδ R as in the FPP. with θ = . Personal use is permitted. an additional the minimum value of the selected energy thresholds Ethr.j+1 .M/G/s λδ mδ OG. The θ 9+θ s−1 operator determines the Ethr.δ mδ = tgδ (t)dt. the charging times depend on the amount of vδ2 of the service time.M/G/s of customers waiting in the queue in an and charging times. As shown in drivers’ preferences or/and overestimated needs.M/G/s ≈ 2vδ2 1−vδ2 (20) + is likely to be high because it is mainly determined by the Lδ. the amount of obtained energy Lδ. Therefore. ω ∈ {Ω} during ρδ αδs X αr αs  αδ −1 which the waiting time Wω is higher than a maximum allowed δ δ Lδ.δ are derived as follows: r varδ Z "s−1 #−1 vδ = .j = Ethr. ρδ ) = + 1− (28) s!(1 − ρδ ) r=0 r! s! s Furthermore. but has not been fully edited. (19) Lδ. S CHEDULED P RICING P OLICY analyzed in [25]. the queue waiting time may rise to proximated through (22)-(25): unacceptably high levels. the utilization rate of the system is: Lδ.

Content may change prior to final publication. (37) αc1 .max computed by the integral Πc. while their recharging pattern and charging time CDF are denoted. since subclass c1 drivers decide to charge their EVs distribution of each subclass. while their recharging pattern In the following analysis we derive the operator’s revenue is obtained by truncating the function fc (x) in the interval and expenses during a time interval Iω where the SPP is [∆SoCc.thr.ω.j + ac3 .j ≤ t ≤ Tc. 0 the EVs that belong to subclasses c1 recharge at price fc1 .j fc (x)dx . while a 0 otherwise.j quit the FCS without recharging.δ . but has not been fully edited.j . ∆SoCc. which refers to the DC system. subclass c3 consists of the proportion Nc3 . request the same amount of energy as in the FPP 0 otherwise because it is necessary for reaching their destination.j mc2 . we assume that a percentage fc3 . c2 and c3 .j = ∆SoCc.j (t) = Bc fc1 . This type of EVs do not alternate their recharging pattern is necessary. the energy provided to the EVs at price R is: where xc (t) = (PDC /Bc )t and Tthr. while the load of subclass c2 is αc2 .ω.j =λc3 .j arrival rate. See http://www.ω.j =λc1 .j ).j.R = τω PDC (ac2 .j =Nc2.  1 x = ∆SoCc.html for more information. percentage Σc.j (t) = Bc fc1 . For these EVs. it is procedure described in Section II-C.min . due to the fact that the price change may affect under the SPP because they are charged with the same price. ∆SoCc.δ .j (xc (t)).ω.j /PDC . .j = Πc.min ≤ t ≤ Tthr.j Σc.j . Furthermore.max ]: activated. Tthr.ω. The arrival rate of subclass c2 EVs is λc2.j ≤x ≤∆SOCc. ∆SoCc.j of the EVs that decide to obtain exactly Ethr.thr.j fc (x)dx Rj .min≤x ≤∆SOCc.j (t) = (33) 0 otherwise.j kWh.max SPP.ω. IEEE Transactions on Smart Grid ZENGINIS et al. as it is noticed in Section II-B.j .j = Π same process can be used for the analysis of the AC system.thr.j increase. Finally. Activated energy thresholds depending on the waiting time level. the load of subclass c1 is EV.min fc (x)dx of those EVs Ξc. the determination of the  1 t = Tthr.j ) percentage of EVs which. Fc2 . 1. and the operator’s daily profit margin.j . the division of each EV class to 3 subclasses FPP.: PERFORMANCE EVALUATION OF A MULTI-STANDARD FAST CHARGING STATION FOR ELECTRIC VEHICLES 5 Subclass c2 consists of the proportion Nc2.j λc.j mc3 . By C X defining as mc1 the mean of gc1 (t).j = λc2 .j .j λc. SPP Subclasses of c-class EVs that would request up to Ethr.j kWh is not at price R.j . the number of the EVs that will enter the FCS and the amount The arrival rate of subclass c3 EVs is λc3.0 fc1 . The proportion of c-class their recharging pattern is obtained by truncating the function EVs that would request more than Ethr. and charging time distribution of each subclass are presented in the following analysis. but republication/redistribution requires IEEE permission. Under the ( fc (x) R ∆SoCc.2017.ω.thr.j (xc (t)). Tc. the arrival rate of subclass B.j − Σc.j (t).j (x) = (32) 0 otherwise we divide each EV class into 3 additional subclasses c1 .thr.j =Nc3. the (30) load of each class (or subclass) represents the mean number of The above normalization is necessary so as the integral of CSs occupied by the EVs belonging to this class (or subclass). Therefore.j kWh under the Under the SPP. the corresponding charging time PDF is: Subclass c1 consists of the remaining Nc1.j.R0 = τω PDC ac1 .j kWh under the FPP is fc in the interval [∆SoCc.j = Ethr. Having determined the arrival rate and the charging time Hence. Furthermore. despite the price gc3 . It should (35) be noted that the proposed SPP dictates that if the EV drivers By defining as mc3 the mean of gc3 . Due to ( fc (x) R ∆SoCc.thr. while the EVs that belong to subclasses c2 and c3 recharge 0 otherwise. The rationale of this division. then they c3 EVs is αc3 . they can satisfy their urgent needs by obtaining as much energy as the arranged threshold. it is assumed that requesting more than Ethr.δ . c=1 1949-3053 (c) 2016 IEEE. as well as the derivation of the recharging pattern Gc2 .ω.ω.j =1−Πc.j the price increase under the SPP. while of energy they will eventually obtain. The The mean charging time of subclass c2 EVs is mc2 . respectively.max .j ]: R ∆SoCc.This article has been accepted for publication in a future issue of this journal.1109/TSG.j (x) to become equal to 1. Daily Profit Margin when the SPP is activated c1 EVs is λc1.j queue waiting time and the operator’s profit under the SPP. in order to avoid the higher price. as: Fig.thr.2660584.j mc1 . 0 have to pay the higher price Rj for each kWh obtained.j =Tthr. the amount of energy provided to the EVs at price Rj of subclass c1 EVs is derived by following the same procedure during each Iω is: used in Section II-A for the derivation of gc (t) from fc (x): XC  PDC EV.j (t) Hence.ω.j A.thr.j (x) = ∆SoCc. ∆SoCc.j λc. considered that they are not strongly motivated to change their recharging pattern.j decide to obtain exactly as much energy as (34) the operator’s threshold.ω.j = Nc1.j (1 −  PDC Ξc. the load of subclass charge more than the arranged energy threshold.ω.j (36) j gc1 .ω. Personal use is permitted. The charging time PDF gc1 . the mean and the tail of the queue under a higher price (because they need more energy for waiting time under the SPP can be derived by following the reaching their destination than the arranged threshold).max c=1 0 otherwise (31) while.ω. Citation information: DOI 10.j (x) = ∆SoCc.

The aforementioned PDF is appropriate In this section. (42) their jobs).00-16. DC system. 2. interval that the SPP is activated is computed by: Hω TABLE I. the operator’s daily revenue RVNAC I3 .R + Rj EV. We also consider the 4 most TABLE II. which is the only AC class. we consider that the EVs’ recharging patterns fc (x) with c = 1.j.00-22.00 0.j = Rh dh PV.ω.8 0.j (40) I1 = 16. . The FCS 3 i-MiEV 16 14 10.6 11.00-18. the only difference I2 = 22.00 and I3 =08.49 0.5 1. R ECHARGING PATTERNS OF THE EV CLASSES whereas the AC system consists of CAC =1 EV class. the superposed arrival rate of the whole DC system. while λAC. the operator’s revenue during Iω under the SPP is: 0 RVNω.ω.ω.00 Z1 = 16. Recharging patterns of the DC classes.min .j. P ERFORMANCE E VALUATION follow the Beta PDF. λ1 takes integer values. which is DC outlets is PDC =45 kW. E NERGY TARIFFS OF THE S PANISH RETAILER [29] X Time Intervals Time slots Energy price (e/KWh) EXPω.22 1949-3053 (c) 2016 IEEE. its arrival rate is used as reference for process. while the arranged energy thresholds this end. Citation information: DOI 10. Table I summarizes the number of time slots Zh each PV.1 Z2 = 00.00-00.δ (the power charged to the EVs under the FPP) is I3 = ∆SoCc. Content may change prior to final publication.21 I1 =16. Personal use is permitted.00 0.00-16.2 3. λ1 (Leaf) λ2 (i3) λ3 (i-MiEV) λ 9.00-22. the times other EV models are computed by dividing their population by of SPP activation and the arranged energy thresholds for the the population of Leaf [28].58 0. Therefore. respectively.00-08.j + ac2 .j + ac3 . ∀δ∈{∆}−{Ω} ∀ω∈{Ω} during night hours I2 .1 replaced by the power charged to the EVs under the SPP: C X [29]. AC) [28].5 3.2660584. In contrast.119 Note that (40) has the same form with (12). The operator’s expenses during a time Fig.00-08.5 5 kWh. we evaluate the operation of a FCS that for modeling random variables that are limited to intervals of consists of s=5 multi-standard CSs. For the third time interval For the AC system. PDC = 45 kW The operation of the FCS is evaluated during a day. while the power output of the limited to the interval [∆SoCc. 2. 0 0 0 0 When the FPP is activated during the whole day.j (39) ∀δ∈{∆}−{Ω} ∀ω∈{Ω} where the two summations correspond to the operator’s rev- enue during the time intervals that the FPP or the SPP is activated.j ) (41) TI contains. To is unacceptably high.δ λAC. TABLE III.00 0. (43) RVNDC + RVNAC this case. fast charging option: DC). while the arrival rates of the depend on the level of the waiting time.21 2 i3 18. the traffic is light.8 9.ω.39 0.00-08.j . but has not been fully edited.This article has been accepted for publication in a future issue of this journal.8].7 1 I3 6 1.5 0.1 3 the fast charging option and the battery capacity of these EV models. AC outlet is PAC =43 kW [17]. and is Finally.j = R EV. each TI is characterized by a Hence.00 0. the FCS 0 (RVNDC + RVNAC ) − (EXPDC + EXPAC ) operator charges the served EVs with a constant price R. DC). and hence the arrival 0 rates are considered to be low. which are: Nissan TI Mean arrival rates (EVs/h) Leaf (battery capacity: B1 =24 kWh. the DC system consists of CDC =3 EV classes. the arrival rates are 0 and expenses EXPAC when the SPP is activated during assumed to take intermediate values.00.8 12 0. 0. the operator’s daily expenses for the DC system when Poisson procedure for the EV arrivals. Table II presents the the SPP is activated during the intervals Iω and the FPP during mean arrival rates of the EV classes for each TI expressed the intervals Iδ . DC) and Renault ZOE (B1 =22 kWh.ω.072 is that PV.max ] = [0. for the DC system and c = 1 for the AC system IV.ω. Note also that λDC. and therefore the arrival rates are high.8 kWh.j = PDC (ac1 .00-22. such as the EVs’ ∆SoC in an FCS. The SPP is applied when the queue waiting time the computation of the arrival rates of the other EV models. See http://www.1 0.ω.00 0. Mitsubishi i-MiEV (B3 =16 I1 10 2.00-16. Furthermore. 3. Given that Leaf is the specific parts of a day are calculated by following the same most popular EV model. PAC = 43 kW 1 Zoe 22 8.δ tackles the arrival rate of computed by the following relation: Zoe. IEEE Transactions on Smart Grid 6 IEEE TRANSACTIONS ON SMART GRID Therefore.html for more information. A RRIVAL R ATES OF THE EV CLASSES popular EV models of the Spanish market.1 h=1 Z2 = 18.00. but republication/redistribution requires IEEE permission.42 0.2017.5 6 18.δ (Zoe) BMW i3 (B2 =18. I2 =22. It is assumed that during I1 0 X X the traffic is heavy (a great number of people returns from EXPDC = EXPδ + EXPω.3 6. Given I2 2 0. and the price Rh of the supplied energy during c=1 these time slots.1109/TSG. the total daily profit margin of the operator is computed through (7). In Γ = 0 0 .R0 (38) j while the daily revenue for the DC system is obtained by: 0 X X RVNDC = RV Nδ + RV Nω. which corresponds to working hours.δ represents AC system may be different than those for the DC system.00 Z1 = 08.00 Z1 = 22. which c Model Ec (kWh) αc bc mean ∆SoCc mc (EVs/h) is considered to be divided into ∆=3 Time Intervals (TI) 1 Leaf 24 4. The power output of the finite length [30].21 operator is supplied energy from a Spanish energy retailer AC system. δ ∈ {∆} − {Ω} is determined as follows: in number of EVs per hour.

2660584. Content may change prior to final publication. bc of each Beta PDF. but has not been fully edited.) Tail (%) Wδ (min. as well as the resulting mean values are summarized in Table III. respectively.162 I2 0.018 0. through (18). Ethr. it is considered that they also obtain 9. the EVs’ arrival rates are given in Table 0.002 0.2017.29 2.15 0.δ Wδ (min. 5.14 0.75 0.) Tail (%) ρAC. . Citation information: DOI 10. 2. Table III also presents the corresponding mean charging times of the EV classes. For the present our model is the utilization of real-case statistical data for the numerical evaluation we assume that TL = 4 min.77 5.3 kWh [16]). The values of αc . but republication/redistribution requires IEEE permission.δ Wδ (min.) Tail (%) ρAC. 3.031 0. 4. Based on the energy tariffs (Table I).015 0. Furthermore.δ . The shape parameters αc . which is derived to be R=0. when the EVs’ mean waiting time in the queue is less than in both cases. IEEE Transactions on Smart Grid ZENGINIS et al. by activate the SPP during this specific interval in order to reduce using (19).146 (e/KWh).04 0 0 0 0 I3 0.δ Wδ (min.html for more information. the EVs’ mean waiting time in the DC queue analytical results. Furthermore.15 0.46 0. This is because implemented. Activated energy thresholds depending on the waiting time level. The resulting recharging patterns of the three DC classes are shown in Fig. Personal use is permitted. we present analytical and simulation results model. Mean waiting time reduction.29 2.: PERFORMANCE EVALUATION OF A MULTI-STANDARD FAST CHARGING STATION FOR ELECTRIC VEHICLES 7 TABLE IV. As shown in case are considered to follow the exponential distribution with Fig.This article has been accepted for publication in a future issue of this journal.013 0. WQ =2min. See http://www.031 0.) Tail (%) I1 0. Due to lack of data regarding the other EV classes.δ and ρAC.004 0. which are calculated based on the analysis of Section II-A.018 0..53 0.) Tail (%) I1 0. which is not possible when the compares the results of our multi-class M/G/s model with M/M/s model is applied. the price is computed the superposed charging time distribution vδ .01 M/M/s model vδ =1 Analysis Simulation vδ =1 Analysis Simulation [15] Time Intervals ρDC. In denotes the probability that the customers’ queue waiting time the M/M/s model vδ = 1.32 34.) Tail (%) Wδ (min.087 Analysis Simulation vδ =0.1 =11 kWh. by definition.001 0. as well as a desirable daily profit margin of Γ=30%. as less than 1%.001 0. For a fair comparison of the two We proceed by assessing the effectiveness of the proposed models.22 0. the AC system operates effectively when the considers the same arrival and charging procedures with the FPP is implemented during the entire duration of the day.91 5.103 0.04 0 0 0 0 I3 0. The FCS operator can to utilization rate values ρDC.46 34. Table IV also contains the coefficient of variation of both the EVs’ mean waiting time in the DC queue (WDC.004 0. in [16] regarding the recharging pattern of Nissan Leaf in a FCS. bc have been selected based on the data presented Fig. Ethr. Based on this QoS criterion and the results pre- The simulation results are obtained by our EV simulator that sented in Table IV.15 0.15 0. the recharging patterns of the EV classes in the M/M/s SPP in terms of EVs’ waiting time reduction.δ are mapped exceeds the targeted value during I1 .28 33.161 0.013 0.54 0. the M/M/s model in [15].1 =11 25.103 I2 0.002 0. Given the recharging patterns.) Tail (%) Wδ (min. Table IV EVs’ recharging patterns [16].063 Analysis Simulation M/G/s model Time Intervals ρDC. Tail of the queue waiting time reduction. Another advantage of will be longer than a specified time period TL . 3.01 0.13 0. the arrival rates λDC. Recall that the tail of the queue waiting time the sensitivity introduced by vδ is not taken into account.22 0.δ and λAC.13 0. we consider that the FCS provides qualitative services the same mean ∆SoCc values as in Table III. the results indicate that the M/M/s well as the tail of the queue waiting time when the FPP is model overestimates the queue waiting time.46 0. The mean value of ∆SoC1 in Table III corresponds to the mean energy obtained by the Leaf drivers during a fast charging session (9. Q UEUE WAITING TIME RESULTS DC system AC system Multi-class vδ =0. the analysis of Section II-B can be used for the determination of R given the operator’s desirable profit margin Γ.3 kWh on average during a fast charging session.55 3. On corresponding analytical models.003 0. The results prove the high precision of our In what follows. the EVs’ arrival rates Fig. For the derivation of the the other hand. (Table II) and mean charging times (Table III).1 ) 1949-3053 (c) 2016 IEEE.015 Fig. since the difference between analysis and simulation is for the EVs’ mean waiting time in the DC and AC queue. Furthermore.δ Wδ (min.021 0.) Tail (%) Wδ (min.45 4.021 0.1109/TSG.77 3 25.

Fig. For presentation purposes.1 =Ξ3. By multiplying of the SPP coincides with the peak energy price period (Table the aforementioned quantities. The case where Σ1=Ξ1=0 corresponds to the queue waiting time when the FPP is applied. Profit margin when the SPP is activated. Profit margin when the SPP is activated during peak energy cost.j denotes the amount of the EVs of each class that quit the FCS without recharging.1 =11 kWh. which reflects the EVs’ arrival rate i. Ethr. while Σc. with the increase of Ξ1 .1 =Σ1 .1109/TSG. we consider values provided by Table II. is Ethr. but has not been fully edited. leads to both lower operator’s revenue and expenses.j . the fact that the operator buys Ethr. Mean waiting time reduction.2660584. Next.1 =Σ3.1 > TL )). the profit margin Γ is higher than Γ = 30% (profit the FPP. the selected energy threshold according to Fig. the other hand. which result in a queue can be interpreted as follows: greater values of Ξ1 correspond waiting time during the interval I1 . Fig. The aforementioned case studies prove at a constant price during the0 whole day. to its minimum value. Figs. it is considered that the Despite the decrease in revenue.2 kWh/km [31]. . which consists of 5 CSs. as the amount Σ1 of the EVs that Fig. 6. result capacity. 8 shows. Citation information: DOI 10.1 =Ξ2. IEEE Transactions on Smart Grid 8 IEEE TRANSACTIONS ON SMART GRID Fig. the operator sets the energy threshold to be equal in turn.2 =9 kWh. the operator can effectively reduce it to acceptable margin when the FPP is applied during the whole day) as far limits by activating the SPP and by appropriately selecting as there are EVs belonging to subclasses c1.. On 0 reduction. but republication/redistribution requires IEEE permission. It should be The scenarios examined in this Section refer to the operation noted that the EVs’ arrival rates are assumed to be equal to the of a single FCS. decide to obtain less energy under the SPP compared to the FPP case increases.j and Σc. compared to the traveled distance between two consecutive charging events and intervals I2 and I3 (Table I).html for more information. 8.1 = 11 kWh.3 is Ethr. those belonging to subclasses c1. However. and We now proceed0 to the determination of the operator’s daily hence they are charged with a higher price when the SPP is profit margin Γ when the SPP is activated during the interval activated. combined with the decrease in the the energy consumption of the EV class with the largest battery energy need to be bought during this specific interval.This article has been accepted for publication in a future issue of this journal. This outcome higher than those presented in Table I. under this case. it is considered that the EVs’ arrival rates are 10% the increase of the EVs Ξ1 that quit the FCS.1 . the queue waiting time is reduced. compared to the case where the FPP is activated which based on (19) leads to lower utilization rate values. WDC. We also consider another case study where the SPP is equal to 9 kWh. I1 and the FPP during the intervals I2 and I3 . this is because the utilization rate and the queue waiting time are further reduced operator charges the EVs that obtain more than Ethr. Specifically. 7 presents the while the arrival rates of the latter (8 CSs station) are 40% profit margin Γ versus the Σ1 − Ξ1 pairs of Figs. Personal use is permitted. In [27].1 = 5. Recall that Ξc. On the one hand. Fig. different energy thresholds. but the operator buys energy this energy threshold. 4 and 5. this case. We also consider that under the SPP. 6 evaluates the effectiveness of SPP for activated during the interval I1 . the energy cost is flat-rate.2017. See http://www. and hence the energy threshold that the same operator owns two additional neighboring FCSs. In to lower amount of energy drawn by the EVs during I1 . with a higher price. This is It is observed that the activation of the SPP leads to higher due to the decrease of the system’s superposed charging time profit margins. it is also observed that Γ gradually rises with This is because those EVs obtain more than Ethr. the mean traveled distance is reported to be in higher daily profit margins.1 kWh.1 kWh.7 mins. 4 and 5 evaluate the effectiveness of the proposed SPP versus the parameters Ξc. the energy threshold is set to be I). respectively. higher than the arrival rates presented in Table II.2 value is selected by taking into account the EVs’ mean energy at a higher price over the interval I1 . while the average energy consumption of a Leaf EV The aforementioned outcome is derived when the activation according to real-case tests is 0.1 = the arrival rates of the former (4 CSs station) are 40% lower. 11 kWh is set 0to be 5% higher than R. we assume that Ξ1. 45 km.1 =Ξ1 and Σ1.. in that given the expected level of the queue waiting time.j denotes the amount of the EVs of each class that decide to obtain as much energy as the selected energy threshold.1 (Ξ1 + Σ1 < 1). consisting of 4 and 8 CSs. Since the queue waiting under the FPP is 3 It is also assumed that 0 the price R1 for those EVs that obtain more than Ethr. 7. This. and the tail of the queue waiting time (P (TQ. In this new 1949-3053 (c) 2016 IEEE.e.1 =Σ2. Content may change prior to final publication. As Fig. The during the whole day.

pt/PresentationLayer/ResourcesUser/Cat\%c 3\%\ proposed Scheduled Pricing Policy (SPP) during peak-traffic \A1logos\%202014/Mobilidade/CS119I1307B1. May 2013. Yilmaz and P. “Capacity planning We analyze the operation of a Fast Charging Station (FCS) frameworks for electric vehicle charging stations with multiclass cus- tomers. Hence.” Journal of 0 R. for Electric Vehicles (EVs) and we propose a new model that [12] J. Dziechciaruk. no. 2016. 2014. vol. Mar. pp. 1993.” account. Smart Grid. Personal use is permitted. 2012. 111.” Idaho National Laboratory. Under the assumption that the EVs are allowed to [15] S. Zhang.1 at price R1 . vol 31. system-proves-tough-127731. S. 1461-1471. 2151-2169. H.” Applied Energy. pp. Brown and G.2660584. For this new case study. 155-164.” IEEE Trans. AC Level 2. Nanaki and C. 2016. of DC fast charging deployment in California. which denotes Appl. IEEE can accommodate up to 25% of the drivers coming from the 53rd An. 4. X. “Unsplittable load queue waiting time in all stations lower than the QoS criterion.2017. balancing in a network of charging stations under QoS guarantees. 251-266. Krein. “Modeling of plug-in hybrid electric vehicle charging demand in probabilistic power flow calculations.html for more information. vol. M. [4] L. .html The present work evaluates the performance of a single FCS [19] https://chargedevs. Smart and D. Vezzini. 94. The various EV models [13] P. no. 53.: PERFORMANCE EVALUATION OF A MULTI-STANDARD FAST CHARGING STATION FOR ELECTRIC VEHICLES 9 R EFERENCES [1] E. the tail Quality of Service Evaluation of a Fast Charging Station for Electric Vehicles. the performance of pp.pdf periods leads to not only lower queue waiting times. and infrastructure for plug-in electric and hybrid vehicles. of the neighboring FCSs and obtain as much energy as they [6] I. Dec. 2015. 111-122. “Electric vehicles charging management in communi- considers that two queues are formed. La Parra. J. G. N. pp.html order to analyze the impact of multiple FCSs on the distribu. into account the EVs’ arrival rates and recharging patterns. 1292-1302. in terms of customers’ Quality of Service (QoS) and operator’s encouraging-efficient-usage/ profits. 2. Kawashima. 1949-3053 (c) 2016 IEEE. U.. and DC by providing financial disincentives to those EVs that want to Fast Charging Units. charging power levels. Power Electron. T.This article has been accepted for publication in a future issue of this journal. Ren. Bodo and E. 2013 [2] P. Verikoukis. 5. Gorner. “Niche accumulation and initial FCS have three choices. vol. 1.” Journal of Cleaner Production. since the FCS consists cation controlled fast charging stations. Dec. Vardakas.” In Proc. “The optimization Fig. Content may change prior to final publication.. as well as the operator’s expected” in Proc. IEEE SPEEDAM. 2012. P. In this scenario technical objectives EV charging station demonstrator. hybrid and electric vehicles-the case study of Greece.1109/TSG. 2015. while the large FCS can accommodate up to [9] G. Subotic. A. S. [3] M. 1. S. 9. vol. van Lente. M.” IEEE Trans. 2011. Kwasinski. A. B. “Review of battery charger topologies. pp. want at price R. IEEE CAMAD. The FCS operator can mitigate this problem Station Utilization at a Work Site with AC Level 1. 157. A. MA. Bae and A.K. charge more than fixed energy thresholds.1819-1826.blinknetwork. Decision and Control (CDC). 6. D. “Workplace Charging Case Study: Charging peak-traffic hours. In our future work the proposed model will be used in [20] http://www. pp. Germany. 2012 (load variance minimization.efacec. [21] https://fastned. during the interval I1 versus the parameter U . “Operation Analysis of Fast Charging are divided into classes based on their battery size and their Stations With Energy Demand Control of Electric Vehicles”. or c) visit one Cleaner Production. Gusrialdi. Zenginis. but also [18] http://www. “Global EV outlook 2016. S. G. Akimaru and K. V. May 2015. “Spatial and temporal model of electric vehicle charging demand. the percentage of EVs that leave the initial FCS and join any [8] A. 3.” OECD/IEA. queue waiting time may rise to unacceptably high levels during [16] J. Bayram. pp. energy needs to the proper stations. the operator can keep the [10] I. V. Tajer. J. 14th Eur. vol. and A. 1934-1943. pp. Fan. no. by optimally routing the EV drivers with high Grid. “Comparative economic and envi- ronmental analysis of conventional.” IEEE Trans. The proposed model takes Trans. voltage quality) will be taken into [23] H. no. “Scheduling and cooperative control of the additional FCSs. Vardakas. the two additional FCSs when all customers quit the initial [7] K. Qaraqe. 9 presents the queue waiting time stochastic charging model. As Fig. vol. as well as the participation of the FCS operators in [22] H. 1390- 1395. pp. together with financial and QoS objectives. no. Abdallah. optimal charging management of EVs. 6. vol 112. Teter and W.1 at price standardizationthe case of electric vehicle recharging plugs. Berlin. S. Z. Cazzorla.4. 6465-6471. Koroneos. pp. 28. higher profits margins for the FCS operator.” IEEE Trans. 9 indicates the small FCS of electric vehicles’ charging at highway service stations. Aug. it is considered that the EV drivers visiting the [5] S. Smart Grid. pp. IEEE fast charging option (DC or AC). 115-119. The activation of the [17] http://www. initial station. vol. Smart Grid. Jul. Simaan. b) charge more than Ethr. S.” IEEE Trans. Smart Grid. N.” Energy. Scoffield. Reza. pp. pp. Beyond one million electric cars. Leguijt and H. T.” in Proc. [14] I. Levi. Birmingham. the 394-403. vol. See http://www. Power Electron. Yunus. pp. Conf. J. a) charge up to Ethr.” In Proc. Bakker. IEEE Transactions on Smart Grid ZENGINIS et al. Michailidis. pp. 2014. of multi-standard Charging Spots(CSs) that can simultaneously 2014. no. 3. 669-678. no. vol. C ONCLUSION [11] I. 2015 case study. of the queue waiting time. 2016. Zelaya and M. Conf. Devetsikiotis. Fig. Bayram. Zhang. Citation information: DOI 10. “Teletraffic Theory and tion grid. “An ultrafast demand response schemes. “Analysis and and determines the mean waiting time in each queue. 3. (EPE).” IEEE Trans. Yi. but has not been fully edited. “Distribution grid impact FCS (Ξ1 = 1) seeking for being served to the neighboring of Plug-In Electric Vehicles charging at fast charging stations using stations is evaluated. 2015. serve EVs requesting DC or AC power. Qu. Hoimoja. 492-499. Jun.plugincars. Smart 75%. but republication/redistribution requires IEEE permission. Power Electron. “An EV Drive-Train With Integrated Fast Charging Capability. Zorba. K. Sainbayar. Oct. C. Rufer. for the Springer Verlag. 6. Samuelsen. Li. Waiting time in the two additional FCSs. obtain as much energy as they want at a flat-rate price.

org/publications_standards/publications/rights/index. Schey. [27] J. 3-4. but republication/redistribution requires IEEE permission.” SAE Techn. S.consumerreports. “Approximations for multi-server queues: system interpola- tions. [28] http://www.2660584.motorpasion. furgonetas-electricas-en-2014 [29] https://www. Tal. Content may change prior to final publication.htm 1949-3053 (c) 2016 IEEE. no. Nicholas. 53-60. “Charging Behavior Impacts on Electric Vehicle Miles Traveled: Who Is Not Plugging In?. Hahn. .2017. Kimura. Powell.html for more information. 2454. [30] Eugene.This article has been accepted for publication in a future issue of this journal. Woodjack. W. vol. See http://www.” Springer Science & Business Media.somenergia. Citation information: DOI 10. Personal use is permitted. Narayan. “Bayesian Methods for Management and Business: Pragmatic Solutions for Real Problems. 17. test-results-are-in/index. IEEE Transactions on Smart Grid 10 IEEE TRANSACTIONS ON SMART GRID [24] Bhat.1109/TSG. pp.”2014. 1994. J. Smart. Paper. pp. 2014. “An introduction to queueing theory: modeling and analysis in applications. M. J. U. “Extended Range Electric Vehicle Driving and Charging Behavior Observed Early in the EV Project. 2008. [26] G. [31] http://www.” Queueing Systems. [25] T. 347-382.” Transportation Research Record: Journal of the Transportation Research Davies. but has not been fully edited.