IEEE SmartGridComm 2013 Symposium - Architectures and Models for the Smart Grid

Online Scheduling for Vehicle-to-Grid Regulation

Service
Junhao Lin, Ka-Cheong Leung, and Victor O. K. Li
Department of Electrical and Electronic Engineering
The University of Hong Kong
Pokfulam Road, Hong Kong, China
E-mail: {jhlin, kcleung, vli}@eee.hku.hk

Abstract—Electric vehicle (EV) fleets can provide ancillary base load or non-EV load to perform the optimal EV charging
services, such as frequency regulation, to the utility grid, if their control.
charging/discharging schedules are coordinated appropriately. In
this paper, a multi-level architecture for bidirectional vehicle-
to-grid regulation service is proposed. In this architecture, ag- V2G seeks to utilize the battery packs installed on EVs
gregators coordinate the charging/discharging schedules of EVs as energy storage to provide energy and ancillary services to
in order to meet their shares of regulation demand requested the grid [5]. It can be both unidirectional when EVs provide
by the grid operator. Based on this architecture, the scheduling ancillary services by modulating their charging rates, and
problem of V2G regulation is then formulated as a convex bidirectional when EVs are also allowed to discharge their
optimization problem, which in turn degenerates to an online
scheduling problem for charging/discharging of EVs. It requires batteries to inject energy back to the grid. Most research efforts
only the current and past regulation profiles, and does not depend on V2G ancillary services focus on frequency regulation
on the accurate forecast of regulation demand. A decentralized service. Frequency regulation is a zero-energy service that
algorithm, which enables every EV to solve its local optimization compensates the minute-to-minute fluctuations of generation
problem and obtain its own schedule, is applied to solve the online and demand [6]. Distributed control strategies based on local
scheduling problem. Based on the household driving pattern
and regulation signal data from the PJM market, a simulation frequency measurement have been proposed in [7], [8]. The
study of 1,000 EVs has been performed. The simulation results control strategy proposed in [7] ensures that an EV will be
show that the proposed online scheduling algorithm is able to charged to a desired level of state-of-charge (SOC), but it
smooth out the power fluctuations of the grid by coordinating the cannot achieve the global optimum among the EV fleet since
EV schedules, demonstrating the potential of V2G in providing the control is for a single EV and the discharging/charging
regulation service to the grid.
schedules of EVs are not coordinated. In [8], each EV decides
its own charging/discharging schedule in response to its locally
I. I NTRODUCTION
measured frequency deviation, but there is no guarantee on
Electric vehicles (EVs) can provide ancillary services, charging needs. Again, global optimum is not achievable.
such as regulation services, to the utility grid if their In [9], [10], centralized scheduling is employed where an
charging/discharging schedules are coordinated appropri- aggregator acts as the central controller. They try to optimize
ately. Therefore, the scheduling problem for EV charg- the schedules of the EV fleets so as to maximize the revenue of
ing/discharging is an important research topic in recent years. regulation service. However, they do not consider whether the
Most research on this topic falls into two categories: EV EVs are adequately charged. The concept of EV aggregators is
charging control and vehicle-to-grid (V2G) scheduling. applied in [9], [10], but no existing work has clearly specifiied
Studies on EV charging control regard the EV charging the structure or architecture of the V2G system consisting of
load as controllable load. Research on smart/coordinated EV the utility grid, aggregators, and EVs. Moreover, the functions
charging can be categorized as centralized control [1], [2], and relations of different parts of the system have not been
and decentralized control [3], [4]. They try to determine the well defined.
optimal EV charging schedules to minimize the distribution
system losses by flattening the total load without sacrificing the In this paper, we try to tackle the problems of V2G
charging needs of EVs. However, the centralized algorithms scheduling mentioned above. In Section II, we first propose
proposed in [1], [2] are inadequate due to high computational a multi-level architecture for bidirectional V2G regulation
complexity as the number of EVs scales up. This is one of service with three types of operation protocols. The scheduling
the key reasons why the decentralized control strategies [3], problem is then formulated as a convex optimization problem
[4], which distribute the computation to EVs and allow them which may be operated online with guarantees of adequate
to determine their schedules locally, are preferred, since the charging for EVs in Section III. A decentralized algorithm
penetration of EVs is expected to be high in the future. We is designed to solve the optimization problem in Section IV.
regard the formulation and algorithms proposed in [1]–[4] as Simulation results are presented and discussed in Section V.
forecast-based, since they all require accurate forecasts of the Finally, Section VI draws the conclusions.

978-1-4799-1526-2/13/$31.00 ©2013 IEEE 43

 IEEE SmartGridComm 2013 Symposium - Architectures and Models for the Smart Grid

II. S YSTEM A RCHITECTURE

In this section, we propose a multi-level system architecture,
the schematic diagram of which is shown in Fig. 1, for
the provision and operation of bidirectional V2G frequency
regulation service. The architecture consists of three key com-
ponents: the grid operator, the aggregators, and a set of EVs.
It has a hierarchical structure with multiple levels of nodes.
The utility grid operator is the root node. The aggregators
directly connected to the grid operator are called level-1
aggregators. An aggregator directly connected to a group of
level-(l + 1) aggregators is called a level-l aggregator, where
l = 1, 2, · · · , NL , and NL is the number of aggregator levels
in the system. For convenience, the aggregators directly con-
nected to EVs are called aggregators of EVs. Correspondingly,
all the other aggregators are called aggregators of aggregators.
Each aggregator node can be viewed as the “root node” of
a subtree of aggregators and EVs. The size of a subtree
is determined by the size of its subordinate EV fleets and Fig. 1. Multi-level Architecture of Bidirectional V2G.
other geographical, economic, and/or technical factors, such
as the communication radius, delay, and cost between nodes at
different levels. For instance, a parking lot or a certain area of guarantees for EVs, and does not depend on the accurate
a large parking lot can install an aggregator of EVs. A number forecast of regulation demand.
of such parking areas can be controlled by an aggregator of
aggregators. A. Model and Constraints
As illustrated in Fig. 1, there are three types of oper- Consider a scenario where an aggregator of EVs coordinates
ation protocols, namely, grid operator-aggregator protocol, NEV EVs to schedule their charging/discharging profiles to
aggregator-aggregator protocol, and aggregator-EV protocol. meet the share of regulation demand assigned to the aggregator
Each protocol operates between a node and its immediate and fulfill the charging requirements of the EVs over a par-
subordinate nodes. The grid operator-aggregator protocol gov- ticipation period [Tbegin , Tend ], which is divided equally into
erns how the grid operator assigns the regulation requests NT time slots of length ∆t. Let T := {Tk |k = 1, 2, . . . , NT }
to and coordinates the level-1 aggregators to meet regulation be the set of the slotted participation period, R(Tk ) be the
demand. The regulation requests are the aggregators’ shares assigned share of regulation demand at time slot Tk , and
of regulation demand according to their signed contracts for Pn (Tk ) be the charging/discharging power of EV n at Tk ,
regulation service. The aggregator-aggregator protocol governs for Tk ∈ T and n ∈ N := {1, 2, . . . , NEV }. We assume
how an aggregator of aggregators coordinates its immedi- that the share of regulation demand of an aggregator accounts
ate subordinate aggregator to meet regulation requests. The for a fixed proportion of the total regulation demand in the
aggregator-EV protocol specifies the process and algorithm grid during T . R(Tk ) > 0 means that the aggregator should
for an aggregator of EVs to coordinate its connected EVs to coordinate the NEV EVs to provide regulation up or deliver
decide their charging/discharging schedules. In other words, active power to the grid. Similarly, R(Tk ) < 0 provides
the aggregators act as the interface between the utility grid regulation down or absorbs excessive power from the grid.
and EV fleets so that the grid operator does not need to When Pn (Tk ) > 0, EV n is charging or consuming power.
care about the individual charging/discharging profiles of EVs. When Pn (Tk ) < 0, it is discharging its battery or delivering
These EVs can collectively form a massive energy storage power back to the grid. Denote the plug-in time and plug-out
system to provide regulation service. In this paper, the focus time of EV n as Tn,in and Tn,out , respectively. Define the
is on the design of the aggregator-EV protocol. lower bound Pn (Tk ) and upper bound Pn (Tk ) of Pn (Tk ) for
Tk ∈ T and n ∈ N as:
III. P ROBLEM F ORMULATION (
Pn,discharge if Tk ∈ [Tn,in , Tn,out ]
The scheduling problem of EVs, which determines the Pn (Tk ) := (1)
0 if Tk ∈
/ [Tn,in , Tn,out ]
charging/discharging scheduling of EVs, is the most important
problem for the operation of the proposed architecture for the and (
V2G regulation service. In this section, we focus on the design Pn,charge if Tk ∈ [Tn,in , Tn,out ]
Pn (Tk ) := (2)
of the aggregator-EV protocol and derive a practical formu- 0 if Tk ∈
/ [Tn,in , Tn,out ]
lation of the online scheduling for V2G frequency regulation
for an aggregator of EVs. The proposed formulation jointly where Pn,discharge and Pn,charge denote the limits of dis-
considers the provision of regulation service and the charging charging power and charging power of EV n, respectively.

44
 IEEE SmartGridComm 2013 Symposium - Architectures and Models for the Smart Grid

Thus coordinates a group of EVs to update their schedules from

Pn (Tk ) ≤ Pn (Tk ) ≤ Pn (Tk ), t ∈ T , n ∈ N (3) Tk to TNT by the following optimization:
X
min (R(Tk ) + Pn (Tk ))2
{Pn (Ti )|n∈N ,k≤i≤NT }
Let SOCn,0 , SOCn (Tk ), and Cn be the initial SOC, SOC n∈N
at the end of Tk , and capacity of the battery pack of EV n, NT
X X (10)
respectively. We propose two constraints for the SOC of the + (E(R(Ti ) | {R(Tj )|1 ≤ j ≤ k}) + Pn (Ti ))2
battery pack during the plugged-in period of EV n, where i=k+1 n∈N
n ∈ N as follows:
Although (10) allows for real-time update of the sched-
SOCn (TNT ) ≥ SOCn,M inCharged (4) ules and does not need accurate forecast of regulation de-
SOCn,min ≤ SOCn (Tk ) ≤ SOCn,max , Tk ∈ T (5) mand, it still requires the calculation of the conditional ex-
pectation of every future regulation request, i.e., the term
E (R(Ti )|{R(Tj )|1 ≤ j ≤ k}) , for i = k + 1, k + 2, · · · , NT .
From (4), SOCn,M inCharged denotes the minimum value Unfortunately, this calculation requires the distribution of
of SOC that EV n needs to reach before it is plugged out. regulation demand which is not known a priori. Nonetheless,
The minimum amount of energy to charge for EV n before it since frequency regulation is a zero-energy service, which
is plugged out, En,min , can be expressed as: means the expectation of the total energy that the regulation
En,min := Cn (SOCn,M inCharged − SOCn,0 ) (6)
service requires is zero over a long period of time, we can
make the following assumption:
X
E( R(Tk )) = 0 (11)
From (5), SOCn,min and SOCn,max denote the lower and Tk ∈T
upper SOC limits, respectively, of EV n for all Tk ∈ T . This
constraint is to avoid deep-discharging or over-charging of the By applying the Cauchy Inequality, we can derive a lower
battery so as to protect the longevity of the battery. bound of the second summation in the objective function of
(10) when k ≤ NT − 1 as follows:
B. Formulation of the Scheduling Problem NT
X X
According to the model in Section III-A, an aggregator (E(R(Ti ) | {R(Tj )|1 ≤ j ≤ k}) + Pn (Ti ))2
receiving a regulation request R(Tk ) would coordinate its i=k+1 n∈N
managed EVs to determine their charging/discharging power NT
1 X X
to smooth out the fluctuation by minimizing: ≥ (E( R(Ti ) | {R(Tj )|1 ≤ j ≤ k}) + F Pn (Ti ))2
NT − k n∈N
X i=k+1
|R(Tk ) + Pn (Tk )| (7) (12)
n∈N

where
or X NT
2 X
(R(Tk ) + Pn (Tk )) (8) F Pn (Tk ) := Pn (Ti ) (13)
n∈N i=k+1
P
We call the term n∈N Pn (Tk ) in (7) and (8) the aggre- is the sum of future charging/discharging profile of EV n.
gated EV power at Tk ∈ T . Since (7) is non-convex, it entails By (11), the conditional expectation of the sum of the future
a higher computational complexity for optimization than the regulation requests in (12) can be calculated as:
convex one (8). Thus, we choose (8) instead and introduce a
NT
forecast-based or offline formulation of the scheduling prob- X k
X
lem over the participation period T . Assume that, before the E( R(Ti ) | {R(Tj )|1 ≤ j ≤ k}) = − R(Tj ) (14)
participation period T , the aggregator receives the forecasting i=k+1 j=1
profile of its assigned regulation demand {R(Tk )|Tk ∈ T }.
Then, it coordinates the EVs to determine the optimal schedule Let Qn (Tk ) := (Pn (Tk ), F Pn (Tk )) be the schedule of EV
by the following optimization: n ∈ N , and Q(Tk ) := (Q1 (Tk ), Q2 (Tk ), . . . , QNEV (Tk ))
min
X
(R(Tk ) +
X
Pn (Tk ))2 (9)
denote the schedules of all EVs at time slot Tk ∈ T . We apply
{Pn (Tk )|n∈N ,Tk ∈T }
Tk ∈T n∈N
the lower bound derived in (12) and propose the formulation
of online scheduling for V2G frequency regulation as follows.
The optimization result of the forecast-based formulation (9) For any Tk ∈ T ,
provides the best possible schedule if the forecasting profile of min U (Q(Tk )) (15)
Q(Tk )
{R(Tk )|Tk ∈ T } is accurate. However, in reality, the forecast
of regulation demand is highly inaccurate and unreliable be- such that
cause regulation demand is vulnerable to forecasting errors of En,min X
k−1

generation and load. Therefore, the forecast-based formulation Pn (Tk ) + F Pn (Tk ) ≥ − Pn (Ti ) (16)
∆t i=1
is not appropriate or practical for frequency regulation.
Considering that regulation demand is derived from the Cn
Pn (Tk ) ≥ (SOCn,min − SOCn (Tk−1 )) (17)
regulation signals measured in real time, an online formula- ∆t
tion, which schedules the EV power in response to the real- Cn
Pn (Tk ) ≤ (SOCn,max − SOCn (Tk−1 )) (18)
time input of R(Tk ), is more realistic. At time slot Tk , the ∆t
aggregator receives the real-time signal of R(Tk ). It then Pn (Tk ) ≤ Pn (Tk ) ≤ Pn (Tk ) (19)

45
 IEEE SmartGridComm 2013 Symposium - Architectures and Models for the Smart Grid

NT NT
X X Algorithm 1 Online Scheduling
Pn (Tk ) ≤ F Pn (Tk ) ≤ Pn (Tk ) (20)
i=k+1 i=k+1
Input: At any time slot Tk ∈ T , the aggregator knows the
total number of time slots, NT , and the number of EVs,
where NEV , and it has received the requests of regulation demand
P
2 {R(Ti )|1 ≤ i ≤ k}. Each EV n ∈ N knows about its own
(Pn∈N Pn (TNT ) + R(TNT ))
 if k = NT
constraint set (16) – (20) and the constraint parameters.
2
U (Q(Tk )) := ( n∈N Pn (Tk ) + R(Tk )) + NT1−k ·
 (P
 Pk 2
Output: Schedule Q(Tk ) = (Q1 (Tk ), . . . , QNEV (Tk ).
n∈N F Pn (Tk ) − j=1 R(Tj )) otherwise
Choose a parameter β satisfying 0 < β < 2N1EV .
(21)
Initialize the schedule Q0n (Tk ) of every EV n ∈ N as:
(
The constraint set (16) – (20) is a simple transformation of (0, 0) k=1
the definitions and constraints (1) – (6) introduced in Section Q0n (Tk ) := (22)
(0, F Pn (Tk−1 )) otherwise
III-A.
It can be shown that the objective function (21) and the Set the iteration number m ← 1, repeat Steps 1) – 3).
set of feasible solutions under the constraint set (16) – (20) 1) The aggregator calculates the control signal sm (Tk ) as
follows. When k 6= NT :
are both convex. Therefore, the proposed formulation (15) –
(21) is a convex optimization problem. Although (15) only ∂U (Q(Tk )) ∂U (Q(Tk ))
sm (Tk ) := β( P , P )
seeks to optimize a lower bound of (10), it is tractable and ∂(
n∈N Pn (Tk )) ∂( n∈N F Pn (Tk ))
X m−1
more practical than (10). In addition, (15) requires much lower = 2β(R(Tk ) + Pn (Tk ),
computational complexity than (9) and (10) since it reduces n∈N
the number of variables needed significantly. 1
k
X X
( R(Tj ) + F Pnm−1 (Tk )))
NT − k j=1
IV. D ECENTRALIZED S CHEDULING A LGORITHM n∈N
(23)
In this section, we propose a decentralized algorithm to
solve the proposed convex optimization problem (15) – (21) When k = NT :
for V2G scheduling. ∂U (Q(TNT ))
Let h·, ·i represent the dot product operation and k·k denote sm (TNT ) := β( P , 0)
∂(n∈N Pn (TNT ))
the Euclidean norm. The proposed algorithm is presented in X m−1 (24)
= 2β(R(TNT ) + Pn (TNT ), 0)
Algorithm 1, which is inspired by the decentralized algorithm n∈N
proposed in the work of the optimal EV charging control [3].
The stopping criterion of Algorithm 1 can be based on the Then, it broadcasts the control signal sm (Tk ) to all
number of iteration performed, and/or the convergence of the EVs.
control signal sm within the convergence tolerance. At each 2) Each EV n ∈ N calculates a new schedule Qm n (Tk )
iteration, each EV n ∈ N solves an convex optimization as:
problem with only two variables Pn (Tk ) and F Pn (Tk ). Hence, Qm m
n (Tk ) := arg min (hs (Tk ), Qn (Tk )i
Qn (Tk )
Algorithm 1 requires very low computational cost. In addition, 1 2
+ Qn (Tk ) − Qm−1 (Tk ) ) (25)
user information privacy can be preserved since the data 2
n

related to the driving patterns of the EV owner, including the s.t. (16), (17), (18), (19), (20) hold
SOC of the EV’s battery, the plug-in time, and plug-out time,
and reports Qmn (Tk ) to the aggregator.
do not need to be reported in advance to the aggregator.
3) If the stopping criterion is not met, set m ← m + 1
Theorem 1. The schedule Qm converges to the optimal and go to Step 1).
solution for the convex optimization problem (15) – (21) as Return Qn (Tk ) = Qm n (Tk ), ∀n ∈ N .
m → ∞.
The proof of Theorem 1 is similar to that of Theorem 3
in [3]. Therefore, we skip the proof due to the constraint in A. V2G Scheduling Algorithms
space. The forecast-based scheduling and the online scheduling
will be studied.
V. C ASE S TUDY As indicated in Section III-B, the forecast-based or offline
formulation (9) will provide the best possible scheduling
In this section, the performances of the proposed online results only when the forecast profile of the regulation requests
formulation and decentralized algorithm are studied by sim- is accurate. The performance of the forecast-based scheduling
ulation. We first outline a set of scheduling algorithms for with inaccurate forecast will also be tested. The forecast error
the provision of the V2G regulation service, followed by e(t) is introduced to the actual regulation request R(t) as:
a performance metric introduced for the comparison of the R(t) = (1 + e(t))Rf (t), t ∈ T (26)
scheduling algorithms. The simulation setup is then specified.
Finally, the simulation results are presented and discussed.

46
 IEEE SmartGridComm 2013 Symposium - Architectures and Models for the Smart Grid

where e(t) ∼ N (0, 0.3) and Rf (t) is the forecast regulation

profile. The forecast-based scheduling will be solved by the
algorithm Optimal Decentralized Charging proposed in [3].
In order to show that the proposed online formulation (15)
– (21) can jointly handle provision of regulation service and
charging requirement of EVs, a so-called “myopic” online
scheduling which only considers the current regulation de-
mand is introduced as follows:
X
min ( Pn (Tk ) + R(Tk ))2
{Pn (Tk )|n∈N }
n∈N (27)
s.t. (16), (17), (18), (19), (20) hold
Fig. 2. Simulation results of forecast-based scheduling when θbi = 1.
Both the proposed online scheduling and the myopic online
scheduling (27) will be studied. They will be solved by the
proposed decentralized scheduling algorithm, Algorithm 1.
B. Performance Metric
First, we define the total power Ptotal (t), which is the sum
of the regulation request and the aggregated EV power, as
follows: X
Ptotal (t) := R(t) + Pn (t), t ∈ T (28)
n∈N

The variance of the total power profile, Var(Ptotal (t)), is

used as the performance metric, which is defined as follows:
NT NT
1 X 1 X
Var(Ptotal (t)) := Ptotal (Tk )2 − 2 ( Ptotal (Tk ))2 Fig. 3. Simulation results of online scheduling when θbi = 1.
NT NT
k=1 k=1
(29)
15 January 2013, because the regulation signal and regulation
A smaller variance Var(Ptotal (t)) indicates a more flattened demand are related linearly [14].
profile of the total power, which means that the fluctuations of
D. Simulation Results
the regulation requests are better absorbed by the aggregated
EV power, resulting in a better scheduling performance. The results of the forecast-based or offline scheduling in (9)
are shown in Fig. 2. The proportion θbi of bidirectional V2G
C. Simulation Setup is 1, which means all the EVs participate in bidirectional V2G.
The simulation scenario is an aggregator coordinating 1,000 It can be observed that, ideally, the total power profile with
EVs to decide their schedules from 19:00 on a weekday to accurate forecast (the dash-dotted curve) is flat indicating that
7:00 on the following day. This 12-hour period of time is the power fluctuations are smoothed out. We note that the
divided equally into NT = 144 slots of length ∆t = 5 dash-dotted curve is close to a constant positive load of about
minutes. All the EVs are assumed to have been contracted 366 kW. This phenomenon is due to the charging requirement
to provide V2G regulation, either unidirectionally or bidirec- of the EVs. Because of the zero energy assumption (11) of
tionally. θbi ∈ [0, 1] denotes the proportion of the EVs that regulation demand, the constant positive load is approximately
participate in bidirectional V2G. According to the standard equal to the power consumption for satisfying the charging
Level 2 charging in the U.S.A [11], we assume that the power needs of EVs. The total power profile with inaccurate forecast
of the EVs that participate in bidirectional and unidirectional (the dashed curve) reflects a more realistic case. Since the
V2G can vary from -4.0 kW to 4.0 kW and from 0 to 4.0 schedules obtained by the offline scheduling are not able to
kW, respectively. According to [12], the distribution of plug- react to the change of real-time regulation demand, the total
in time of EVs is close to a normal distribution. Hence, in the power profile has frequent and significant fluctuations.
simulation, the plug-in time of the EVs is assumed to follow The simulation results of the proposed online scheduling
a normal distribution with the mean at 19:00 and the standard (15) and the myopic scheduling (27) when θbi = 1 are shown
deviation is equal to 1 hour. In addition, the plug-out time is in Fig. 3. Since the myopic scheduling just tries to meet the
also assumed to follow a normal distribution with the mean current regulation request at each time slot without considering
at 7:00 and the standard deviation is equal to 1 hour. Any the influence of the EVs’ charging requirements, it results in a
plug-in time before 19:00 and plug-out time after 7:00 is set huge peak load (the dashed curve) for charging the EVs during
to be 19:00 and 7:00, respectively. The profiles of frequency the last two hours of the simulation period. The proposed
regulation demand used in the simulation are scaled data of online scheduling outperforms the myopic scheduling and the
the fast response regulation signal of the PJM market [13] offline scheduling with forecasting errors. The total power
from 19:00 on Monday, 14 January 2013 to 7:00 on Tuesday, profile (the dash-dotted curve in Fig. 3) of the proposed online

47
 IEEE SmartGridComm 2013 Symposium - Architectures and Models for the Smart Grid

protocols introduced for this architecture, the EV fleets are

able to provide frequency regulation service to the utility grid
under the coordination of the aggregators. Focused on the
design of the aggregator-EV protocol, an online formulation
for the EV charging/discharging scheduling problem is then
proposed. We apply a decentralized algorithm to solve the
proposed scheduling problem. Our simulation results show
that the proposed online scheduling algorithm can perform
nearly as well as offline scheduling with accurate forecasts
of regulation requests and outperforms off-line scheduling
with inaccurate forecasts. Future work will focus on the
algorithm design of the grid operator-aggregator protocol and
the aggregator-aggregator protocol for the distributed control
of regulation demand.

Fig. 4. Influence of θbi on the proposed online scheduling. ACKNOWLEDGMENT

This work is supported in part by the Collaborative Research
TABLE I Fund of the Research Grants Council, Hong Kong Special Ad-
P ERFORMANCE OF VARIOUS S CHEDULING A LGORITHMS WHEN θbi = 1 ministrative Region, China, under Grant No. HKU10/CRF/10.
Forecast-Based Online R EFERENCES
Algorithm
Accurate Inaccurate Proposed Myopic
Var(Ptotal (t)) 1.7 6765.6 296.1 696955.7 [1] K. Clement-Nyns, E. Haesen, and J. Driesen, “The Impact of Charging
Plug-In Hybrid Electric Vehicles on a Residential Distribution Grid,”
IEEE Transactions on Power Systems, Vol. 25, No. 1, pp. 371–380,
Feb. 2010.
scheduling is almost as flat as that (the dash-dotted curve [2] E. Sortomme, M. Hindi, S. MacPherson, and S. Venkata, “Coordinated
in Fig. 2) of the offline scheduling with accurate forecast, Charging of Plug-In Hybrid Electric Vehicles to Minimize Distribution
System Losses,” IEEE Transactions on Smart Grid, Vol. 2, No. 1, pp.
although it has some minor fluctuations during the final one 198–205, Mar. 2011.
hour because the EVs start to be plugged out and thus become [3] L. Gan, U. Topcu, and S. H. Low, “Optimal Decentralized Protocol
unavailable for providing regulation. for Electric Vehicle Charging,” IEEE Transactions on Power Systems,
Vol. 28, No. 2, pp. 940–951, May 2013.
Table I compares the performance of the forecast-based [4] Z. Ma, D. Callaway, and I. Hiskens, “Decentralized Charging Control
scheduling and the online scheduling when θbi = 1 based on for Large Populations of Plug-In Electric Vehicles: Application of the
the Var(Ptotal (t)) metric. It again shows that our proposed Nash Certainty Equivalence Principle,” Proceeding of IEEE CCA 2010,
pp. 191–195, Sep. 2010.
online scheduling algorithm can perform nearly as well as [5] W. Kempton and J. Tomić, “Vehicle-to-Grid Power Implementation:
the forecast-based scheduling with accurate forecasts, and From Stabilizing the Grid to Supporting Large-Scale Renewable En-
outperforms the forecasting-based scheduling with forecast ergy,” Journal of Power Sources, Vol. 144, No. 1, pp. 280–294, Jun.
2005.
errors and the myopic online scheduling. [6] B. J. Kirby, “Frequency Regulation Basics and Trends,” Oak Ridge
Fig. 4 presents the influence of the participation ratio θbi National Laboratory, Tech. Rep., Dec. 2004.
of bidirectional V2G on the performance of the proposed [7] Y. Ota, H. Taniguchi, T. Nakajima, K. Liyanage, J. Baba, and
A. Yokoyama, “Autonomous Distributed V2G (Vehicle-to-Grid) Satis-
online scheduling. The total power profile (the pointed curve) fying Scheduled Charging,” IEEE Transactions on Smart Grid, Vol. 3,
when θbi = 0, which means that all the EVs participate No. 1, pp. 559–564, Mar. 2012.
in unidirectional V2G, has many positive spikes concurrent [8] H. Yang, C. Y. Chung, and J. Zhao, “Application of Plug-In Electric Ve-
hicles to Frequency Regulation Based on Distributed Signal Acquisition
with the regulation up requests because the EVs cannot Via Limited Communication,” IEEE Transactions on Power Systems,
discharge the batteries to provide energy back to the grid. Vol. 28, No. 2, pp. 1017–1026, May 2013.
The total power curve (the dashed curve) when θbi = 0.5 [9] S. Han, S. Han, and K. Sezaki, “Development of an Optimal Vehicle-
to-Grid Aggregator for Frequency Regulation,” IEEE Transactions on
is similar to that (the dash-dotted curve) when θbi = 1, except Smart Grid, Vol. 1, No. 1, pp. 65–72, Jun. 2010.
for the performance degradation in the final two hours of [10] E. Sortomme and M. El-Sharkawi, “Optimal Scheduling of Vehicle-to-
the simulation period. Such small performance degradation Grid Energy and Ancillary Services,” IEEE Transactions on Smart Grid,
Vol. 3, No. 1, pp. 351–359, Mar. 2012.
suggests that it may not always be necessary to have all [11] SAE Electric Vehicle and Plug-in Hybrid Electric Vehicle Conductive
EVs enabled with bidirectional V2G and there may exist a Charge Coupler, SAE Std. J1772, 2010.
minimum θbi to satisfy the regulation requests. Considering [12] S. Shao, M. Pipattanasomporn, and S. Rahman, “Grid Integration of
Electric Eehicles and Demand Response with Customer Choice,” IEEE
the high infrastructure cost of bidirectional V2G, it is desired Transactions on Smart Grid, Vol. 3, No. 1, pp. 543–550, Mar. 2012.
to determine the minimum θbi . However, due to the constraint [13] “Fast response regulation signal,” PJM, Feb.
in space, we skip the discussion in this paper. 2013. [Online: http://www.pjm.com/markets-and-operations/ancillary-
services/mkt-based-regulation/fast-response-regulation-signal.aspx]
VI. C ONCLUSIONS [14] Y. Rebours, D. Kirschen, M. Trotignon, and S. Rossignol, “A Survey of
Frequency and Voltage Control Ancillary Services — Part I: Technical
A multi-level architecture for bidirectional V2G regulation Features,” IEEE Transactions on Power Systems, Vol. 22, No. 1, pp.
service is proposed. Based on the three types of operation 350–357, Feb. 2007.

48