IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing

Design of a V2G Aggregator to Optimize PHEV

Charging and Frequency Regulation Control
Ran Wang, Yifan Li, Ping Wang and Dusit Niyato
School of Computer Engineering
Nanyang Technological University, Singapore

Abstract—The key elements in an indirect V2G system archi- of vehicles and their limited individual storage capacity make
tecture are aggregators. They act as an interface between the the direct architecture impractical in the existing contracting
grid and a group of plug-in hybrid electric vehicles (PHEVs). In framework (e.g., with minimum 1 MW threshold for ancillary
this paper, we design an optimal vehicle to grid (V2G) aggregator
to control the charging and frequency regulation processes of a services hourly contracts) [3]. The alternative indirect V2G
group of PHEVs. We consider a problem that an aggregator has system architecture involves several aggregators. In this regard,
to minimize the overall cost of PHEV fleet in a multiple time each aggregator aggregates the services provided by individual
slot horizon and meet the required battery level when PHEVs PHEVs to make a single controllable power resource. The
plug out. We adopt summation of PHEVs’ expenditure in a aggregator is an intermediate interface between the vehicles
finite number of time slots as our objective function, which is
a quadratic optimization problem. A model predictive control and the grid operator [3].
based (MPC-based) PHEV charging and regulation algorithm Due to advancement of the power storage technology,
is proposed to schedule the charging and regulation processes. PHEVs’ battery capacity can range from 16 kWh to 53 kWh
Through the numerical experiments, we obtain the optimal and the frequency regulation is performed on the MW basis
charging and frequency regulation sequences for each PHEV, in most electricity market [4], [5]. Therefore, an aggregator is
the effect of price prediction error on PHEV’s cost as well as the
impact of penalty factor to plug-out State of Charge (SOC). It necessary to deal with hundreds to thousands of vehicles while
is also shown that by taking the optimal control sequences, the simultaneously providing the regulation service on large-scale
PHEV owner can reduce his cost and depart with desired SOC. power delivery [3].
In general, an aggregator may take two roles in the V2G
I. I NTRODUCTION system. With regard to PHEVs, the aggregator represents the
The widespread use of plug-in hybrid electric vehicles grid operator, trying to coordinate charging and regulation
(PHEVs) over the next few decades will result in a great processes to fulfill charging the batteries for PHEV owners
number of benefits to the electric power industry. In USA, while maximizing the overall revenue of providing services.
the goal has been set to achieve a penetration of 1 million For a grid system operator, the aggregator represents agent of
PHEVs by 2015 [1]. The new plug-in concept is enabling PHEVs, regarded as a dedicated regulation provider and elec-
the utilization of vehicle batteries for grid-side benefit, which tricity consumer. Therefore, designing an efficient aggretator is
is referred to as Vehicle-to-Grid or V2G [2]. Utilizing the challenging, due to the fact that the conditions of each vehicle
V2G technology, the PHEV will be able to feed power into including current state of charge (SOC), expected parking time
the grid and the PHEV users can earn revenues. Specially, and expected plug-out battery level differ from each other.
the V2G system can help to regulate the frequency in a Thus, it is necessary to propose an efficient V2G operation
power system. Unbalanced active power will lead to the method that provides the frequency regulation service in an
current frequency variation in the power grid. Currently, such optimal way while charging each vehicle to a satisfactory SOC
regulation is achieved mainly by increasing fast response level before departure.
generators, which are very costly. Alternatively, PHEVs can For V2G aggregator design problems, a few related works
help by charging their batteries and increase their load demand can be found in the recent literature. [6] presents a set of
when the frequency is too high. While on the contrary, if the schemes to distribute the power for V2G regulation service
frequency is too low, by terminating charging or discharging when this service is provided by electric vehicles (EVs)
a number of PHEVs, the adjustment can be done [3]. managed by an aggregator. The focus of this paper is on
There are two types of V2G system architecture, i.e., the fair distribution of power among the EVs. [5] and [7]
direct and indirect architecture. In the direct architecture, investigate the charging and regulation optimization problem
there exists a direct line of communication between the grid from the perspective of a single vehicle. They propose that
system operator and the vehicle, so that each vehicle can the regulation signal has a zero mean distribution and the
be treated as a deterministic resource to be commanded by frequency regulation itself does not affect the change of SOC
the grid system operator. In this case, as the operator needs under this circumstance. [8] introduces a discrete dispatch
to directly interact with millions of individual PHEVs, the algorithm for EVs performing frequency regulation. This
amount of signals and control tasks on the grid operator will be algorithm switches EVs on and off to meet the aggregators
overwhelming. In addition, geographically distributed nature total regulation dispatch required based on each EV’s charging

978-1-4799-1526-2/13/$31.00 ©2013 IEEE 127

 IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing

expected departure time, battery capacity, charging rate and

some other parameters. The prerequisite for PHEV owners to
participate in this activity is that their vehicles are charged to a
"
"
satisfactory level at the plug-out time. Other incentives include
direct payment, credit points or battery maintenance. However,
if the driver does not follow the contract, which means that he
"
!
!
drives away before the pre-notified departure time, he has to
accept the fact that the battery may not be charged enough at
the plug-out time. However, in practice, since the aggregator
will have contract with hundreds to thousands of vehicles, the
proportion of unexpected departure will be a constant. The
Fig. 1: System architecture. studies have shown that under such a condition, the early
departure is negligible in the regulation process [5]. Then, the
aggregator will make another contract with the grid operator
on how much regulation capacity that the aggregator can
priority. [9] studies the problem of stimulating self-interested
provide to the power grid. Alternatively, the aggregator may
EVs in providing ancillary services to the power grid.
report the amount of power that it needs from the grid operator
In this paper, we propose a new scheme from the perspective
for charging vehicles. Note that the regulation contract size
of an aggregator for organizing the charging and frequency
should not exceed the total available capacity of a PHEV fleet.
regulation processes for PHEVs. We consider the problem
Once these contracts are established, the aggregator can be
that an aggregator has to optimally schedule the charging and
regarded as a dedicated energy consumer and service provider
regulation processes of a group of PHEVs to minimize the
for grid operator, which simplifies the control task of the grid
cost, and at the same time meet the required battery level when
operator significantly.
each PHEV plugs out. The main challenge of this problem
is that the number of PHEVs in the system is dynamic and B. System Model
aggretator does not know when a new vehicle will arrive. To
In this paper, the intended time cycle for the operation of
solve this problem, a model predictive control based (MPC-
the aggregator is divided into H time slots, where H = |H|,
based) PHEV charging and regulation algorithm is proposed to
and H is the set of all time slots. An adaptive pricing scheme
schedule the charging and regulation processes. The aggregator
is considered where the exact price values (charging price,
will plan charging and regulation sequences at the beginning
regulation up price and regulation down price) for each time
of each time slot by solving the quadratic programming
slot are calculated in real time and are announced only at the
problem. Numerical results demonstrate the optimal charging
beginning of each operation period. For the next H − 1 time
and regulation sequences of the vehicles. We also present the
slots, we are only aware of the predicted prices (H is also
impact of price prediction error and compare the cost for
the price prediction horizon). The goal of the aggregator is
utilizing and not utilizing the optimal charging result.
to optimize charging and regulation processes to satisfy the
The rest of this paper is organized as follows. Section II
charging requirements of PHEV owners and simultaneously
presents the system overview (e.g., V2G charging and fre-
minimize the total expenses of the PHEV fleet. For the
quency regulation) and system model under our consideration.
charging process, the PHEV should pay to the power grid due
The problem formulation and model predictive control (MPC)-
to buying energy, whereas for regulation up (discharging) and
based algorithm is discussed in Section III. Section IV gives
regulation down (charging), PHEV can earn revenue. Note that
the performance evaluation and analysis. Finally, conclusions
even though charging and regulation down are physically the
are drawn in Section V.
same for the PHEV battery, they have different intentions. For
II. S YSTEM OVERVIEW the charging process, PHEVs are energy consumers. However,
for the regulation down, PHEVs are service providers.
A. Roles and Operations of Aggregator The aggregator control scheme works as follows. In a time
An aggregator should be able to provide the frequency slot, each vehicle has three states, i.e., idle, charging and
regulation service for the grid operator in a desired scale regulation. The operation period of the PHEV is the period
by organizing the PHEVs rationally [5]. In this case, the in which vehicle charges its battery or provides regulation
vehicle cannot discharge by itself, but only regulation up is service for the power grid (in charging or regulation state,
commanded from the grid operator [7]. The architecture of the respectively). Since we divide the time into multiple time
system is simplified as shown in Fig. 1. We can simply observe slots, we define the operation period of PHEV as the set of
that the aggregator acts as an interface between vehicles and continuous time slots that fall between the arrival time and
grid operator. The process of the system works as follows. departure time. Without loss of generality, we assume that at
First, the aggregator will make a contract with each PHEV each time slot, the cost of buying electricity is proportional
for using its battery based on the current state of the vehicle. to the energy amount. However, the revenue due to providing
The PHEV will announce to the aggregator its current SOC, frequency regulation service is proportional to the available

128
 IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing

energy capacity. This is because the fluctuation of power fn charging time needed for PHEV n to fill the gap
changes between positive and negative is almost evenly dis- between the initial SOC and desired plug-out SOC
tributed. Also, the amount of frequency regulation is expected α penalty factor for not charging the vehicle to the
to have a zero mean distribution. Thus, frequency regulation desired SOC
will not affect the SOC of the battery in a long period. It xmax
n maximum permitted battery level of vehicle n
is also important to note that what the PHEVs provide is xmin
n minimum permitted battery level of vehicle n.
the regulation capacity, not the truly charging (discharging) βn the plug-out time slot of PHEV n.
H
energy. Therefore, intuitively, the aggretator should allow the CN and RH N are the matrices of cn (h) and rn (h) for n =
vehicles to charge when the electricity price is low and to join 1, . . . , N and h = 1, . . . , H, respectively (i.e., matrices of
the regulation service when the regulation price is high. decision variables).
For the objective function defined in (1), the first two terms
III. P ROBLEM F ORMULATION AND MPC-BASED
denote the sum of charging expenses of all the vehicles during
C HARGING AND R EGULATION A LGORITHM
time horizon H. qn is a constant representing the maximum
A. Objective Function and Constraints charging rate of PHEV n. The third and fourth terms denote
The aim of the proposed algorithm is to find an optimal the revenue obtained from providing frequency regulation ser-
vehicle charging and regulation control schedule to satisfy vices. Here, the frequency regulation revenue is proportional
PHEV plug-out battery level and minimize the cost of the to the effective capacity of the battery. The reason is that the
whole fleet over H time slots. Thus, we propose the following regulation up and regulation down signals are fairly distributed
quadratic optimization formulation: with zero mean distribution. The goal of frequency regulation
  is to balance the generator supply with user consumption.
X In this case, PHEVs provide a battery capacity instead of
min pc (k)  cn (k)qn  (1) real energy. Also notice that the vehicle’s SOC will affect
H ,RH
CN N
n∈N (k) the regulation up and regulation down capacities. However,
 
H+k−1
X X it is unnecessary to separate regulation up and regulation
+ p̂c (h)  cn (h)qn  down processes. The probabilities of having regulation up
h=k+1 n∈N (k) and regulation down signals are all closely equal to 0.5. The
X regulation down capacity is xmax − xn (h) and regulation up
xmax − xmin

n
−pr (k) rn (k)
n n
capacity is xn (h) − xmin
n , where xn (h) is vehicle n’s SOC at
n∈N (k)
time slot h. Thus, the total regulation capacity is in fact equal
H+k−1
X X to the effective battery capacity.
p̂r (h) xmax − xmin


− n n rn (h)
The last term is the total penalty cost due to failing
h=k+1 n∈N (k)
!β #2 to achieve the desired battery level. The penalty factor α
X X n
represents the significance for vehicles to achieve the desirable
+ α cn (h) − fn battery level when vehicle departs. Without loss of generality,
n∈N (k) h=k we assume that all vehicles have the same penalty factor. fn
s.t. 0 ≤ cn (h) ≤ 1 (2) is equal to the battery gap divided by the charging rate, which
0 ≤ rn (h) ≤ 1 (3) the aggregator can obtain at the beginning of each time slot.
If we use xdn to denote the desired plug-out battery level of
cn (h) + rn (h) = 1 if h ≤ βn (4)
PHEV n and xin to denote the initial battery level, then we
cn (h) = 0, if h > βn (5) have
rn (h) = 0, if h > βn (6) xdn − xin
fn = . (7)
where: qn
k current time slot cn (h) and rn (h) are decision variables representing the pro-
h time index, e.g., hour portion of time that PHEV n is charging and regulating during
N set of PHEVs time slot h. For instance, cn (h) = 0.6 and rn (h) = 0.4
H set of time slots, H = [k, k + 1, . . . , k + H − 1] indicate that 40% of the whole time slot is used for charging
pc (k) charging price at time slot k and the remaining 60% is used for regulation. Thus, it is
p̂c (h) predicted charging price at time slot h obvious that cn (h) and rn (h) should vary within the range
pr (k) regulation price at time slot k between 0 and 1. Note that charging and frequency regulation
p̂r (h) predicted regulation price at time slot h are two processes that have totally different purposes, and
qn maximum charging rate of PHEV n they cannot happen simultaneously. We assume that during
cn (h) charging control parameter, denoting the proportion the controlled time slots, PHEVs will either charge or regulate
of time that PHEV n charges during time slot h for the grid operator within one time slot, which is captured
rn (h) regulation control parameter, denoting the proportion in the constraint defined in (4). Obviously, with the aim of
of time that PHEV n regulates during time slot h maximizing revenues, PHEVs want to regulate as long as they

129
 IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing

do not need to charge. Also, it is unnecessary to consider the 0.32

Charging Prices /S$

order of charging and regulation, since charging and regulation 0.3

prices remain the same within a time slot. We can assume that 0.28

in a time slot, PHEVs will first charge and then regulate for the 0.26
0 2 4 6 8 10 12
power grid. (5) and (6) indicate that after the PHEV departs, −3
Time slots index
x 10
the aggregator would stop controlling the vehicle. Thus we set 15

Regulation Prices /S$

10
cn and rn equal to 0 after time slot βn . 5

Also, the aggretator has to ensure that the desired battery 0

level should not exceed the maximum charging capacity of −5

0 2 4 6 8 10 12
Time slots index
vehicle during its parking period. In other words, if the vehicle
is charging during the entire parking period, it can definitely Fig. 2: Charging and regulation prices.
reach the pre-set desired battery level, which can be expressed
as follows:

β n − k ≥ fn . (8) (IV) Solve the quadratic programming optimization problem

(1) constrained by (2) ∼ (8) for the entire time horizon.
H
By solving the above optimization problem expressed in Solve for the solution matrices, i.e., CN and RHN , which
(1)-(8), we can obtain the optimal charging and regulation give an optimal operation schedule from time slot k to
decisions of the system for the entire H time slots. Unfor- time slot k + H.
tunately, due to vehicle arrival and departure, the system is (V) Apply the optimal solutions cn (k) and rn (k), for n =
k
dynamic and the number of vehicles and their parameters will 1, . . . , N , which is from the matrices CN and RkN ,
change over time. It is impossible for the aggregator to make respectively.
a stationary long term schedule sequence for all time slots. (VI) Set k = k + 1. Update the predicted model for the next
Also, the aggregator cannot greedily make a decision for a iteration. Update the current vehicles’ information. Go
single time slot because the plug-out SOC constraint is for to Step IV.
the whole plug-in duration. The above MPC-based charging and regulation algorithm
To address the above conditions, we propose the model shows the following advantages:
predictive control (MPC)-based charging and regulation algo- • The optimization solving is efficient, fast and easy to im-
rithm. plement, the time complexity of the algorithm is O(n3 ).
• Short-term price prediction is much more precise than
B. MPC-Based Charging and Regulation Algorithm the long-term one. The most updated price information
The basic idea of the MPC is that a finite horizon opti- could be effectively used in the schedule decision making
mization problem determining the series of optimal control process to minimize the overall cost.
operation is solved but only the first step of control sequence • The case that vehicles may join and leave the system
is actually implemented [10] [11]. The state variables over is considered. Vehicles’ information could be updated
the prediction horizon are described by a predictive model, quickly and each vehicle will be guaranteed to get a
with the initial state being the measured state of the actual satisfactory charging level when it departs.
system. After the first step of control sequence is applied, the Note that the control sequences that we obtained may not
actual state of the system will be updated and observed. Also, be the truly optimal schedule. The reason is that the control
the future states will be predicted [11]. Then, with the new sequence obtained from the algorithm is optimal given the
observation of system states, the optimization problem will current system state and the predicted prices at the decision
be solved again. In this paper, the predicted variables are the making time slot. However, the system is dynamic and predic-
charging and regulation prices for the upcoming H − 1 time tion error exists. As a result, the control sequence may not be
slots. Note that since the focus of this paper is to design the truly optimal. The truly optimal decision can only be obtained
optimal charging and regulation processes, the predicted states after the whole time horizon ends (when we can look back).
are assumed to be available, in which any prediction methods Therefore, the control sequence obtained from the proposed
can be applied. algorithm is the most practical given the available state and
With the MPC, a charging and regulation scheduling algo- vehicle information, which from the evaluation, yields the
rithm can be expressed as follows: good performance.
(I) Set k = 0.
(II) Select the time horizon H (e.g., 6 or 12 hours), the IV. P ERFORMANCE E VALUATION AND A NALYSIS
length of each time slot (e.g., 1 hour). Select prediction In the experiment, we consider a case of 800 vehicles to
model for charging and regulation prices. be scheduled at the beginning of the time slot. This fits the
(III) Obtain the current vehicles’ information (e.g. current requirements that the aggregator should have contracts with
SOC, expected departure time, battery capacity, and hundreds to thousands of vehicles and frequency regulation
charging rate). should be carried out at least on the M W basis. When new

130
 IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing

Extra Cost due to Imperfect Price Forecast

Charing result for vehicle 735 at each time slot 0.035
1

0.8
0.03
0.6
hours

0.4 0.025
0.2

Extra Cost /S$

0 0.02
1 2 3 4 5 6 7 8 9
Time slots index
Charing result for vehicle 326 at each time slot
1 0.015

0.8
0.01
0.6
hours

0.4
0.005
0.2

0
1 2 3 4 5 6 7 0
Time slots index 0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
Price Prediction Error

Fig. 3: Charging results for vehicle 735 and vehicle 326. Fig. 4: Mean extra cost due to price prediction error.

vehicles enter the system, they will be scheduled according B. Cost Comparison Due to Price Prediction Error
to the proposed algorithm. Vehicles will leave the system The accuracy of price prediction will affect the cost. The
after the prior noticed plug-out time. Here, three different control sequence from the proposed charging and regulation
types of vehicles including Sedan, Compact and Roadster algorithm is optimal for the predicted prices, but not for the
are considered. The charging rates of these types of vehicles real prices. We compare the average costs of two scenarios,
are 3 kW , 5.5 kW , and 6 kW and battery capacities of and the difference between these costs are called cost gap.
these types of vehicles are 16 kW h, 35 kW h and 53 kW h, The fist cost is from when the aggregator charges and regulates
respectively. The initial number of vehicles are 241, 232 according to the proposed MPC-based algorithm (the predicted
and 327, respectively. The current SOC of the vehicles are prices are with error). The second cost is from when the
uniformly distributed between 0.2 and 0.7. The expected plug- aggregator charges and regulates according to the truly optimal
out SOC of the vehicles are all 0.9. The schedule horizon is control sequences. This cost is obtained when the prices are
12 hours and is partitioned into 12 time slots with the constant perfectly known (e.g., at the end of all time slots). Basically,
duration. The parking duration of the vehicles ranges from 3 the second cost is the lower bound for the aggregator. The
to 12 hours. The charging and frequency regulation prices in average cost gap is shown in Fig. 4. Compared with the
the first time slot are shown in Fig. 2. However, only the prices truly optimal control sequence, the extra cost will be incurred
in the first hour is the actual prices, while those in the other due to the prediction error. Thus, accurate price prediction is
11 hours are the predicted prices. By solving the optimization important which can help the aggregator to determine a control
problem at the beginning of each hour, we obtain the charging sequence approximating the truly optimal one.
and regulation sequences of all the vehicles.
C. Cost Comparison for Different Charging Schemes

A. Optimal Charging and Regulation Results Next, we compare vehicles’ cost among three charging
schemes. In the first scheme, we control the vehicles according
We first assume that the price in the 12 hour time hori- to the control sequence obtained from the above MPC-based
zon is precisely predicted as shown in Fig. 2. To ease the algorithm. In the second scheme, all the vehicles will keep
presentation, we select to present the charging results for two charging until they reach the desired battery level and then
PHEVs. The selected vehicles are n = 735 and n = 326. Both start to regulate. For the third one, the vehicles will first try
vehicles are Roadsters and the initial SOC of the vehicles are to regulate as long as they can and then charge to the desired
0.42 and 0.53. The parking durations of the vehicles are 9 SOC. We present the costs of four randomly selected vehicles
hours and 7 hours. The charging results are shown in Fig. 3. and the mean costs of all the vehicles for aforementioned three
We observe that both vehicles plan to charge during the first schemes. The results are shown in Fig. 5. From Fig. 5, we
and fourth hours and regulate during the second and fifth can observe that the cost (bar height) for the first scheme is
hours. In the seventh hour, both vehicles will charge about less than the second and third ones. By taking the optimal
20 minutes and regulate for the rest of the hour. This verifies control sequence obtained from the proposed algorithm, the
the previous statement that the aggregator will let the vehicles costs of the individual vehicle and the cost of the vehicle
charge (regulate) when the charging and regulation price is fleet will be reduced. In this case, PHEV owners can save
low (high). Also we observe that if the electricity price varies costs by permitting the aggregator to control their vehicles
in accordance with the total load in the grid, the vehicles will and at the same time by waiting until their vehicles reach the
charge during the off-peak hours. If regulation demand is high, desired SOC before departing from the aggregator. For the
the grid operator can set higher regulation prices to motivate grid operator, by using electric power from a PHEV fleet with
vehicles to regulate for the power grid. In this case, through an optimal control sequence, the fast response but expensive
adjusting the prices, the grid operator can change the charging power generator can be avoided, saving a significant cost and
behaviors of PHEVs. environmental impact.

131
 IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing

Cost Comparison of Three Schemes

7
Scheme 1
low but the regulation price is high. We observe that when the
Scheme 2
6 Scheme 3 penalty factor is small, the vehicle will choose to regulate for
5
almost the entire time slot to gain more revenue. However, if
4
the penalty factor increases, the vehicle will use more time to
Cost /S$

charge to meet the required battery level since the punishment

3
for not achieving the requirement is high.
2

1
V. C ONCLUSION
0
In this paper, we have designed a V2G aggregator to
Vehicle 1 Vehicle 2 Vehicle 3 Vehicle 4 Vehicle 5 Average
Vehicles effectively schedule the charging and frequency regulation pro-
cesses for a group of PHEVs. We have considered the problem
Fig. 5: Cost comparison.
that the aggregator has to minimize the cost of all the vehicles
in the system during finite time horizon while simultaneously
0.905
ensuring that vehicles are charged to the desirable level at the
plug-out time. Since the system is dynamic (e.g., vehicles can
Average Plug−out SOC of the Vehicle Fleet

0.9

join or leave the system), we have proposed the MPC-based

0.895
charging and frequency regulation algorithm to achieve the
0.89 near truly optimal control sequences for the PHEVs given the
0.885
available state and vehicle information. Also, the advantages
of the proposed algorithm have been analyzed. Numerical
0.88
results have shown that the proposed algorithm is beneficial
0.875
0 5 10 15 20 25 30
to the individual vehicle as well as the whole PHEV fleet.
Penalty Factor α
Also, the impacts of price prediction error and penalty factor
Fig. 6: The impact of penalty factor α to the average plug-out have been also discussed. For the future work, the stochastic
SOC. programming and robust optimization frameworks could be
used to take into account the uncertainties of the charging and
regulation prices.
D. The Impact of Penalty Factor α R EFERENCES
We vary α (i.e., penalty factor from missing desired SOC [1] A. Bandyopadhyay, L. Wang, V. K. Devabhaktuni, and R. C. G. II,
“Aggregator analysis for efficient day-time charging of plug-in hybrid
at plug out time) from 1 to 30. Fig. 6 presents the SOC when electric vehicles.” 2011 IEEE Power and Energy Society General
PHEVs are plugged out. It is clearly shown that plug out Meeting, 24-29 July 2011, pp. 1–8.
SOC will converge to the desired SOC when α increases. We [2] S. Jang, S. Han, S. H. Han, and K. Sezaki, “Optimal decision on contract
size for v2g aggregator regarding frequency regulation.” International
observe that when α is larger than 5, the plug out SOC can Conference on Optimization of Electrical and Electronic Equipment, 15
satisfy the users’ requirements. However, when α is relative July 2010, pp. 54–62.
low, e.g., α = 1, the vehicles will spend more time on [3] C. Wu, H. Mohsenian-Rad, and J. Huang, “Vehicle-to-grid
systems: ancillary services and communications.” [Online]. Available:
regulation and less time on charging since the revenue gained http://ncel.ie.cuhk.edu.hk/sites/default/files/WuHamedHuangBook2011.pdf
by regulating is higher than the penalty. In this case, we can [4] A. Ipakchi and F. Albuyeh, “Grid of the future,” IEEE Power and Energy
control the penalty factor α to tighten or to relax the plug-out Magazine, vol. 7, pp. 52–62, 24 Feb. 2009.
[5] S. Han, S. Han, and K. Sezaki, “Development of an optimal vehicle-to-
SOC requirement. Fig. 7 shows the charging results of vehicle grid aggregator for frequency regulation,” IEEE Transactions on Smart
n = 735 when α = 3 and α = 10. Note that the key difference Grid, vol. 1, pp. 65–72, 27 May 2010.
occurs at the sixth time slot when the charging price is relative [6] J. J. Escudero-Garzás, A. Garcı́a-Armada, and G. Seco-Granados, “Fair
design of plug-in electric vehicles aggregator for v2g regulation,” IEEE
Transactions on Smart Grid, vol. 61, pp. 3406–3419, October 2012.
[7] S. Han, S. Han, and K. Sezaki, “Optimal control of the plug-in electric
1
Charging Results of Vehicle 735 when α=3 vehicles for v2g frequency regulation using quadratic programming.”
0.8
IEEE Innovative Smart Grid Technologies, 1-9 Jan. 2011, pp. 1–6.
0.6
[8] E. Sortomme and K. Cheung, “Intelligent dispatch of electric vehicles
Hours

0.4 performing vehicle-to-grid regulation,” in IEEE International Electric

0.2 Vehicle Conference. IEEE, 2012, pp. 1–6.
0
1 2 3 4 5 6 7 8 9
[9] Y. Gao, Y. Chen, C.-Y. Wang, and K. Liu, “Optimal contract design for
Time slots index ancillary services in vehicle-to-grid networks,” in IEEE International
Charging Results of Vehicle 735 when α=20
1 Conference on Smart Grid Communications. IEEE, 2012, pp. 79–84.
0.8 [10] Y. Xu, L. Xie, and C. Singh, “Optimal scheduling and operation of
0.6 load aggregator with electric energy storage in power markets.” North
Hours

0.4 American Power Symposium, 26-28 Sept. 2010, pp. 1–7.

0.2 [11] L. Xie and M. Ilić, “Model predictive economic/environmental dispatch
0
1 2 3 4 5 6 7 8 9 of power systems with intermittent resources.” IEEE Power and Energy
Time slots index
Society General Meeting, 26-30 July 2009, pp. 1–6.
Fig. 7: The impact of penalty factor α to charging results.

132