You are on page 1of 5

Economic and Battery Health Conscious

Vehicle-to-Grid Electric Vehicle Operation


Maigha and M. L. Crow
Department of Electrical and Computer Engineering,
Missouri University of Science and Technology
Rolla, Missouri
Email: mmck6@mst.edu, crow@mst.edu

Abstract—Maximizing profits through vehicle-to-grid interac- prices for maximizing utility for the customer. The amount
tions are in conflict with the minimizing battery degradation of energy supplied or received, and the power at which it
costs, and thus a major customer concern. A holistic scheduling is supplied/received depends on the charger with which the
problem is, therefore, required to consider the impact of the
charging power and that of the switching between vehicle-to- vehicle is equipped, the state-of-charge (SOC) of the battery,
grid and grid-to-vehicle modes to account for battery health. and its capacity. Battery degradation costs can be resolved
In this paper, we propose a mixed-objective approach toward into two components: one pertaining to capacity fade and
vehicle-to-grid interactions of a residential parking lot complex. the other to power fade [2], both of which affect its usable
The solution makes accommodation for battery health while automotive life application and are affected by the number of
considering the cost/profit model for the electric vehicles. We
discuss the importance of system constraints in achieving both, charge/discharge cycles it is exposed to. Since battery cost is
valley filling and peak shaving. A moving horizon model is the major component of vehicle cost, its health is of critical
used to determine real-time schedules for the vehicles. Customer importance to the owner.
convenience is the central focus of the proposed methodology. In [3], electrochemical models are used to optimize charging
profiles considering charging costs and establish their con-
I. I NTRODUCTION
flicting nature. This study focuses on the battery chemistry
Along with renewable energy resources, transportation elec- and charging cost. The aggregate load imposed by battery
trification can mitigate some of the problems related to health conscious charging of plug-in electric vehicles has been
fossil fuels including environmental, economic, and political investigated in [4]. These articles do not consider the V2G
concerns. Despite the recent impetus and federal support to mode or any system constraints. In [1], battery degradation
their rapid adoption, research has shown that transportation costs and vehicle charging costs have been combined together
electrification could impact the electric grid adversely unless to find optimal charging schedules for the vehicles. Intuitively,
controlled strategically. Increases in peak load demand can these two objectives are conflicting and thus require a mixed-
result in a need to deploy high cost generators, increases in objective analysis. Here, we use a simple weighted sum
thermal stresses of network components including lines and approach to understand this aspect of G2V/V2G dynamics.
transformers, network congestion, and other power quality The major contributions of this study include:
problems. Thus there is a need for scheduled charging of 1) Analyzing the impact of using battery degradation and
electric vehicles. charging costs under system constraints independently
Charge coordination of electric vehicle (EV) load can and using a mixed-objective approach to devise an
mitigate the aforementioned impacts on the grid. Further, optimal charging/discharging framework for a residential
the aggregated battery storage of a fleet may be used for parking lot
renewable energy support and for providing ancillary services, 2) Developing an adaptable, scalable, and computationally
thereby improving the operation of the grid. Accordingly, efficient solution focused on customer convenience, en-
electric vehicles may be used as demand-response resources, compassing the driving habits obtained from National
components in transactive-energy models, and energy-storage Household Travel Survey (NHTS-2009 [9])
business models. With the advancements in battery and ve- 3) Integrating intermittent/continuous energy exchange in
hicular technologies, automotive research is driven lower cost V2G/G2V modes.
solutions to battery chemistries capable of providing higher
energy and power densities. II. P ROBLEM F ORMULATION AND M ETHODOLOGY
Vehicle-to-grid (V2G) refers to supplying energy from the The residential parking lot controller (PLC) can be modeled
vehicle to the grid by discharging the battery. Grid-to-vehicle as a single entity that optimizes, schedules, and controls the
(G2V) refers to the normal operation of charging the vehicle available vehicles. The PLC identifies each vehicle with its
battery using the energy from the grid. G2V incurs cost to the unique VID that helps it retrieve the vehicle characteristic
customer whereas V2G is expected to generate revenue for information (VCI): bcapi , SOCi,min , SOCi,max . The VCI-
the customer. This aligns with the availability of time-varying tuple corresponds to the total battery capacity and the allow-

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


able minimum and maximum SOC of the vehicle respectively. Ppeak : forecasted peak load demand
We formulate the problem as a day-ahead scheduling problem Pmax : maximum power rating of vehicle charger in two
where the vehicle profile information (VPI) is made available quadrants for charging and discharging modes
to the PLC. The VPI for each vehicle i is a tuple containing the nV eh: total number of vehicles in the parking lot
expected driving profile. This includes ti,arr , ti,dep , SOCi0  ti,avail : time available for charging vehicle i
which indicate the expected time of arrival, time of depar- SOCi,min &SOCi,max : maximum and minimum allowable
ture for next day, and the SOC on arrival depending on SOC for vehicle i
the approximate miles planned. The PLC can then map the
VPI for each vehicle i to the corresponding VCI using the A. Battery Degradation Cost Model (BDCM)
VID (Figure 1). We define two objective functions: 1) cost As defined in [1], we use the sum of two com-
of battery lifetime degradation for EV i, and 2) cost of ponents to quantify battery degradation due to continu-
charging the battery. Following an initial mapping of all vehi- ous/repeated/frequent charging/discharging of EV battery. The
cles (nV eh), the PLC generates an optimal charging policy total cost Ψi,t is defined as a sum of 1) charging/discharging
Π and communicates the corresponding charging schedule power at time instant t: Ψai,t , and 2) changes in the charg-
{Πi : i ∈ {1, 2, ..., nV eh}, ∀t ∈ tavail } to each vehicle, where ing/discharging power or modes in any two consecutive time
tavail is the total parking time of the vehicle in the facility. intervals: Ψbi,t . Since we are considering a time interval of 1
hour, the power is numerically equal to the energy supplied.
  


ti,avail

ti,avail

ti,avail
Ψi (x) = Ψti = Ψai,t + Ψbi,t

t=1 t=1 t=2

ti,avail
 2 
ti,avail
 2
    
  
= β xti + α xti − xt−1
i (1)
t=1 t=2

∀x ∈ X where x = {xti : i ∈ [1, nV eh], t ∈ [1, tavail ]}


    
  
where β and α are model parameters expressed in $/kW h2 ,
  
xti is the power supplied or received during time interval t.

      
When extending it for the fleet, we get:

nV eh
Fig. 1. Schematic of the Proposed Scheme
argmin f1 (x) = Ψi (x) (2)
x
i=1
It is assumed that the PLC maintains a secure VCI database
and is equipped with the communication infrastructure to send The above function is convex and can be solved using
signals to the EVs. In a day-ahead scenario, each customer can quadratic optimization techniques.
log onto a secure web portal through a portable device and B. Customer Charging-Discharging Cost Model (CDCM)
log their expected VPI or this information can be extrapolated
from previous behaviors. After running the optimization, the Aside from battery life anxiety, a customer would expect
PLC can furnish the expected charging profile and data on to earn revenue while providing its battery for grid support.
costs incurred vs. revenues earned and battery degradation They would also like to charge the battery at the least cost
costs. Since this is a day-ahead local control scheme, neither possible. This can be achieved by selling energy during the
computational nor communication overheads are of major peak hours while buying energy during the valley period. To
concern. Nonetheless, the simple problem formulation is both incentivize customers and drive their behavior, utilities are
scalable and efficient. We also assume that the charger has devising special pricing plans for EV owners. Apart from
a unity power factor and is ideal (100% efficient). Since the time-of-use (TOU) rates, real-time pricing (RTP), peak-time
daily driving habits do not vary widely, this would provide a rebate, and critical peak pricing, some hybrid plans such as
tool for making an informed choice by the customer enroll in tiered-TOU, and business policies such as demand-bidding,
varied demand response programs. capacity-bidding, etc. have been introduced [7]. The total cost
In our discussion for any vehicle i and time instant t, the of charging/discharging the vehicle battery is:
following terms will be used: ti,avail nV eh 
   
δ: a variable in the range [0, (Pavg − Prest
)] argmin f2 (x) = xti λt (3)
x
tdep : time of next-day departure of vehicle i t=1 i=1

P (x)tsys : total load of the parking lot and residential complex where λt is the rate for buying/selling at time t which could
Pres
t
: non-responsive residential load be selected as the RTP or the TOU structure. We will be
Pavg : average of the forecasted residential load profile considering a net-metering framework in which the energy
Pi,req : total energy requirement of vehicle i is sold at the retail electricity price. It is assumed that the

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


meter will roll-back (reverse) for the equivalent amount of where w1 and w2 are the weights for functions f1 (BDCM) and
energy supplied by the source (the vehicle fleet). The function f2 (CDCM) respectively, and w1 + w2 = 1. Functions f1 and
in (3) is a linear function and can be solved using a linear f2 have been normalized using their respective optimal values
optimization technique such as the Simplex method. Note that f1∗ and f2∗ , obtained from BDCM and CDCM independently.
the profit earned has been treated as a negative cost and hence The problem preserves convexity and can be solved as a
the minimization function. quadratic optimization problem along with constraint sets
In addition to the cost, valley filling can be added to the (5) and (10). We use IBM-ILOG’s CPLEX [8] to solve the
objective function in the following form: optimization problem.
24
  2 III. S IMULATION R ESULTS
argmin f3 (x) = P (x)tsys − Pavg (4)
x
t=1 A typical summer load profile has been used from PGE [7]
C. Vehicle and System Constraints for demonstrating the optimization results. Time-of-use pricing
was also obtained from currently implemented rate structures
Broadly, the constraints can be categorized into: a) sys-
for PGE (Table I). We consider a three-level TOU structure
tem and b) vehicle constraints, giving due consideration to
which is designed using a clustering method proposed in
customer convenience and load serving entity’s needs. The
[6]. The average load data is scaled up to accommodate 245
optimization problem is subject to the following charging
residential customers with a 60% EV penetration. This size
constraints (5)-(9):
customer set has been considered to demonstrate the scalability
−Pmax ≤ xti ≤ Pmax ∀i, t (5) of the proposed schemes. The sample driving characteristics

ti,avail were obtained from the NHTS database [9]. Battery and
xti Δt = Pi,req ∀i (6) charger assignments to the vehicles were done in accordance
t=1 with the distributions given in [5]. We consider Type I and
xti Δt Type II chargers, as per vehicle needs and 5 battery sizes.
SOCit = SOCit−1 + ∀i, t (7) Since we are considering a parking lot, only the cumulative
bcapi
t effect of charging will be observed on the system.
SOCi dep = 1 ∀i (8)
SOCi,min ≤ SOCit ≤ SOCi,max ∀i, t (9) TABLE I
TOU R ATE S TRUCTURE [7]
and the following system constraints (10)-(11):

nV eh Rate Type Energy Charge (¢/kWh)
P (x)tsys = Pres
t
+ xti Δt ∀t (10) TOU 3-tier
13.101 20.779 32.306
2:00-10:00 11:00-13:00 14:00-20:00
i=1
21:00-1:00

nV eh
 
xti Δt ≤ Ppeak − Pres
t
+δ ∀t (11)
i=1 The following cases have been considered for analyzing and
The maximum power that the charger can supply or receive comparing the proposed schemes:
can vary according to (5). At each time instant, the SOC is a 1) Base Case: Uncoordinated charging of electric vehicles
function of the energy supplied or received by the battery in immediately upon arrival in G2V mode only
that hour (7). Since Δt = 1, it can be dropped explicitly from 2) Case 1: Valley filling using fitness function in (4) in
the equations. If a smaller time frame e.g. 15 minutes is used, V2G/G2V modes
this term will have to be included. The total amount of energy 3) Case 2: Minimizing costs incurred vs. revenues earned
supplied to the battery should be equal to its requirement (CDCM) in V2G/G2V modes using (3)
specified by the miles driven (6). The battery must be fully 4) Case 3: Minimizing battery degradation cost (BDCM)
charged by departure time (8) and the SOC must remain within in V2G/G2V modes using (1)
an appropriate range (9). The total load of the residence is the 5) Case 4: Minimizing weighted combination of Cases 1
sum of the non-responsive and the EV load at each hour (10). & 2 (BHPM) in V2G/G2V mode using (2)
Since the objective is valley filling, we want our EV load to lie It is assumed that the battery can discharge to a SOC
within the load window defined in (11) and thus never exceed level of 0.2. The parameter α = 10−3 $/kW h2 and β =
the non-responsive peak load demand. 5 × 10−4 $/kW h2 [1].
D. Battery Health Conscious Profit Model (BHPM) Figure 2 clearly shows that uncoordinated charging has
Since BCDM and CCDM are conflicting in nature, we use a the detrimental effect of causing an increase in system peak
simple weighted function approach to quantify the dynamics of load. With coordinated scheduling, the vehicles are charged
one against the other. The optimization function is expressed during the valley period and are discharged during the peak
as: hours. A peak load reduction of 15.5% is observed for the test
system under consideration, thus improving the load factor of
argmin f4 (x) = w1 (f1 (x)/f1∗ ) + w2 (f2 (x)/f2∗ ) (12) the system. Instead of demanding energy from the grid, the
x

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


Base Case Case 1 Case 2
1.5 1.5 1.5

1
1

1 0.5
0.5
0
0
0.5 −0.5

−0.5
−1

0 −1 −1.5
4 8 12 16 20 24 4 8 12 16 20 24 4 8 12 16 20 24

Case 3 Case 4−a


1.5 Case 4−b
1.5
1.5

1 1
1

0.5 0.5
0.5

0 0
0

−0.5 −0.5
−0.5

−1 −1
4 8 12 16 20 24 4 8 12 16 20 24 −1
4 8 12 16 20 24

Time of Day

Fig. 3. Sample vehicle energy profile with VID=V91, VCI:18, 0.1, 1, VPI:18 : 05, 7 : 15, 0.675

500
nearly the same for the coordinated cases. The PLC optimizes
the centralized objective function by controlling the charge
450 Base Case (Uncoordinated) schedule for each vehicle in each hour separately.
Load Demand Forecast (kWh)

Observation: Individually some vehicles might be in a dis-


400
Coordinated Charging Load charging mode while others are in the charging mode unless
350
Discharging
the fleet is controlled as a single entity.

300
Charging
3000 Base Case
250
Residential Non−Responsive Load Case 1
Charging/Discharging of Vehicle Fleet ($)

2500
Case 2
2000 Case3
200

2 4 6 8 10 12 14 16 18 20 22 24 1500
Aggregate Cost of

Time of Day 1000

500
Fig. 2. Resulting load demand forecast under V2G/G2V modes 0

−500

aggregated battery storage can be used to serve peak demand, −1000

and hence support the system. The sample profile for a vehicle −1500

(Fig. 3) shows that the amount of energy sent/received is −2000


0 5 10 15 20 25
different for different cases. Furthermore, the vehicle under- Time of Day

goes charging and discharging in continuous blocks instead of


intermittently to reduce penalty due to frequent switching. Fig. 5. Charge/discharge cost of vehicles over a 24 hour period

Figure 5 shows that base case leads to maximum charging


160
costs $239.32 since vehicles charge as soon as they arrive
Charging/Discharging Vehicle Load (kWh)

140
Base Case Case 1 Case 2 Case 3 Case 4 Case 4 without considering the TOU at that time. When considering
120
V2G mode, case 2 results in the cheapest costs $86.54
100
followed by case 1 $86.62. When considering the battery
80

60
degradation fitness function alone, we observe higher costs
40
$86.70. In the weighted mixed-objective approach, the costs
20
lie between those obtained in cases 2 & 3. It is to be noted
0 that we are considering only the aggregate cost of charg-
−20 ing/discharging for the PLC here using the TOU structure.
−40 These do not include any battery health information. Each
−60 scheduling scheme results in valley filling during the low-
0 5 10 15 20 25 rate period (night hours) and peak shaving during the day
Time of day
(high-rate period). In a centralized control paradigm it is not
Fig. 4. Charging and discharging energy of the aggregated EV fleet possible to select any one scheme that is best for all vehicles.
In fact, some vehicles earn greater profits in case 2 while others
The total energy requested or provided by the aggregated perform better with case 3 or 4. Nonetheless, the observations
EV fleet is shown in Fig. 4. No energy is supplied to the described above are relevant for an aggregated fleet. An
grid in the base case. The total energy bought or sold is investigation into this aspect can help in improving results by

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


selection of appropriate fitness function for individual vehicles.
Observation: A system optimum does not necessarily indicate
an individual optimum for each vehicle.
IV. C ONCLUSION
In this paper, we investigated a day-ahead charge scheduling
scheme that accounts for battery degradation and charging
costs in a V2G model. Results show that an effective technique
should consider battery and system health while maximizing
customer comfort. Also, even for a small system, the avail-
ability of V2G capability can be leveraged to earn profits
and in turn support the system. Using the day-ahead service,
customers can monitor their battery charging profile, make
their driving plans and prioritize their charging schedules. In
real-time, this would provide the system operator with a good
insight into the next-day’s load and bid into the market for
resources. Furthermore, new business and transactive energy
models can be developed based on customer habits using this
system.
ACKNOWLEDGMENT
The authors gratefully acknowledge the financial support
of the National Science Foundation under the project ECCS-
1068996.
R EFERENCES
[1] Y. He, B. Venkatesh and L. Guan, “Optimal Scheduling for Charging and
Discharging of Electric Vehicles,” IEEE Trans. Smart Grid, vol. 3, no. 3,
pp. 1095-1105, Sept. 2012.
[2] A. Hoke, A. Brissette, K. Smith, A. Pratt and D. Maksimovic, “Account-
ing for Lithium-Ion Battery Degradation in Electric Vehicle Charging
Optimization,” IEEE Journal of Emerging and Selected Topics in Power
Electronics, vol. 2, no. 3, pp. 691-700, Sept. 2014.
[3] S. J. Moura, J. L. Stein and H. K. Fathy, “Battery-Health Conscious Power
Management in Plug-In Hybrid Electric Vehicles via Electrochemical
Modeling and Stochastic Control,” IEEE Trans. Control Systems Tech-
nology, vol. 21, no. 3, pp. 679-694, May 2013.
[4] S. Bashash, S. J. Moura and H. K. Fathy, “On the aggregate grid load
imposed by battery health-conscious charging of plug-in hybrid electric
vehicles,” Journal of Power Sources, vol. 196, no. 20, pp. 8747-8754, Oct
2011.
[5] Maigha and M. L. Crow, “Economic Scheduling of Residential PHEV
Charging”, Energies, vol. 7, pp. 1876-1898, March 2014.
[6] Maigha, M. L. Crow, “Clustering-based methodology for optimal resi-
dential time of use design structure,”, North American Power Symposium,
Pullman, WA, Sept. 2014.
[7] (2015, Oct.) PGE website. [Online]. Available:
http://www.pge.com/en/mybusiness/save/energymanagement/index.page
[8] IBM Corp., ILOG CPLEX 12.6. [Online]. Available: https://www-
304.ibm.com/ibm/university/academic/pub/page/software all
[9] (2014, Aug.)2009 National Household Travel Survey. [Online]. Available:
http://nhts.ornl.gov

978-1-5090-5167-0/16/$31.00 ©2016 IEEE

You might also like