Professional Documents
Culture Documents
123
123
I. I NTRODUCTION
Electric Vehicle (EV) charging is expected to become a
key aspect for the integration of electromobility in the future
electricity distribution grids and nowadays ICT technologies
allow to implement the needed distributed monitoring an
control architecture [1], [2], [3]. On one hand a massive
penetration of EV technology will result in a significant
increase of the magnitude and volatility of load on the
distribution lines [4], [5], [6], on the other the possibility
of combining the control of the charging process with the
control of reverse energy flows from the EVs to the grid
(Vehicle to Grid (V2G)) [7] appears as a promising opportunity for balancing demand, Distributed Generation (DG) and
Renewable Energy Sources (RES) [8], [9] through proper
Demand Side Management (DSM) strategies [10], [11].
In this paper, both load management and V2G power
control are taken into account for the design of an EV
aggregator, named Charging Station Control Centre (CSCC),
aimed at optimizing the charging operations of EVs along a
proper portion of the low voltage distribution grid, in the
following named Load Area (LA) [12]. More specifically,
we outline an event driven Model Predictive Control (MPC)
strategy aimed at finding a proper trade-off between the need
of providing a cost-effective charging service in respect of
drivers preferences, and the one of guaranteeing the tracking
of a proper aggregated power profile, also supporting the
provisioning of ancillary services. Moreover, proper constraints are designed in order to keep bounded the cost
for the single user and guarantee that technical limitations
(both of the EV, Charging Station (CS) and the grid) are
always respected. In particular, the control action is limited in
compliance with the international standard IEC 61851 [13].
This work is partially financed by the European Union FP7-2011-ICT-GC
SMARTV2G project, grant agreement no. 284953. A. Di Giorgio, F. Liberati
and S. Canale are with the Department of Computer, Control and Management Engineering A. Ruberti, Sapienza University of Rome, ITALY
{digiorgio,liberati,canale}@dis.uniroma1.it
978-1-4799-0997-1/13/$31.00 2013 IEEE
1329
1330
The k-th component of vector P , denoted by P [k], represents the controlled aggregated power at the generic k-th
time interval for k I:
X
P [k] = P s [k] +
Pm Um [k].
(5)
V. P ROBLEM FORMALIZATION
Mk = {m M : I k Em }
A. Assumptions
The mathematical formulation of problem has been
achieved based on the following assumptions:
1) EV charging power is a semi-continuous variable (compliant with standard IEC 61851).
2) V2G is modeled as a semi-continuous variable too, with
the same domain of existence.
3) The cost of the energy absorbed by EVs and the price of
V2G energy are equal.
4) Input and output battery efficiency coefficients are assumed to be coincident and known a priori for each EV.
B. Target function
Let T R denote the discretisation step of the control
problem and I N the first time interval of problem definition
(the first after the triggering event). Then, from time interval
I on, we define the multi-objective target function
mMk
C. Prediction Model
J = Jcost + Jreg ,
(1)
EX
m 1
Pm T C[k]Um [k].
(2)
(7)
mM k=I
i=1
D. Control constraints
The first constraints are related to the nature of the control
variables. In compliance with the standard IEC 61851, for
the generic m-th EV we have
(3)
(4)
1331
(10)
mM
cm [I]+
k=I
m M
k [I, Em 2],
(12)
max
where Xm
is the maximum allowed level of charge and
min
Xm
represents the allowed depth of discharge.
Finally, we consider a termination constraint for each EV,
ref
aimed at guaranteeing the desired state of charge Xm
at
the end of the stop at the CS
ref
max
Xm
xm [Em ] Xm
m M.
(13)
Um ,mM
J,
(14)
m 1, .., N
zmk qmk
ymk pmk
zmk m qmk k 1, .., A. (15)
ymk m pmk
qmk {0, 1}
pmk {0, 1}
It is easy to see that when pmk = 0 the only value that ymk
can assume is 0. On the other side, when pmk = 1 then
m ymk 1. Analogously for zmk and qmk . Of course,
for each m {1, .., N }, we set ymk = 0, pmk = 0, zmk = 0
and qmk = 0 for all k [Em , E].
In order to formulate the objective function (1) as a linear
function by linearizing the regulation term (3), we indicate
by w 0 the maximum weighted deviation of the aggregated
power P [k] with respect to the reference P ref [k] when k =
1, .., A (w = maxk=1,..,A |k (P [k] P ref [k])|). Then we
introduce two subsets of non negative variables. The former
subset of variables uk 0 represents the positive part of
P [k] P ref [k], the latter subset of variables vk 0 the
negative one, for each EV k {1, .., A}. Then we set
uk vk = P [k] P ref [k].
(16)
(17)
where the argument of the function max is a linear expression. As customary in linear programming, we can
1332
TABLE I
N X
A
X
Pm T C[k]ymk
m=1 k=1
N X
A
X
Pm T C[k]zmk +
m=1 k=1
(19)
+ w.
m=1 Pm ymk
m=1 Pm zmk
P
[k]
P
[k]
k {1, .., A}
PA
PA
k=1 Pm T C[k]ymk
k=1 Pm T C[k]zmk
(1 + )cm cm [I]
m {1, .., N }
PN
PN
P
y
P
m mk
m zmk uk + vk =
m=1
m=1
ref
s
P [k] P [k]
k {1, .., A}
(u
+
v
)
k
=
1, .., A
k k
k
p y p
m {1, .., N }, k {1, ..A}
m mk
mk
mk
p
+
q
1
m
{1,
.., N }, k {1, ..A}
mk
mk
Pi
Pi
min
max
Xm
m {1, .., N } i {2, .., A 1}
PA
PA
ref
m {1, .., N }
Xm
qmk {0, 1}
m {1, .., N }, k {1, ..A}.
VII. S IMULATION RESULTS
A. Experimental setup
Simulations have been performed on a 2.4 GHz, 2GB
RAM computer. The simulation environment has been built
in MATLAB. The MILP problem defined in Section VI has
been solved by using the solver built in IBM ILOG CPLEX
v12.2, which relies on the Branch-and-Cut method [23].
For ease of discussion, EV technical parameters are the
same for all the simulated EVs: X max = 12.83 kW h,
X min = 3.2 kW h, P = 3.3 kW , = 0.1875 and = 2%.
We assume that all the users are flexible (their charging
profiles can always be rescheduled). The used sequence of
CRs, related UPs and initial state of charge are reported in
the first four columns of Table I. We simulate the sequential
arrival of 63 EVs, from 06:00 am to 23:00 pm. The first
four EVs are included to simulate the presence of EVs
which have reached the desired state of charge and still
have time to perform V2G (we call them stationing EVs).
The considered sequence has been chosen in such a way
Time interval
06:00-17:00
06:00-17:00
06:00-17:00
06:00-17:00
06:15-09:35
06:30-09:55
06:45-09:30
07:00-10:30
07:15-10:40
07:30-10:10
07:45-11:05
08:00-10:30
08:15-10:55
08:30-10:55
08:45-11:25
09:00-11:05
09:15-11:55
09:30-11:30
09:45-12:30
10:00-12:30
10:15-11:50
10:30-12:05
10:45-12:05
11:00-12:25
11:15-12:55
11:30-13:05
11:45-13:15
12:00-13:25
12:30-14:50
13:00-14:50
13:30-15:35
14:00-16:20
14:30-16:25
15:00-16:45
15:30-16:35
16:00-17:45
16:10-18:45
16:20-18:40
16:30-18:50
16:40-19:15
16:50-19:30
17:00-19:05
17:10-19:20
17:20-20:20
17:30-19:35
17:40-20:25
17:50-19:55
18:00-20:50
18:10-20:20
18:20-20:30
18:30-21:05
18:40-21:15
18:50-21:15
19:00-21:05
19:10-21:35
19:20-22:15
19:30-21:50
19:40-21:45
19:50-22:05
20:00-22:10
20:15-22:50
20:30-22:55
20:45-22:55
X 0 [kW h]
10.00
10.00
10.00
10.00
6.41
6.13
6.05
5.66
5.99
5.92
6.46
6.35
7.52
7.31
6.06
6.19
6.63
6.88
7.59
6.89
7.51
7.31
7.00
7.17
6.51
7.78
6.28
7.68
6.49
6.39
6.95
7.17
6.57
6.76
6.11
7.87
6.94
6.32
7.06
6.53
7.50
6.46
7.65
6.16
7.92
7.63
6.80
6.86
6.53
7.74
6.29
6.70
6.15
6.37
6.10
6.98
7.80
7.56
6.81
7.88
6.12
7.64
6.34
X ref [kW h]
10.00
10.00
10.00
10.00
12.13
12.10
12.50
12.50
12.50
12.50
12.50
12.50
10.49
9.34
9.55
10.65
10.90
9.76
9.37
10.29
8.55
8.33
9.92
8.45
9.01
9.92
8.30
8.51
9.36
8.00
8.20
8.60
10.01
9.64
9.56
8.76
9.02
10.59
9.33
10.31
9.90
10.83
10.08
9.89
9.01
10.74
9.52
10.82
9.29
10.16
10.71
10.03
9.48
9.48
10.81
9.98
9.74
9.78
9.19
10.91
9.47
9.03
10.30
cm [e]
0.12
0.09
0.02
0.02
0.44
0.48
0.53
0.58
0.56
0.57
0.53
0.55
0.27
0.18
0.31
0.40
0.38
0.26
0.13
0.28
0.09
0.08
0.25
0.10
0.19
0.16
0.14
0.06
0.20
0.11
0.09
0.11
0.26
0.22
0.26
0.07
0.16
0.36
0.19
0.34
0.22
0.40
0.22
0.34
0.10
0.28
0.25
0.37
0.26
0.23
0.41
0.31
0.31
0.29
0.45
0.28
0.19
0.21
0.23
0.28
0.30
0.12
0.37
cm [e]
-0.06
-0.08
-0.07
-0.08
0.44
0.48
0.53
0.58
0.56
0.57
0.53
0.55
0.27
0.18
0.31
0.40
0.38
0.25
0.13
0.28
0.09
0.08
0.25
0.10
0.18
0.16
0.14
0.06
0.20
0.11
0.09
0.11
0.26
0.22
0.26
0.07
0.16
0.36
0.19
0.34
0.22
0.40
0.22
0.34
0.10
0.29
0.25
0.37
0.26
0.23
0.41
0.31
0.31
0.29
0.45
0.28
0.19
0.21
0.23
0.28
0.30
0.12
0.37
cDSM
[e]
m
-0.07
-0.08
-0.07
-0.08
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
0.15
0.29
0.09
0.09
0.24
0.11
0.19
0.16
0.15
0.06
0.20
0.11
0.09
0.11
0.26
0.22
0.26
0.08
0.19
0.38
0.21
0.34
0.22
0.40
0.22
0.34
0.10
0.29
0.25
0.37
0.26
0.23
0.41
0.31
0.32
0.29
0.45
0.28
0.18
0.21
0.23
0.28
0.30
0.12
0.35
1333
the budgeted costs and the final costs for all the EVs are
reported, respectively, in columns 5 and 6 of Table I. The
total cost is 16.00 e, while the average error is 2.2 % of the
power reference value (the maximum error is 16.76 % of
the reference). The cost is not far from the theoretical bound
achieved by a greedy controller obatained working with
= 0 (15.44 e), but less than the uncontrolled case (16,39
e). Hence, thanks to proper load shifting and V2G power
control (see Figure 1.b) the aggregator manages to optimize
costs while providing good power reference tracking and
charging curve flattering and valley filling capabilities.
In the present case, the aggregator manages to fulfill user
requests at a cost always equal to the budgeted one. It is
also interesting to see how the control action is updated as
time goes on and new events trigger the aggregator. In Figure
2, the charging power profile is reported in correspondence
of three different times of the day. It is seen how tracking is
accurate before the line of current time (where the control
has been already actuated), while it is not after that, because:
1) there is still not enough power demand after that time to
track the reference (the EVs arrival is sequential); 2) after
the line of current time, power is distributed depending
on the electricity tariff and depending on the choice of .
In particular, we can see from Figure 2.a how part of the
charging power is shifted far ahead of the line of current
time in order to take advantage of a convenient tariff. Since
k is monotonically decreasing, the tracking is accurate near
the line of the current time only, as desired. By properly
choosing k it is possible to define the length of the moving
window ahead of the current time line in which the
requirement of reference tracking is strict. The same ability
of the aggregator to dynamically adjust control, event after
event, can be seen by analyzing the control sequences and the
state sequences of the single EVs. As an example, in Figure
3 the evolution of control of EV no 2 at three different time
points is reported. Control constraints are respected.
Finally, regarding the computational effort, the average
number of variables along the iterations is 2823 ,the average computational time is 0.98 seconds and the maximum
computational time is 30.26 seconds.
Fig. 1.
Charging power (a), discharging power (b) and net power (c).
Fig. 2.
Fig. 3.
Fig. 4.
1334
Fig. 5.
D. Sensitivity analysis
The main parameter governing the equilibrium between
the two objectives of minimizing costs and tracking reference
is . In the above sections, has been chosen with the
objective of guaranteeing a good power reference tracking.
Moreover, we have evaluated how two indicators of the
opposite objectives, respectively, the overall cost, and the
mean tracking error, vary with . To this end, we perform
a battery of tests, varying from 0 to 20000. Results are
displayed in Figure 5, in which we plot the pairs error-cost
for different values of the parameter. The cost for = 0 is
the cost we obtained in the case of the greedy controller.
VIII. C ONCLUSIONS AND FUTURE WORKS
In this paper, an event driven MPC approach for the
management of EVs charging in distribution grids has been
presented, enabling the provisioning of the charging service
at a competitive cost and in respect of drivers preferences.
Also technical limitations on the range of control actions and
both grid and market requirements have been considered.
In particular, the amplitude of control is limited according
to the international standard IEC 61851 and the controller
is designed to let the aggregated power profile track a
reference resulting from the trading in the electricity markets.
Simulation results show that the control system is flexible
enough to properly react to sequences of asynchronous CRs
and DSM signals, then adapting its behavior and exploiting
the V2G power to meet mobility dynamics and short term
requirements for a feasible operation of the distribution grid.
Future works regard ad hoc strategies for solving the MILP
problem, a more detailed model of the EV battery and the
formulation of an extended problem aimed at balancing EV
charging and generation from RES [25], [26] at LA level.
R EFERENCES
[1] Brenna, M., Falvo, M.C., Foiadelli, F., Martirano, L., Massaro, F.,
Poli, D. and Vaccaro, A. (2012), Challenges in energy systems for
the smart-cities of the future, in Energy Conference and Exhibition
(ENERGYCON), 2012 IEEE International, pp. 755 -762.
[2] Falvo, M.C. and Foiadelli, F. (2010), Preliminary analysis for the
design of an energy-efficient and environmental sustainable integrated
mobility system, in Power and Energy Society General Meeting, 2010
IEEE, pp. 1-7.
[3] Castrucci, M., Delli Priscoli, F., Pietrabissa, A., and Suraci, V., (2011),
A cognitive future internet architecture, Springer-Verlag, 168, 91-102.
[4] Clement-Nyns, K., Haesen, E., and Driesen, J. (2010), The impact of
charging plug-in hybrid electric vehicles on a residential distribution
grid, IEEE Transactions on Power Systems, 25, 371-380.
[5] Putrus, G., Suwanapingkarl, P., Johnston, D., Bentley, E., and Narayana,
M. (2009), Impact of electric vehicles on power distribution networks,
in Proceedings of the Vehicle Power and Propulsion Conference, pp.
827-831.
[6] Richardson, P., Flynn, D., and Keane, A. (2010) Impact assessment
of varying penetrations of electric vehicles on low voltage distribution
systems, in Proceedings of the Power and Energy Society General
Meeting, pp. 1-6.
[7] Kempton, W., and Letendre, S.E. (1997), Electric vehicles as a new
power source for electric utilities, Transportation Research Part D:
Transport and Environment, 2, 157-175.
[8] Kempton, K., and Tomic, J. (2005), Vehicle-to-grid power fundamentals: Calculating capacity and net revenue, Journal of Power Sources,
144, 268-279.
[9] Kempton, K., and Tomic, J. (2005), Vehicle-to-grid power implementation: From stabilizing the grid to supporting large-scale renewable
energy, Journal of Power Sources, 144, 280-294.
[10] Tomic, J. and Kempton, W. (2007), Using fleets of electric-drive
vehicles for grid support, Journal of Power Sources, 168, 459-468.
[11] Dallinger, D., Krampe, D., and Wietschel, M. (2011), Vehicle-togrid regulation reserves based on a dynamic simulation of mobility
behavior, IEEE Transactions on Smart Grid, 2, 302-313.
[12] ADDRESS Consortium (2011), Prototypes and algorithms for network management, providing the signals sent by the dso to aggregators
and the markets, enabling and exploiting active demand, in Deliverable
3.1. [Online]. Available: http://www.addressfp7.org
[13] IEC Technical Committee 69, IEC 61851-1 Electric vehicle conductive charging system - Part 1: General requirements.
[14] Peeters, E., Belhomme, R., Batlle, C., Bouffard, F., Karkkainen,
S., Six, D., and Hommelberg, M. (2009) Address: Scenarios and
architecture for active demand development in the smart grids of
the future, in Proceedings of the 20th International Conference and
Exhibition on Electricity Distribution (CIRED), pp. 1-4.
[15] Cusani, R., Delli Priscoli, F., Ferrari, G., Torregiani, M., A novel
MAC and scheduling strategy to guarantee QoS for-the new-generation
WIND-FLEX wireless LAN, Wireless Communications, IEEE , vol.9,
no.3, pp.46, June 2002.
[16] Deilami, S., Masoum, A.S., Moses P. S., and Masoum, M.A.S. (2011),
Real-Time Coordination of Plug-In Electric Vehicle Charging in Smart
Grids to Minimize Power Losses and Improve Voltage Profiles, IEEE
Transactions on Smart Grid, 2, 456-467.
[17] Richardson, P., Flynn, D., and Keane, A. (2012) Optimal charging
of electric vehicles in low-voltage distribution systems, IEEE Transactions on Power Systems, 27, 268-279.
[18] Bashash, S., and Fathy, H. (2011), Robust Demand-side Plug-in
Electric Vehicle Load Control for Renewable Energy Management,
in Proceedings of the American Control Conference, pp. 929-934.
[19] Studli, S., Crisostomi, E., Middleton, R., and Shorten, R. (2012): A
flexible distributed framework for realizing electric and plug-in vehicle
charging policies, International Journal of Control, 85, 1130-1145.
[20] Fan, Z., (2011) A Distributed Demand Response Algorithm and Its
Application to PHEV Charging in Smart Grids, IEEE Transactions on
Smart Grid, 3, 1280-1290.
[21] Ehsani, M., Gao, Y., and Emadi, A. (2009), Modern Electric, Hybrid
Electric, and Fuel Cell Vehicles: Fundamentals, Theory, and Design,
Boca Raton, FL: CRC Press.
[22] Di Giorgio, A., Pimpinella, L., Liberati, F., (2012), A model predictive
control approach to the load shifting problem in a household equipped
with an energy storage unit, in Proceedings of the 20th Mediterranean
Conference on Control and Automation (MED), pp. 1491-1498.
[23] Nemhauser, G.L., and Wolsey, L.A. (2000), Integer and Combinatorial
Optimization, Hoboken, NJ: John Wiley & Sons.
[24] Gestore Mercati Energetici, http://www.mercatoelettrico.org/It/.
[25] Di Giorgio, A., Pimpinella, L., Mercurio, A., (2010), A feedback
linearization based Wind turbine control system for ancillary services
and standard steady state operation, in Proceedings of the 18th Mediterranean Conference on Control and Automation (MED), pp. 1585-1590.
[26] Di Giorgio, A., Mercurio, A., Liberati, F., (2013), Regulation of
Angular Speed and Reactive Power for a Wind Turbine applying Robust
Feedback Linearization and H-infinity Control, in Proceedings of the
21st Mediterranean Conference on Control and Automation (MED),
Chania, 25-28 June 2013.
1335