Académique Documents
Professionnel Documents
Culture Documents
AbstractFrequency control is essential to maintain the stability and reliability of power systems. Traditionally, generation
side controllers have been used to stabilize the system frequency
upon contingencies; however, they incur high operational cost.
Enabling demand response in next generation smart grids
is thus a promising alternative to reduce the power system
dependency on expensive controllers. In particular, demand
response of distributed electric vehicles (EVs) via controlled
charging/discharging power from/to the grid can be an effective
method to help stabilize the system frequency. In this paper, we
propose a new distributed frequency control algorithm for EVs
with randomized responses and characterize its performance in
a large-scale dynamic power system in terms of the mean of the
system frequency over time, the mean frequency recovery time,
and the expected number of responded EVs upon a contingency.
Finally, we validate our analysis via simulations under a practical
power system setup.
I. I NTRODUCTION
Integrating renewable sources such as wind and solar into
power systems offers both economic and environmental advantages by lowering operational costs and also reducing
greenhouse gas emissions. However, the massive deployment
of renewable sources as well as storage-capable loads such as
electric vehicles (EVs) results in power system frequency and
voltage fluctuations due to their intermittent characteristics.
Frequency and voltage fluctuations need to be stabilized in
a very short time in order to prevent catastrophic damages
such as blackouts. In next generation smart grids, distributed
EVs that can charge/discharge power from/to the grid provides
an effective method to restore the system frequency and
voltage [1], [2], when they deviate from their nominal values
due to e.g. unexpected outages of power plants, intermittent
characteristics of renewable generations, and sudden spikes in
demand power consumption. Practically, both generation and
demand sides can be used to restore the system frequency.
At the generation side, fast response units that are equipped
with automatic generation controllers provide reserve services
to regulate power imbalances between demand and supply for
restoring the system frequency. However, employing reserve
services are expensive and can increase the annual operation
cost of power systems up to ten percent [3]. On the other hand,
EVs at the demand side can help restore the system frequency
by controlling their charging or discharging power according
to the system frequency [4]. Using EVs to recover the system
frequency decreases the required amount of reserve services
and hence reduces the power system operational costs [5].
It is worth noting that both centralized and distributed control paradigms for EV-based frequency control in smart grids
have been studied in the literature. Centralized algorithms for
the power system operator to optimally control the charging or
discharging power of individual EV according to the system
frequency have been proposed in [6], [7]. To implement centralized control algorithms, the system operator needs to rely
on an overlaid bidirectional communication network to collect
information from EVs and distribute commands to them,
which may not be available yet in todays power systems. The
complexity of solving the required optimization problems for
large-scale power systems is another challenge in centralized
control. On the other hand, distributed algorithms for EVs
to independently adjust their charging or discharging power
levels in response to the locally measured system frequency
have also been proposed in [8], [9]. As shown in Fig. 1, the
proposed power charging and discharging policy in [8], [9]
is a piece-wise linear function over the system frequency and
thus needs to be implemented continuously in time by varying
over a wide range of power values. As a result, designing
rechargeable batteries that are highly efficient for such wide
operational ranges is a practical challenge. The continuous
responses of EVs are also difficult to implement due to cost
considerations. Furthermore, although the prior studies have
shown promising aspects of utilizing EVs to replace traditional
reserve services for frequency control, there has been so far no
rigorous analysis available on characterizing the performance
of these algorithms in large-scale dynamic power systems.
In this paper, we propose a novel distributed frequency
139
IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing
demand power consumption is unchanged. Under this assumption, we have Pg (t) Pd (t) = A0 1{t>0} , where 1{} denotes
an indicator function. Therefore, the system frequency after a
contingency can be obtained by solving (1) as
f0
A0 (1 et ), t 0,
(2)
f (t) = f0 +
Kf P0
where = (Kf P0 )/(2H). In the above frequency equation,
we have assumed |A0 |/(Kf P0 ) 1 to assure that the power
system remains in the synchronous operation regime.
III. D ISTRIBUTED F REQUENCY C ONTROL VIA
R ANDOMIZED R ESPONSE
We consider a power system consisting of M grid-connected
EVs, denoted in the set V = {1, . . . , M }, with a fixed charging
power AC,v 0 and a fixed discharging power AD,v 0 for
each v V. It is assumed that all EVs monitor the system
frequency locally and respond independently. As shown in Fig.
2, let Sv (t) {1, 0, 1} represent three operational modes of
EV v over time t, where 1 indicates the charging mode
when the EV draws AC,v amount of power from the grid,
0 indicates the idle mode when the EV has zero power
exchanged with the grid, and 1 indicates the discharging
mode when the EV delivers AD,v amount of power to the grid.
Our proposed threshold-based frequency control algorithm
with a given pair of lower and upper frequency thresholds,
denoted by fmin f0 and fmax f0 , respectively, is then
described by
Sv (t), otherwise
where t+ denotes the time immediately after monitoring the
system frequency at time t. We assume that each EV can
charge or discharge its battery upon a contingency. This
assumption is justified in practice since contingencies do not
occur frequently in power systems and each contingency event
lasts at most for a couple of minutes before the system
frequency is restored by certain means. Define Av (t) as the
power exchanged between EV v and the grid, i.e.,
AC,v , if Sv (t) = 1
0, if Sv (t) = 0
Av (t) =
(4)
AD,v , if Sv (t) = 1
Accordingly, the power responses of each EV over time can
be characterized by Xv (t) , Av (t) Av (t+ ), which has five
140
IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing
Fig. 3.
process: Nv (t) =
n=1 1{tn,v <t} . We then define the nth
inter-response time of EV v as Tn,v = tn,v tn1,v , with
t0,v = 0 by default. Furthermore, we design the inter-response
time {Tn,v : n = 1, 2, } of EV v as independent exponentially distributed random variables with mean of 1/C,v
when Sv (t) = 1, 1/I,v when Sv (t) = 0, or 1/D,v when
Sv (t) = 1. This implies that the counting process Nv (t)
becomes a Poisson process with variable rates over time.
In practice, we can set I,v C,v D,v for each v
to minimize the number of responses of each EV for cost
reduction. In Fig. 3, we illustrate our proposed threshold-based
algorithm for one particular EV that is assumed to be initially
in the idle mode for a given system frequency function and a
set of frequency monitoring time instants. Consequently, the
frequency equation given in (2) is modified as
f0
f (t) = f0 +
A0 (1 et )+
K f P0
M Nv (t)
f0 X X
X(tn,v )(1 e(ttn,v ) ), t 0. (5)
Kf P0 v=1 n=1
A0
In this case, f (t) is given by (5) for t t0 . Next, we consider
the impact of Aa on f (t) by defining the following power
threshold:
Kf P0
Aa,min =
(fmin f0 ) A0 .
(8)
f0
From (5), it follows that if Aa < Aa,min , then f (t) does not
recover back to fmin even though all EVs respond after t t0 .
However, if Aa Aa,min , then f (t) recovers back to fmin at
a certain time Tr , which is termed frequency recovery time
and defined as the smallest t > t0 which solves f (t) = fmin ,
with f (t) given by (5).
IV. F REQUENCY M EAN A NALYSIS
With the introduced frequency control algorithm, the system frequency given in (5) becomes a random process.
To characterize the system frequency dynamics, we derive
its mean function over time for given EVs response rates
{C,v , I,v , D,v } in the following proposition.
Proposition 4.1: Given A0 < A0,min and Aa Aa,min , the
mean of the system frequency over t0 t Tr is given by
f0
E[f (t)] = f0 +
A0 (1 et )+
Kf P0
X
h (AC,v , AD,v , C,v , I,v , s)+
vVC
h (AD,v , 0, I,v , 0, s) ,
(9)
vVI
es es
, if 6=
u(, , s) =
(11)
s
(s + 1)e
, if = .
141
IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing
Fig.
4.
Frequency
h (AC,v , AD,v , C,v , I,v , s).
mean
characterization
function,
(12)
where MC,c and MI,c denote the number of EVs from Class c
that are initially in the charging and idle modes at time t = 0,
respectively.
To be consistent
PC
PC with our previous notation, let
M
=
M
and
C,c
C
c=1
c=1 MI,c = MI . We then state our
result in the following proposition.
Proposition 5.2: Given A0 < A0,min and Aa Aa,min ,
the expected number of EVs from Class c C that are in the
charging, idle, or discharging mode at time t, t0 t Tr , is
given by
E[NC,c (t)] = MC,c q1 (C,c , I,c , s),
(13)
E[NI,c (t)] = MC,c q2 (C,c , I,c , s) + MI,c q1 (I,c , 0, s),
(14)
E[ND,c (t)] = MC,c q3 (C,c , I,c , s) + MI,c q2 (I,c , 0, s),
(15)
where s = t t0 0, and q1 (C , I , s) is given by
q1 (C , I , s) = 1 q2 (C , I , s) q3 (C , I , s),
(16)
142
TABLE I
PARAMETERS OF THE WSCC POWER SYSTEM FOR SIMULATIONS .
System
WSCC
Parameters
P0 (MW)
f0 (Hz)
Kf
500
60
1.5
0.1
IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing
Fig. 6.
Fig. 5.
VII. C ONCLUSION
AD,v (1 e
(sT3,v )+
) , (19)
vVC
143
IEEE SmartGridComm 2013 Symposium - Demand Side Management, Demand Response, Dynamic Pricing
+
AD,v (1 E[e(sT3,v ) ]) . (20)
vVC
C,v I,v
(eC,v x eI,v x ), x 0.
I,v C,v
(21)
C,v es eC,v s
if C,v 6= ;
=
(22)
C,v
s
C,v s
C,v se
+e
if C,v = .
For our convenience, u(, , s) is defined in (11) to represent
the result in (22). Furthermore, we have
h
i
+
E e(sT3,v )
Z
+
C,v I,v
=
e(sx)
(eC,v x eI,v x )dx
I,v
C,v
0
C,v
I,v
u(, C,v , s) +
u(, I,v , s).
=
I,v C,v
C,v I,v
(23)
By substituting (22) and (23) into (20), the proof is thus
completed fo the case of MI = 0. The above proof can be
easily extended to incorporate the case of MI 6= 0, for which
the details are omitted due to the space limitation. Proposition
4.1 is thus proved.
B. Proof of Proposition 5.2
Consider first the case of MI,c = 0. Let q2 (C,c , I,c , tt0 )
and q3 (C,c , I,c , t t0 ) represent the probabilities that each
EV from Class c switches from the charging mode to idle and
discharging modes by time t, t0 t Tr , respectively. By
assuming C,c 6= I,c and defining s = t t0 0, we have
q2 (C,c , I,c , s) = Pr{T1,v < s T1,v + T2,v > s}
Z s Z +
=
C,c eC,c x I,c eI,c y dy dx
0
sx
(24)
and
q3 (C,c , I,c , s) = Pr{T1,v < s T1,v + T2,v < s}
Z s Z sx
C,c eC,c x I,c eI,c y dy dx
=
0
1
m=1
MC,c 1
= MC,c z z + 1 z
= MC,c z,
(27)
where z = q1 (C,c , I,c , s). Then the expected number of EVs
from Class c that are in the idle and discharging modes can be
shown similarly. The above proof can also be applied similarly
for the case of MI,c 6= 0. The proof of Proposition 5.2 is thus
completed.
R EFERENCES
[1] S. Shao, M. Pipattanasomporn, and S. Rahman, Grid integration of
electric vehicles and demand response with customer choice, IEEE
Trans. Smart Grid, vol. 3, no. 1, pp. 543-550, Mar. 2012.
[2] P. Zhang, K. Qian, C. Zhou, B. G. Stewart, and D. M. Hepburn, Demand
response for optimization of power systems demand due to EV charging
load, IEEE Conf. Power and Energy Engineering, pp. 1-4, Mar. 2012.
[3] S. Riedel and H. Weigt, German electricity reserve markets, WP-EM20, 2007.
[4] M. A. Ortega-Vazquez, F. Bouffard, and V. Silva, Electric vehicle
aggregator/system operator coordination for charging scheduling and
services procurement, IEEE Trans. Power Sys., to appear.
[5] A. Saffarian and M. Sanaye-Pasand, Enhancement of power system
stability using adaptive combinational load shedding methods, IEEE
Trans. Power Sys., vol. 26, no. 3, pp. 1010-1020, Aug. 2011.
[6] H. Sekyung, H. Soohee, and K. Sezaki, Development of an optimal
vehicle-to-grid aggregator for frequency regulation, IEEE Trans. Smart
Grid, vol. 1, no. 1, pp. 65-72, June 2010.
[7] M. D. Galus, S. Koch, and G. Andersson, Provision of load frequency
control by PHEVs, controllable loads, and a cogeneration Unit, IEEE
Trans. Industrial Electronics, vol. 58, no. 10, pp. 4568-4582, June 2010.
[8] Y. Ota, H. Taniguchi, T. Nakajima, K. M. Liyanage, J. Baba, and A.
Yokoyama, Autonomous distributed V2G (vehicle-to-grid) satisfying
scheduled charging, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 559564, Mar. 2012.
[9] J. A. P. Lopes, F. J. Soares, and P. M. R. Almeida, Integration of electric
vehicles in the electric power system, Proc. IEEE, vol. 99, no. 1, pp.
168-183, Jan. 2011.
[10] M. R. V. Moghadam, R. T. B. Ma, and R. Zhang, Distributed frequency
control via demand responses in smart grids, IEEE Int. Conf. on
Acoustics, Speech, and Signal Processing (ICASSP), to appear.
[11] P. Kundur, Power System Stability and Control, McGraw-Hill, 1994.
[12] PowerWorld Simulator, Computer Software, Ver. 16, available at
www.powerworld.com.
[13] A. Papoulis and S. U. Pillai, Probability, Random Variables, and
Stochastic Processes, McGraw-Hill, 2002.
144