You are on page 1of 12

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.


IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

Storage-Reserve Sizing With Qualified Reliability for


Connected High Renewable Penetration Micro-Grid
Jiaojiao Dong, Student Member, IEEE, Feng Gao, Member, IEEE, Xiaohong Guan, Fellow, IEEE,
Qiaozhu Zhai, Member, IEEE, and Jiang Wu, Member, IEEE
AbstractThe major challenge in high renewable penetration
microgrid is the power mismatch between stochastic renewables
and demand. Energy storage and reserve purchase are main
techniques in reducing that. Storage sizing problem is widely
investigated in literatures without reserve capacity co-optimizing.
And due to innate no loss-of-load assumption, a large proportion
of capacity is used to handle inessential energy deficiency with
small probability. In this paper, storage-reserve sizing problem
with qualified reliability is raised to integrate reserve sizing and
loss-of-load probability (LOLP) index into existing storage sizing
problem. Two-stage probabilistic model is established to minimize
total cost with optimizing storage capacity during first-stage and
reserve strategy during second-stage. Since the time-consuming
Monte Carlo simulation and stage iteration are usually required
in problem solving, Markovian steady-state sizing method is proposed to improve efficiency. Probability constraint is tested by
mathematical quantile. And two-stage model is transformed to
single-stage one attributed to analytical solution of second-stage.
Meanwhile obtained relationship among storage capacity, reserve
capacity, and LOLP index can help designers balance between
capacity and reliability. Numerical test shows: needed capacity
is significantly reduced with little sacrifice of reliability; storage
reserve combination is economical, since they are probabilistically
complementary; proposed solution method is fast and accurate.
Index TermsMarkov chain, micro-grid, reliability, steadystate method, storage-reserve sizing.

N OMENCLATURE
Input parameters
Reserve price.
c1
Storage price.
c2
Selling price.
c3
D (t)
Demand.
M
Maximum state of (t).
p (m)
Probability of (t) m
Physical limit of reserve.
Rlim
T
Number of time periods.
Manuscript received April 12, 2015; revised August 11, 2015 and September
21, 2015; accepted October 30, 2015. This work was supported in part by
the National Natural Science Foundation under Grants 61174146, 61221063,
61304212, U1301254, and 61473218, 863 High Tech Development Plan
(2012AA011003), and in part by the 111 International Collaboration Program,
of China. Paper no. TSTE-00266-2015.
J. Dong and F. Gao are with the State Key Laboratory for Manufacturing
System Engineering of Xian, Jiaotong University, Xian 710049, Shaanxi,
China (e-mail: jjdong@sei.xjtu.edu.cn; fgao@@sei.xjtu.edu.cn).
X. Guan, Q. Zhai, and J. Wu are with the Ministry of Education Key
Lab for Intelligent Networks and Networks Security of Xian, Jiaotong
University, Xian 200240, Shaanxi, China (e-mail: xhguan@sei.xjtu.edu.cn;
qzzhai@sei.xjtu.edu.cn; jwu@sei.xjtu.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSTE.2015.2498599

T0
U
w (t)
X1
X lim

(t)

i
o
x

Periodical length.
Rated power of energy storage.
Renewable energy.
Initial stored power.
Physical limit of storage.
Qualified LOLP index.
Effective power transfer.
Threshold for charging and discharging activation.
Charging ratio.
Discharging ratio.
Discrete interval.

Variables
E
E
Expected excess energy of infinite horizon.
Expected excess energy of steady-state.
ESE
R
Expected reserve of infinite horizon.
E
Expected reserve of steady-state.
ESR
U
Expected unsatisfied energy of infinite horizon.
E
Expected unsatisfied energy of steady-state.
ESU
K
Maximum state of reserve capacity.
Maximum state of optimal reserve capacity.
K
l (t)
Power mismatch after storage.
l
Realization of l (t).
n
Current state of stored power.
Next state of stored power.
n
N
Maximum state of storage capacity.
Steady-state probability of excess energy.
PSE
P trans Transition matrix of stored power.
Steady-state probability of unsatisfied energy.
PSU
q(l)
Optimization parameter under constant price.
q  (t, l) Optimization parameter under periodical price.
r (t)
Reserve amount.
R
Reserve strategy R = {r(t), t = 1, 2 . . . }.
Reserve capacity.
Rcap
t
Time period.
Charging power.
ui (t)
Discharging power.
uo (t)
v
Minimum state transition.
Maximum state transition.
v
X (t)
Stored power.
Storage capacity.
X cap
y (t)
Residual power mismatch.
Infinite horizon LOLP index.

S
Steady-state LOLP index.
Stand-alone steady-state LOLP index.


LOLP index decrement.

Steady-state distribution of stored power.

1949-3029 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

I. I NTRODUCTION

UE TO ENERGY crisis and environmental pollution,


great importance has been attached to renewable energy
worldwide [1]. But because of the uncontrollable and stochastic feature, it is a significant challenge to maintain the reliability
of high renewable penetration system [2]. One such challenge
is called power mismatch meaning generation is usually not
equal to demand. Main methods to handle that are: renewable
energy mitigation via complementary characteristic, controllable generation, demand response and energy storage. In particular, energy storage can provide more flexibility of renewable
generation and decrease its impact on utility grid [3][5].
The settings we focus on are high renewable penetration
micro-grid with energy storage, where renewables are the main
generation source to satisfy local demand. Storage is employed
to handle power mismatch. The above system can be found
in homes or industries that own renewable generators such as
solar panels or wind turbines. It can be categorized into standalone and grid-connected according to whether or not energy
deficiency can be covered from reserve market.
One of the most important decisions in system design level
is the optimal storage capacity, since storage cost will still be
expensive in the near future. It is uneconomical for micro-grid
with a relatively large storage, while system reliability may be
disturbed with a small one. Storage sizing problem [6][17] is
raised to minimize reserve cost and storage cost, if any.
Stand-alone and grid-connected storage sizing researches
are reviewed respectively. Stand-alone storage sizing problem
[6][11] is investigated to determine the function of storage
capacity and reliability. After research of probabilistic density
function (PDF) of wind power forecast error, stochastic storage sizing method is presented and compared with deterministic
ones [6]. Single objective is then extended to multi-objectives
including LOLP index, environmental impact and economical
benefit [7]. Load shifting strategy is introduced to alleviate
storage capacity need [8]. Close form of residual power mismatch is found under independent and identical distribution
(i.i.d.) Laplace PDF assumption [9]. Component failure is studied to provide a more realistic quantification of LOLP index
[10]. Grid-connected storage sizing problem [12][17] is investigated to find a trade-off between expected reserve cost and
storage cost. Deterministic linear model is established with
substation limit [12]. Cost-benefit optimal sizing method is
then formulated with micro-turbine [13]. The complementary
characteristic of wind and solar is focused on with defined fluctuation rate [14]. Close form of LOLP index is found under i.i.d.
Laplace PDF assumption [15]. Dynamic pricing is studied and
storage marginal value is proved to be decreasing [16].
The above research mainly discussed storage sizing problem
with a given physical reserve limit. But with the increasing penetration level of stochastic renewable energy, there is a need
to estimate the adequate reserve limit to match the stochastic renewable generation and demand. And optimized reserve
limit, i.e. reserve capacity, can be employed to represent the
disturbance caused by micro-grid on utility grid.
Reserve capacity induced by renewable energy is defined
as the minimum reserve requirement from auxiliary service to

match the stochastic renewable generation and load. Reserve


capacity problem [18][24] is mainly studied in large-scale
renewable integration. Arizona public service operation is simulated for one typical year and reserve capacity for wind is
estimated [18]. A new set of indices is introduced to measure
different types of reserve [19]. Complementary characteristic of
different renewables is focused on and a decrement in reserve
capacity is found [20]. Risk map is introduced as an intuitive
method to quantify LOLP index with different net load level
and reserve capacity level [21]. Reserve capacity problem is
also studied as stochastic unit commitment, where thermal units
are used to complement renewable generation [23], [24].
In our paper, reserve sizing is co-optimized with storage sizing to minimize the total cost of micro-grid, i.e. reserve cost
plus storage cost minus selling profit. Most relevant work [25]
is carried on about stand-alone storage sizing with storage cost
minimization, and associated reserve requirement under the
optimal storage capacity is calculated without co-optimizing.
One interesting phenomenon in the proposed storage-reserve
sizing problem has drawn our attention: in order to increase reliability index from 99% to 100%, storage capacity or reserve
capacity is almost doubled. It means that a large proportion of
capacity is used to handle small probability energy deficiency;
and a little sacrifice of system reliability would lead to a surprising capacity reduction. The unsatisfied load can be covered
by more flexible but expensive reserve, or cut through load
shedding in circumstance with no critical load. In order to help
designers balance between capacity and system reliability, relationship among storage capacity, reserve capacity and LOLP
index should be investigated.
Based on the above two motivations, both reserve sizing
problem and LOLP index are integrated into the existing storage sizing problem. Probabilistic model is established to seek
for a trade-off between reserve cost, storage cost and selling
profit, in which LOLP index constraint is used to substitute
previous demand equation constraint. After that relationship
among storage capacity, reserve capacity and LOLP index is
obtained. Then the effects of energy storage and reserve purchase on reducing power mismatch are examined respectively.
Combination of the above two is effective since they are probabilistically complementary, i.e. reserve is economical for small
probability energy deficiency while storage is beneficial for
large probability energy deficiency.
Existing storage sizing problem considering reliability in
above literatures is hard to solve due to probability constraint.
Monte-Carlo simulation [6][8], [10], [14] is often employed.
But it is time-consuming since the time span of sizing problem
is long. Analytical method [9], [15] is used under a certain PDF
assumption. Researches [26][28] are carried out using Markov
chain with i.i.d. assumption of power mismatch and stationary storage control strategy to solve it efficiently. Calculation
method of LOLP index under a given storage capacity is
obtained in stand-alone micro-grid.
Proposed storage-reserve sizing problem is more difficult
to solve since reserve strategy is co-optimized with storage
capacity under probability constraint. It is formulated as two
stage probabilistic infinite horizon dynamic model with both
deterministic and stochastic variables in this paper. First-stage

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
DONG et al.: STORAGE-RESERVE SIZING WITH QUALIFIED RELIABILITY

Fig. 1. Structure of high renewable penetration Micro-gird.

problem is storage capacity optimization, while second-stage


one is reserve strategy optimization. Markovian steady-state
sizing method is further investigated to avoid stage iteration.
Probability constraint is efficiently tested by mathematical
quantile and solution of second-stage is achieved. Two-stage
model is equivalently transformed into single stage one.
The rest of the paper is organized as follows. In section II,
mathematical model and solution procedure is proposed. In
section III, relationship of storage capacity and reserve capacity is obtained. The extensions and numerical results are in
section IV and section V respectively. Section VI gives the
conclusion.

II. P ROBLEM D ESCRIPTION


In this section, the working mechanism and reliability indices
for micro-grid are described. Mathematical model is then established. Solution procedure is also proposed.

U
in (2); and expected value of
value of unsatisfied energy E
E
excess energy E in (3). Related functions are in (4) and (5).
  T

1 
= lim E
I{y(t)<0}
(1)
T
T t=1


T
1

U
(2)
y(t)
E = lim E
T
T t=1


T
1
+
E
(3)
y(t)
E = lim E
T
T t=1

1 if x < 0
I{x<0} =
(4)
0 otherwise

x+ = max {x, 0} ; x = min {x, 0}

(5)

Equation (1), i.e. LOLP index, is calculated as the frequency


of infinite horizon loss-of-load event y(t) < 0. It is used as
qualified reliability for micro-grid in this paper and can be
transformed to no loss-of-load research when it is set to zero.
Equation (2) and (3) are calculated by the expected value of
positive mismatch and negative mismatch respectively.
C. Mathematical Model of StorageReserve Sizing Problem
Probabilistic model of storage-reserve sizing problem
is established with decision variables X cap and R =
{r(t), t = 1, 2, . . . }. Reserve capacity is defined as Rcap =
lim max {r(t), t = 1 . . . T }.

1) Objective Function:
Original program:
R
E
+ c2 X cap c3 E
min J(X cap , R) = c1 E
 

T
1
= c1 lim E
r(t)
+ c2 X cap
T
T t=1
 

T
1
+
c3 lim E
y(t)
T
T t=1

X cap ,R

A. Working Mechanism for Micro-Grid


System structure is in Fig. 1. If reserve is available, the power
mismatch between renewable energy and demand is covered by
both storage and reserve. Storage is used prior with a given stationary policy to decrease reserve cost. Then a certain amount
of reserve is provided within physical limit. Excess energy is
also allowed to exchange with benefit or penalty.
If micro-grid is accompanied by a relatively large storage
and offered with unlimited reserve, the reliability is high. But a
small storage capacity can reduce system cost, and decrement
on reserve capacity will make the integration of micro-grid
more friendly. So major focus of this paper is establishing the
analytical relationship between capacity and system reliability.
The treatment for this power mismatch may include: load
adjusting strategy in the demand of smart grid; or the more
expensive but flexible reserve purchase and so on. The first
treatment is studied in [29] to ensure that grid-connected microgrid can operate reliably when total generation is less than
demand. Reserve capacity can be viewed as the physical device
limit of a dc-ac converter [29] or a current limiters [30].

The first term in (6) is expected reserve cost. And the


remaining terms are storage cost and expected selling profit.
It is modeled as an infinite horizon problem, since the
granularity of discretizations (e.g. hourly) is relatively small
compared with the life-cycle of storage devices (e.g. a few
years) [16]. In another word, a few years is a relatively infinite number compared with hours in engineering, but it is not
convenient for mathematical analysis. So infinite horizon modeling is used to represent a relatively long time period in system
design. Error due to infinite horizon modeling is analyzed in
section V-E.
2) Constraints of LOLP Index:
  T

1 
I{y(t)<0}
(7)

lim E
T
T t=1
y(t) = w(t) D(t) ui (t) + uo (t) + r(t)

B. Reliability Indices for Micro-Grid


Reliability indices [6][15], [26][28] are defined in three
ways: probability of unsatisfied energy in (1); expected

(6)

(8)

LOLP index constraint is in (7) with residual power mismatch y(t) in (8). Original power mismatch w(t) D(t) is
supported by both storage ui (t)/uo (t) and reserve r(t).

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

3) Constraints of Energy Storage:


X(t + 1) = X(t) + ui (t)i uo (t)/o
X(1) = X1

ui (t) = min (w(t) D(t)) , (X cap X(t)) /i , U





uo (t) = min (w(t) D(t)) , X(t)o , U

(9)


(10)
(11)
(12)

Suppose unit scheduling period is 1 hour, state transition of


stored power is in (9), with initial power in (10). Control policy
[6][10], [15], [26][28] is in (11)(12). It is proved to be a
stationary optimal policy when the reserve price is constant [9].
Both storage capacity and rated power are considered.
Charging power in (11) means: charging power is no more than
the positive value of power mismatch; storage capacity cannot
be exceeded after this charging action; charging power is no
more than rated power. Explanation of discharging power in
(12) is almost the same as (11): discharging power is no more
than the absolute value of negative power mismatch; minimum
stored power should be ensured after the discharging action;
discharging power is no more than rated power.
In reality, storage cannot be completely discharged, and a
limit on minimum stored power exists. It is used as a reference
and is assumed to be zero without loss of generality [15].
4) Constraints of Physical Limit:
0 r(t) Rlim
0X

cap

lim

(13)
(14)

Physical limits cannot be exceeded as in (13) and (14).


D. Solution Procedure
Original program in (6)(14) is two-stage probabilistic
discrete-time average cost infinite horizon dynamic model with
deterministic variable X cap and stochastic variable R.
Note that Original program is complicated, but once the
variable X cap is fixed to a given value, Original program is
transformed to an easier program with variable R. It coincides
with the basic motivation of decomposition technique [31].
After decomposition, Original program is split into Master
program and Sub-program. Master program in (15) is to optiR
is calculated in Sub-program.
mize X cap , where E
Master program:
R
E
c 1 E
+ c2 X cap c3 E
min
cap

s.t. 0 X cap X lim

(15)

Sub-program in (16) is to find the optimal expected reserve


R
E
with a given X cap . It is difficult to solve because of the
probability constraint and state transition of stored power.
Sub-program:


T

1
R
= min lim E
r(t)
E
R T
T t=1

  T
1 

I{y(t)<0}
s.t.
lim E
T
T t=1
 

y(t) = w(t) D(t) ui (t) + uo (t) + r(t)


storage constraint (9)(12)

(16)

Fig. 2. Solution procedure of Markovian steady-state sizing method.

Generally, the above problem in (15)(16) can be solved


by decomposition algorithm, such as benders decomposition.
Master program is easily solved with enumeration method [6],
[10], [14], [26][28] due to the limited discrete value space of
X cap . Solution procedure of enumeration method is in Fig. 2.
But it is time-consuming since it is a two-stage iteration
approach and Sub-program is always solved by Monte-Carlo
simulation due to the probability constraint. In the following section, Markovian steady-state analysis is carried out to
obtain the analytical solution of Sub-program with the benefit
of speeding up the solution procedure.

III. M ARKOVIAN S TEADY-S TATE A NALYSIS :


R ELATIONSHIP B ETWEEN S TORAGE C APACITY AND
R ESERVE C APACITY U NDER Q UALIFIED LOLP I NDEX
In this section, steady-state analysis is proposed to obtain the
solution of Sub-program with a given storage capacity.
We find that there exists a similarity of input under some
circumstances below. When operation of micro-grid is multitimescale, the system consists of a scheduled part and a random part [9], [15]. The scheduled part is deterministic operation
from the last timescale optimization. And the random part is
power mismatch between the prediction value of this timescale
and the one of last timescale. It is the input of the system in this
paper, representing renewable generation prediction error and
demand prediction error, which can be assumed as i.i.d. under
good prediction. This input characteristic also holds when the
action of both users in the system and the renewable generation have little relationship with time [27]. So in this paper, we
focus on the micro-grid system with the input which can be
assumed as i.i.d. Analytical solutions can be obtained with this
assumption.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
DONG et al.: STORAGE-RESERVE SIZING WITH QUALIFIED RELIABILITY

A. Key Idea of Markovian Steady-State Analysis


Due to the probability constraint and state transition of stored
power, Sub-program is hard to solve. Markov chain method is
used to solve it efficiently, since the dynamic state transition of
stored power is a Markov Chain with steady-state distribution.
The solution procedure includes three steps.
First, equivalence between infinite horizon reliability indices
and steady-state ones is proved in Equivalent lemma. The
benefit for this lemma is: original infinite horizon probability
constraint is always calculated through time-consuming MonteCarlo simulation [6], [7], [10], [14]. After transforming it into
the steady state one, the solution time can be reduced, since the
steady state one can be easily tested by mathematical quantile,
as long as the PDF of y(t) is obtained.
After that, PDF of l(t) is analyzed with results in
Proposition 1, since y(t) = l(t) + r(t). It represents the power
mismatch after storage, i.e. stand-alone power mismatch.
Finally, since the quantile of y(t) is constrained, decision
variable r(t) is optimized to achieve the minimum reserve with
results in Proposition 2. It is the solution of Sub-program.
B. Steady-State Equivalent Lemma
Equivalent lemma. If sequence of stochastic input w(t)
D(t) is i.i.d. process and energy storage control strategy is stationary, then infinite horizon reliability indices in (1)(3) are
equal to steady-state ones in (17)(19) respectively.
S = lim Pr {y(t) < 0}
t



U
ES = lim E y(t)
t


+
E
ES = lim E y(t)
t

(17)
(18)
(19)

Based on above lemma, infinite horizon reliability indices,


which are always calculated through Monte-Carlo simulation
[6], [7], [10], [14], can be efficiently achieved by mathematical
quantile or expectation, after obtaining the PDF of y(t).
C. Effective Power Transfer, Variable Discretization and State
Transition of Stored Power
An imperfect round efficiency of storage leads to effective power transfer of stochastic input, meaning if the effective power interchange is w(t) D(t), real one is (w(t)
D(t))/i for charging and (w(t) D(t))o for discharging.
Denote (t) as the effective power transfer in (20).

(w(t) D(t)) i ,
if w(t) D(t) 0
(t) =
(20)
(w(t) D(t)) /o , else
After this transformation, the behavior of imperfect round
efficiency energy storage with input w(t) D(t) is the same
as that of a perfect round energy storage with input (t). In the
following paragraph, (t) is used in steady-state analysis.
In order to obtain the transition matrix of stored power,
(t), ui (t), uo (t), X(t), U are all discretized with x. Above
notations are used to represent their states in the following.
Continuous variables in sizing problem (6)(14), i.e. storage

Fig. 3. The relationship between transition probability and PDF of (t), where
value of v and v  (t) are U + n and U + n respectively in this figure.

capacity X cap and reserve capacity Rcap , are transformed to


discrete variables of N and K in (21).
X cap = N x
Rcap = Kx

(21)

For a perfect round storage with input (t), state transition is


rewritten in (22) after substituted (11)(12) into (9).

min{X(t) + min{(t), U }, X cap }, if (t) > 0


X(t + 1) =
max{X(t) + max{(t), U }, 0} , else
(22)
Suppose current state is n, and next state is n, where
n, n [0, N ]. Calculation method of P trans (n, n ) is in Fig. 3.
Horizontal ordinate is X(t + 1), and vertical one is p (m).
P trans (n, n ) is listed above horizontal ordinate.
The minimum state and maximum one for n are v =
max(0, U + n) and v  = min(N, U + n) respectively, since
both storage constraints and rated power constraints should be
satisfied. The arrow line in Fig. 3 represents that transition probability for v and v  are the summation of probability, since the
event set (t) < v n and (t) > v  n can only reach to v
and v  respectively.
Transition matrix P trans is in (23) with size (N + 1)
(N + 1). And the element in row n and column n is
P trans (n, n ) in (24).


p (0)
0

0
p (1) p+ (U )
0

.
.
.
.
.
..
..
..
..
..
..
..
P trans =

.
.

0
p (U ) p+ (1)
0

0
p+ (0)

trans

(n, n ) =

p (n n) ,

p (v n) =

(23)
vn


if n (v, v  )
p (m), if n = v

m=M
M


p+ (v  n) =
p (m), if n = v 

m=wn

0,
else
(24)

Steady-state distribution is solved by P trans = .

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

D. Calculation Method for the PDF of Stand-Alone Power


Mismatch l(t) and Stand-Alone Steady-State LOLP Index 
If rated power is unlimited, l(t) is in (25). First situation in
(25) means unsatisfied energy; the second one is excess energy;
and the third one means no power mismatch after storage.

if X(t) + (t) < 0


(X(t) + (t)) o ,
l(t) = (X(t) N + (t)) /i , if X(t) + (t) > N (25)

0
else
PDFs are calculated by summation of stochastic variables
(t) and X(t) in (26) and (27) respectively.
PSU (l) = lim Pr {l(t) = l}
t

M +l/o

= lim

Pr {X(t) = n} Pr {(t) = n + l/o }

n=0

M +l/o

(n) p (n + l/o ), if l < 0

Fig. 4. Relationship between storage capacity and reserve capacity under qualified LOLP index in grid-connected micro-grid, where Charge and Discharge is
the working condition of storage; Excess is excess energy and Loss is residual
energy deficiency.

(26)
ESE =

n=0

PSE (l) = lim Pr {l(t) = l}


M
li

M
li

l PSE (l)

Pr {X(t) = N n} Pr {(t) = n + li }

n=0

(N n) p (n + li ), if l > 0

(27)

n=0

Proposition 1: Considering stand-alone micro-grid system


with energy storage satisfying (9)(12). If rated power is
limited,
1) The probability of unsatisfied energy PSU (l) and excess
energy PSE (l) are (28) and (29) respectively.
2) Stand-alone steady-state LOLP index  is (30).
3) Steady-state expected energy ESE is (31).
PSU (l)
M +l/
 o

(n) p (n + l/o ), if M o l < M o + U

n=0

(U 1)/
 o
=
(n) p (n + l/o ) + p (U + l/o )

n=0

N


(n),
if M o + U l < 0

n=U/o

(28)
PSE (l)

E. Optimal Reserve Strategy With Qualified LOLP Index


Proposition 2: Considering grid-connected micro-grid system with storage satisfying (9)(12). If qualified LOLP index
is , the stochastic power mismatch after storage is l(t) with
PDF in (28)(29) and its realization denoted as l,
1) The optimal value space of reserve strategy to minimize
expected reserve is 0 and l, that is to say, the optimal
reserve strategy is full amount of neglecting or purchasing
respectively.
2) Minimum expected reserve is (32) with optimal
(K , ) reserve strategy (33). K and are in
(34)(35).
ESR

K
1


PSU (k) k + K

M
li

(N n) p (n + li ), if M/i U < l M/i

n=0

(U 1)
 i
=
(N n) p (n + li ) + p (U + li )

n=0

N


(N n),
if 0 < l M/i U

n=U i

(29)
 =

l=1

PSU (l)

(30)

(32)

k=1

0, if l > 0 or l < K ;

Pr {r(t) = l|l(t) = l} = 1, if K < l 0;

/PSU (K ), if l = K .
(33)

K
1

PSU (k)

<

k=1

M


(31)

l=1

= lim

M


K


PSU (k)

(34)

k=1


K
1

PSU (k)

(35)

k=1

K is the maximum reserve amount, i.e. optimal reserve


capacity. And is the LOLP decrement which should be covered by r(t) = K . It is appeared due to discretization and will
approach to zero when discrete interval is very small.
(K , ) reserve strategy is graphically shown in Fig. 4
Reserve, where Reserve is energy deficiency after storage and
should be covered to maintain system reliability. And the low
bound of Reserve is purchased with probability .

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
DONG et al.: STORAGE-RESERVE SIZING WITH QUALIFIED RELIABILITY

F. Tradeoff Between Storage Capacity and Reserve Capacity


Under Qualified LOLP Index
Another feature reflected in Fig. 4 is the trade-off between
storage capacity and reserve capacity with qualified LOLP
index. And qualified LOLP index is the probability area of Loss
divided by that of rectangle. When N becomes larger, i.e. the
area of rectangle increases, K can be decreased since LOLP
index becomes smaller with the previous K .
IV. E XTENSIONS
In this section, the model is extended in several directions to
cover broader applications.
A. Applying a Threshold-Based Control Policy
A threshold-based control policy has been studied to avoid
fast switching events [25], [32]. Key idea is that energy storage
stays in idle when power mismatch is in the threshold interval.
This threshold-based control policy is in (11+ ) and (12+ ). The
mathematical model is the same as the one defined in Section IIC after replacing (11) and (12) with (11+ ) and (12+ ).

0, if0 < w(t) D(t) or w(t) D(t) 


0
ui (t) =
+
cap
X(t)) /i , U , else
min (w(t) D(t)) , (X

uo (t) =

(11+ )
0, if- w(t) D(t) < 0 or w(t)
 D(t) 0

min (w(t) D(t)) , X(t)o , U , else


(12+ )

Reserve is required to meet the power mismatch within the


threshold interval , since storage is neither charged nor
discharged. Modification of solution procedure is as follows.
Effective power transfer is 0, when the actual one is in the
threshold interval, as in (20+ ). Expected reserve is the original one plus the expected reserve in the threshold interval, as
in (31+ ). Expected excess energy is the original one plus the
expected excess energy in the threshold interval, as in (32+ ).
The steady-state analysis is the same as the one in Section III
after replacing (20), (31), (32) with (20+ ), (31+ ), (32+ ).

if w(t) D(t)
(w(t) D(t)) i ,
(t) = (w(t) D(t)) /o , if w(t) D(t)
(20+ )

0,
else
ESE =

M


l PSE (l) +

ESR

k=1

PSU (k)

m p (m)

(31+ )

m=1

l=1

K
1

k + K +

1

m=


T
1
c1 (t) r(t)
+ c2 X cap
min lim E
X cap ,R T
T t=1
 

T
1
+
lim E
c3 (t) y(t)
T
T t=1
 

Since control policy of storage is still stationary, the Markov


property of micro-grid system has not changed. Equivalence
between infinite horizon reliability indices and steady-state
ones, i.e. Equivalent lemma, still exists. So the analysis is
still carried on in steady-state system. The power mismatch
after storage l(t) and stand-alone LOLP index, i.e. (28)(30)
in Proposition 1, is the same as that in previous section.
Suppose both reserve price and selling price are periodical, i.e. c1 (t + T0 ) = c1 (t), c3 (t + T0 ) = c3 (t), t = 1, 2, . . . .
Selling profit is changed to (37) due to time-variant selling
price c3 .
T

M
0


E
expected selling profit =
l PS (l)
c3 (t) /T0
(37)
The optimal reserve strategy and reserve cost, i.e. results in
Proposition 2, is modified. Mathematical model is established
to find the optimal reserve strategy under time-variant reserve
price. The expected reserve cost minimization model is in (38)
where decision variable r(t) is periodical, i.e. r(t + T0 ) = r(t).
min expected reserve cost
r(t)

t +T
0
0
c1 (t) E (r(t)) /T0
= lim
t0

t=t0

t
s.t.

lim

t0

0 +T0

Pr {l(t) + r(t) < 0} /T0

r(t, l) = 0, if l 0
Pr {r(t) = l|l(t) = l} = q  (t, l);
Pr {r(t) = 0|l(t) = l} = 1 q  (t, l), else

(39)

The optimization problem (38) is equivalently transformed


to (40) after substituted (39) into it. It is a simple linear programming which is easy to solve with the output of purchase
probability q  (t, l).

T
M
0


U

min
c1 (t)
PS (l) q (t, l) l /T0
s.t.

t=1
T0 
M

t=1 l=1


l=1

PSU (l) q(t, l)/T0 = 

0 q (t, l) 1

When both reserve price and selling price are time-variant,


objective of sizing problem is modified from (6) into (36).

(38)

t=t0

Optimal reserve strategy is full amount of neglecting or purchasing respectively with the same proof of proposition 21) in
Appendix. Optimal format is in (39), where t = 1, 2, . . . T0 .

m p (m)

B. Applying Time-Variant Reserve Price and Selling Price

t=1

l=1

q(t,l)

(32+ )

(36)

(40)

The main solution procedure and key idea of steady-state


analysis are the same except the above modification in the
calculation form of expected selling profit and reserve cost.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

TABLE I
S TORAGE C APACITY AND R ESERVE C APACITY U NDER T WO T YPES OF
C ONSTRAINT W ITH L OOSE P HYSICAL R ESERVE L IMIT

V. N UMERICAL R ESULTS
The proposed model is tested on an Intel Core i5 2.4-GHz
personal computer using Matlab 2009. A case study of windstorage system is established with a small wind turbine as main
generation source for 300 households.
Specifications [26], [27] of wind power and demand
are as follows. PDF of wind speed V is Weibull distrik1 (V /c)k
e
with parameter
bution f (V ; c, k) = (k/c) (V /c)
k = 2.85, c = 6.52. Power curve of wind turbine is modeled by a quadratic function of wind speed V [33] with
parameter Vin = 3, Vrated = 10.5, Vout = 20. PDF of demand
2
2
is Rayleigh distribution f (x; ) = xex /2 / 2 , x 0 with
parameter = 161.76. Maximum value and mean value of
unsatisfied energy are about 700 kW and 78 kW respectively.
Storage inventory cost is chosen referred to WB-LYP1000
AHA lithium ion 1000Ah battery modules with 3.2 V discharge voltage listed $1000 [34] and reserve price is $40/MWh.
The selling price c3 is set to zero except in section G. 175200
samples of power mismatch, i.e. data length of 20 years, are
generated through Monte-Carlo simulation.

Fig. 5. Relationship among storage capacity, reserve capacity and reliability


index.

Fig. 6. Relationship between storage capacity and reliability index with simulation range of X cap from 0 to 5000 kWh when Rcap = 100 kW.

A. Effect of LOLP Index Constraint


Influence of LOLP index constraint is discussed compared
with demand equation constraint. Optimization results are different depending on activeness of physical reserve limit.
When the limit is loose, optimization results are in Table I.
It shows that optimal storage capacity is not increased much,
since it is not economical to decrease LOLP index by storage due to high cost of storage inventory. But the optimal
reserve capacity is increased by 51.12%, which means a larger
disturbance is emerged to cover the last 1% energy deficiency.
When the limit is tight, the phenomenon is mentioned in
introduction. And it is graphically represented in Fig. 6 zoomed
in. Simulation range of X cap is set to 10005000 kWh for a
clear observation. It shows needed capacity for 99% reliability
is around 2600 kWh, while the one for 100% is larger than 5000
kWh. Implied meaning is that a large proportion of storage is
used only to handle small probability deficiency.
From the above analysis, a significant decline of optimal
reserve or storage capacity is observed under LOLP index.

Fig. 7. Relationship between storage capacity and reserve capacity when


= 1%.

An intersection between the surface in Fig. 5 and a surface


of Rcap = 100 kW is in Fig. 6 with simulation range of X cap
from 0 to 5000 kWh. It shows that the contribution of storage
capacity to reliability is significant at first and tends to saturate.
Range of X cap from 1000 to 5000 kWh is in Zoomed in.
Another intersection between the surface in Fig. 5 and a surface of = 1% is shown in Fig. 7. As we can see, needed
reserve capacity and storage capacity are no more than 400 kW
and 3500 kWh respectively. Designers can balance between
capacity and reliability and avoid capacity redundancy.
C. Solution of Storage-Reserve Sizing Problem

B. Relationship Among Storage Capacity, Reserve Capacity


and Reliability Index
Relationship in Fig. 5 can be obtained by enumerating X cap
from 0 to 1500 kWh and Rcap from 0 to 700 kW respectively,
where reliability index is 1 .

Solution of storage-reserve sizing problem with = 1% is


shown in Table II. Three settings are examined:
P1: Grid-connected with energy storage;
P2: Stand-alone with energy storage;
P3: Grid-connected without energy storage.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
DONG et al.: STORAGE-RESERVE SIZING WITH QUALIFIED RELIABILITY

TABLE II
S OLUTION OF S TORAGE -R ESERVE S IZING P ROBLEM

Fig. 10. The PDF of finite horizon stand-alone LOLP index with a given =
1% and X cap = 364.8 kWh after 1000 trails.

Fig. 8. Relationship between cost and storage capacity.


Fig. 11. The PDF of finite horizon reserve capacity with a given = 1% and
X cap = 364.8 kWh after 1000 trails.

In our paper, combination mechanism for reducing power


mismatch is: storage is used first; and the rest small probability
energy deficiency, which is not economical for storage, is dealt
by reserve with the permission of a certain loss-of-load.

E. Modeling Error of Infinite Horizon Compared With a


Relatively Large Finite One
Fig. 9. PDF of original unsatisfied energy and unsatisfied energy after storage
when X cap = 364.8 kWh.

A decrement in optimal total cost is seen in setting P1, which


means that the combination of storage and reserve is effective.
Trade-off between storage cost and expected reserve cost is
shown in Fig. 8. As we can see, total cost is a convex function
of X cap . It means that benefit from storage is large at first, but
after the optimal storage capacity, total cost increases meaning
no benefit can come from storage.
D. Characteristic of Two Techniques, i.e. Energy Storage and
Reserve Purchase, in Reducing Power Mismatch
Characteristics of the two techniques are discussed to show
the reason for combination effectiveness. And the simulation
results imply that they are probabilistically complementary.
Reserve purchase is economical for small probability energy
deficiency since expected reserve cost is calculated through
mathematical expectation. And energy storage is beneficial
to handle large probability deficiency. It can be vividly seen
from the decrement of PDF of energy deficiency in Fig. 9.
A large decrement can be observed at energy deficiency of
about 60 kW which is the value of original energy deficiency
with large probability; while the energy deficiency with small
probability is not dealt properly such as around 400 kW.

In reality, the horizon of a real sizing problem is always finite,


and modeling error is examined. Key issue in sizing problem is
the accuracy of stand-alone LOLP index and reserve capacity
with a given storage capacity and qualified reliability.
Finite horizon simulation is carried out using 8760 simulated
data of power mismatch as an input, i.e. one year data length.
The outputs are stand-alone LOLP index and reserve capacity.
Since the power mismatch is stochastic, it is simulated for 1000
trails. PDF of stand-alone LOLP index and reserve capacity are
in Figs. 10 and 11 respectively. As we can see, the output of
finite horizon modeling has some contingency in one trail since
the power mismatch is stochastic.
Mean values of LOLP index and reserve capacity in finite
horizon modeling are 0.2320 and 329.8 kW respectively. For
comparison, infinite horizon LOLP index and reserve capacity
are also calculated. The outputs are 0.2323 and 330.4 kW. As
we can see, the outputs by two types of modeling are very close.

F. Calculation Time and Accuracy of Markovian Steady-State


Sizing Method Versus Historical Method
Historical method [35] using 8760 simulated data is compared with steady-state sizing method using 175200 one.
Performances are in Table III.
Penalty factor for physical reserve limit is 5000. Simulated
reserve capacity and total cost are obtained by Monte-Carlo

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

TABLE III
P ERFORMANCE OF H ISTORICAL S IZING M ETHOD AND P ROPOSED
M ARKOVIAN S TEADY-S TATE S IZING M ETHOD W HEN = 1%,
Rlim = 100 K W

TABLE IV
I MPACT OF S ELLING P RICE

convergent property of sequence, infinite horizon indices are


equivalent to steady-state ones.
Proof of LOLP index equivalence is in (41) as an example.



T
T
1
1
I{y(t)<0} = lim E
Y (t)
lim E
T
T
T t=1
T t=1

T

T

1
1
E
Y (t) = lim
E {Y (t)}
= lim
T T
T T
t=1
t=1

= lim E {Y (t)} = lim Pr {y(t) < 0}


t

(41)


Proof of Proposition 1: 1) l(t) = w(t) D(t) ui (t) +


uo (t).
When (t) < 0, unsatisfied energy l(t) < 0 is in (42), since
uo (t) = l(t) (w(t) D(t)) < U when l(t) < M o + U .
simulation. It shows that total cost using proposed method
is decreased, since the probability of reserve limit constraint
violation is smaller.
G. Impact of Selling Price c3
Impact of selling price c3 is in Table IV. Three settings are
examined individually. It shows that when c3 = $10/MWh,
i.e. penalty for selling, there are few impact on grid since large
storage is preferred leading to less reserve and less selling.
VI. C ONCLUSION
Storage-reserve sizing problem with qualified reliability is
proposed to quantify the needed capacity for micro-grid. It is
formulated as two-stage probabilistic discrete-time average cost
infinite horizon dynamic model. Markovian steady-state sizing
method is proposed to solve efficiently. Key idea is that infinite horizon reliability indices are equivalent to steady-state
ones. State transition of stored power is studied and relationship among storage capacity, reserve capacity and LOLP index
is obtained. Two-stage model is transformed to single stage one
to avoid iteration. Probabilistically complementary characteristic of energy storage and reserve purchase in reducing power
mismatch is found, which reveals the economical reason of this
combination mechanism. More results about storage-reserve
sizing problem with dynamic pricing can be further analyzed
with periodical PDF assumption.
A PPENDIX
Proof of Equivalent lemma: Since storage control strategy
is stationary, X(t) is an ergodic Markov Chain with stationary
distribution [16], [26][28].
Write Y (t) = I{y(t)<0} . Because of the Markov property,
there exists a stationary distribution of y(t).

So sequence of {E (Y (t)) , t = 1 . . . T }, {E(y(t) ), t =


+
1 . . . T } and {E(y(t) ), t = 1 . . . T } is convergent. Due to
the exchange property of mathematical expectation and the

l(t) = w(t) D(t) + uo (t)

(X(t) + (t)) o , if M (t) + X(t) < M + U/o


= (min (X(t), U/o ) + (t)) o ,

if M + U/o (t) + X(t) < 0


(42)
Probability of l(t) is calculated by probabilistic summation
of stochastic variables (t) and X(t) in (28).
When U is unlimited, the first situation in (28) always holds
for all l < 0 and (28) turns into (26).
Proof of PSE (l) is almost the same as above with no details.
2)-3)  is the summation of probability of independent
events l [M, 1]. ESE is calculated by the PDF of excess
energy after storage.

Proof of Proposition 2: Expected reserve minimization
model is in (43) where decision variable r(t) is stationary. It
is the same as Sub-program (16) since l(t) = (t) ui (t) +
uo (t).
ESR = min lim E (r(t))
r(t) t

s.t.

lim Pr {l(t) + r(t) < 0}

(43)

1) If l 0, r(t) = 0 meaning if there is no energy deficiency


after storage, no benefit can come from reserve;
If l < 0 and a feasible solution r(t) (0, l) exists, an optimized r (t) = 0 can be used to replace original one with no
constraint violation and a smaller objective function value.
Optimal format of r(t) is in (44), where q(l) is undetermined
optimization parameter.
r(t) = 0, if l 0
Pr {r(t) = l|l(t) = l} = q(l);
Pr {r(t) = 0|l(t) = l} = 1 q(l), else

(44)

2) The optimization problem (43) is equivalently transformed


to (45) after substituted (44) into it.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
DONG et al.: STORAGE-RESERVE SIZING WITH QUALIFIED RELIABILITY

min
q(l)

s.t.

M


11

PSU (l) q(l) l

l=1
M


PSU (l) q(l) = 

l=1

0 q(l) 1

(45)

Optimal decision of (45) is q(1), . . . , q(K 1) = 1,


q(K ) = /PSU (K ) and q(K + 1), . . . , q(M ) = 0. K
and are in (34) and (35).

ACKNOWLEDGMENT
The authors would like to thank the anonymous referees for
their valuable comments and suggestions.
R EFERENCES
[1] F. Birol, World energy outlook, Int. Energy Agency, Paris, France,
pp. 2329, 2014.
[2] D. Infield and W. Lei, The challenges of high wind energy penetration in the UK power system, in Proc. Int. Conf. Power Syst. Technol.
(POWERCON14), 2014, pp. 29993003.
[3] J. Eyer and G. Corey, Energy storage for the electricity grid: Benefits
and market potential assessment guide, Sandia Natl. Labs, Albuquerque,
NM and Livermore, CA, Tech. Rep. SAND2010-0815, Feb. 2010.
[4] F. S. Barnes and J. G. Levine, Impacts of intermittent generation, in
Large Energy Storage Systems Handbook, Boca Raton, FL, USA: CRC
Press, 2011, pp. 1750.
[5] P. Medina, A. W. Bizuayehu, J. P. S. Catalo, E. M. G. Rodrigues, and
J. Contreras, Electrical energy storage systems: Technologies stateof-the-art, techno-economic benefits and applications analysis, in Proc.
Hawaii Int. Conf. Syst. Sci. (HICSS14), 2014, pp. 22952304.
[6] C. V. T. Cabral, D. O. Filho, A. S. A. C. Diniz, J. H. Martins,
O. M. Toledo, and L. D. V. B. Machado Neto, A stochastic method for
stand-alone photovoltaic system sizing, Solar Energy, vol. 84, pp. 1628
1636, Sep. 2010.
[7] D. Abbes, A. Martinez, and G. Champenois, Eco-design optimisation of
an autonomous hybrid wind-photovoltaic system with battery storage,
IET Renew. Power Gener., vol. 6, pp. 358371, Sep. 2012.
[8] A. Arabali, M. Ghofrani, M. Etezadi-Amoli, and M. S. Fadali, Stochastic
performance assessment and sizing for a hybrid power system of
solar/wind/energy storage, IEEE Trans. Sustain. Energy, vol. 5, no. 2,
pp. 363371, Apr. 2014.
[9] H. Su and A. E. Gamal, Modeling and analysis of the role of energy
storage for renewable integration: Power balancing, IEEE Trans. Power
Syst., vol. 28, no. 4, pp. 41094117, Nov. 2013.
[10] M. Theristis and I. A. Papazoglou, Markovian reliability analysis
of standalone photovoltaic systems incorporating repairs, IEEE J.
Photovoltaics, vol. 4, no. 1, pp. 414422, Jan. 2014.
[11] W. Zhou, C. Lou, Z. Li, L. Lu, and H. Yang, Current status of research
on optimum sizing of stand-alone hybrid solarwind power generation
systems, Appl. Energy, vol. 87, pp. 380389, Feb. 2010.
[12] R. Chedid and S. Rahman, Unit sizing and control of hybrid wind-solar
power systems, IEEE Trans. Energy Convers., vol. 12, no. 1, pp. 7985,
Mar. 1997.
[13] S. X. Chen, H. B. Gooi, and M. Q. Wang, Sizing of energy storage for
microgrids, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 142151, Mar.
2012.
[14] L. Xu, X. Ruan, C. Mao, B. Zhang, and Y. Luo, An improved optimal
sizing method for wind-solar-battery hybrid power system, IEEE Trans.
Sustain. Energy, vol. 4, no. 3, pp. 774785, Jul. 2013.
[15] H. I. Su and A. E. Gamal. (2012, Apr.). Limits on the Benefits of Energy
Storage for Renewable Integration [Online]. Available: http://arxiv.org/
abs/1109.3841

[16] P. Harsha and M. Dahleh, Optimal management and sizing of energy


storage under dynamic pricing for the efficient integration of renewable
Energy, IEEE Trans. Power Syst., vol. 30, no. 3, pp. 11641181, May
2015.
[17] O. Erdinc and M. Uzunoglu, Optimum design of hybrid renewable
energy systems: Overview of different approaches, Renew. Sustain.
Energy Rev., vol. 16, pp. 14121425, Apr. 2012.
[18] T.L. Acker, R Zavadil, C. Potter, and R. Flood, Wind integration cost
impact study for the Arizona public service company: Modeling approach
and results, Wind Eng., vol. 32, pp. 339353, Jun. 2008.
[19] A. M. L. Da Silva, W. S. Sales, L. A. Da Fonseca Manso, and R. Billinton,
Long-term probabilistic evaluation of operating reserve requirements
with renewable sources, IEEE Trans. Power Syst., vol. 25, no. 1,
pp. 106116, Feb. 2010.
[20] D. A. Halamay, T. K. A. Brekken, A. Simmons, and S. McArthur,
Reserve requirement impacts of large-scale integration of wind, solar,
and ocean wave power generation, IEEE Trans. Sustain. Energy, vol. 2,
no. 3, pp. 321328, Jul. 2011.
[21] M. S. Modarresi and L. Xie, An operating reserve risk map for quantifiable reliability performances in renewable power systems, in Proc. IEEE
PES General Meeting Conf. Expo., 2014, pp. 15.
[22] H. Holttinen et al., Methodologies to determine operating reserves due
to increased wind power, IEEE Trans. Sustain. Energy, vol. 3, no. 4,
pp. 713723, Oct. 2012.
[23] J. Xiao, B. S. Hodge, J. F. Pekny, and G. V. Reklaitis, Operating reserve
policies with high wind power penetration, Comput. Chem. Eng., vol. 35,
pp. 18761885, Sep. 2011.
[24] N. G. Paterakis, O. Erdin, A. G. Bakirtzis, and J. P. S. Catalo,
Qualification and quantification of reserves in power systems under high
wind generation penetration considering demand response, IEEE Trans.
Sustain. Energy, vol. 6, no. 1, pp. 88103, Jan. 2015.
[25] T. K. A. Brekken, A. Yokochi, A. von Jouanne, Z. Z. Yen, H. M. Hapke,
and D. A. Halamay, Optimal energy storage sizing and control for wind
power applications, IEEE Trans. Sustain. Energy, vol. 2, no. 1, pp. 69
77, Jan. 2011.
[26] J. Song, V. Krishnamurthy, A. Kwasinski, and R. Sharma, Development
of a Markov-chain-based energy storage model for power supply availability assessment of photovoltaic generation plants, IEEE Trans.
Sustain. Energy, vol. 4, no. 2, pp. 491500, Apr. 2013.
[27] J. Li and W. Wei, Probabilistic evaluation of available power of a renewable generation system consisting of wind turbines and storage batteries:
A Markov chain method, J. Renew. Sustain. Energy, vol. 6, p. 013139,
Jan. 2014.
[28] S. Lu and H. M. Kim, Hybrid power generation system design optimization based on a Markovian reliability analysis pattern, in Proc. ASME
Int. Des. Eng. Tech. Conf./Comput. Inf. Eng. (IDETC/CIE11), 2011,
pp. 12011211.
[29] L. Xu and D. Chen, Control and operation of a DC microgrid with variable generation and energy storage, IEEE Trans. Power Del., vol. 26,
no. 4, pp. 25132522, Oct. 2011.
[30] U. A. Khan, J. K. Seong, S. H. Lee, S. H. Lim, and B. W. Lee, Feasibility
analysis of the positioning of superconducting fault current limiters for
the smart grid application using Simulink and SimPowerSystem, IEEE
Trans. Appl. Supercond., vol. 21, no. 3, pp. 21652169, Jun. 2011.
[31] A. J. Conejo, E. Castillo, R. Mnguez, and R. Garca-Bertrand,
Motivating examples, in Decomposition Techniques in Mathematical
Programming: Engineering and Science Applications, New York, NY,
USA: Springer, 2006, pp. 38.
[32] F. Marra, G. Yang, C. Traeholt, J. Ostergaard, and E. Larsen, A decentralized storage strategy for residential feeders with photovoltaics, IEEE
Trans. Smart Grid, vol. 5, no. 2, pp. 974981, Mar. 2014.
[33] B. Bhandari, S. R. Poudel, K. Lee, and S. Ahn, Mathematical modeling
of hybrid renewable energy system: A review on small hydro-solar-wind
power generation, Int. J. Precis. Eng. Manuf.-Green Technol., vol. 1,
pp. 157173, Apr. 2014.
[34] H. Akhavan-Hejazi and H. Mohsenian-Rad, Optimal operation of independent storage systems in energy and reserve markets with high wind
penetration, IEEE Trans. Smart Grid, vol. 5, no. 2, pp. 10881097, Mar.
2014.
[35] S. Sinha and S. S. Chandel, Review of software tools for hybrid renewable energy systems, Renew. Sustain. Energy Rev., vol. 32, pp. 192205,
Apr. 2014.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
12

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY

Jiaojiao Dong (S12) received the B.S. degree in


information engineering and the M.S. degree in
automation control from Xian Jiaotong University,
Xian, China, in 2008 and 2011, respectively. She is
currently pursuing the Ph.D. degree at the System
Engineering Institute, Xian Jiaotong University,
Xian, China. Her research interests include power
system optimization and scheduling, and renewable
energy integration.

Feng Gao (M07) received the B.S. degree in automatic control, and the M.S. and Ph.D. degrees in systems engineering from the Xian Jiaotong University,
Xian, China, in 1988, 1991, and 1996, respectively.
He is currently a Professor with Systems Engineering
Institute, Xian Jiaotong University, Xian, China. He
visited Harvard University, Cambridge, MA, USA,
as a Postdoc from February 2000 to June 2001. His
research interests include power system optimization,
scheduling, and prediction.

Xiaohong Guan (M93SM95F07) received the


B.S. and M.S. degrees in automatic control from
Tsinghua University, Beijing, China, and the Ph.D.
degree in electrical engineering from the University
of Connecticut, Storrs, CT, USA, in 1982, 1985,
and 1993, respectively. He was a Senior Consulting
Engineer with PG&E from 1993 to 1995. He visited the Division of Engineering and Applied Science,
Harvard University, Cambridge, MA, USA, from
January 1999 to February 2000. Since 1995, he has
been with the Systems Engineering Institute, Xian
Jiaotong University, Xian, China. He was appointed as the Cheung Kong
Professor of Systems Engineering in 1999, and the Dean of the School of
Electronic and Information Engineering in 2008. Since 2001, he has been
the Director of the Center for Intelligent and Networked Systems, Tsinghua
University, and served as the Head of the Department of Automation (2003
2008). His research interests include scheduling of power and manufacturing
systems, bidding strategies for deregulated electric power markets, and security
of complex network systems.

Qiaozhu Zhai (M07) received the B.S. and M.S.


degrees in applied mathematics and the Ph.D.
degree in systems engineering from Xian Jiaotong
University, Xian, China, in 1993, 1996, and 2005,
respectively. He is currently a Professor with
the Systems Engineering Institute, Xian Jiaotong
University. His research interests include optimization of large-scale systems and integrated resource
bidding and scheduling in the deregulated electric
power market.

Jiang Wu (S07M09) received the B.S. degree


in electrical engineering and the Ph.D. in systems
engineering from Xian Jiaotong University, Xian,
Shaanxi, China, in 2002 and 2008, respectively. He
is currently a Lecturer with Systems Engineering
Institute, Xian Jiaotong University. His research
interests include optimization and simulation for
renewable generation, smart grids, and electricity
markets.

You might also like