1386 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO.

7, JULY 2014

Cooperation and Storage Tradeoffs in Power Grids

With Renewable Energy Resources
Subhash Lakshminarayana, Member, IEEE, Tony Q. S. Quek, Senior Member, IEEE, and
H. Vincent Poor, Fellow, IEEE

Abstract—One of the most important challenges in smart grid penetration of renewable energy this technique is no longer
systems is the integration of renewable energy resources into its cost effective [2]. As a result, there is a need for more cost
design. In this paper, two different techniques to mitigate the time- effective solutions such as the use of energy storage [3], and
varying and intermittent nature of renewable energy generation
are considered. The first one is the use of storage, which smooths load scheduling by demand-response [4].
out the fluctuations in the renewable energy generation across Prior work on design and analysis of renewable energy with
time. The second technique is the concept of distributed genera- storage includes [5], which formulates the finite horizon opti-
tion combined with cooperation by exchanging energy among the mal power flow problem with storage as a convex optimization
distributed sources. This technique averages out the variation in problem. This work shows that for the special case of a single
energy production across space. This paper analyzes the tradeoff
between these two techniques. The problem is formulated as a generator and single load, the optimal policy is to charge the
stochastic optimization problem with the objective of minimizing battery at the beginning and discharge towards the end of the
the time average cost of energy exchange within the grid. First, an time horizon. References [6] and [7] formulate the problem as
analytical model of the optimal cost is provided by investigating a dynamic programming problem and derive threshold based
the steady state of the system for some specific scenarios. Then, control policies for battery charging and discharging deci-
an algorithm to solve the cost minimization problem using the
technique of Lyapunov optimization is developed, and results for sions. In [8], storage is used as a means to reduce the time
the performance of the algorithm are provided. These results show average electricity bills in data center applications. Using a
that in the presence of limited storage devices, the grid can benefit Lyapunov optimization based approach, this work shows that
greatly from cooperation, whereas in the presence of large storage increasing the storage capacity results in a greater reduction
capacity, cooperation does not yield much benefit. Further, it is in the electricity bills. Other relevant works include [9] and
observed that most of the gains from cooperation can be ob-
tained by exchanging energy only among a few energy-harvesting [10] (and references within). On the other hand, prior work on
sources. demand response includes [11], which formulates the problem
of scheduling the power consumption as a Markov decision
Index Terms—Renewable energy, micro-grids, cooperation,
storage, Lyapunov optimization. problem, where the scheduler has access to the past and the cur-
rent prices, but only statistical knowledge about future prices.
It is shown that incorporating the statistical knowledge into the
I. I NTRODUCTION scheduling policies can result in significant savings. References
[12] considers the problem of demand response in a finite
R ENEWABLE energy provides a greener alternative to
traditional fossil fuel based electric power generation.
Thus, there has been significant emphasis on integration of
time domain and solves the problem using convex optimization
based techniques. The multi-period power procurement is han-
renewable energy into smart grid design [1]. However, a signif- dled in [13], where it is solved using stochastic gradient based
icant challenge lies in the inherently stochastic and intermittent techniques. The combination of storage and demand-response
nature of renewable energy production. A popular technique has been examined in [14], and is solved using Lyapunov
to compensate for this is the use of expensive fast-ramping optimization based approach with the conclusion that storage
fuel-based generators as a back-up. However, with greater combined with demand response can give greater cost savings.
One of the other techniques to combat the intermittent nature
of renewable energy that has been explored relatively less,
is the use cooperation in distributed power generation units
Manuscript received March 27, 2014; accepted May 10, 2014. Date of
[15]. The idea is to exploit the averaging effect produced
publication June 19, 2014; date of current version August 13, 2014. This work by diversity in renewable energy production across different
was supported in part by the SUTD-MIT International Design Centre under geographical areas. By enabling cooperation, areas that have
Grant IDSF1200106OH.
S. Lakshminarayana is with the Singapore University of Technology and
excess production can transfer energy to areas that are deficient.
Design, Singapore 138682 (e-mail: [email protected]). For example, studies have been conducted by monitoring the
T. Q. S. Quek is with the Singapore University of Technology and De- renewable energy production across 5 different states in the
sign, Singapore 138682, and also with the Institute for Infocomm Research,
Singapore 138632 (e-mail: [email protected]). United States (Arizona, Colorado, Nevada, New Mexico, and
H. V. Poor is with the Department of Electrical Engineering, Princeton Wyoming). These studies have shown that while the variability
University, Princeton, NJ 08544 USA (e-mail: [email protected]). of the load greatly increases with an increase in penetration
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. of renewable energy in the individual states, aggregating the
Digital Object Identifier 10.1109/JSAC.2014.2332093 diverse renewable resources over these geographic areas leads

0733-8716 © 2014 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.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
LAKSHMINARAYANA et al.: COOPERATION AND STORAGE TRADEOFFS IN POWER GRIDS 1387

Fig. 1. Power grid consisting of micro-grids and a macro-grid.

to only a slight increase in the load variability [15]. This the macro-grid. We analyze the optimal cost as a function of
also leads to a very substantial reduction in the operating the storage capacity and the number of cooperating MGs. We
cost of the grid as well ($2 billion in this case). In terms of also investigate the following question: for a given number of
analytical results, the impact of aggregation of wind power has cooperating MGs, what is the storage capacity needed in order
been considered in the framework of coalitional game theory to be self sufficient (i.e. to eliminate the need for energy transfer
in [16] and [17], where it is shown that independent wind from the macro-grid).
producers can benefit by aggregating their harvested energy. Our result shows that when the storage capacity is low,
Distributed energy production has also been studied within cooperation among the MGs yields a significant reduction in
the framework of micro-grids (MGs) [18], with a focus on the time average cost of energy exchange. However, when the
distributed storage and decentralized control of MG networks MGs have a large storage capacity, cooperation does not yield
[19], [20]. Energy sharing among MGs has also been studied much benefit. This is because each MGs can simply store all
in [21] via simulations, and shown to reduce energy losses in its excess harvested energy and use it during the time slots
the network. when it is deficient. Further, most of the gains are obtained by
While all of the above mentioned works address the issues cooperation among only a few neighboring MGs.
of storage, demand response and aggregation individually, the The rest of the paper is organized as follows. We present
combination of the dual averaging effect produced by storage the system model and provide the problem formulation of
and cooperation by energy sharing has not been explored. The minimizing the time average cost of energy transfer across the
objective of this work is to provide an analytical framework for grid in Section II. We provide the analytical modeling of the
studying the trade-off present between storage and cooperation. optimal cost for some special cases in Section III. Then, in
We consider a scenario consisting of MGs that are powered by Section IV, we present an algorithm to solve the time average
harvesting renewable energy and are serving their respective cost minimization problem using the technique of Lyapunov
loads. The MGs have finite capacity storage units and can optimization, and provide results for the algorithm perfor-
cooperate by transferring energy among themselves. For any mance. Numerical results are presented in Section V followed
excess load, the MGs can borrow energy from the macro-grid. by conclusions in Section VI. Finally, the proofs of some results
The objective is to minimize the time average cost of energy in the paper are presented in Appendices A, B, and C.
exchange within the entire grid. We first provide an analytical
characterization of the optimal cost by examining the steady
II. S YSTEM M ODEL
state behavior of the system for some particular settings. We
then provide an online algorithm to solve the optimization We consider an inter-connected power grid consisting of N
problem using the technique of Lyapunov optimization [22]. MGs and a macro-grid as shown in Fig. 1. The MGs are capable
The control decision to be taken at each time slot is how to of harvesting renewable energy (e.g. wind, solar energy etc.). In
divide the excess renewable energy optimally between storage addition, the MGs are equipped with batteries in which they can
and cooperation, and how much energy is to be borrowed from store the harvested energy for future use.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
1388 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 7, JULY 2014

A. Energy Supply, Demand and Distribution Models Accordingly, at each time t, we have
1) Energy Generation: MGi harvests Xi [t], i = 1, . . . , N 
Yi [t] + Bi,j [t] ≤ X̃i [t] ∀ t, i = 1, . . . , N. (2)
units of energy during the time slot t, which is the only source j=i
of energy generation at the MG. We assume that the harvested
energy Xi [t] evolves according to an independent and identi- 3) Energy Storage: Next, we consider the energy model for
cally distributed (i.i.d.) random process across time.1 However, the battery at the MGs. At MGi , the battery evolves according
the energy harvesting process can be arbitrarily correlated to the following rule:
across different MGs. The macro-grid generates energy from
convention energy sources. We assume that the macro-grid has Ei [t + 1] = Ei [t] − Bi,i [t] + Yi [t] ∀ t, i = 1, . . . , N,
a very large supply of energy (and do not impose any constraint (3)
on its energy generation).
2) Load Serving: MGi serves a set of users whose aggregate where the energy availability constrains the battery at each MGi
energy2 demand is Li [t] units of energy per time slot. The to satisfy
energy demand is bounded as Li [t] ≤ Lmax , for finite Lmax .
The energy demand is met in the following manner. Bi,i [t] ≤ Ei [t] ∀ t, i = 1, . . . , N. (4)
Firstly, the harvested energy is used to serve the load Li [t].
We consider the two cases as follows: We also impose a battery discharge constraint during every time
• If Xi [t] < Li [t], then MGi uses all the harvested energy to slot t, namely,
serve its load. The unsatisfied load is denoted by L̃i [t] =
Bi,i [t] ≤ Bmax
s
∀ t, i = 1, . . . , N, (5)
(Li [t] − Xi [t])+ and the MGi does the following to serve
the unsatisfied load: where Bmaxs
is the maximum discharge of the battery per time
1) Draw energy stored in its own battery slot. Furthermore, the energy storage device has finite capacity
The MGi uses Bi,i [t] units of energy from the en- of Emax units as follows:
ergy stored in its own battery to serve the unsatisfied
load to its respective users. Ei [t] ≤ Emax ∀ t, i = 1, . . . , N. (6)
2) Exchange energy among the MGs
In addition, MGi can borrow Bj,i [t] units of en- Additionally, we make the following practical assumption on
ergy from MGj such that j = i where Bj,i [t] is the battery capacity:
bounded as Bj,i [t] ≤ Bmax ex ex
, for some Bmax < ∞.
s
Note that Bj,i [t] > 0 only when X̃j [t] = (Xj [t] − Emax > Ymax + Bmax . (7)
Lj [t])+ > 0, i.e., MGj has excess harvested energy
(i.e. the harvested energy is greater than its demand). The constraints (4) and (5) can be combined as follows: we have
3) Transfer energy from the macro-grid
In case the energy from the battery and the energy Bi,i [t] ≤ min (Ei [t], Bmax
s
) ∀ t, i = 1, 2, . . . , N. (8)
borrowed from neighboring MGs is insufficient to
satisfy the demand, MGi can borrow Gi [t] units of Similarly the battery input energy constraint Yi [t] ≤ Ymax and
energy from the macro-grid. (6) can be combined as
The sum of energy drawn from the battery, energy
Yi [t] ≤ min (Emax − Ei [t], Ymax ) ∀ t, i = 1, 2, . . . , N.
exchange with the neighboring MGs, and the energy
(9)
borrowed from the macro-grid must satisfy the residual
demand, i.e., 4) Cost Model and Problem Formulation: We consider that
 transferring energy from MGi to MGj during the time slot t
Bi,i [t] + Bj,i [t] + Gi [t] = L̃i [t] ∀ t, i = 1, . . . , N. incurs a cost of pi,j [t] per unit. Similarly, transferring energy
j=i from the macro-grid to MGi incurs a cost of qi [t] per unit.
(1) For simplicity, we assume that the costs {pi,j [t], qi [t]} are
• Now we consider the second case, in which the harvested i.i.d. across time slots.3 Further, we assume that the costs are
energy exceeds the energy demand, i.e., if Xi [t] ≥ Li [t], bounded as pi,j [t] ≤ pmax ∀i, j, t for some finite pmax , and
then the MG does the following: qi [t] ≤ qmax ∀i, t for some finite qmax .
1) As previously mentioned, it can donate an amount The total cost incurred for energy transfer to MGi during the
Bi,j [t] to satisfy the load of MGj . time slot t (denoted by Costi [t])is given by
2) Store an amount Yi [t] ≤ Ymax (where Ymax < ∞) 
in its own battery to be used at a later time. Costi [t] = qi [t]Gi [t] + pj,i [t]Bj,i [t], i = 1, . . . , N.
j=i
1 The i.i.d. assumption is made for the sake of convenience of illustrations (10)
and technical proofs. We note that the algorithm developed in this paper can
also be extended to the case where the energy harvesting process in Markovian.
2 With slight abuse of terminology, we use the terms power and energy 3 Once again, we point out that the algorithm developed in this work can be
interchangeably. generalized to the case when {pi,j [t], qi [t]} are Markovian.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
LAKSHMINARAYANA et al.: COOPERATION AND STORAGE TRADEOFFS IN POWER GRIDS 1389

The objective of the controller is to design the system parame- • A more practical set-up in which the MGs are connected
ters in order to minimize the time average cost of energy trans- to different buses, and energy exchange performed via a
fer across the grid subject to the renewable energy generation constrained transmission network is not considered this
and battery constraints, stated as follows: work. This is done in order to clearly illustrate the trade-
N  offs between cooperation and storage, without the compli-
T −1
1   cations arising from physical power flow constraints. The
min lim sup E Costi [t] more practical case incorporating transmission network
T →∞ T t=0 i=1 constraints is left for future investigation.

s.t. Bi,i [t] + Bj,i [t] + Gi [t] = L̃i [t], ∀t, ∀i
j=i
III. A NALYTICAL C HARACTERIZATION

Yi [t] + Bi,j [t] ≤ X̃i [t], ∀t, ∀i OF THE O PTIMAL C OST
j=i Before solving the optimization problem in (11), we first
Ei [t + 1] = Ei [t] − Bi,i [t] + Yi [t], ∀t, ∀i provide an analytical characterization of the time average cost
for certain specific scenarios.
Bi,i [t] ≤ s
min (Ei [t], Bmax ) ∀t, ∀i
Yi [t] ≤ min (Emax − Ei [t], Ymax ) ∀t, ∀i A. Single MG Scenario
Bi,j [t] ≤ ex
Bmax ∀t, ∀j = i, ∀i. (11) First, we consider the analysis only for a single MG which
is equipped with a storage device. There is no energy to be
During each time slot t, the decision variables are Yi [t], Bi,i [t], exchanged, and any unsatisfied load must be fulfilled by using
Bi,j [t](∀j = i), Gi [t] ∀i. We denote the optimal value of the the energy stored in storage device, or by borrowing energy
 −1 N
cost function limT →∞ (1/T ) Tt=0 E[ i=1 Costi [t]] over all from the macro-grid. We make the following simplifications.
∗
possible control actions by fN . Note that we have explicitly 1) Modeling the Energy Arrival Process: We assume that
mentioned the subscript N to denote the optimal solution when the energy arrivals per time slot are integer random variables.
N MGs cooperate. In addition, we consider the probability mass function (p.m.f.)
of the excess energy arrival process4 i.e.,

B. Discussion of the System Model X̃[t] = X[t] − L (12)

We now provide some remarks on the system model.
to be given by
• Note that in this work, we assume that the excess renew- ⎧
⎪ M w.p. aM
able energy from MGi can be exchanged with MGj only ⎪
⎪
⎪
⎪ M −1 w.p. aM −1
to satisfy the load of MGj during the same time slot. ⎪
⎪ .
In other words, MGj cannot use the excess renewable ⎪
⎨ ..
energy of MGi to charge its battery. Further, we assume fX̃ (x) = 0 w.p. a0 (13)
⎪
⎪ .
that the energy stored in the battery of MGi can only be ⎪
⎪ ..
⎪
⎪
used to serve its own load during a future time slot, and ⎪
⎪ −(M − 1) w.p. dM −1
⎩
there is no energy exchange possible from the battery. −M w.p. dM ,
These restrictions are imposed in order to clearly exhibit
the trade-off between storing energy for future use and where
cooperation in the current time slot (by exchanging en-
ergy in the current time slot with neighboring MGs). The dM + dM −1 + · · · + a0 + · · · + aM −1 + aM = 1. (14)
framework developed in this work can be easily extended
Here, we have assumed that −M ≤ X̃[t] ≤ M . Note that the
to incorporate all these cases.
random process X̃[t] can be used to model discrete random
• Transferring energy between different elements of the
processes like the Poisson process.
grid incurs energy losses. The cost of energy transfer
2) Modeling the Storage: As before, let us assume that
{pi,j [t], qi [t]} considered in this work can be interpreted
the MG has storage capability represented as a virtual energy
as a price paid for these power losses. Typically, the
queue whose queue-length at time t is given by E[t], and the
distance between the MGs is much less compared to their
maximum battery capacity being Emax . The evolution of the
distance from the macro-grid. Therefore, in practice, the
battery can be modeled as a random walk which evolves as
cost of energy exchange between the MGs is lower, i.e.,
follows:
pi,j [t] < qi [t]. Moreover, the MGs can be connected by
a short distance DC power line which incurs much less E[t + 1] = min max E[t] + X̃[t], 0 , Emax . (15)
energy loss than the AC power line connection between
the MGs and the macro-grid [23]. The objective function
can then be viewed as minimizing the cost associated with 4 In this section, we suppress the subscript i since we are considering a
the energy losses. single MG.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
1390 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 7, JULY 2014

Fig. 3. State diagram for the Markov chain modeling of the energy storage
Fig. 2. State diagram for the Markov chain modeling for the energy storage with p.m.f. of the arrival process as in (19).
model with arrival process given in (13).
B. Case of Multiple Micro-Grids
Assuming that the arrivals are i.i.d. across time, it can be
The characterization of the optimal cost in the case of mul-
verified that the random process (E[t], t ≥ 0) is Markovian.
tiple MGs is complicated due to the fact the excess energy
The state diagram of the Markov chain corresponding to the
arrivals in the grids are correlated because of the ability to
random process E[t] is shown in Fig. 2. We denote πt [i] =
share energy among themselves. Therefore, we provide a closed
P(E[t] = i). Let us assume that the limiting state distribution
form analytical characterization only in the special case of a
function π(i) = limt→∞ πt (i) exists for all 0 ≤ i ≤ Emax and
completely symmetric two MG set-up.
denote π = [π(1), π(2), . . . , π(Emax )]. We denote the transi-
Consider a symmetric scenario consisting of 2 MGs (MG1
tion matrix of this Markov chain by P. The Markov chain corre-
and MG2 respectively), such that the cost of energy exchanges
sponding to the random process E[t] has a limiting distribution
p1,2 = p2,1 = pmax and q1 = p2 = qmax . Each MG has an
π(i) = limt→∞ πt (i) for all 0 ≤ i ≤ Emax , independent of the
excess energy arrival process whose p.m.f. is given as in (19).
initial distribution, if the system of equations
Further, we consider a special case when the energy arrival
π = πP (16) process is independent across the two MGs. In order to model
the energy transfer between the grid, we consider the following
π1 = 1 (17) policy. Whenever, MG1 produces excess energy and MG2 has
an energy deficit, MG1 transfers its excess energy to MG2 with
has a strictly positive solution [24]. probability α ∈ [0, 1]. Otherwise, MG1 stores the excess energy
If the limiting distribution exists, then an analytical expres- into its battery with a probability 1 − α. Since the system is
sion for the cost incurred in borrowing energy from the macro- perfectly symmetric, MG2 does the same in the case when it
grid (denoted by Cost) can be derived as follows: In the steady over produces and MG1 has an energy deficit. In all other cases,
state, when the excess renewable energy arrival is negative, i.e., there is no requirement to exchange energy between them. Let
X̃ = i for i = −1, . . . , −M (which happens with probability us denote the random variable Z̃1 and Z̃2 representing the
di , i = 1, . . . , M , respectively), and when the battery is in the effective excess energy arrival. It is related to X̃1 and X̃2 as
state j for j ≤ i, (which happens with probability πj ), the MG follows:
has to borrow i − j units of energy from the macro-grid at the • If X̃1 = 1 and X̃2 = −1, then
price qmax per unit. Mathematically we can write, 
1 w.p. (1 − α)ad

M 
i Z̃1 =
0 w.p. αad,
(22)
Cost = qmax (i − j)di π(j). (18)

i=1 j=0 −1 w.p. (1 − α)ad
Z̃2 = (23)
0 w.p. αad.
Example 1: Consider the special case in which the excess
energy arrival process has the following distribution:
• If X̃1 = −1 and X̃2 = 1
−1 w.p. d 
−1 w.p. (1 − α)ad
fX̃ (x) = 0 w.p. 1 − a − d (19) Z̃1 = (24)
0 w.p. αad,
1 w.p. a.

1 w.p. (1 − α)ad
The Markov chain associated with the random process E[t] Z̃2 = (25)
0 w.p. αad.
in this case is illustrated in Fig. 3. In this case, the stationary
distribution of the Markov chain corresponding to the random • In all other cases, Z̃1 = X̃1 and Z̃2 = X̃2 .
walk has a simple form given by
Therefore, the unconditional p.m.f. of Z̃1 and Z̃2 can be

1−r computed by integrating out the other variable. It is given as
π(i) = ri (20) ⎧
1 − r(Emax +1) ⎨ −1 w.p. d(1 − αa)
fZ̃1 (z) = 0 w.p. 2αad + (1 − a − d) (26)
where r = a/d. Therefore, the cost of borrowing energy from ⎩
1 w.p. a(1 − αd).
the macro-grid is given by
The evolution of the battery can now be modeled as a random
Cost = qmax P(X̃ = −1)π(0) = qmax dπ(0)
 walk which evolves as
1−r
= qmax d . (21) Ei [t + 1] = min max Ei [t] + Z̃i [t], 0 , Emax . (27)
1 − r(Emax +1)

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
LAKSHMINARAYANA et al.: COOPERATION AND STORAGE TRADEOFFS IN POWER GRIDS 1391

The steady state distribution of the Markov chain corresponding from very high complexity and require a-priori knowledge of
to this random walk is given by the statistics of all the random processes in the system.
 But first, we note that the technique of Lyapunov optimiza-
1−r
π(j) = rj , (28) tion is not directly applicable for solving (11). This is due to
1 − r(Emax +1) the presence of constraints (8) and (9), which have the effect
where of coupling the control decisions across time slots. In order to
circumvent this issue, we consider an approach similar to [8],
a(1 − αd) and formulate a slightly modified version of this problem, stated
r= . (29)
d(1 − αa) as follows:
T −1
N 
Therefore, the cost of energy exchange within the grid has 1  
two components: energy exchanged among the MGs and the min lim E Costi [t]
T →∞ T
t=0 i=1
energy borrowed from the macro-grid. Mathematically, this is
T −1 T −1
given as 1  1 
s.t. lim E [Yi [t]] ≤ lim E [Bi,i [t]] ∀i
T →∞ T T →∞ T
Cost(α) = 2αadpmax + 2d(1 − αa)π(0)qmax . (30) t=0 t=0

Bi,i [t] + Bj,i [t] + Gi [t] = L̃i [t], ∀t, ∀i
The minimum cost is then obtained by optimizing over the j=i
choice of α. Let us define 
Yi [t] + Bi,j [t] ≤ X̃i [t], ∀t, ∀i
α∗ = arg min Cost(α), (31) j=i
α
Bi,i [t] ≤ Bmax
s
; Bi,j [t] ≤ Bmax
ex
∀t, ∀j = i, ∀i,
and hence, the minimum cost of energy exchange is given by Yi [t] ≤ Ymax ∀t, ∀i. (35)
Cost(α∗ ). From (30), we can analyze the following two extreme
cases, namely, the case with no storage and the case with infinite Note that in (35), all the constraints associated with the bat-
storage. tery are relaxed and a constraint of the form limT →∞ (1/T )
T −1 T −1
t = 0 E[Yi [t]] ≤ lim T → ∞ (1/T ) t = 0 E[B i , i [t]] ∀ i are
• Emax = 0—No storage
In this case π(0) = 1. Therefore, added. Effectively, this constraint represents the condition for
stability of the virtual energy queue associated with the battery.
Cost = 2dqmax + adα(−qmax + pmax ), (32) We will henceforth address this problem as the relaxed problem.
∗
Let us denote by gN , the optimal value of the cost function
and the cost is minimized when,
 of the relaxed problem over all possible control decisions.
∗ ∗
α=
1 if pmax < qmax
(33) First, it is easy to see that gN ≤ fN , i.e. the solution to the
0 else. relaxed problem acts as a lower bound on the original problem
Thus, in the absence of storage, the MGs must always (since the relaxed problem has fewer constraints compared to
share the excess available energy as long as pmax < qmax . the original problem and any feasible solution of the original
The result is quite intuitive since in the absence of storage, problem is feasible for the relaxed problem as well).
one can always reduce the cost by exchanging energy We now focus on solving the relaxed problem. It can be
locally between the MGs. shown that the optimal solution to the relaxed problem can be
• Emax = ∞—Infinite storage obtained by the method of stationary randomize policy, stated
In this case, π(0) = (d − a)/d(1 − αa). Therefore, in the following theorem.
Theorem 1: There exists a stationary and randomized policy
Cost = 2d(d − a)qmax + 2adα. (34) Π that achieves
N 
Thus, the cost is minimized when α = 0. This implies that 
∗
E Π
Costi [t] = gN ∀t, (36)
in the presence of infinite storage, any excess energy must i=1
always be stored rather than exchanging among them.
In what follows, we provide a practical algorithm to solve the and satisfies the constraints:
time average cost minimization problem across the grid using  
E YiΠ [t] ≤ E[Bi,i Π
[t]] ∀i, ∀t
the technique of Lyapunov optimization technique. 
 Π   Π   
E Bi,i [t] + E Bj,i [t] + E GΠ
i [t] = L̃i [t], ∀t, ∀i
j=i
IV. A LGORITHM D ESIGN BASED ON
    Π   
LYAPUNOV O PTIMIZATION E YiΠ [t] + E Bi,j [t] ≤ E X̃i [t] , ∀t, ∀i
The Lyapunov optimization method provides simple online j=i
solutions based only on the current knowledge of the system  Π   Π 
E Bi,i [t] ≤ Bmax
s
; E Bi,j [t] ≤ Bmax
ex
∀t, ∀j = i, ∀i,
state as opposed to traditional approaches such as dynamic  
programming and Markov decision processes which suffer E YiΠ [t] ≤ Ymax ∀t, ∀i.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
1392 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 7, JULY 2014

⎛
The existence of such a policy can be proved by using the  
Caratheodory theorem, similar to the arguments in [22] and +V ⎝ pi,j Bi,j [t] + qi [t]
omitted here for brevity. Note that due to the high dimensional- i,j i
⎛ ⎞⎞ ⎤
ity, it is not practical to solve the problem using the method of


stationary randomized policies. × ⎝L̃i [t] − Bi,i [t] − Bj,i [t]⎠⎠ E[t]⎦
In what follows, we apply Lyapunov optimization to solve j=i 
the relaxed problem (35). Further, we will show that the so-
(41)
lution developed for (35) by our method also satisfies all the 
 
constraints associated with battery, hence making it applicable =C + E V qi [t]L̃i [t] + Ẽi [t]Yi [t]
for solving the original problem (11). i i
We proceed by considering the Lyapunov function associated 
+ V Bi,j [t] (pi,j [t] − qj [t])
with the virtual energy queues defined as follows:
i j=i
1  
 
Ψ[t] = (Ei [t] − θ)2 (37) 
2 i − Bi,i [t] Ẽi [t] + V qi [t] E[t] . (42)

i
s
where θ is a perturbation which is given by θ = Bmax +
From the theory of Lyapunov optimization (drift-plus penalty
V qmax . The exact rationale for the choice of the value of θ will
method), the control actions are chosen during each time slot
be specified later when we analyze the algorithm performance.
to minimize the bound on the modified Lyapunov drift function
We will now examine the Lyapunov drift which represents
(on the right hand side of (42)) [22]. Before we proceed, we pro-
the expected change in the Lyapunov function from one time
vide the main intuition behind the solving the relaxed problem
slot to the other, which is defined as
using this approach. Notice that the relaxed problem can viewed
Δ[t] = E [Ψ[t + 1] − Ψ[t]|E[t]] , (38) as minimizing the time average cost of energy exchange in the
grid while maintaining the stability of the virtual energy queue
where the expectation is with respect to the random processes (battery). The modified Lyapunov drift has two components,
associated with the system, given the energy queue-length val- the Lyapunov drift term Δ[t], and V × Cost[t] term. Intuitively,
ues E[t] = [E1 [t], . . . , EN [t]]. Using the equation for evolution minimizing the Lyapunov drift term alone pushes the queue-
of the virtual energy queue associated with the battery in (3), length of the virtual energy queue to a lower value. The second
and some standard manipulations, it can be shown that the metric V × Cost[t] can be viewed as the penalty term, with the
Lyapunov drift can be bounded as parameter V representing the trade-off between minimizing the
  queue-length drift and minimizing the penalty function. Greater

Δ[t] ≤ C − E (Ei [t] − θ) (Bi,i [t] − Yi [t]) |E[t] (39) value of V represents greater priority to minimizing the cost
i metric at the expense of greater size of the virtual energy queue
and vice versa. This is indeed the rationale behind minimizing
where C < ∞ is a constant. For completeness, we provide the the modified Lyapunov drift ΔV [t] during each time slot.
proof of this step in Appendix A. The control algorithm using the aforementioned rule can
 Ẽi [t] = Ei [t]−θ.
We will henceforth denote  Adding the per- be described as follows. During each time slot t, one must
formance metric V E[( i,j pi,j Bi,j [t] + i qi [t]Gi [t])|E[t]] choose the control decisions as a solution to the following linear
(where V is another control parameter which will be specified programming problem (obtained by minimizing the right hand
later) to both the sides and denoting side of (42)):
⎡ ⎤
   
ΔV [t] = Δ[t] + V E ⎣ pi,j Bi,j [t] + qi [t]Gi [t]|E[t]⎦ , min Ẽi [t]Yi + V Bi,j (pi,j [t] − qj [t])
Yi ,Bi,i ,Bi,j
i,j i i j=i

− Bi,i Ẽi [t] + V qi [t]
we have
i
⎡  
 s.t. Yi + Bi,j ≤ X̃i [t], Bi,i + Bj,i ≤ L̃i [t]
ΔV [t] ≤ C − E ⎣ Ẽi [t] (Bi,i [t] − Yi [t]) j=i j=i

⎛
i
⎞ ⎤ 0 ≤ Yi ≤ Ymax , 0 ≤ Bi,i ≤ Bmax
s
∀i,
 0 ≤ Bi,j ≤ Bmax ∀j = i, ∀i.
ex
   (43)
−V ⎝ pi,j Bi,j [t] + qi [t]Gi [t]⎠ E[t]⎦ . (40)
i,j i  Let us denote the solution corresponding to (43) as Yi∗ [t], Bi,i
∗
[t]
∗ ∗
and Bi,j [t]. The value of Gi [t] is then given by

Using (1), we have Gi [t] = L̃i [t] − Bi,i [t] − j=i Bj,i [t]. ⎛ ⎞+
Substituting for Gi [t] in the right hand side (40), we have, 
⎡ G∗i [t] = ⎝Li [t] − Bi,i
∗
[t] − ∗
Bj,i [t]⎠ , (44)
 j=i
ΔV [t] ≤ C + E ⎣− Ẽi [t] (Bi,i [t] − Yi [t])
i where (x)+ = max(x, 0).

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
LAKSHMINARAYANA et al.: COOPERATION AND STORAGE TRADEOFFS IN POWER GRIDS 1393

A. Algorithm Performance Analysis

We will now analyze the performance of the algorithm
described in the previous section.
Lemma 1: By choosing the parameters V and θ as
Emax − (Ymax + Bmax
s
)
0 <V ≤ (45)
qmax
s
θ = Bmax + V qmax (46)

the following hold true:

1) If Ei [t] > Emax − Ymax , then Yi∗ [t] = 0.
s ∗
2) If Ei [t] < Bmax , then Bi,i [t] = 0.
Proof: See Appendix B. 
From the result of Lemma 1, it can be seen that the battery
charging decisions are non-zero only when Ei [t] < Emax −
Ymax and the discharge decisions are non-zero only when
s Fig. 4. Time average cost versus storage capacity based on analytical result
Ei [t] < Bmax . As a consequence of this lemma, it can be and Lyapunov optimization based method.
verified that the algorithm developed satisfies all the constraints
associated with the battery in the original problem, i.e., con-
straints (8) and (9). Therefore, the algorithm developed is a the excess energy arrival process considered in the simulations
feasible algorithm for the original optimization problem, and it is provided in (19), with d = 0.5, a = 0.2. We observe that
can be applied to the original system under consideration (with there is a good match between the two curves and the time
finite battery capacity constraints). This also justifies the choice average cost decreases with an increase in storage capacity.
of the perturbation parameter θ in our design. Next, we consider the normalized time average cost for
 different combinations of the number of cooperating MGs and
Let us denote the cost function fN [t] = N i=1 Costi [t]. We
will now provide the main result related to the performance of the storage capacity. We consider the following setting. We
our algorithm. assume that the actual harvested renewable energy consists
Theorem 2: For the algorithm developed in the previous of two components, namely the predicted component which
subsection, the virtual energy queue-lengths can be bounded as is deterministic, and a prediction error which is random i.e.,
follows: Xi [t] = X̂i [t] + Wi [t]. For the sake of simplicity, we assume
that the predicted component X̂i [t] perfectly matches the ag-
0 ≤ Ei [t] ≤ Emax ∀ t, i = 1, . . . , N, (47) gregate load Li [t]. We model the excess renewable energy pro-
duction X̃i [t] = Xi [t] − Li [t] (also the prediction error Wi [t]
and the time average cost function achieved by this algorithm in this case) by the truncated normal distribution in accor-
satisfies dance with some previous studies in this field [25]. Therefore,
T −1 excess renewable energy production is a zero mean random
1  ∗ B̃ process with its distribution given by X̃i [t] ∼ N (0, ν 2 ), where
lim sup E [fN [t]] ≤ gN + (48)
T →∞ T t=0 V ν is the standard deviation. Note that following this model
of energy production (i.e., X̂i [t] = Li [t] and mean value of
where B̃ < ∞ is a constant. X̃i [t] = 0), we ensure that the aggregate energy production of
Proof: See Appendix C.  each MG matches the aggregate load, and the only reason to
Theorem 2 implies that performance of our algorithm can store/exchange energy is to compensate for the fluctuations in
be made arbitrarily close to the optimal value by increasing renewable energy (thus capturing the most essential aspect of
the value of parameter V . However, this comes at the cost of the problem considered in this work). In our numerical results,
increasing the battery capacity Emax (due to the bound on the we choose the value of Li [t] = Li = 10 MW, ∀ t, i and the
∗ ∗
value of V in (45), given Emax ). Also, note that since gN ≤ fN , variance ν = 3 MW.
∗
the performance bounds of (48) hold with respect fN as well In order to construct a wind farm of MGs, we choose a
(and are in fact tighter). square grid of dimensions 10 × 10 km. The positions of the
micro-grids are chosen by generating uniform random numbers
within this grid. We consider that the macro-grid is located at
V. N UMERICAL R ESULTS
the coordinate (20, 20) km. A random snapshot the power grid
In this section, we present some numerical results to examine with N = 5 MGs is shown in Fig. 5.
the analytical modeling of the cost function and the Lyapunov We assume that the cost of transporting energy among differ-
optimization based algorithm described in the previous section. ent elements of the grid is directly proportional to the distance
First, we plot the optimal time average cost as a function of between them. Therefore, pi,j [t] = βdi,j , ∀t, units (where di,j
the storage size for the single MG case based on the analytical is the distance between MGi and MGj in km) and qi [t] =
expression of (21), and by running the Lyapunov optimization βDi , ∀t, (where Di is the distance between MGi and the
based algorithm for T = 5000 iterations in Fig. 4. The p.m.f. of macro-grid in km) and β being the proportionality constant. In

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
1394 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 7, JULY 2014

ex
10, 20, and 50 MWh) respectively. The value of Bmax chosen
in 10 MW.
The following observations can be made.
• For a given storage capacity, the cost of energy exchange
decreases with an increase in the number of cooperating
MGs. This is due to the fact that greater number of
cooperating MGs leads to a greater diversity in energy
production and hence greater possibility of sharing energy
among the MGs (hence reducing the need for borrowing
energy from the macro-grid).
• The decrease in the normalized cost (as a function of the
number of cooperating MGs) is greater for lower values
of storage capacity. For higher values of storage capacity,
the normalized cost does not reduce with increasing N .
This is due to the fact that with greater storage capacity,
the MGs are able to store any excess harvested energy
Fig. 5. A snapshot of the power grid with 5 MGs and the macro-grid. The red
dots indicate the position of the MGs and the black dot indicates the macro-grid. (during the time slots when the harvested energy is greater
The dotted lines represent the boundary of the wind farm. than the aggregate load) and use it during the time slots
when the harvested energy is deficient, thereby, eliminat-
ing the need for energy cooperation.
• Further, most of the reduction in the time average cost
is achieved by cooperation among only a few neighbor-
ing MGs. The incremental gain obtained by cooperation
among large number of MGs is not significant.
Next, we try to examine the following question: for a given
number of cooperating MGs, what is the storage capacity
needed per MG to eliminate the need for borrowing energy from
the macro-grid? In order to do so, we consider a hypothetical
scenario in which pi,j [t] = pi,j = β unit ∀i, j, t and qi [t] =
qi = 3β units ∀i, t (both being independent of the distance
between elements of the grid). Once again, we choose Li [t] =
Li = 10 MW, ∀t, i and ν = 3 MW. We look for the combina-
tion of the storage and number of cooperating MGs that yields
a normalized time average cost

Fig. 6. Normalized cost versus the number of cooperating MGs for different C̄
values of storage capacity. = β × 107 units
N
our numerical results, we use β = 1. For each random snapshot as the bench mark (The choice of β × 107 units comes from the
the power grid, we run the simulation for T = 5000 time slots. fact that pi,j [t] = β units and 107 corresponds to the 10 MW
Let us consider the time average cost incurred demand per time slot). We plot the optimal value of N and
T −1 N Emax required to make the normalized cost below 1 unit in
1 
C̄N = Ci [t] Fig. 7. In our numerical results, we choose β = 1. It can be
T t=0 i=1 seen that for a given number of cooperating MGs, there exists
⎛ ⎞ an optimal storage capacity requirement to eliminate the need
−1 −1
1  ⎝  
N T T
for borrowing energy from the macro-grid. It is evident that as
= βdi,j Bi,j [t] + qi Di Gi [t]⎠
T i=1 t=0 t=0
the number of cooperating MGs increases, the optimal storage
j=i
capacity requirement reduces.
in transferring energy across the grid, where N is the number A more practical case in which the algorithm is implemented
of cooperating MGs. In order to average out the position of on the renewable energy data provided by National Renewable
the elements of the grid, we generate 100 random snapshots Energy Laboratory (NREL) of the United States is presented in
of the power grid and obtain an average of the normalized cost [26] and similar results have been obtained.
for these 100 positions. We plot the normalized cost incurred
C̄N /N as a function of N , for different values of storage capac-
VI. C ONCLUSION AND F UTURE W ORK
ity (2, 5, 10, 20, and 50 MWh) in Fig. 6. In each of these cases,
s
we choose Bmax and Ymax to satisfy the constraint (7). Specif- In this work, we explored the benefits of energy storage and
s
ically, we choose (Bmax , Ymax ) = (0.5, 0.5), (1, 1), (2, 2), cooperation among interconnected MGs as a means to combat
(5, 5), (10, 10) MW in the five cases of storage capacity (2, 5, the uncertainty in harvesting renewable energy. We modeled

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
LAKSHMINARAYANA et al.: COOPERATION AND STORAGE TRADEOFFS IN POWER GRIDS 1395

A PPENDIX B
P ROOF OF L EMMA 1
Let us first focus on statement 1. First, note that since the
objective is to minimize (43), it can be easily inferred that
Yi∗ [t] = 0 when Ẽi [t] > 0, i.e., Ei [t] − V qmax + Bmax
s
> 0.
∗
This implies that Yi [t] = 0 when
s
Ei [t] > V qmax + Bmax . (51)

Using the bound on the value of V from (45), in equation(51),

we can conclude thatYi∗ [t] = 0 when

Ei [t] > Emax − (Bmax

s s
+ Ymax ) + Bmax
= Emax − Ymax . (52)

Let us now turn to statement 2. Once again, from the objective

∗
Fig. 7. Storage capacity versus number of cooperating MGs to achieve a time function in (43), it is clear that Bi,i [t] = 0 if Ẽi [t] + V qi [t] < 0.
average cost of 107 units. Substituting for Ẽi [t], we have,
the set-up as an optimization problem to minimize the cost of Ei [t] − (V qmax + Bmax
s
) + V qi [t] < 0
energy exchange among the grid for a given storage capacity
Ei [t] − Bmax
s
< 0, (53)
at the MGs. First, we provided an analytical expression for the
time average cost of energy exchange in the presence of storage where the last step follows since qi [t] ≤ qmax . Therefore,
by analyzing the steady state of the system for some special ∗
Bi,i [t] = 0 when Ei [t] − Bmax
s
< 0.
cases of interest. We then developed an algorithm based on
Lyapunov optimization to solve this problem, and provided per-
A PPENDIX C
formance analysis of this algorithm. Our results show that in the
P ROOF OF T HEOREM 2
presence of limited storage devices, the grid can benefit greatly
by cooperating even among only a few distributed sources. We will first prove the bound on the battery size 0 ≤ E[t] ≤
However, in the presence of large storage, cooperation does not Emax . We use the analysis similar to [8] to obtain a bound on
yield much benefit. Our solution can be useful for the power the battery size. We consider four cases.
grid designer in terms of choosing the optimal combination of • Case 1: V qmax + Bmax s
≤ Ei [t] ≤ Emax . In this case,
storage size and cooperation in order to meet a specific cost ∗
we have that Yi [t] = 0. Therefore, Ei [t + 1] ≤ Ei [t] ≤
criterion. Since this work is a first step towards exploring the Emax .
trade-offs between cooperation and storage, we ignored the • Case 2: Ei [t] < V qmax + Bmaxs
. In this case, we have that
transmission network constraints throughout. The analysis of ∗
Yi [t] ≤ Ymax . Therefore, E[t + 1] ≤ V qmax + Bmax s
+
how these constraints impact the trade-off between cooperation Ymax ≤ Emax , where the last inequality follows from the
and storage is left for future investigation. range of values of V as considered in (45).
∗
• Case 3: 0 ≤ E[t] ≤ Bmaxs
. In this case, Bi,i [t] = 0. There-
A PPENDIX A fore, E[t + 1] ≥ E[t] ≥ 0.
∗
P ROOF OF (39) • Case 4: E[t] > Bmaxs
. In this case, Bi,i [t] ≤ Bmax
s
and
First note that therefore, E[t] ≥ 0.
We next proceed to prove the result on time average perfor-
Ei [t + 1] − θ = Ei [t] − θ − Bi,i [t] + Yi [t]. mance of the algorithm. Consider the bound on the Lyapunov
drift function of (42). It is clear that the control actions chosen
Squaring both sides of the equation, we obtain, according to the solution of (43) minimize the bound on the
Lyapunov function over all possible control actions. Comparing
(Ei [t + 1] − θ)2 = (Ei [t] − θ + Yi [t] − Bi,i [t])2
it with the control action chosen according to any stationary and
= (Ei [t] − θ)2 + (Bi,i [t] − Yi [t])2 randomized policy (which we will denote with the superscript
− 2 (Ei [t] − θ) (Bi,i [t] − Yi [t]) . (49) Π), we have,

The term (Bi,i [t] − Yi [t])2 ≤ Ymax
2 2
+ Bmax
Δ
= C. Using this  
bound and rearranging (49), we obtain ΔV [t] ≤ C + E V qi [t]L̃i [t] + Ẽi [t]Yi∗ [t]
i i

(Ei [t + 1] − θ)2 − (Ei [t] − θ)2 ≤ C + ∗
V Bi,j [t] (pi,j [t] − qj [t])
− 2 (Ei [t] − θ) (Bi,i [t] − Yi [t]) . (50) i,j=i


Summing it over i = 1, . . . , N and taking the conditional ex- ∗
− Bi,i [t] Ẽi [t] + V qi [t] |E[t] (54)
pectation on both sides given E[t], we obtain the bound in (39). i

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
1396 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 7, JULY 2014


  [10] E. Bitar, R. Rajagopal, P. Khargonekar, and K. Poolla, “The role of
≤C + E V qi [t]L̃i [t] + Ẽi [t]YiΠ [t] co-located storage for wind power producers in conventional electricity
i i
markets,” in Proc. Amer. Control Conf., 2011, pp. 3886–3891.
 [11] T. T. Kim and H. V. Poor, “Scheduling power consumption with price
+ V Π
Bi,j [t] (pi,j [t] − qj [t]) uncertainty,” IEEE Trans. Smart Grid, vol. 2, no. 3, pp. 519–527,
i,j=i Sep. 2011.
 [12] N. Li, L. Chen, and S. Low, “Optimal demand response based on utility
 maximization in power networks,” in Proc. IEEE PES Gen. Meet., 2011,
− Π
Bi,i [t] Ẽi [t] + V qi [t] |E[t] . (55) pp. 1–8.
i [13] L. Jiang and S. Low, “Multi-period optimal energy procurement and de-
mand response in smart grid with uncertain supply,” in Proc. IEEE Conf.
Rearranging, we obtain Decision Control Eur. Control Conf., 2011, pp. 4348–4353.
⎡ ⎛ ⎞ [14] L. Huang, J. Walrand, and K. Ramchandran, “Optimal demand response
with energy storage management,” in Proc. IEEE Int. Conf. Smart Grid
 
ΔV [t] ≤ C + E ⎣V qi [t] ⎝L̃i [t] − Bi,i [t]⎠
Commun., 2012, pp. 61–66.
Π
[t] − Π
Bj,i [15] Western Wind, Solar Integration Study, May 2009. [Online]. Available:
i j=i http://www.nrel.gov/electricity/transmission/western_wind.html
⎤ [16] E. Baeyens, E. Bitar, P. Khargonekar, and K. Poolla, “Wind energy ag-
    gregation: A coalitional game approach,” in Proc. IEEE Conf. Decision
+ Ẽi [t] YiΠ [t] − Bi,i
Π
[t] + Π
V pi,j [t]Bi,j [t]|E[t]⎦ Control Eur. Control Conf., 2011, pp. 3000–3007.
i i,j=i [17] W. Saad, Z. Han, and H. Poor, “Coalitional game theory for cooperative
 micro-grid distribution networks,” in Proc. IEEE Int. Conf. Commun.
 Workshops, 2011, pp. 1–5.
=C +E V qi [t]GΠ
i [t] [18] N. Hatziargyriou, H. Sano, R. Iravani, and C. Marnay, “Microgrids: An
overview of ongoing research development and demonstration projects,”
i
⎤ IEEE Power Energy Mag., vol. 5, no. 4, pp. 78–94, Aug. 2007.
    [19] S. Nykamp, M. Bosman, A. Molderink, J. Hurink, and G. Smit, “Value of
+ Ẽi [t] YiΠ [t] − Bi,i
Π
[t] + Π
V pi,j [t]Bi,j [t]|E[t]⎦ , storage in distribution grids—Competition or cooperation of stakehold-
i i,j=i ers?” IEEE Trans. Smart Grid, vol. 4, no. 3, pp. 1361–1370, Sep. 2013.
[20] J. Guerrero, M. Chandorkar, T. Lee, and P. Loh, “Advanced control archi-
(56) tectures for intelligent microgrids—Part I: Decentralized and hierarchi-
cal control,” IEEE Trans. Ind. Electron., vol. 60, no. 4, pp. 1254–1262,
where the last step we have used the fact that GΠ i [t] = L̃i [t] −
Apr. 2013.
Π Π [21] T. Zhu et al., “Sharing renewable energy in smart microgrids,” in Proc.
Bi,i [t] + j=i Bj,i [t]. In particular, let us consider the station- ACM/IEEE Int. Conf. Cyber-Phys. Syst., Apr. 2013, pp. 219–228.
ary and randomized policy Π from Theorem I, which achieves [22] L. Georgiadis, M. J. Neely, and L. Tassiulas, “Resource allocation and
∗
the cost gN . Using this policy in (56), we obtain cross-layer control in wireless networks,” Foundations and Trends in
Networking, vol. 1, no. 1, pp. 1–144, Apr. 2006, Now Publishers.
∗ [23] D. M. Larruskain, I. Zamora, A. J. Mazón, O. Abarrategui, and
ΔV [t] ≤ C + V gN . (57) J. Monasterio, “Transmission and distribution networks: AC versus DC,”
in Proc. 9th Spanish–Portuguese Congr. Elect. Eng., 2005, pp. 1–6.
Taking the expectation on both the sides and summing from [24] R. G. Gallager, Discrete Stochastic Processes. Boston, MA, USA:
t = 0, . . . , T − 1, normalizing by T and taking the limit, it can Kluwer, 1996.
[25] Y. Makarov, C. Loutan, J. Ma, and P. de Mello, “Operational impacts of
be shown that wind generation on California power systems,” IEEE Trans. Power Syst.,
T −1 vol. 24, no. 2, pp. 1039–1050, May 2009.
1  ∗ B̃ [26] S. Lakshminarayana, T. Q. S. Quek, and H. V. Poor, “Combining coopera-
lim sup E [fN [t]] ≤ gN + . (58) tion and storage for the integration of renewable energy in smart grids,” in
T →∞ T t=0 V
Proc. IEEE INFOCOM Workshop Commun. Control Smart Energy Syst.,
Toronto, ON, Canada, 2014, pp. 622–627.

R EFERENCES
[1] “Accommodating high levels of variable generation,” Atlanta, GA, USA,
Tech. Rep., 2009.
[2] E. K. Hart and M. Z. Jacobson, “A Monte Carlo approach to generator
portfolio planning and carbon emissions assessments of systems with
large penetrations of variable renewables,” Renew. Energy, vol. 36, no. 8,
pp. 2278–2286, Aug. 2011.
[3] B. Roberts, “Capturing grid power,” IEEE Power Energy Mag., vol. 7,
no. 4, pp. 32–41, Jul./Aug. 2009. Subhash Lakshminarayana (S’07–M’12) received
[4] “Assessment of demand response and advanced metering,” Washington, the M.S. degree in electrical and computer engineer-
DC, USA, Aug. 2006 (revised 2008), Staff Rep. ing from The Ohio State University, Columbus, OH,
[5] K. Chandy, S. Low, U. Topcu, and H. Xu, “A simple optimal power flow USA, in 2009 and the Ph.D. degree from the Alcatel
model with energy storage,” in Proc. IEEE Conf. Decision Control, 2010, Lucent Chair on Flexible Radio and the Department
pp. 1051–1057. of Telecommunications, SUPELEC, Gif-sur-Yvette,
[6] H.-I. Su and A. El Gamal, “Modeling and analysis of the role of fast- France, in 2012. He is currently with the Singapore
response energy storage in the smart grid,” in Proc. Allerton Conf. University of Technology and Design, Singapore.
Commun., Control, Comput., 2011, pp. 719–726. He has held a visiting research appointment at
[7] I. Koutsopoulos, V. Hatzi, and L. Tassiulas, “Optimal energy storage Princeton University from August to December
control policies for the smart power grid,” in Proc. IEEE Int. Conf. Smart 2013. He has also been a Student Researcher at
Grid Commun., 2011, pp. 475–480. the Indian Institute of Science, Bangalore, during 2007. He has served as
[8] R. Urgaonkar, B. Urgaonkar, M. J. Neely, and A. Sivasubramaniam, a Technical Program Committee Member for IEEE PIMRC in 2014, IEEE
“Optimal power cost management using stored energy in data centers,” VTC in 2014, and IEEE WCNC in 2014. His research interests broadly span
in Proc. ACM SIGMETRICS, 2011, pp. 221–232. wireless communication and signal processing with emphasis on small cell
[9] J. V. Paatero and P. D. Lund, “Effect of energy storage on variations in networks, cross-layer design wireless networks, MIMO systems, stochastic
wind power,” Wind Energy, vol. 8, no. 4, pp. 421–441, Oct.–Dec. 2005. network optimization, energy harvesting, and smart grid systems.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.
LAKSHMINARAYANA et al.: COOPERATION AND STORAGE TRADEOFFS IN POWER GRIDS 1397

Tony Q. S. Quek (S’98–M’08–SM’12) received the H. Vincent Poor (S’72–M’77–SM’82–F’87) re-

B.E. and M.E. degrees in electrical and electron- ceived the Ph.D. degree in electrical engineering
ics engineering from Tokyo Institute of Technology, and computer science from Princeton University,
Tokyo, Japan, respectively, and the Ph.D. degree in Princeton, NJ, USA, in 1977.
electrical engineering and computer science from the From 1977 until 1990, he was on the faculty of the
Massachusetts Institute of Technology, Cambridge, University of Illinois at Urbana-Champaign. Since
MA, USA. He is currently an Assistant Professor 1990, he has been on the faculty at Princeton, where
with the Information Systems Technology and De- he is the Michael Henry Strater University Professor
sign Pillar, Singapore University of Technology and of Electrical Engineering and Dean of the School of
Design, Singapore. He is also a Scientist with the Engineering and Applied Science. His research inter-
Institute for Infocomm Research, Singapore. His ests are in the areas of information theory, statistical
main research interests are the application of mathematical, optimization, and signal processing and stochastic analysis, and their applications in wireless
statistical theories to communication, networking, signal processing, and re- networks and related fields, including social networks and smart grid. Among
source allocation problems. Specific current research topics include cooperative his publications in these areas are the recent books Principles of Cognitive
networks, heterogeneous networks, green communications, smart grid, wireless Radio (Cambridge University Press, 2013) and Mechanisms and Games for
security, compressed sensing, big data processing, and cognitive radio. Dynamic Spectrum Allocation (Cambridge University Press, 2014).
Dr. Quek has been actively involved in organizing and chairing sessions Dr. Poor is a member of the National Academy of Engineering, the Na-
and has served as a Technical Program Committee Member (TPC) as well as tional Academy of Sciences, and Academia Europaea and is a Fellow of the
a Symposium Chair in a number of international conferences. He is serving American Academy of Arts and Sciences, the Royal Academy of Engineering
as the TPC Co-Chair for IEEE ICCS in 2014, the Wireless Networks and (U.K.), and the Royal Society of Edinburgh. He received the Marconi and
Security Track for IEEE VTC Fall in 2014, the PHY and Fundamentals Track Armstrong Awards from the IEEE Communications Society in 2007 and 2009,
for IEEE WCNC in 2015, and the Communication Theory Symposium for respectively. Recent recognition of his work includes the 2010 IET Ambrose
IEEE ICC in 2015. He is currently an Editor of the IEEE T RANSACTIONS ON Fleming Medal for Achievement in Communications, the 2011 IEEE Eric E.
C OMMUNICATIONS and the IEEE W IRELESS C OMMUNICATIONS L ETTERS Sumner Award, and honorary doctorates from Aalborg University, the Hong
and an Executive Editorial Committee Member of the IEEE T RANSACTIONS Kong University of Science and Technology, and the University of Edinburgh.
ON W IRELESS C OMMUNICATIONS . He was a Guest Editor of the IEEE
C OMMUNICATIONS M AGAZINE (Special Issue on Heterogeneous and Small
Cell Networks) in 2013 and the IEEE S IGNAL P ROCESSING M AGAZINE
(Special Issue on Signal Processing for the 5G Revolution) in 2014.
Dr. Quek was honored with the 2008 Philip Yeo Prize for Outstanding
Achievement in Research, the IEEE Globecom 2010 Best Paper Award, the
2011 JSPS Invited Fellow for Research in Japan, the CAS Fellowship for Young
International Scientists in 2011, the 2012 IEEE William R. Bennett Prize, and
the IEEE SPAWC 2013 Best Student Paper Award.

Authorized licensed use limited to: Md. Taohidur Rahman Shovon. Downloaded on February 17,2021 at 20:22:24 UTC from IEEE Xplore. Restrictions apply.