IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO.

2, APRIL 2018 729

A Methodology for Optimal Distributed Storage

Planning in Smart Distribution Grids
Mohammad Ghasemi Damavandi , Student Member, IEEE, José R. Martı́ , Life Fellow, IEEE,
and Vikram Krishnamurthy, Fellow, IEEE

Abstract—This paper proposes a methodology for optimal profit by storing the excess energy that would be spilled oth-
planning of distributed storage systems (DSS) in smart grids erwise. Furthermore, in virtue of DSS, the SGO will be able
with high penetration of renewable sources. The DSS can store energy at off-peak hours and sell it at peak hours when the
provide distribution systems with various benefits, including
arbitrage gain, reduction in the system losses, system resilience demand is higher. In markets with Time of Use (ToU) energy
enhancement, reduction in nondispatchable energy curtailment, pricing [1], that naturally translates into an arbitrage gain for
and peak load shaving. In particular, by alleviating the peak load, the SGO. The power losses in the system will also decrease by
the system upgrade time can be deferred to future years, resulting shaving the load at peak hours. If the system undergoes a fault,
in noticeable financial gains for the system operator. In this the SGO will also have the degree of freedom to take advantage
paper, the problem of DSS planning for optimizing the discounted
economic gain of the system operator is formulated and solved as a of these DSS during the system restoration time to minimize
mixed-integer convex program. Various economic gains are taken the energy-not-supplied [2]. In other words, the resilience of the
into account and the stochasticity of the loads and renewable system is improved by means of DSS. Moreover, the SGO can
generations is accounted for by evaluating the total expected gain. employ the DSS for peak load shaping [3] and thereby defer the
Numerical results for DSS planning on a 69-node, 11-kV smart upgrade of the system which will be inevitable due to constant
grid and using real data of smart meters and renewable energy
sources are presented and discussed. load growth in distribution systems [4]. If the budget required
for the system upgrade is noticeable, the deferral of the system
Index Terms—Distributed generators (DG), distributed storage upgrade may result in considerable financial gains due to inter-
systems (DSS), renewable energy, smart grids (SG).
est rates. Other advantages of the DSS include voltage control
[5], [6], ancillary services, and power smoothing for solar ar-
I. INTRODUCTION rays [7]. As a result, several authors have considered the use of
DSS for distribution systems with high penetration of wind and
MART distribution systems can benefit from high penetra-
S tion of renewable energy sources distributed in the grid.
Parts of the future distribution systems are even envisioned to
photovoltaic generation [8], [9].
The optimal DSS planning has been considered in several re-
cent works [10]–[19]. In [10], a framework is presented which
work as microgrids where during the islanded mode the de-
optimizes the capacity and power rating of DSS to ensure that
mand power should be completely supplied by local Distributed
the renewable energy generated by DGs never spill. Nonethe-
Generators (DGs). These DGs are often highly intermittent and
less, this work does not consider various other advantages that
can cause dramatic changes in the demand/generation of the
the DDS introduce to the system. In [11], the potential of DSS
nodes.
in the low-voltage distribution grid for deferring upgrades
When the amount of non-dispatchable renewable energy gen-
needed to increase the solar penetration level is investigated.
eration is forecast to violate a technical or contractual system
In [12], the optimal allocation of DSS in distribution systems
constraint, the Smart Grid Operator (SGO) will have to cur-
using a multi-objective optimization approach is considered.
tail the excess power or store it in Distributed Storage Systems
However, [11] and [12] do not consider the role of DSS in im-
(DSS). Therefore, using DSS the SGO will be able to gain some
proving the resilience of the system. In [13], a methodology for
DSS allocation in distribution systems is proposed which aims
Manuscript received October 4, 2016; revised February 2, 2017, May 14,
2017, and August 19, 2017; accepted October 2, 2017. Date of publication Oc- at cost-effective improvement of the system reliability. In [14],
tober 6, 2017; date of current version March 20, 2018. This work was supported the optimal planning of DSS using the point estimate method
in part by the National Science Foundation under Grant 1714180, and in part is considered. In [15] and [16], the optimal placement of DSS
by the Schmidt Sciences. Paper number TSTE-00757-2016. (Corresponding
author: Mohammad Ghasemi Damavandi.) considering the benefits due to system upgrade deferral is inves-
M. G. Damavandi and J. R. Martı́ are with the Department of Electrical and tigated using genetic algorithm. Genetic algorithm is also used
Computer Engineering, University of British Columbia, Vancouver, BC V6T in [17] for DSS planning in smart grids. In [18] and [19], DSS
1Z4, Canada (e-mail: mghasemi@[Link]; jrms@[Link]).
V. Krishnamurthy is with the School of Electrical and Computer Engineering, planning methodologies are proposed where the stochasticity of
and the Cornell Tech, Cornell University, New York, NY 10044 USA (e-mail: the loads and renewable energy sources is treated using scenario
vikramk@[Link]). based approaches.
Color versions of one or more of the figures in this paper are available online
at [Link] This paper proposes a new methodology for optimal DSS
Digital Object Identifier 10.1109/TSTE.2017.2759733 planning in smart distribution systems using linearized power

1949-3029 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See [Link] standards/publications/rights/[Link] for more information.

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
730 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 2, APRIL 2018

flow equations [20]. In particular, the aim of this paper is to For real distribution systems it is a known fact that the elements
present a novel formulation that is simultaneously: of f are very small (i.e., the voltage angles with respect to the
1) Comprehensive, in that all the economic gains due to in- slack node are very small.) The operational constraints of distri-
stallation of DSS are incorporated. bution systems also require that the elements of e remain small
2) Statistically rigorous, in that the stochasticity of the loads (e.g., within ±0.06 p.u.) at peak hours.2 At off-peak hours, the
and renewable energy sources is treated by explicitly eval- elements of e are even smaller which make the linearized power
uating the mathematical expectations. flow equations more accurate [23].
3) Tractable, in that the problem is formulated as a mixed- Let YN ×N = G + jB be the bus admittance matrix of the
integer convex1 program. system where G and B are the bus conductance and bus sus-
The methodology proposed in the present paper jointly opti- ceptance matrices, respectively. By removing the first row and
mizes the number of storage units to be installed, their location, column of Y, an (N − 1) × (N − 1) submatrix correspond-
their power rating and capacity, and their optimal active/reactive ing to the non-slack nodes is obtained which we denote by
strategy of charging and discharging. The financial gains due to Ỹ = G̃ + j B̃. Using these matrices and assuming that no shunt
price arbitrage, reduction in the system losses, reduction in the capacitor is installed in the system, the linearized power flow
renewable energy curtailed, system resilience enhancement, and equations can be written as [21]:
system upgrade deferral are formulated in the presented method-      
p pDG e
ology. The stochasticity of the loads and renewable energy − = A−1 (1)
q qDG f
sources is accounted for by evaluating the expected discounted
gains using the Law of Large Numbers (LLN). In addition, the where p = [p2 , p3 , . . . , pN ]T and q = [q2 , q3 , . . . , qN ]T are the
optimal planning problem is formulated as a mixed-integer con- vector of active and reactive power consumptions of the nodes,
vex program to be solved using branch and bound algorithms. respectively. Similarly, pDG = [pDG,2 , pDG,3 , . . . , pDG,N ]T and
The proposed methodology has been tested on a typical distri- qDG = [qDG,2 , qDG,3 , . . . , qDG,N ]T are the vector of active and
bution system using real data of smart meters and renewable reactive power generation by local DGs, respectively. Moreover,
energy sources and the results are presented and discussed. A is defined as:
The rest of the paper is organized as follows. Section II  −1
provides some preliminaries about the linearized power flow G̃ −B̃
A= . (2)
equations in distribution systems. Section III formulates vari- −B̃ −G̃
ous financial gains due to installation of DSS in distribution Rearranging the matrix form of the linearized power flow
systems. Section IV presents a mixed-integer convex formu- equations, (1) reads as:
lation of the problem of distributed storage planning in smart
distribution systems. Section V provides the numerical results of (p − jq) − (pDG − jqDG ) = (G̃g − B̃f ) + j(B̃g + G̃f )
the optimal DSS placement in a test system. Finally, Section VI = (G̃ + j B̃)(g + jf ), (3)
concludes the paper.
or, equivalently,
II. LINEARIZED POWER FLOW EQUATIONS IN s∗ − s∗DG = Ỹṽ, (4)
RECTANGULAR COORDINATES
where s = p + jq is the vector of complex power consumption
This section reviews the linearized power flow equations in of the nodes, sDG = pDG + jqDG is the vector of complex power
rectangular coordinates for application in distribution systems injections by local DGs, and (·)∗ denotes complex conjugation.
[20]–[22]. The section also provides some numerical results to Equation (4) is the complex form derived in [20] with appro-
demonstrate the effectiveness of these equations in distribution priate modifications to account for DGs. In [20], the authors
systems with DGs. Throughout the paper, constant active and have presented the mathematical and practical conditions under
reactive power consumption is assumed for the nodes. which the linearized power flow equations provide valid solu-
Consider a distribution system with N nodes and let the volt- tions. They have also used the complex form of (4) to theoreti-
age of the nth node be represented in rectangular coordinates cally bound the approximation error of the linearized equations.
as vn = 1 + ṽn = 1 + en + jfn . Without loss of generality, as- The interested reader is referred to to [20] for full mathematical
sume that the first node is the slack node with a given voltage of analysis of the linearized power flow equations.
1 p.u. Let ṽ = [ṽ2 , ṽ3 , . . . , ṽN ]T denote the vector of deviations Using linearized power flow equations, the active power loss
from the flat voltage profile at the remaining nodes of the sys- in the system can also be approximated as a quadratic function
tem. Also, let e = [e2 , e3 , . . . , eN ]T and f = [f2 , f3 , . . . , fN ]T of power injections. To see this, define the net demand vector
represent the real and imaginary parts of ṽ. These vectors can d as:
be approximated by linearized power flow equations in rectan-    
p pDG
gular coordinates provided that their elements are small [21]. d= − (5)
q qDG
1 With slight abuse of terminology, the term mixed-integer convex program-
ming is used in this paper to refer to the problems that are non-convex merely 2 Although the operational constraints are usually imposed on the magnitude
due to the integrality of some variables. Therefore, these problems become of the nodal voltages, the voltage magnitudes in distribution systems are very
convex if the integrality constraints are relaxed. well approximated by their real parts.

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
DAMAVANDI et al.: METHODOLOGY FOR OPTIMAL DISTRIBUTED STORAGE PLANNING IN SMART DISTRIBUTION GRIDS 731

Based on (5), the linearized equations can be written in the NLE averaged over all simulations turns out to be 3.05%. These
following compact form: numbers show that the linearized power flow equations can
  provide good approximations in terms of nodal voltages and
e
= Ad. (6) system power losses.
f
Using (6), it is shown in the Appendix A that the total active III. FORMULATION OF THE ECONOMIC GAINS OF THE
power loss in the system can be written in terms of the net DISTRIBUTED STORAGE SYSTEMS
demand vector as:
This section formulates the various economic gains of the
1
L(d) = dT Md, (7) DSS for smart distribution grids. Due to the stochasticity of the
2 loads, wind and solar generations, the expected economic gain
where of the SGO due to installation of storage units is considered.
  To ensure a longer lifetime for the DSS, similarly to [10], [25],
T G̃ 0
M = 2A A, (8) it is assumed that the storage units are charged and discharged
0 G̃
once a day. In particular, for each day, there is one charging
is a 2(N − 1) × 2(N − 1) symmetric real matrix that depends cycle corresponding to off-peak hours and one discharging cy-
merely on the bus conductance and bus susceptance matrices of cle corresponding to peak hours. This cycling strategy is also
the system. Using (7) and (8), it is also shown in Appendix A considered fixed during the planning horizon. Once the storage
that the loss function is convex in d. units are optimally planned for the system, their daily charging
To illustrate the effectiveness of the linearized power flow and discharging periods can be further optimized based on the
equations, we consider a modified version of the 69-node, 11 kV day-to-day profile of the loads and renewable energy sources.
distribution system presented in [24]. For this system, inclined This idea is explored in [1], where the contributions of [10]
solar cells are considered on half of the nodes selected randomly. and [25] are extended to include flexible charging/discharging
Moreover, two wind turbines are installed on the nodes 21 and periods for daily operation of the storage units.
50 with power ratings of 400 kW and 800 kW, respectively. The proposed methodology in this paper seeks an optimal
Real data of the loads, wind and solar generations are exploited DSS planning which optimizes the number of storage units,
for this illustration. Details about the test system and the real their location, their power rating and capacity, and their optimal
data used will be provided in Section V. All the numbers re- charging and discharging strategy jointly. In addition, the pro-
ported here are obtained by simulating the system for 535 days, posed methodology presents a framework in which the stochas-
24 hours each. For each simulation, the error of the linearized ticity of the real data of loads and renewable energy sources is
power flow equations is computed in comparison with the New- taken into account. Analysis of the real data of smart meters,
ton’s AC Power Flow (ACPF) in terms of Average Magnitude wind turbines, and solar cells demonstrates that the load and re-
Error (AME) in p.u., Average Angle Error (AAE) in degrees, newable energy sources have hourly and seasonal patterns. As a
Maximum Magnitude Error (MME) in p.u., Maximum Angle result, the optimal charging and discharging strategy of the DSS
Error (MAE) in degrees, and Normalized Loss Error (NLE). In will inevitably be different for different hours of the day and
particular, these metrics are defined as follows: different segments of the year. Therefore, the proposed method-
ology allows for different charging and discharging strategies
1   
N
 for different hours of the day and different segments of the year.
AME = |vn ,lin | − |vn ,ACPF |
N n =1 To that end, the planning horizon consists of Y years and each
year is divided into S segments. For hour h of segment s, let
1   
N
 pDSS
s,h be the vector of charging/discharging active powers by
AAE = ∠vn ,lin − ∠vn ,ACPF 
N n =1 the storage units installed in the system. Similarly, let qDSS
s,h be
  the vector of charging/discharging reactive powers by the stor-
 
MME = max |vn ,lin | − |vn ,ACPF | age units at hour h of segment s. If the result of the optimal
n
  planning does not assign a storage unit to a node, the corre-
  sponding active and reactive powers will equivalently remain
MAE = max ∠vn ,lin − ∠vn ,ACPF 
n zero at all time. Thus, the number of storage units is optimized.
   DSS T 
  To simplify the notation, define dDSS DSS T T
Llin − LACPF  s,h = [ps,h ] , [qs,h ]
NLE = , (9) for the storage units at hour h of segment s. In general, the
LACPF charging/discharging strategy of each storage unit can be op-
where the subscript lin indicates the solution of the linearized timized not only for each hour and each segment, but also for
power flow equations. For the test system under study and using each year. However, we restrict ourselves to the case where
the real data of loads, wind, and solar generation, the AME and the optimized strategy for each hour and segment remains the
AAE indexes averaged over all simulations are 3.84 × 10−4 same for all years. That is because even though the load and
p.u. and 1.04 × 10−3 degrees, respectively. Also, the MME and renewable generations grow over years, the capacity of the DSS
MAE indexes averaged over all simulations turn out to be 1.45 × remains fixed for the whole planning horizon. Nonetheless, the
10−3 p.u. and 9.31 × 10−3 degrees, respectively. Moreover, the proposed methodology can simply accommodate the case where

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
732 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 2, APRIL 2018

  1  
the DSS strategy is optimized for each year, in addition to op-
cys,h = E Lys,h (0) = E [dys,h ]T M [dys,h ] , (13)
timizing for each hour and each segment. We will denote the 2
operation strategy of the DSS in the sth segment of each year
 
DSS T T
is the expected hourly power loss in the sth segment of the yth
by π s = [dDSS
s,1 ] , [ds,2 ] , · · · , [ds,24 ]
T DSS T
. Also, the matrix year when no DSS is installed in the system.
of DSS operation strategies for the whole planning horizon will Leveraging the Law of Large Numbers (LLN), it is possible
be denoted by Π = [π 1 , π 2 , . . . , π S ]. to use the real data of the loads and renewable generations to
approximate d̄ys,h with the empirical mean which is an unbiased
A. The Arbitrage Gain estimator. Let d0s,h (l) be the lth random element of the net
To evaluate the arbitrage gain of the SGO due to installation demand vector d0s,h at the first year. Also, assume that at the
of DSS, a Time of Use (ToU) pricing scheme is considered in planning time a total number of Ks real data points of the load
y
this paper. Let ηs,h be the price of electricity per kWh at hour h and renewable generation is available for each hour of segment s.
of the sth segment of the yth year of the planning horizon. The With slight abuse of notation, let d0s,h (l, k) be the kth measured
average arbitrage gain in the yth year can be written as: data corresponding to the lth element of the net demand vector
at hour h and segment s. Then, based on LLN, d¯0s,h (l) can be
S 
365   y T DSS  y approximated as:
arb
Γy (Π) = ηs,h 1 ps,h − ηs,h 1T pDSS
s,h ,
S s=1
1 
h∈Hp h∈Ho Ks
(10) d¯0s,h (l) = E{d0s,h (l)} ≈ d0s,h (l, k). (14)
Ks
k =1
where Hp is the set of candidate discharging hours (peak hours)
and Ho = {1, 2, . . . , 24} \ Hp is the set of candidate charging In a similar way, one can approximate the expected hourly power
hours (off-peak hours). Observe from (10) that the average ar- loss in the system when no DSS is installed. Note, however,
bitrage gain of the SGO is linear in Π. that computation of cys,h is not necessary for the optimal DSS
placement problem as will be explained soon. Nonetheless, the
B. The Expected Reduction in Active Power Loss relevant details are provided here for completeness. Using real
data of loads and renewable generation, the expected hourly
In this section, the expected economic gain of the SGO due to power loss at the planning year can be approximated as:
reduction in the active power loss is formulated. The expected
daily power loss of the system in the sth segment of the yth year 1  0 T 
c0s,h = E [ds,h ] M [d0s,h ]
as a funciton of the DSS operation strategy can be written as: 2
2(N −1) 2(N −1)

24     1    
E Lys,h dDSS = E L dys,h + dDSS = ml,r E d0s,h (l)d0s,h (r)
s,h s,h 2 r =1
l=1
h=1 h∈Ho
    2(N −1) 2(N −1) K s
  
+ E L dys,h − dDSS , 1
s,h ≈ ml,r d0s,h (l, k)d0s,h (r, k) (15)
h∈Hp 2Ks r =1
l=1 k =1
(11)
where ml,r is the (l, r) element of M.
where the L(·) function is defined in (7) and E{·} denotes the Computation of the above-mentioned expectations for the
 T
expectation operator. Here, dys,h = [pys,h ]T , [qys,h ]T is the whole planning horizon requires the statistics of the vector
vector of stochastic net demands at hour h of the sth segment of of stochastic net demands in future years. This can be com-
the yth year. Expanding (11) using (7) and considering the ToU puted based on its current statistics as well as the antici-
pricing scheme, the expected daily cost due to the active power pated growth rate of the load and renewable generations. Let
loss in the sth segment of the yth year can be written as: the net demand data points at the planning time be decom-
posed as d0s,h (l, k) = d0,load 0,DG
s,h (l, k) − ds,h (l, k), corresponding

24
y 1  y  DSS T
24
  to the load and DGs. Assume a fixed annual growth rate of
E ηs,h Lys,h dDSS
s,h = ηs,h ds,h M dDSS
s,h
2 γload and γDG for the the load and renewable generation, re-
h=1 h=1
 y spectively. Then d̄ys,h and cys,h for y ≥ 1 can be computed
+ ηs,h [d̄ys,h ]T M dDSS
s,h by (14) and (15), respectively, with d0s,h (l, k) replaced by
h∈Ho dys,h (l, k) = [γload ]y d0,load y 0,DG
s,h (l, k) − [γDG ] ds,h (l, k).

− y
ηs,h [d̄ys,h ]T M dDSS Finally, the expected economic gain of SGO due to the re-
s,h
h∈Hp
duction in the active power loss in the yth year of the planning
horizon is given by:

24
y
+ ηs,h cys,h , (12) 365   y y
S 24 
24
y
h=1 Γloss
y (Π) = ηs,h cs,h − E ηs,h Lys,h dDSS
s,h .
S s=1
  h=1 h=1
where d̄ys,h = E dys,h , and (16)

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
DAMAVANDI et al.: METHODOLOGY FOR OPTIMAL DISTRIBUTED STORAGE PLANNING IN SMART DISTRIBUTION GRIDS 733

Observe from (16) and (12) that the terms corresponding to cys,h   
+2 [d̄ys,h ]T ATe Ae dDSS
s,h
cancel out in Γloss
y . Therefore, Γy
loss
contains only quadratic and h∈Ho
linear terms of Π. This is indeed expected as Γloss y (0) should   
be equal to zero. In other words, when no DSS is installed −2 [d̄ys,h ]T ATe Ae dDSS
s,h . (17)
in the system, there will be no economic gain. Also note that h∈Hp
−1
the Hessian of Γlossy (Π) with respect to Π is equal to S × In (17), d̄ys,h can be approximated using LLN as in (14). Ob-
y y y y
diag{η1,1 M, η2,1 M, . . . , ηS,1 M, . . . , ηS,24 M}, which is neg-
y serve that the Hessian of Γvol (Π) with respect to Π is equal
ative semi-definite, because M  0 and ηs,h > 0, ∀h, ∀s, ∀y.
loss
to −2
SY × diag{Ae Ae , Ae Ae , . . . , Ae Ae } which is negative
T T T
Therefore, Γy (Π) is concave in Π. semi-definite, because Ae Ae  0 by structure. Therefore,
T

Γvol (Π) is concave in the operation strategy of the DSS.

C. The Reduction in Expected Price of Renewable Energy The limits on the amount of reverse power flow from distribu-
Curtailed tion system to the sub-transmission system may also result in the
Due to the intermittent nature of renewable energy sources, curtailment of renewable energy sources [10]. Thus, the reverse
there may be times when the technical system constraints are power flow limits in the system may affect the planning of DSS
about to be violated due to excess distributed generation. Un- [26]. If the reverse power limit at the substation is high enough,
der such circumstances, the SGO may have to curtail the non- however, there is no need for DSS in terms of prevention of the
dispatchable renewable generations. However, if the system is non-dispatchable energy spillage [26]. For instance, the Ontario
equipped with DSS, they can be optimally operated to reduce Power Authority allows for a 60% reverse power flow through
the curtailment of renewable energy sources. This section ad- the substation transformer [10]. This amount of reverse power
dresses the economic gain of DSS installation in terms of re- limit will accommodate many scenarios of high generation of
ducing the non-dispatchable energy curtailment. The average non-dispatchable DGs. Nonetheless, the installation of DSS can
economic gain due to lower energy curtailment in year y will be still be financially justifiable if other economic gains are no-
refered to as Γcurt ticeable. Depending on whether or not reverse active/reactive
y (Π).
It is difficult to explicitly formulate the expected economic power flow is allowed in the system, the smart grid may also
value of the spilled energy due to violation of system constraints. have different operating states [27], [28]. Similarly to [10], [27],
Instead, an indirect approach is taken in this paper. Here, we and the references therein, the operating state considered in this
regularize the objective function of the DSS placement with paper only allows for the reverse active power flow to the sub-
a virtual term associated with the improvement in the voltage transmission system. We further assume that the reverse active
profile of the system. With this virtual benefit included in the power flow to the sub-transmission system is allowed without
objective function, the DSS placement routine will try to avoid any rejection [29]. Note that the amount of excessive power in
the violation of system constraints. Once the storage units are the system is random and will be known only at the operation
optimally placed in the system and their operation strategy is stage. Nevertheless, at the planning stage which is the focus
optimized, the actual economic value of the spilled energy will of this paper, one can require that the reverse reactive power
be approximated through Monte Carlo simulations. constraint is respected on average. That is,
To regularize the objective function with regards to the volt- 1T E {qs,h } − qDSS ≥ 0, ∀s, h, (18)
s,h
age profile of the system, first note that in distribution systems
the imaginary part of the voltages are very small (the voltage where E {qs,h } is the average net reactive demand of the system.
angles are very small). Therefore, one can approximate the mag- Once the optimal DSS planning strategy is obtained for the
nitude of the nodal voltages by their real parts. As a result, the system, the expected price of renewable energy curtailed will be
total expected deviation in the real parts of the nodal voltages, computed for the system with and without DSS. Without loss
with respect to the flat voltage profile, can serve as an index of generality, assume that the wind turbines are candidates of
for evaluation of the voltage profile. Now consider the decom- energy curtailment as they are more likely to be utility-owned.3
position A = [ATe ATf ]T for the matrix in (2), where Ae and Using linearized power flow equations, the optimal curtailment
Af are (N − 1) × 2(N − 1) matrices corresponding to the real strategy seeking the minimum power curtailment to satisfy the
and imaginary parts, respectively. Let eys,h (pDSS
s,h ) be the vector
system constraints, can be formulated as:
of deviations in the real parts of the nodal voltages at hour h,
min 1T dcurt (19)
segment s, and year y. Similarly to the way that the expected
reduction in the power loss was computed in Section III-B, the subject to.
average improvement in the voltage profile during the planning
horizon due to installation of DSS can be formulated as: 0 ≤ pcurt ≤ pwind (20)

365  
24
 y   2 0 ≤ qcurt ≤ qDSS (21)
Γvol (Π) = E e (0)2 − ey (dDSS
s,h )

s,h s,h
SY s,y
h=1
1 T
qs,h − q DSS
≥ 0, (22)

−1    DSS T T  
24
= ds,h Ae Ae dDSS
s,h 3 The formulation presented in the paper can simply accommodate the cur-
SY s,y
h=1 tailment of solar generation as well.

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
734 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 2, APRIL 2018

vm in ≤ Ae d + dcurt ≤ vm ax (23) during one year. Let μ̄y be the expected interruption cost per
1 kWh of energy in year y of the planning horizon. In practice,
|Il (dcurt )|2 ≤ |Ilm ax |2 , l = 1, 2, . . . , L, (24) the interruption cost per unit kWh is a function of the interrup-
T T tion duration as well [30]. However, one can use the Customer
where dcurt = [(pcurt ) , (qcurt ) ]T is the vector of curtailed ac-
Average Interruption Duration Index (CAIDI) to find an approx-
tive and reactive powers and pwind is the vector of power gener-
imate value for the expected interruption cost. Let ξs,h be the
ations by wind turbines. Constraints (21)–(22) suggest that the
fraction of the interruptions that occur at hour h during segment
reactive power injection of the DSS may need to be curtailed to
s of each year, see [30]. The average economic value of the
respect the reverse reactive power constraint of the substation.
DSS in terms of enhancing the resilience of the system can be
However, this constraint is not likely to be active as the reverse
approximated as:
reactive power constraint of the substation was taken into con- 
sideration in (18) during the planning stage. In (23), vm in and y (Π) = μ̄y × SAIFI ×
Γres ξs,h Es,h . (25)
vm ax are the vector of minimum and maximum allowable volt- h s
age limits, respectively. Also, d is the vector of forecast total
y (Π) is linear in Es,h and, hence, in Π.
Note that Γres
demand of the nodes for the next hour. When DSS are installed
in the system, the vector d also includes the optimized injection
E. The Economic Gain of the System Upgrade Deferral
and consumption of the storage units. In (24), |Il | is the magni-
tude of the current flowing in branch l, Ilmax is the ampacity of Due to the constant load growth in distribution systems, the
branch l, and L is the total number of branches in the system. system will require an upgrade in the feeder ampacities and
It is shown in Appendix B that the quadratic term (24) is con- substation capacity at some point in the future. However, if DSS
vex in dcurt . Therefore, (19) is a convex optimization problem. are installed in the system, they can be employed to shave the
Note, however, that due to the inherent approximation of the load at peak hours and thereby defer the required upgrade of the
linearized power flow equations, the solution of (19) is a rough system [4], [15].
estimate of the curtailed powers. To find an accurate solution, In order to defer the upgrade of the system, the objective
one can solve (19) for coarse tuning the curtailed powers and function of the DSS placement can include a term to shave the
then do grid search along with ACPF to fine tune the results. peak load in the system. To that end, form the expected total
active power drawn from the substation during peak hours as:
D. The Improvement in the System Resilience 
p eak
Ps,h (Π) = E 1T (pYs,h − pDSS s,h ) , ∀s, h ∈ Hp , (26)
In this section the economic value of the DSS in terms of
improving the system resilience is formulated.
where 1 and 0 are of size (N − 1) × 1. The optimization routine
Let Ho (h) ∈ Ho be the set of off-peak hours from the be-
ginning of the off-peak period up to the hour h. Similarly, let will try to lower the maximum of Phpeak (Π) over peak hours as
Hp (h) ∈ Hp be the set of peak hours from the beginning of will be explained in the next section. Note that the expected total
the peak period up to the hour h. Also, let 0 < βch < 1 and active load in (26) has been computed for the last year of the
0 < βdis < 1 be the charging and discharging efficiencies of the planning horizon because the last year is supposed to have the
DSS technology used, respectively. Hence, βr t = βch βdis < 1 maximum demand.
is the round-trip efficiency of the storage units. Denote the total Once the DSS is optimally planned in the system, the eco-
energy that the DSS can supply to the grid in segment s and at nomic gain of deferring the system upgrade due to installation
the end of hour h by Es,h (Π). Since Es,h is equal to the total of DSS will be computed. As per common practice of distribu-
energy stored in the storage units times βdis , one can write: tion system planning, the system constraints have to be satisfied
⎧  for all combinations of the load and distributed generation. To
T DSS
⎨βr t h ∈Ho (h) 1 ps,h , h ∈ Ho evaluate the latest possible time for the system upgrade, the
Es,h = histograms of the available real data are examined to find the
⎩  
s,h −
βr t h ∈Ho 1T pDSS h ∈Hp (h) 1 ps,h , h ∈ Hp
T DSS
maximum values of the load and minimum values of the dis-
tributed generation based a predetermined confidence interval.
Upon imposing appropriate charging/discharging constraints,
Then, for each year of the planning horizon a power flow is
as will be done in Section IV, Es,h is guaranteed to be non-
run according to the the extreme loading and generation to find
negative. In the event that the primary supply of the system is
the required ampacity of the feeders and substations. Then, the
interrupted, the distribution system may be operated as an is-
upgrade requirements of the feeders and substation of the sys-
landed microgrid. In such a case, the local DGs as well as the
tem are calculated for the system with and without DSS [15].
energy stored in the DSS can supply the loads of the system for
Finally, the average yearly difference in the system upgrade cost
a limited time. The non-critical loads of the system may need
for the yth year, denoted by Γupy (Π), is computed.
to be shed depending on the availability of the energy by local
sources. Therefore, Es,h can be viewed as the additional amount
of load that can be preserved each time the supply from the IV. THE OPTIMAL DISTRIBUTED STORAGE
PLANNING PROBLEM
primary source has failed. In distribution systems, the System
Average Interruption Frequency Index (SAIFI) measures the This section aims at formulating the optimal DSS planning
average number of power interruptions a customer experiences problem in smart distribution grids. To that end, the regularized

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
DAMAVANDI et al.: METHODOLOGY FOR OPTIMAL DISTRIBUTED STORAGE PLANNING IN SMART DISTRIBUTION GRIDS 735

expected discounted gain of the SGO due to installation of DSS positive and negative signs. Therefore, the following con-
is defined as: straint should be added to the optimal planning problem:
Γ (Π) = ω1 Γvol (Π) − ω2 Δ
s,h ≥ 0,
pDSS ∀s, h. (31)

Y  
+ λy Γarb loss res
y (Π) + Γy (Π) + Γy (Π) , (27) 2) The charging/discharging capacity of the storage units:
y =1 Discharging the storage units completely affects their
lifetime adversely [31]. In particular, the lifetime of the
where ω1 , ω2 > 0 are two regularization factors and Δ is an
storage units is dependent on their maximum Depth Of
auxiliary variable used for shaving the peak load as will be
Discharge (DOD) [17], [31]. Let γDOD be the maximum
explained shortly (see (34)). Also, 0 < λ < 1 is the discount
DOD of the storage units. The parameter γDOD along with
factor. If the interest rate is int and a uniform inflation rate of inf
βch and βdis determines how much energy is needed to
is assumed for the energy price and the interruption cost, then
charge a storage unit of a given capacity during off-peak
λ = (1 + inf)/(1 + int). The regularization parameters ω1 and
hours. The first time that a storage unit of size bn is con-
ω2 will be tuned based on simulations to yield the best trade-off
nected to node n, a total energy of (1 − γDOD )bn kWh is
between the investment cost and the economic gain of the DSS.
absorbed which remains in the unit and never discharges.
The tuning of the regularization parameters can be done using
Next, an amount of up to γDOD bn kWh is absorbed from
a grid search. Note that based on the formulations presented in
the grid during the next charging cycle. Therefore, for
Section III, Γ (Π) is concave in Π.
charging cycles (off-peak hours) one can write:
In order to formulate the investment cost of the DSS, let
b = [b2 , b3 , . . . , bN ]T be the capacity of the installed storage 
βch pDSS
s,h = γDOD b, ∀s. (32)
units in kWh. Also, let κ1 $/kWh be the unit cost of the
h∈Ho
DSS technology used and κ2 $/kVA be the associated power
electronics and O&M costs. Then, the initial investment cost Similarly, at peak hours when the storage units get dis-
of DSS in the system is equal to κ1 1T b + κ2 1T r, where charged, the total energy absorbed by the units during
r = [r2 , r3 , . . . , rn ]T is the vector of rated powers of the storage off-peak hours times the discharging efficiency will be
units. Now, assume that the planning horizon is greater than the equal to the total energy injected to the grid. That is,
lifetime of the DSS with a factor of K, i.e., Y = K × YDSS , where
 
YDSS is the DSS lifetime in years. Then, the storage units need pDSS pDSS ∀s.
s,h = βdis βch s,h , (33)
to be replaced in the system K times. Therefore, the discounted h∈Hp h∈Ho
investment cost of storage units will be given by:
Since the values of the charging and discharging pow-
Ω(b, r) = κ1 1T b + κ2 1T r, (28)
ers are assumed to be non-negative in our formulation,
K – 1 kYDSS
where κ1 = κ1 × k=0 λ . Observe from (28) that Ω(b, r) the constraints (32) and (33) require that the charg-
is linear in b and r. ing/discharging power of each node be zero if no storage
The optimal DSS planning aiming at maximizing the eco- unit is installed on the node.
nomic gain minus the investment cost can now be cast as the 3) The maximum peak load: The parameter Δ in (27) serves
following optimization problem: to represent the maximum peak load over peak hours.
Hence, the following set of linear constraints has to be
max Γ (Π) − Ω(b, r) (29) added to the optimal DSS placement problem:
b,r,Π

subject to a series of constraints as follows. peak

Ps,h (Π) ≤ Δ, ∀s, h ∈ Hp (34)
1) The power capability of the storage units: The active and
reactive power injection of the DSS are jointly constrained peak
where Ps,h (Π) is defined in (26).
by the power capability curve of the DSS [12], [25]. This
Once the optimal DSS planning problem (29) is solved,
can be modeled by the following quadratic, convex con-
the total discounted economic gain of the SGO will be
straint:
computed as:
2 2
pDSS
s,h + qDSS
s,h ≤ r2 , ∀s, h. (30)

Y 
Note that the notation is slightly abused in (30) as (·)2 is Γ(Π) = λy Γarb loss res
y (Π) + Γy (Π) + Γy (Π)
used to denote element-wise exponentiation of the vec- y =1
tors. In practice, one could use a set of arbitrarily accurate 
linearized constraints equivalent to (30) as done in [12]. + Γcurt up
y (Π) + Γy (Π) . (35)
Recall that to ensure a longer lifetime for the storage
up
units, they are restricted to charge during off-peak hours Note that the explicit formulation of Γcurt
y (Π) and Γy (Π) in
and discharge during peak hours [10], [25]. Throughout terms of Π is intractable and, hence, the regularized expected
Section III, the charging and discharging strategies of the discounted gain Γ (Π) is considered for optimal DSS planning
active power of the DSS are differentiated by appropriate instead of Γ(Π).

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
736 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 2, APRIL 2018

V. NUMERICAL RESULTS Also, μ is considered 15 $/kWh based on the survey reported in

This section presents the numerical results of the optimal DSS [30] and SAIFI is 1.5.
planning methodology on a modified 69-node, 11 kV distribu- The system upgrade includes upgrading the distribution sys-
tion system [24]. tem feeders and substation. In addition, similarly to [4], it is
The simulations presented in the paper were done using MAT- assumed that a 25 km transmission line between the distribution
POWER [32] on a computer with 8 GB memory and a Core i5 substation and the HV/MV primary substation needs to be up-
CPU of frequency 2.7 GHz. The optimal DSS planning prob- graded too. The upgrade costs of lines, feeders and substations
lem was solved using the branch and bound algorithm, which are obtained from [40].
was implemented using the CVX package [33] bundled with
Mosek [34]. B. Results
The optimal DSS planning methodology on the system un-
A. The Setting of the Simulations
der study results in the installation of one storage unit of size
The modified test system is radial (all tie line switches are 200 kWh with an inverter power rating of 100 kVA. The opti-
open) with four feeders where node 1 is considered as the slack mal location of this unit is node 29 which is on feeder 2. Note
node. The total nominal demand of the system in the first year that the optimization routine was not enforced to place only
of the planning period is 4468 + j3059 kVA. The power factor one storage unit in the system. In other words, the optimal DSS
of each node is derived from the nominal active and reactive de- planning could have conceivably resulted in the placement of
mand of the node based on the information reported in [24]. For more than one unit, possibly on different nodes of the system.
this system, inclined solar cells are considered on 34 randomly The placement of only one storage unit, its location, its power
selected nodes of the grid. The nodes equipped with solar cells rating and capacity, and its charging and discharging strategy
are 3, 4, 6, 8, 10, 13, 17, 19, 20, 22, 23, 30, 31, 32, 33, 35, 36, are rather the optimal setting to yield the maximum expected
37, 40, 44, 45, 46, 49, 51, 53, 55, 57, 58, 61, 62, 63, 65, 67, discounted reward. Fig. 1 illustrates the test system under study
68. A total rated power of 1000 kW is considered for the solar with the optimal location of the storage unit. The locations of
cells in this test system. This total rated power is allocated to the two wind turbines are also shown in Fig. 1. In this figure,
the solar cell-equipped nodes proportionally to their secondary the nodes equipped with solar cells are indicated with shadow.
transformer power rating. In addition, two wind turbines are Fig. 2 depicts the optimal State Of Charge (SOC) of the
installed on the nodes 21 and 50 with power ratings of 400 kW installed storage unit for three different segments of the year. It
and 800 kW, respectively. The reverse power flow is allowed for is apparent from Fig. 2 that the optimal SOC of the storage unit
the active power only without any rejection. is different for different segments of the year. This is because
In our simulations, the Advanced Metering Infrastructure the statistics of the loads and renewable energy sources change
(AMI) data released by the Commission for Energy Regulation over different hours of the day and different segments of the year.
(CER) [35] is utilized to model the loads. The dataset provided Accordingly, the optimal charging and discharging strategy of
by CER is from the Electricity Customer Behaviour Trail study the storage unit is different for different hours and segments. To
and has been collected from 5000 smart meters in Ireland from see how the statistics of the load and renewable energy sources
July 14, 2009 to December 31, 2010. This dataset was received impact the optimal charging and discharging strategy of the
by authors from Irish Social Science Data Archive (ISSDA) storage unit, the expected total net demand of the first year of
[36]. The reactive power demands are modeled by assuming a the system is also depicted in Fig. 3. Here, the total expected
constant power factor for the nodes. A growth of 5% [17] in ned demand refers to the arithmetic average of the total load
the demands of the nodes is considered over the planning hori- minus total renewable generation in the system among all the
zon. The time series of wind generation and solar generation are days belonging to the segment. It can be seen from Figs. 2
obtained from [37] and [38], respectively. and 3 that the optimized strategy makes the storage unit charge
The DSS technology used is lead-acid with specifications when the expected total net demand is low (off-peak hours) and
provided in [39]. The DSS are assumed to come in 100 kVA– discharge when the expected total net demand is high (peak
100 kWh units with charging and discharging efficiency of hours.) Inspection of Figs. 2 and 3 reveals that the sharp peak of
βch = βdis = 0.85. In order to guarantee that the lifetime of the Segment 5 causes the storage unit to discharge quickly over
the storage units is not shortened, the DOD of the storage units just two hours, while a smoother peak of the Segment 11 results
is assumed to be γDOD = 0.75 [31]. The minimum and maxi- in a smoother discharge over 6 hours. In contrast to Segment
mum voltage limits for all nodes are considered to be 0.95 p.u. 11 which has a smooth off-peak demand, in Segments 5 and 8,
and 1.05 p.u., respectively. The DSS costs are 305 $/kWh for the demand at h ≈ 1 is very higher than the minimum demand
the storage units and replacements, 175 $/kVA for the inverter, which happens at h ≈ 5. As a result, the storage unit does not
and 15 $/kVA for annual maintenance. The planning horizon is charge at h = 1 for Segments 5 and 8, and starts charging only
assumed to be twice as the lifetime of the DSS and each year is at h = 2. During mid hours when the demand is not very high or
divided into 12 segments. very low, the storage unit neither absorbs nor injects any power
The price of energy is considered to be 8.29 ¢/kWh for off- in any of the segments.
peak hours and 12.43 ¢/kWh for peak hours. The interest rate The total investment cost for this storage unit will be
and the inflation rates are assumed 5% and 1%, respectively. $141,310, which includes the following:

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
DAMAVANDI et al.: METHODOLOGY FOR OPTIMAL DISTRIBUTED STORAGE PLANNING IN SMART DISTRIBUTION GRIDS 737

Fig. 1. The test system under study with wind turbines and solar cells and the storage unit optimally located on node 29.

Fig. 2. The optimal SOC of the installed storage unit for three different
segments of the year. Fig. 3. The expected total net demand of the system for three different seg-
ments of the first year.
1) Initial investment cost: $78,500
2) Replacement cost: $43,006 1) Expected discounted arbitrage gain: $6,135
3) Total discounted maintenance cost: $19,804 2) Expected discounted gain due to reduction in the system
With the optimal DSS placement in the system, the SGO is losses: $8,181
expected to obtain the following gains during the entire planning 3) Expected discounted economic value of system resilience
horizon: enhancement: $24,854

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
738 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 2, APRIL 2018

4) Expected discounted gain due to reduction in the renew- TABLE I

IMPACT OF THE NUMBER OF CANDIDATE NODES AND THE DUALITY GAP ON
able energy curtailment: $235 THE COMPUTATIONAL TIME OF THE OPTIMIZATION
5) Total economic gain due to deferring the upgrade of the
system: $176,038
Number of CPU Time (s) CPU Time (s)
The upgrade time of the system is determined by running the Candidate Nodes (Duality gap = 0) (Duality gap < 0.5%)
Newton’s AC power flow for the future years and under the
9 5.3 1.9
following two cases: the maximum load along with minimum 17 17.6 3.9
renewable generation and the minimum load along with maxi- 34 101.0 10.2
mum renewable generation. The minimum and maximum values 69 1571.6 28.7
of the loads and renewable generation are obtained by examin-
ing the available real data set. If some system equipment need to
be upgraded, the required increase in their capacity is evaluated. time of the solution obtained for a 0.5% duality gap is sig-
Based on our simulations, the feeders of the test system will nificantly smaller than that of the the zero duality gap. More
need to be upgraded in a year. Feeder 2, in particular, will be importantly, the computational time of the solution obtained for
upgraded to accommodate an additional 1.5 MW. According to a 0.5% duality gap scales much better in the number of can-
the upgrade costs reported in [40], this requires a total invest- didate nodes compared with the solution obtained for a zero
ment of $336,000. Based on our simulations, the installation duality gap.
of DSS can postpone this task for another two years. There- We finish this section by noting that increasing the num-
fore, considering the discount factor λ = 1.01/1.05 = 0.9619, ber of segments per year can potentially lead to a better DSS
a financial gain of (λ − λ3 ) × 336, 000 = $24, 159 is achiev- planning with a better cost-benefit trade-off. On the other hand,
able. In addition, the existing capacity of the substation of increasing the number of segments will increase the size and the
the system is 4.5 MW. With a 5% annual increase in the de- computational costs of the optimization problem. The authors
mands, the system should accommodate 10.83 MW at the end will explore this aspect of the problem in their future research.
of the planning horizon. Therefore, an additional capacity of
6.33 MW will be needed. According to the upgrade costs re- VI. CONCLUSION
ported in [40], this would require an investment of $516,500.
This paper presented a methodology for optimal planning of
Based on our simulations, the optimal planning of DSS can
DSS in smart distribution grids. The problem of optimizing the
postpone the upgrade cost of the substation by one year, re-
expected economic gain of the system operator was formulated
sulting in a financial gain of (1 − λ) × 516, 500 = $19, 679.
as a mixed-integer convex program. In particular, the optimal
Finally, the transmission line between the distribution substa-
planning problem incorporated the arbitrage gain, the reduction
tion and the HV/MV primary substation needs to be upgraded in
in the active power loss, the reduction in the non-dispatchable
two years. According to [40], this would require an investment
energy curtailment, the improvement in the system resilience,
cost of $3,750,000. With the help of DSS, this investment can
and the financial gain due to deferring the system upgrade
be postponed from two years to three years, which would re-
to future years. The stochasticity of the loads and renewable
sult in a financial gain of (λ2 − λ3 ) × 3, 750, 000 = $132, 200.
energy sources was treated in a statistically rigorous way by
Therefore, the total saving in the upgrade costs would be
evaluating the mathematical expectations using the Law of
19, 679 + 24, 159 + 132, 200 = $176, 038.
Large Numbers. Numerical results using real data of smart
Based on the above-mentioned numbers, the total discounted
meters and renewable sources on a typical distribution system
gain that the SGO obtains due to installation of DSS will be
was presented which demonstrated the effectiveness of using
$215,443. Given that the total investment and maintenance cost
DSS in future smart grids.
of DSS installation is $141,310, we conclude that energy storage
in this system results in a total discounted saving of $74,133.
APPENDIX A
C. Computational Cost and Scalability In this Appendix we derive the quadratic form of (7) for the
active power loss of the distribution systems using the results
In terms of the computational cost of the optimization, the presented in [41] and show that the loss function is convex in
bottleneck is the number of integer variables. Since the storage the demand vector.
units are assumed to have integer capacity and integer power For ease of notation define the augmented vectors
rating, the number of integer variables in the proposed method- ê = [0 eT ]T , f̂ = [0 f T ]T , and p̂ = [pslack , p2 − pDG,2 , p3 −
ology is 2N . However, the computational cost can be reduced if,
pDG,3 , . . . , pN − pDG,N ]. Also, let î be the vector of current in-
similarly to some existing literature [14]–[16], only some candi-
jections to the system. If the voltage of the slack node is 1 p.u.,
date nodes are considered for DSS installation rather than all the
the active power loss in the system can be written as [41]:
nodes. Moreover, the computational time can be significantly re-
duced if, instead of a zero duality gap, a very small duality gap L = 1T p̂
(e.g., 0.5%) is considered satisfactory. Table I demonstrates this  
fact by showing the impact of the number of candidate nodes = (1 + ê + j f̂ )T î∗
and the desired duality gap on the computational time of the  
optimization. It can be seen from Table I that the computational = (1 + ê + j f̂ )T [Y(1 + ê + j f̂ )]∗

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
DAMAVANDI et al.: METHODOLOGY FOR OPTIMAL DISTRIBUTED STORAGE PLANNING IN SMART DISTRIBUTION GRIDS 739

 
= (ê + j f̂ )T Y∗ (ê − j f̂ ) (36) = Ỹbr (e + jf )
 
= (ê + j f̂ )T (G − jB)(ê − j f̂ ) = (G̃br e − B̃br f ) + j(G̃br f + B̃br e)
 
= êT Gê + f̂ T Gf̂ = (G̃br Ae − B̃br Af ) + j(G̃br Af + B̃br Ae ) (d + dcurt )
= Tre (d + dcurt ) + jTim (d + dcurt ), (43)
= eT G̃e + f T G̃f , (37)
where Tre = G̃br Ae − B̃br Af , Tim = G̃br Af + B̃br Ae . Also,
∗ T ∗
where (36) follows from the fact that Y 1 = 0 and 1 Y = 0. Ỹbr = G̃br + j B̃br is the branch admittance matrix with the
Equation (37) can also be written in a matrix form as: first column removed. Moreover, (42) stems from the fact that
 T    Ybr 1 = 0. Let tTre,l and tTim,l be the lth row of Tre and Tim ,
e G̃ 0 e respectively. Then, the magnitude of the currents in the branches
L(e; f ) = . (38)
f 0 G̃ f of the system as a function of the curtailment vector dcurt will
be given by:
Equation (7) then follows by substituting the the linearized 2 2
power flow equations (6) into (38). |Il (dcurt )|2 = tTre,l (d + dcurt ) + tTim,l (d + dcurt )
To establish the convexity of the loss function L(d) in the = (d + dcurt )T tre,l tTre,l (d + dcurt )
demand vector d, we need to show that ∇2 L(d) = M  0. To
that end, first note that the system admittance matrix is positive + (d + dcurt )T tim,l tTim,l (d + dcurt )
semi-definite, that is G  0. In [41], this is shown using the
fact that the active power loss in the system is non-negative for = (d + dcurt )T Tl (d + dcurt ), (44)
all values of the nodal voltages. Next, note that for a positive where Tl = tre,l tTre,l + tim,l tTim,l . Now observe that the matrices
semi-definite matrix, all the principal submatrices have to be
tre,l tTre,l and tim,l tTim,l are positive semi-definite by structure.
positive semi-definite. Therefore, it follows from the positive
Therefore, Tl is positive semi-definite, too.
semi-definiteness of G that G̃ is positive semi-definite, too. As
a result, for any x and y belonging to RN −1 , we have:
REFERENCES
 T   
x G̃ 0 x [1] A. Gabash and P. Li, “Flexible optimal operation of battery storage systems
= xT G̃x + y T G̃y ≥ 0. (39) for energy supply networks,” IEEE Trans. Power Syst., vol. 28, no. 3,
y 0 G̃ y pp. 2788–2797, Aug. 2013.
[2] C. P. Nguyen and A. J. Flueck, “Agent based restoration with distributed
Because G̃ is also symmetric and real, the following Cholesky energy storage support in smart grids,” IEEE Trans. Smart Grid, vol. 3,
no. 2, pp. 1029–1038, Jun. 2012.
decomposition exists: [3] T. Jiang, Y. Cao, L. Yu, and Z. Wang, “Load shaping strategy based on
  energy storage and dynamic pricing in smart grid,” IEEE Trans. Smart
G̃ 0 Grid, vol. 5, no. 6, pp. 2868–2876, Nov. 2014.
= UUT , (40) [4] A. Oudalov, D. Chartouni, C. Ohler, and G. Linhofer, “Value analysis
0 G̃
of battery energy storage applications in power systems,” in Proc. Power
Syst. Conf. Expo., 2006, pp. 2206–2211.
where U is a real, lower triangular matrix with positive diagonal [5] K. Christakou, D.-C. Tomozei, M. Bahramipanah, J.-Y. Le Boudec, and
entries. Therefore, for any z ∈ RN −1 , we have: M. Paolone, “Primary voltage control in active distribution networks via
broadcast signals: The case of distributed storage,” IEEE Trans. Smart
zT Mz = 2zT AT UUT Az Grid, vol. 5, no. 5, pp. 2314–2325, Sep. 2014.
[6] P. Wang, D. H. Liang, J. Yi, P. F. Lyons, P. J. Davison, and P. C. Tay-
lor, “Integrating electrical energy storage into coordinated voltage con-
= 2(UT Az)T (UT Az) trol schemes for distribution networks,” IEEE Trans. Smart Grid, vol. 5,
no. 2, pp. 1018–1032, Mar. 2014.
= 2 UT Az 2
2 ≥ 0, (41) [7] B. P. Roberts and C. Sandberg, “The role of energy storage in development
of smart grids,” Proc. IEEE, vol. 99, no. 6, pp. 1139–1144, Jun. 2011.
which demonstrates that M is positive semi-definite. [8] A. Nagarajan and R. Ayyanar, “Design and strategy for the deploy-
ment of energy storage systems in a distribution feeder with penetra-
tion of renewable resources,” IEEE Trans. Sustain. Energy, vol. 6, no. 3,
APPENDIX B pp. 1085–1092, Jul. 2015.
[9] Q. Li, R. Ayyanar, and V. Vittal, “Convex optimization for DES plan-
In this appendix we show that the quadratic form of (24) is ning and operation in radial distribution systems with high penetration
convex in dcurt . of photovoltaic resources,” IEEE Trans. Sustain. Energy, vol. 7, no. 3,
pp. 985–995, Jul. 2016.
Let i = ire + jiim be the vector of currents in the branches [10] Y. M. Atwa and E. El-Saadany, “Optimal allocation of ESS in distribution
of the system. Also, let Ybr be the branch admittance matrix of systems with a high penetration of wind energy,” IEEE Trans. Power Syst.,
the system. Then the vector of branch currents corresponding to vol. 25, no. 4, pp. 1815–1822, Nov. 2010.
[11] J. Tant, F. Geth, D. Six, P. Tant, and J. Driesen, “Multiobjective battery
the total demand of d + dcurt is given by: storage to improve PV integration in residential distribution grids,” IEEE
Trans. Sustain. Energy, vol. 4, no. 1, pp. 182–191, Jan. 2013.
I = Ybr (1 + ê + j fˆ) [12] M. Nick, R. Cherkaoui, and M. Paolone, “Optimal allocation of dispersed
energy storage systems in active distribution networks for energy balance
and grid support,” IEEE Trans. Power Syst., vol. 29, no. 5, pp. 2300–2310,
= Ybr (ê + j fˆ) (42) Sep. 2014.

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
740 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 2, APRIL 2018

[13] A. S. Awad et al.,“Optimal ESS allocation and load shedding for improv- [28] A. Gabash and P. Li, “On variable reverse power flow—Part II: An elec-
ing distribution system reliability,” IEEE Trans. Smart Grid, vol. 5, no. 5, tricity market model considering wind station size and location,” Energies,
pp. 2339–2349, Sep. 2014. vol. 9, no. 4, p. 235, 2016.
[14] M. Sedghi, A. Ahmadian, and M. Aliakbar-Golkar, “Optimal storage [29] A. Gabash, D. Xie, and P. Li, “Analysis of influence factors on rejected
planning in active distribution network considering uncertainty of wind active power from active distribution networks,” in Proc. IEEE Power
power distributed generation,” IEEE Trans. Power Syst., vol. 31, no. 1, Energy Student Summit, 2012, pp. 25–29.
pp. 304–316, Jan. 2016. [30] M. Sullivan, J. Schellenberg, and M. Blundell, “Updated value of service
[15] A. S. Awad, T. H. El-Fouly, and M. Salama, “Optimal ESS allocation for reliability estimates for electric utility customers in the United States,”
load management application,” IEEE Trans. Power Syst., vol. 30, no. 1, Ernest Orlando Lawrence Berkeley Nat. Lab., Berkeley, CA, USA, Tech.
pp. 327–336, Jan. 2015. Rep., 2015.
[16] A. S. Awad, T. H. EL-Fouly, and M. Salama, “Optimal ESS allocation for [31] K. Divya and J. Østergaard, “Battery energy storage technology for power
benefit maximization in distribution networks,” IEEE Trans. vol. 8, no. 4, systems-An overview,” Elect. Power Syst. Res., vol. 79, no. 4, pp. 511–520,
pp. 1668–1678, Jul. 2017. 2009.
[17] G. Carpinelli, G. Celli, S. Mocci, F. Mottola, F. Pilo, and D. Proto, “Opti- [32] R. D. Zimmerman, “MATPOWER: A MATLAB power system simulation
mal integration of distributed energy storage devices in smart grids,” IEEE package.”
Trans. Smart Grid, vol. 4, no. 2, pp. 985–995, Jan. 2013. [33] M. Grant, S. Boyd, and Y. Ye, “CVX: MATLAB software for disciplined
[18] P. Xiong and C. Singh, “Optimal planning of storage in power systems convex programming,” 2008.
integrated with wind power generation,” IEEE Trans. Sustain. Energy, vol. [34] MOSEK ApS, “The MOSEK optimization toolbox for MATLAB man-
7, no. 1, pp. 232–240, Jan. 2016. ual. Version 7.1 (Revision 28),” 2015. [Online]. Available: [Link]
[19] S. Hashemi, J. Ostergaard, and G. Yang, “A scenario-based approach for [Link]/7.1/toolbox/[Link]
energy storage capacity determination in LV grids with high PV penetra- [35] Commission for Energy Regulation (CER). Customer Behaviour
tion,” IEEE Trans. Smart Grid, vol. 5, no. 3, pp. 1514–1522, May 2014. Trials (CBTs). [Online]. Available: [Link]
[20] S. Bolognani and S. Zampieri, “On the existence and linear approximation smart-metering/
of the power flow solution in power distribution networks,” IEEE Trans. [36] Irish Social Science Data Archive. CER Smart Metering Project.
Power Syst., vol. 31, no. 1, pp. 163–172, Jan. 2016. [Online]. Available: [Link]
[21] R. Baldick, “DC power flow in rectangular coordinates,” in Proc. DIMACS gulationcer/
Workshop Energy Infrastruct., Des. Stability Resilience, Feb. 2013. [37] EirGrid. [Online]. Available: [Link]
[22] S. V. Dhople, S. S. Guggilam, and Y. C. Chen, “Linear approximations to erformancedata/
AC power flow in rectangular coordinates,” IEEE Communicat., Control, [38] Solar Radiation Data. [Online]. Available: [Link]
Comput. (Allerton), 53rd Annual Allerton Conf., pp. 211–217, 2015. eng/services/
[23] M. Ghasemi Damavandi, V. Krishnamurthy, and J. R. Martı́, “Robust [39] P. Poonpun and W. T. Jewell, “Analysis of the cost per kilowatt hour
meter placement for state estimation in active distribution systems,” IEEE to store electricity,” IEEE Trans. Energy Convers., vol. 23, no. 2,
Trans. Smart Grid, vol. 6, no. 4, pp. 1972–1982, Jul. 2015. pp. 529–534, Jun. 2008.
[24] D. Das, “A fuzzy multiobjective approach for network reconfiguration of [40] S. Wong, K. Bhattacharya, and J. D. Fuller, “Electric power distribution
distribution systems,” IEEE Trans. Power Del., vol. 21, no. 1, pp. 202–209, system design and planning in a deregulated environment,” IET Gener.,
Jan. 2006. Transm. Distrib., vol. 3, no. 12, pp. 1061–1078, 2009.
[25] A. Gabash and P. Li, “Active-reactive optimal power flow in distribution [41] S. De la Torre and F. D. Galiana, “On the convexity of the system
networks with embedded generation and battery storage,” IEEE Trans. loss function,” IEEE Trans. Power Syst., vol. 20, no. 4, pp. 2061–2069,
Power Syst., vol. 27, no. 4, pp. 2026–2035, Nov. 2012. Nov. 2005.
[26] A. Gabash, M. Alkal, and P. Li, “Impact of allowed reverse active power
flow on planning PVs and BSSs in distribution networks considering
demand and EVs growth,” in Proc. IEEE Power Energy Student Summit,
Bielefeld, Germany, 2013, pp. 23–25.
[27] A. Gabash and P. Li, “On variable reverse power flow—Part I: Active-
Reactive optimal power flow with reactive power of wind stations,” Ener- Authors’ photographs and biographies not available at the time of publication.
gies, vol. 9, no. 3, p. 121, 2016.

Authorized licensed use limited to: Federation University Australia. Downloaded on December 01,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.