This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
1

A Comprehensive Battery Energy Storage

Optimal Sizing Model for Microgrid
Applications
Ibrahim Alsaidan, Student Member, IEEE Amin Khodaei, Senior Member, IEEE, and Wenzhong Gao,
Senior Member, IEEE

L Set of microgrid distribution lines.

Abstract— Microgrids expansion problems with battery energy N Set of maximum depth of discharge segments.
storage (BES) have gained a great attention in recent years. To G Set of dispatchable units.
ensure reliable, resilient, and cost effective operation of
W Set of renewable generation units.
microgrids, the installed BES must be optimally sized. However,
critical factors that have a great impact on the accuracy and Φ Set of uncertain parameters.
practicality of the BES sizing results are normally overlooked. Parameters
These factors include the wide range of characteristics for BL BES investment budget limit.
different technologies, the distributed deployment, the impact of
depth of discharge and the number of charging/discharging cycles
CEia , CPi a BES annualized energy/power capital cost.
on the BES degradation, and the coordination of microgrid CI ia BES annualized installation cost.
operation modes. Thus, this paper proposes a comprehensive BES
sizing model for microgrid applications which takes these critical CM i BES annual operating and maintenance cost.
factors into account when solving the microgrid expansion Dbdh , CDbdh Total load demand and critical load demand at
problem and accordingly returns the optimal BES size,
technology, number, and maximum depth of discharge. The bus b, day d, hour h.
microgrid expansion problem is formulated using mixed integer DRi ,URi Ramp down and ramp up rates.
linear programming. The nonlinear relationship between the BES DTi ,UTi Minimum down and up times.
depth of discharge and lifecycle is linearized using piece wise
linearization technique and implemented to model the BES f l max Maximum power capacity of distribution
degradation. The proposed model is validated using a test lines.
microgrid. The conducted numerical simulation shows that the
proposed model is able to determine the optimal BES size, P M , max Maximum power capacity of the line
technology, number, and maximum depth of discharge and connecting the microgrid to the utility grid.
further enhances the accuracy and practicality of the BES sizing r Interest rate.
solutions. T Project lifetime.
prs Probability of islanding scenarios.
Index Terms—Battery energy storage, distributed energy
resource, microgrid, expansion planning. v Value of lost load, $/kWh.
zdhs Microgrid/utility grid connection state.
NOMENCLATURE α imax , α imin Maximum and minimum BES energy rating to
Indices power rating ratio.
b Index for bus. γ ibm BES maximum depth of discharge.
d Index for day.
h Index for hour.
κ im BES lifecycle.
i Index for distributed energy resources. ρ dh Electricity market price, $/kWh.
l Index for lines. ηi BES round trip efficiency.
m Index for depth of discharge segments. µ ib Element of generation-bus incidence matrix (1 if
s Index for scenarios.
unit i is connected to bus b, 0 otherwise).
~ Index for forecasted parameter.
ψ lb Element of line-bus incidence matrix (1 if line l
Sets
B Set of BES technologies. is connected to bus b, 0 otherwise).
K Set of microgrid buses. Variables
This work was supported in part by US NSF Grant (1429093, 1711951).
Cibdhs Stored energy in the BES at each interval.
I. Alsaidan, A. Khodaei and W. Gao are with Department of Electrical and CibR , PibR BES rated energy and rated power.
Computer Engineering, University of Denver, Denver, CO 80210 USA (e-
ch dch
mails:[email protected],[email protected], Pibdhs , Pibdhs BES charging and discharging power.
[email protected]).
* Corresponding Author: David Wenzhong Gao.

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
2

Fi Cost function of the microgrid local DG units. resources, especially when the connection to the main grid is
lost. By optimally managing the local microgrid resources such
f ldhs Distribution line power flow.
as dispatchable DGs, renewable DGs, energy storage, and
LS bdhs Load curtailment. controllable loads, the microgrids operation costs can be
I idhs Commitment state of dispatchable units. significantly reduced [10]. However, such solution raises
privacy concerns as microgrid owners would need to exchange
Pidhs DER output power. information about their microgrid with either other microgrid
M
Pdhs Power transferred to/from the utility grid. owners or the system operator. Provisional microgrids
on off significantly facilitate the integration of renewable DGs,
Tidh , Tidh Number of consecutive ON and OFF times. however, they require additional investments and control
uibdhs BES operating state. mechanism to ensure a reliable and economic operation. The
xib BES investment state (1 if installed, 0 energy storage, among the rest, is discussed to be the best option
for mitigating the challenges imposed by renewable generation
otherwise).
and improving microgrid reliability while at the same time
wibm Binary variable that represents the chosen value reducing the microgrid operation cost. Energy storage can store
of the BES maximum depth of discharge for the excess renewable generation to be utilized when it is
discharge segment m (1 if chosen, 0 otherwise) beneficial from either an economic perspective (e.g., energy
ξ ibdhs BES cycle indicator. arbitrage) or a technical perspective (e.g., frequency and
voltage regulation) [11]. Energy storage applications in
χ g , χl , χ p Auxiliary binary variables for renewable DGs microgrids can be further categorized into energy applications
generation, load demand, and electricity price. and power applications [12]. Energy storage technologies that
have high power density and fast response are known to be best
I. INTRODUCTION suited for power quality and frequency regulation applications.

T
On the other hand, energy storage technologies that have high
HE urgent need for reducing greenhouse gas emissions,
energy density and long discharging time are well suited for
improving the system reliability and power quality, and
long-term applications including peak shaving and energy
upgrading the aging transmission and distribution
arbitrage. Among these technologies, battery energy storage
infrastructure, have led to a significant increase in the
(BES) technology is considered to be the most attractive option
deployment of distributed energy resources (DERs) in power
due to its technological maturity and ability to provide both
systems. To ensure reliable operation of power systems under
sufficient energy and power densities [13].
high penetrations of DERs, a comprehensive control method
The BES degradation is greatly related to its operation. How
that takes the stochastic nature of DERs into consideration must
deep the BES is discharged and how many
be implemented, as proposed in [1]. These DERs can also be
charging/discharging cycles are performed have a significant
connected to each other to form a microgrid. The U.S.
impact on the BES rate of degradation. The relationship
Department of Energy defines a microgrid as “a group of
between these operation parameters and the BES lifetime must
interconnected loads and distributed energy resources (DERs)
be taken into account when the BES operation or planning
with clearly defined electrical boundaries that acts as a single
problems are investigated. One of the common approaches used
controllable entity with respect to the grid and can connect and
to consider the BES degradation phenomena in the BES
disconnect from the grid to enable it to operate in both grid-
operation problem is to add an extra term to the objective
connected or island modes” [2]. Microgrids can be either AC or
function that represents the BES degradation cost in $/kWh
DC. These two types of microgrids can also be combined into
(i.e., based on its charged/discharged energy) [14]–[17]. In BES
one hybrid AC/DC microgrids where power conditioning
planning problem, however, the Ah-throughput model is
devices should be used to link the AC part and DC part together
normally used to estimate the BES lifetime [18],[19]. In this
and to manage their operation [3]. Microgrids are considered as
model, the total delivered energy by the BES during the
viable enablers of DER integration, and in particular, would
planning time horizon is computed and compared with the
facilitate an efficient and reliable integration of emission free
expected Ah (i.e., current-hour) that the BES can deliver during
renewable distributed generators (DGs) to support the
its lifetime, which is typically provided by the manufacturer.
environmental agenda [4]. Renewable DGs, however, produce
This, however, may yield inaccurate estimation of the BES
a variable output power that may impose several challenges to
lifetime as the relation between the BES depth of discharge and
the microgrid operation and control, especially during the
number of cycles are not taken into consideration.
islanded operation. Various methods are studied to mitigate the
The topic of energy storage sizing in microgrids for both
generation intermittency and volatility associated with
power and energy applications is extensively investigated in
renewable DGs, including but not limited to demand response
literature. Determining appropriate energy storage size for
[5], generation curtailment [6], cluster of microgrids [7],
frequency regulation in an islanded microgrid is presented in
provisional microgrids [8], and energy storage deployment [9].
[20]–[22]. In these works, the BES is sized to perform
The demand response and renewable generation curtailment
frequency regulation services, and thus the economic viability
methods are argued to reduce the microgrid’s economic value
of installing the BES into the microgrid is not investigated.
and/or reliability as they are based on either reducing the
Energy storage sizing for energy applications is studied in [23]–
available renewable DGs generation or supplied demand (e.g.,
[25]. The optimal energy storage size that minimizes the total
load shedding or load shifting). In a microgrid cluster, two or
planning cost of a grid-connected microgrid is determined in
more microgrids are connected together to fully utilize their

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
3

[23] and [24]. The authors use mixed integer programming high maintenance cost [30],[31].
(MIP) to formulate the planning problem. In [25] a cost
B. Number of BES Units
sensitivity analysis for different energy storage technologies
and sizes in an islanded microgrid is presented. In [26] and [27], BES units can be integrated to the microgrid as either
a stochastic programming based model is implemented to study aggregated or distributed [32]. Under the aggregated
the impact of both BES and demand response on the configuration, one large BES is installed, normally close to the
distribution network expansion problem in the presence of utility substation, while under the distributed configuration
renewable DGs. Optimal sizing for a vanadium redox battery is multiple small-sized BES units are installed in different
studied in [28], in which dynamic programming is used to solve locations within the microgrid. Distributed BES units provide
the day-ahead unit commitment problem for the microgrid. the microgrid with the required redundancy and flexibility to
It is found that the reviewed publications have either one or respond to electricity price variations and ensure the desired
more of the following shortfalls: (i) Short time frame (e.g., one load/generation profiles, which will accordingly lead to
day) or static models (i.e., operation snapshots) are used to increase in economic benefits of the installation [33]. With
calculate the optimal BES size, which reduce the accuracy and proper control and management, distributed configuration may
the practicality of the obtained results; (ii) A single BES maximize the installed BES units’ lifetime, efficiency, and
technology is considered while ignoring the wide range of safety [34]. Moreover, increasing the storage capacity tends to
available BES with various technical and economical increase the difficulty of BES manufacturing and control [32].
characteristics; (iii) The impact of some decisive factors on the The performance of both aggregated and distributed
BES lifetime is overlooked, such as the BES depth of discharge, configurations in wind farm applications is investigated in [35]
number of charging/discharging cycles, and centralized vs. and [36]. However, these papers do not consider the economic
distributed installations; and (iv) On merely one operation aspect of the problem. The proposed model in this paper has the
mode (i.e., either grid-connected or islanded) is focused while capability to identify the optimal solution between the
the required coordination is not taken into account. To distributed and the aggregated configurations and to
overcome these shortfalls, a comprehensive model for BES accordingly determine the number of required BES units under
selection and sizing is proposed in this paper. The proposed the distributed configuration. This capability stems from the
model considers upgrading an existing microgrid with BES, fact that the proposed model takes into account the relationship
i.e., an expansion planning problem, in order to reduce its between the BES depth of discharge and lifecycle.
operation cost and improve its supply reliability. The model can C. BES Size
be used by microgrids operators and planners to determine the
The BES investment cost mainly depends on its size, i.e., the
optimal BES technology, size, and maximum depth of
power rating and the energy rating. The oversized BES tends to
discharge, as well as the required number of BES units in case
be economically unattractive while an undersized BES may not
of a distributed deployment. The problem is formulated using
render the desired benefits. Fig. 1 depicts the impact of the BES
mixed-integer linear programming (MIP).
size on the microgrid expansion planning cost [24], where an
The rest of the paper is organized as follows: Section II
increase in the size linearly increases the investment cost while
outlines the important and decisive variables (i.e., BES
reduces the operation cost in a nonlinear fashion. The optimal
technology, number of units, size, and depth of discharge) in
point is commonly the point that the summation of these two
the expansion planning problem. The expansion planning
costs is minimum.
problem is formulated in Section III. Numerical simulations on
a test microgrid are provided in Section IV to demonstrate the
benefits of the proposed model and to validate its applicability.
The discussion on the proposed model and conclusion are
provided in Sections V and VI, respectively.

II. PROBLEM OUTLINE AND VARIABLES

A. BES Technology
There are a variety of BES technologies with different
characteristics such as lead acid, sodium sulphur (NaS),
lithium-ion (Li-ion), nickel cadmium (NiCd), and vanadium
redox (VR). A comprehensive comparison of BES technologies Fig. 1. Microgrid expansion planning cost vs. BES size [24].
for power system applications is presented in [29]. The capital
cost of the BES is normally composed of power rating cost in D. BES Depth of Discharge
$/kW and energy rating cost in $/kWh. Even though these are The BES degradation is mainly caused by two factors:
important factors on selecting the appropriate BES technology calendric aging and cyclic aging [37]. The former occurs even
to be integrated into the microgrid, they are not the only factors if the BES is not used and is affected by the BES cells
that must be considered. For example, it is true that lead acid temperature and voltage, while the latter results from using
battery has the lowest capital cost among other technologies, (cycling) the BES and is greatly affected by BES depth of
however, it may not be the best option for performing the discharge and number of cycles. Disregarding the BES cyclic
applications that require frequent charging/discharging, such as aging in the expansion planning problem results in an
load leveling and energy arbitrage, due to its low lifecycle and inaccurate economical assessment as the BES might need to be

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
4

replaced before the end of the considered project lifetime. programming models have higher computation time compared
Different methods are proposed to estimate the BES lifecycle to linear programming models, especially when binary
[38]–[41]. However, it is not uncommon for BES manufacturer variables are introduced to the problem, which is the case in the
to provide the relationship between lifecycle and depth of proposed microgrid expansion formulation in this paper. In
discharge. This information is normally presented in a curve as general, it can be said that mixed integer nonlinear
the one depicted in Fig. 2. As the depth of discharge increases, programming (MINLP) are hard to be solved and can be
the BES lifecycle decreases. Different BES technologies have numerically intractable [45].
different lifecycle versus depth of discharge relationships. In This section discusses in details the objective function and
lead acid batteries, for example, this relationship tends to constraints associated with the proposed microgrid expansion
exhibit an exponential form whereas in lithium ion batteries a model. The objective of the proposed BES optimal selection
linear relationship is normally observed [38]. The depth of and sizing problem is to minimize the total expansion planning
discharge curves are linearized in this paper cost (1) subject to prevailing operational and budget constraints
(3)-(22).
Min ∑∑∑ Fi ( Pidh0 , I idh0 ) + ∑∑ ρ dh PdhM0
i∈G d h d h

+ ∑ ∑∑∑prs LSbdhs v (1)

s b∈K d h

+ ∑ ∑( ( PibR CPia ) (
+ CM i + CibR CEia + CI ia ))
i∈B b∈K
The first two terms in (1) represent the microgrid operation
cost, which is only determined during the microgrid grid-
connected mode, i.e., normal operation (as during the islanded
Fig. 2. Linearization of the lifecycle as a function of the depth of discharge. mode only the load curtailment is to be minimized). However,
if the microgrid is disconnected from the main grid for any
by using a piecewise linear approximation as depicted in Fig. 2. reason, the third term in (1) determines the cost associated with
Increasing the number of segments reduces the approximation failing to supply the microgrid demand (i.e., reliability
error but at the same time increases the computational evaluation). It must be noted that under normal microgrid
requirements. operation, there will not be any load curtailment as the
difference between the microgrid’s available generation and
III. PROBLEM FORMULATION demand is picked up by the main grid. This explains the
Expansion planning problems are commonly formulated utilization of sub index 0 in (1). That is, index s is used for
using MIP [42]–[44]. In MIP, an objective function is typically islanded scenarios which can change from 0 to S. The first
needed to be either maximized or minimized. This objective scenario (i.e., s= 0) represents the grid-connected mode while
function is composed of variables (continuous, integers, or scenarios s= 1,…,S represent various islanded scenarios. The
binaries) called decision variables and is solved subject to a set value of lost load (VOLL) is used to quantify the economic loss
of constraints. If the studied expansion problem consists of associated with the unserved energy. The VOLL represents a
nonlinear constraints, these constraints must be linearized first customer’s willingness to pay for reliable electricity service
before solving the problem. A commonly used approach to [46]. This value depends on the customer type and location in
solve MIP problems is branch and bound approach. This addition to the outage time and duration. To consider the
approach is based on two processes: 1) bounding process, in probability of occurrence of each islanding scenario, pr s is
which the solution of a relaxed MIP problem (e.g., transforming added as a weighting factor for each scenario. Since there is no
MIP problem into LP problem by removing integrality load curtailment in the grid-connected mode, the cost of
restrictions) is found and imposed as lower bound for unserved energy for the first scenario (i.e., s=0) will be zero.
minimization problems or upper bound for maximization Due to the expected insignificant changes in the microgrid
problems; 2) branching process, in which the problem is split demand, DGs generation, and electricity price over the planning
into a number of subproblems. A comprehensive discussion on horizon, one-year historical data are considered in this work.
the branch and bound approach is given in Appendix B. However, if the annual variation of data needs to be included,
Powerful solvers such as CPLEX, Xpress-MP, and the proposed model can be easily expanded and solved for
SYMPHONEY implement a combination of branch and bound longer periods. Another issue to consider is that in the proposed
techniques and cutting-plane techniques to accelerate the model, the investments are performed in year 1 of the planning
computation time associated with solving MIP problems, which horizon, thus enabling an annualized modeling and simulation.
allows large MIP problems to be solved using personal This is different from bulk power system planning in which new
computers. assets are added in various years, thus requiring multi-year
Compared with MIP, using nonlinear programming to model dynamic planning studies.
the microgrid expansion problem will have two major impacts The BES investment cost is composed of power rating and
on the results: (1) solution optimality, as nonlinear energy rating capital costs, annual maintenance cost, and
programming models may get stuck in a local optimal solution installation cost. It is assumed that the power conversion system
and never reach the global optimal solution, which is not the cost is embedded in the power rating capital cost. The annual
case in linear programming models; (2) solution time, nonlinear maintenance cost is normally given in terms of the BES power

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
5

rating whereas the installation cost is given in term of the BES this in mind, the load curtailment limits can be defined as in (5).
energy rating. Both the BES capital cost and installation cost The power flow in the microgrid distribution network is limited
are annualized using (2) by the lines capacities (6). A radial distribution network is
r (1 + r )T considered, hence (3) and (6) can efficiently model the power
Annualized cost = × One time cost (2) flow in the microgrid distribution network. In the proposed
(1 + r )T − 1 model it is assumed that the microgrid generations and loads are
The installed BES units are used in the grid-connected mode in close proximity, thus active losses are small compared to the
to increase the economic viability of the microgrid as they store power transferred through the microgrid distribution lines and
energy at low price periods and generate the stored energy back therefore are negligible. In this work, low renewable DGs
to the system to be either used by local demand or sold to the penetration is assumed. For this reason, the renewable DGs
utility grid at high price periods. In the islanded mode, however, spillage and the ability of using renewable DGs power
BES units are used to improve the microgrid reliability by conditioning units to supply or absorb reactive power is
minimizing the curtailed load and the cost of unserved energy. ignored. However, it must be known that under high renewable
This objective is subject to several operation and technical DGs penetration, these factors must be included in the
constraints, associated with the microgrid, dispatchable DGs, microgrid expansion problem [51], [52].
and the BES, that must be taken into account as discussed in the
following. B. Dispatchable DGs Constraints
Dispatchable DGs output power is limited by maximum and
A. Microgrid Constraints minimum capacities (7), variations across two successive
Microgrid’s system level constraints include nodal power intervals, i.e., ramp up and ramp down (8), (9), and minimum
balance, power exchange with the utility grid, limits on load up/down time limits (10), (11). Other constraints such as
curtailment and the distribution network power flow (3)-(6). emission and fuel limits can be easily included.
∑ µib Pidhs + ∑(
ch
Pibdhs dch
+ Pibdhs ) ∑ M
+ ψ lb f ldhs + Pdhs Pi min I idh ≤ Pidhs ≤ Pi max I idh ∀i ∈ G, ∀d , ∀h, ∀s (7)
i∈{G, W} i∈B l∈L
(3) Pidhs − Pid ( h−1) s ≤ URi ∀i ∈ G , ∀d , ∀h, ∀s (8)
+ LSbdhs = Dbdh ∀b ∈ K, ∀d , ∀h, ∀s Pid ( h−1) s − Pidhs ≤ DRi ∀i ∈ G , ∀d , ∀h, ∀s (9)
on
Tidh ≥ UTi ( I idh − I id ( h−1) ) ∀i ∈ G, ∀d , ∀h (10)
− P M,max z dhs ≤ Pdhs
M
≤ P M,max z dhs ∀d , ∀h, ∀s (4) off
Tidh ≥ DTi ( I id ( h−1) − I idh ) ∀i ∈ G, ∀d , ∀h (11)
0 ≤ LSbdhs ≤ (Dbdh − CDbdh ) ∀b ∈ K, ∀d , ∀h, ∀s (5)
− f l max ≤ f ldhs ≤ f l max ∀l ∈ L, ∀d , ∀h, ∀s
(6) C. BES Constraints
The nodal power balance (3) ensures that at each bus the Although different BES technologies are considered in this
power generated form DERs located at that bus plus/minus paper, the following equations can accurately model their
power flowing to/from the bus equals local demand. If the operation.
generation is not sufficient, load would be curtailed to satisfy Pi min xib ≤ PibR ≤ Pi max xib ∀i ∈ B, ∀b ∈ K (12)
the power balance. The BES power is positive when
discharging and negative when charging. The utility grid power
α imin PibR ≤ CibR ≤ α imax PibR ∀i ∈ B, ∀b ∈ K (13)
dch
is positive when the power flows from the utility grid to the 0 ≤ Pibdhs ≤ PibR u ibdhs ∀i ∈ B, ∀b ∈ K, ∀d , ∀h, ∀s (14)
microgrid, and negative otherwise. Note that the utility grid
power is zero at all buses except at the point of common
( )
− PibR 1 − u ibdhs ≤ Pibdhs
ch
≤ 0 ∀i ∈ B, ∀b ∈ K, ∀d , ∀h, ∀s (15)
coupling (PCC). It must be noted that the power losses as well ξ ibdhs = (u ibdhs − u ibd ( h −1) s ) u ibdhs
(16)
as the bus voltage magnitude and angle are ignored in this work. ∀i ∈ B, ∀b ∈ K, ∀d , ∀h, ∀s
A linear power flow model is needed to be combined with the 1
proposed model in order to solve the full AC power flow
without introducing nonlinear equations. Thus, existing power
∑∑ pr ξ
d h
s ibdhs ≤
T ∑κ
m∈N
im wibm ∀i ∈ B, ∀b ∈ K, ∀s (17)

flow models presented in literature (e.g.,[47]–[50]) are not

suitable to be used with the proposed model. Equation (4)
∑w ≤ x
m∈N
ibm ib ∀i ∈ B, ∀b ∈ K (18)

imposes a maximum limit on the power transferred through the dch

line connecting the microgrid to the utility grid. This equation Pibdhsτ ch
Cibdhs = Cibd ( h −1) s − − Pibdhsτ
is modified by including a binary parameter z that indicates the ηi (19)
microgrid islanding state. That is, if the value of z is 0, the ∀i ∈ B, ∀b ∈ K, ∀d , ∀h, ∀s
microgrid is disconnected from the utility grid and operated in
the islanded mode; while if it is equal to 1, the microgrid is grid-  
connected. The value of z is set by the microgrid planner before
1 −
 ∑
γ ibm wibm  C ibR ≤ C ibdhs ≤ C ibR

 (20)
solving the expansion planning problem and reflects how many  m∈N
hours in a year the microgrid operates in the islanded mode. ∀i ∈ B, ∀b ∈ K, ∀d , ∀h, ∀s
One of the motivations for microgrid deployment is the The BES power rating is limited by maximum and minimum
continuity of service for critical loads. The critical loads are values (12). For some BES technologies, such as those
typically associated with high VOLL so it is not economically considered in this paper, the energy rating is correlated to the
advisable to consider them for the load curtailment. Keeping power rating and cannot be sized independently. A capacity to

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
6

power ration (α) is used to size the BES capacity and determine
the maximum discharge time at rated power (13). If flow
batteries such as vanadium redox battery are considered, this
constraint can be easily modified to decouple the power rating
and the energy rating. The binary variable x is used to indicate
the investment state of a BES technology. To consider multi
period modeling, this binary should be time dependent.
However, in this work the BES is assumed to be installed in the
first year. The BES charging/discharging powers are limited by
the installed rated power (14), (15), which further impose that
the BES power be negative in the charging mode while positive
in the discharging mode. The binary variable u is used to
represent the BES operating state. The BES can discharge only Fig. 3. Schematic diagram for the proposed comprehensive BES sizing model.
when u equals 1 and can charge when u equals 0. Each BES D. Data Uncertainties Consideration
technology has a specific lifecycle, which depends on its
associated depth of discharge. The BES cycle is typically In the presented microgrid expansion planning formulation
defined as a complete charge and discharge cycle. Equation above, hourly forecasted data for the renewable DGs
(16) is used to determine the BES cycles. The value of ξ will be generation, the load demand, and the electricity price is used.
1 every time the discharging process is initiated, otherwise it is However, forecasting errors may arise as these parameters are
0. The summation of the BES cycles over the planning time affected by uncontrollable factors such as weather conditions,
horizon cannot exceed the determined lifecycle associated with customers’ behavior, and congestion or outage incidents. The
the chosen maximum depth of discharge and desired project proposed model can be extended by applying robust
lifetime (17). That is, the installed BES does not need to be optimization method presented in [53] to address the presence
replaced during the considered project lifetime and therefore of uncertainties in the microgrid expansion problem. Robust
the BES replacement cost is not included in (1). The value of κ optimization determines the worst-case solution by maximizing
is determined based on the chosen maximum depth of discharge the minimum value of the objective function (1) over
(Fig. 2) in which it is assumed that the curve is divided into N uncertainty set Ф (i.e., for renewable DG generation, load
segments. w is a binary variable that represents the chosen demand, and electricity price). The objective function in (1) can
maximum depth of discharge segment. It must be noted that the be rewritten as:
linearization of the BES depth of discharge versus lifecycle max min ∑∑∑ Fi ( Pidh 0 , I idh 0 ) + ∑∑ ρ dh PdhM0
Φ
curve for each considered BES technology must be performed i∈G d h d h
before solving the optimization problem. That is, the number of + ∑ prs ∑∑∑ LS bdhs v (22)
cycle associated with each depth of discharge value is entered s b∈K d h
as an input to the proposed model. Equation (18) ensures that
only one maximum depth of discharge value is considered for
+∑∑ ( (
PibR CPia ) (
+ CM i + CibR CEia + CI ia ))
i∈B b∈K
each installed BES unit. The stored energy in the BES at each
time interval equals the stored energy in the preceding interval
Uncertain parameters are associated with a nominal value
minus the discharged or charged energy (19). The BES cannot
that can be found from the forecast data. These nominal values,
be charged more than its rated energy and cannot be discharged
however, expand around a range of uncertainty which define an
below its minimum value which is defined by the determined
interval within which the uncertain parameter is presumed to
maximum depth of discharge (20). It must be noted that the BES
lie. Thus, the uncertain parameters can be expressed as:
can, however, be discharged at any depth of discharge value less ~
than the determined maximum depth of discharge. Finally, the
g
Pidhs = Pidhs + Pidhs χ idhs − P idhs χ g ∀i ∈ W, ∀d , ∀h, ∀s (23)
idhs
investment cost of the installed BES units is limited by the ~
available budget (21). Dbdh = Dbdh + Dbdh χ bdh l
− D bdh χ l ∀b ∈ K, ∀d , ∀h (24)
bdh

∑ ∑( (
i∈B b∈K
) (
PibR CPia + CM i + CibR CEia + CI ia ≤ BL (21))) ρ~dh − p
ρ dh χ dh ≤ ρ bdh ≤ ρ~
dh
p
+ ρ dh χ dh ∀b ∈ K, ∀d , ∀h (25)
where the inserted bars in (23)-(25) represent the upper and
The problem is solved from a microgrid developer
lower bounds of each parameter. To ensure only one extreme
perspective, which means that savings in the upstream grid,
point is chosen, the following constraints are imposed to the
such as deferred distribution and transmission upgrades as well
microgrid expansion model at each time interval:
as benefits of the reduced congestion, are not included. Figure
g
3 shows a schematic diagram for the proposed BES sizing and
g
χ idhs + χ idhs l
≤ 1 , χ bdh + χ lbdh ≤ 1 , χ dhp + χ dh
p
≤1 (26)
selection model. However, it must be noted that a trade-off between the
solution optimality and robustness must be performed when
robust optimization method is used. This can be achieved by
imposing a higher cap on the maximum number of uncertain
parameters that can reach their bounds in the considered
planning horizon. This cap is known as the budget of
uncertainty [54]. Increasing the budget of uncertainty value will

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
7

increase the robustness of the obtained solution at the expense 2 4 PV 0 0-1 -

3 4 Wind 0 0-1.5 -
of optimality, and vice versa. If the budget of uncertainty is set
to be 0, the problem is solved by ignoring uncertain parameters. TABLE II
MICROGRID LOCAL DEMAND DETAILS (R: RESIDENTIAL, C: COMMERCIAL)
To solve the resulted min-max optimization problem, the
Peak Load Critical Load Load VOLL
duality theory is used to convert the problem into either Load Bus
(MW) (%) Type ($/MWh)
maximization or minimization problem. For more details about 1 3 6.62 60 C 50,000
robust optimization formulation and duality theorem, the 2 5 4.41 30 R&C 50,000
readers are referred to [53].
TABLE III
DISTRIBUTION LINES CONNECTIONS AND CAPACITIES
IV. NUMERICAL SIMULATION Line From Bus To Bus Capacity (MW)
A 5-bus microgrid that contains a gas generator, a wind 1 1 2 8
2 2 3 6
turbine, and a solar photovoltaic unit is used to study the 3 2 4 5
proposed expansion planning model. DGs characteristics and 4 2 5 5
location in the microgrid are given in Table I. The hourly data
TABLE IV
of renewable DGs generation and local loads are obtained from BES TECHNOLOGIES CHARACTERISTICS
[55], whereas the hourly data for the electricity market price are Power Energy Maintenance Installation
taken from [56]. The local load details and location in the η
Technology Rating Cost Rating Cost Cost Cost
(%)
microgrid are given in Table II while the microgrid distribution ($/kW) ($/kWh) ($/kW/yr) ($/kWh)
network lines characteristics are given in Table III. The point of Lead-acid 200 200 50 20 70
NiCd 500 400 20 12 85
common coupling (PCC), which connects the microgrid to the Li-ion 900 600 - 3.6 98
utility grid, is located at bus 1. Four BES technologies are used NaS 350 300 80 8 95
in the simulation: lead acid, NiCd, Li-ion, and NaS. The
TABLE V
characteristics of the BES technologies are borrowed from [29] BES LIFECYCLES FOR VARIOUS DEPTH OF DISCHARGE VALUES
and shown in Table IV. The power rating of each BES Depth of Number of Cycles
technology is constrained by a maximum value, assumed to be Discharge Lead acid NiCd Li-ion NaS
4 MW in this paper. A minimum discharging time of 1 hour and 10 8000 7900 - 100000
a maximum discharging time of 5 hours are considered. The 20 2500 5800 - 60000
30 1500 3400 - 30000
BES manufacturers data sheets are used to determine the 40 950 2000 - 15000
relationship between the depth of discharge and lifecycle of 50 700 1200 8000 10000
each BES technology [57]–[60]. Based on the manufacturer 55 - - 7500 -
data sheet, ten different maximum depth of discharge values are 60 590 900 6900 9000
considered for each BES technology (i.e., N=10) through 65 - - 6200 -
70 500 800 5800 7000
linearization. Increasing N will increase both the accuracy and 75 - - 5000 -
the computational requirements. Table V indicates the lifecycle 80 450 700 4500 6000
of the BES technologies at the considered maximum depth of 85 - - 4100 -
discharge values. In the Li-ion battery case, the given minimum 90 390 600 3700 5000
100 350 500 3000 4000
depth of discharge in the manufacturer data is 50% and no
information is given for lower depth of discharge values. One- Case 0: To accurately assess the benefits of installing the BES
hour islanded scenarios are implemented to evaluate the into the microgrid, the pre-expansion case is solved first in
reliability of the microgrid under islanded modes (i.e., 24 order to enable comparisons to the case of BES installation. In
scenarios for each day), with uniform probability (i.e., this case, the optimization problem is reduced to an optimal
pr=1/24). The following four cases are studied: scheduling problem in which the objective is to minimize the
Case 0: Microgrid optimal scheduling (i.e., the BES units microgrid operation cost. The total microgrid operation cost is
installation is not included). found to be $6,059,253/year. This cost is composed of local
Case 1: Microgrid expansion planning. In this case, the BES units generation cost, power exchanged with main grid cost, and
installation to reduce both the microgrid operation cost and the the expected cost of unserved energy. The amount of expected
cost of unserved energy is considered. unserved energy in this case is 67.5 MWh/year. This of course
Case 2: This case investigates the impact of ignoring the would happen only when the microgrid is disconnected from
relationship between the BES depth of discharge and lifecycle the utility grid and operates in the islanded mode.
on the obtained solution accuracy and practicality.
Case 3: The impact of uncertinities associated with Case 1: In this case, the BES installation is considered and the
renewable DGs generation and load demand on the obtained proposed mathematical model is used to model the microgrid
solution is studied in this case. expansion problem. This case is solved for various project
lifetimes. The BES units capital costs and installation costs are
annualized according to the considered project lifetime with a
4% interest rate. Moreover, the BES lifecycle is expended over
TABLE I the project lifetime to ensure that once the BES is installed, it
DISPATCHABLE GENERATION UNITS CHARACTERISTICS
does not need to be replaced until the end of the project period.
Cost Coefficient Min-Max Min Up/Down
Unit Bus Type
($/MWh) Capacity (MW) Time (hour) The solution of the proposed model includes the optimal BES
1 3 Gas unit 90 0-7 1 technology, or combination of technologies, as well as BES

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
8

size, number of BES units, and maximum depth of discharge $817,125 over the considered expansion timeframe. Both
that minimizes the total microgrid expansion planning cost. In batteries can be discharged up to 80% of their rated energy size.
the grid-connected mode, the BES installation reduces the The expected unserved demand in the islanded operation is
microgrid operation cost by storing energy during low price reduced by 99.68% compared to Case 1.
hours to be used during high price hours toward either For a project lifetime of 20 years, the minimum expansion
supplying local demand (i.e., load shifting) or making economic cost is found when two Li-ion batteries are integrated at buses
benefit from selling the stored energy to the utility grid (i.e., 1 and 4. The optimal size and maximum depth of discharge
energy arbitrage). In the islanded mode, however, the BES values for these two BES units are shown in Table VI. The BES
reduces the unserved energy, which results in improving the installed at bus 4, i.e., where the renewable DGs are located, is
microgrid overall reliability. The obtained results for various used to shift the renewable generation from off-peak periods to
project lifetimes are given in Table VI. It is clear from the the peak periods which will reduce the amount of energy that is
results that installing the BES is economically justifiable, as the needed to be imported from the utility grid during the high price
total expansion cost for all the considered project lifetimes is periods and therefore reduce the microgrid operation cost. The
less than the cost of operating the microgrid without BES. BES located at bus 1 is used for energy arbitrage. The expected
Based on the solution of the optimization problem, the BES unserved energy in this case is reduced by 99.26% compared to
optimal technology, number of units, size, maximum depth of Case 1.
discharge, as well as the number of annual cycles performed by The BES cycles are computed using equation (16). Fig. 4
the BES in the grid-connected mode are given in Table VI. shows how the proposed model can accurately compute the
For the project lifetime of 10 years, a centralized lead acid BES cycles over the planning horizon. It can be seen from the
battery located at bus 2 with the size of 2.905 MW and 5.929 figure that the summation of the BES discharging cycles (ξ)
MWh yields the minimum total expansion cost. However, from over one week equals to the number of performed cycles over
the BES operation analysis, it is found that the lead acid battery the same period. This enables microgrid planners to take the
is mostly installed to improve the microgrid reliability under the impact of the number of BES cycles on its lifetime into
islanded operation as the number of its cycles in grid-connected consideration during the planning stage. Ignoring this impact
operation is low (i.e., 48 cycles). This explains why the lead may require the BES replacement before the expected end of
acid battery is selected as the optimal technology in this case as project which imposes an extra cost to the expansion plan.
it is characterized with low capital cost and lifecycle. The lead The other factor that affects the BES lifetime is the depth of
acid battery is located at bus 2 in order to be available to supply discharge, i.e., the amount of energy that can be taken from the
both microgrid demand in the islanded mode of operation which BES in each cycle. As discussed in Section III, the number of
are located at buses 3 and 5. In order for the lead acid battery to cycles that the BES can perform before it needs to be replaced
perform this number of cycles per year and remains in service is determined by how deep the BES is discharged. Figs. 5-7
for 10 years, its depth of discharge cannot exceed 70%. depict the SOC for the installed BES units for each considered
Installing the lead acid battery is expected to save $7,348/year. project lifetime for a sample one week. It must be noted that the
However, the big saving is noticed in the islanded operation as determined maximum depth of discharge value puts a cap on
the expected unserved energy is reduced by 98.09% compared how deep the BES can be discharged based on the relationship
to Case 1. between the BES depth of discharge and lifecycle. However,
When the project lifetime is increased to 15 years, the the BES can operate with a depth of discharge value that is less
investment in expensive technologies such as Li-ion and NaS than the determined optimal value as can be seen from the state
becomes feasible. In this case, it is found that the optimal of charge curves (Figs. 6 and 7). The determined maximum
solution yields when Li-ion and NaS batteries are installed at depth of discharge value, however, will ensure that the installed

TABLE VI
NUMERICAL SIMULATION RESULTS FOR CASE 1.
Optimal Determined Parameters for the Installed BES Associated Expansion Costs
BES BES Total Local Cost of Expected Cost
Lifetime BES Bus Power Energy Depth of Number of Investment Generation Power Total
of Unserved Expansion
(years) Technology Number Rating Rating Discharge Cycles
(MW) (MWh) (%) (Cycles/year) Cost Cost Exchange Energy Cost ($/year)
($/year) ($/year) ($/year) ($/year)
10 Lead-acid 2 2.905 5.929 70 48 377,682 834,778 1,843,639 64,272 3,120,371
Li-ion 1 1.461 1.886 80 300
15 NaS 3 1.444 1.900 80 396 432,445 834,778 1,796,512 10,680 3,074,416
Li-ion 1 2.527 2.865 90 168
20 Li-ion 4 0.401 0.818 50 396 357,420 834,778 1,815,893 24,960 3,033,053

buses 1 and 3, respectively. As these technologies can perform BES does not need to be replaced during the considered project
a high number of cycles before they reach their end of lifetime, lifetime which is one of the microgrid planner requirements in
they are used to reduce the microgrid operation cost in the grid- this work.
connected mode by purchasing power from the utility grid in The energy not supplied at each BES lifetime is depicted in
low price periods and either use it to supply the demand or sell Fig. 8. Even though the 15-year lifetime yields the minimum
it to the utility grid in high price period. This saves the energy not supplied value, it does not yield the optimal solution
microgrid operator $54,475 per year and will sum up to in terms of microgrid total cost, which is associated with the 20-

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
9

year BES lifetime case. To examine the impact of depth of 120

discharge on the overall microgrid expansion cost and to 100
validate the ability of the proposed model to determine the

BES SOC (%)

optimal maximum depth of discharge value, the numerical 80

simulation is solved again for a 10-year BES lifetime. Fig. 9 60

shows the obtained microgrid expansion cost for a variety of 40
depth of discharge values of lead acid battery. It can be seen
that the microgrid expansion cost decreases as the depth of 20
Lead-acid battery
discharge increases until it reaches its optimal value found by 0
1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161
the proposed model (i.e., 70%) after which the microgrid
Time (hour)
expansion cost starts to slightly increase again. The change in
the microgrid cost is relatively small for depth of discharge Fig. 5. The installed Lead-acid battery SOC for one sample week (10 years
project lifetime).
values larger than 70%, thus not easily visible in the figure. 120

Case 2: In order to accurately estimate the benefits and the

100
optimal parameters of installed BES, the impact of operation
factors such as depth of discharge and number of cycles on the 80

BES SOC (%)

BES lifetime must be included into the microgrid expansion
problem. In this section, the importance of considering such 60

impact is investigated. The microgrid expansion planning

problem is resolved while ignoring the limit on the BES number 40

of cycles. In other words, the relationship between the BES 20

depth of discharge and lifecycle, which is represented by (17), Li-ion battery

is omitted from the proposed formulation. A 10-year BES 0

NaS battery

1 21 41 61 81 101 121 141 161

lifetime case is considered. Table VII shows the obtained Time (hour)
results for this case. Since the BES operation impact on its Fig. 6. The installed Li-ion battery and NaS battery SOC for one sample week
lifetime is not included in the model, the optimal BES (15 years lifetime).
technology would be the less expensive BES candidate, which 120

is lead acid battery. Moreover, the optimal maximum depth of 100

discharge is found to be 100%. This result, however, is
BES SOC (%)

unrealistic as the installed lead acid battery is expected to 80

perform 792 cycles/year. Based on the relationship between the 60

BES depth of discharge and lifecycle, which is given in Table
40
V, the installed lead acid battery must be replaced within the
first 5 months from its installation. This shows how important 20 Li-ion battery 1

it is to consider the BES operation impact on its lifetime in the Li-ion battery 2

0
microgrid expansion problem in order to enhance the accuracy 1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161
and practicality of the obtained results. Time (hour)
TABLE VII Fig. 7. The installed Li-ion batteries SOC for one sample week (20 years
NUMERICAL SIMULATION RESULTS FOR CASE 2. lifetime).

Optimal Case 3: In this case, the forecast errors in renewable DG

BES Number Expected
BES Optimal
Optimal
Maximum
of End of generation and load demand impacts on the obtained solution
Lifetime BES Depth of are investigated. The worst-case scenario occurs when a
Size performed Lifetime
(years) Technology Discharge
(MW/MWh)
(%)
cycles/year (months) reduction in renewable DG generation and increase in load
demand compared to the forecasted data take place. Thus, -20%
Lead-acid 0.823/1.306 100 792
10 5 forecast errors in renewable DGs generation and +10% forecast
Lead-acid 2.105/3.341 100 792
errors in load demand are considered. These forecast
2
BES Power (MW) and Cycle

1.5

0.5
Indicator

-0.5

-1
BES cycles indicator
-1.5
BES power
-2
1 21 41 61 81 101 121 141 161
Time (hour)
Fig. 4. The Li-ion battery power and cycles for 15-year project lifetime.

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
10

BES
1.4 BES Optimal Optimal Number
Optimal
Lifetime BES Maximum Depth of performed
1.2 Size
(years) Technology of Discharge (%) cycles/year
(MW/MWh)
Energy Not Supplied

1
NiCd 2.483/2.922 100 48
10
(MWh/year)

0.8 NaS 1.510/1.987 50 600

0.6 TABLE IX
STUDIED CASES SUMMARY
0.4
Case Pros Cons
0.2
• No BES investment cost as the • High microgrid operation cost
0 0 BES is not installed in this case. and low reliability, especially
10-year 15-year 20-year during islanded operation.
BES Lifetime • Improve the microgrid reliability • Stochastic nature of
by supplying demand during renewable DGs generation and
Fig. 8. Energy not supplied for the considered BES lifetime. load demand is not included in
islanded incidents.
the expansion problem.
• Reduce operation cost by using
1
Microgrid Total Expansion

5
BES to perform energy arbitrage
application.
Cost Million $/year

4
• Impact of BES depth of discharge
on its lifetime is considered.
3
• Microgrid total expansion cost is • Unrealistic results are
reduced as the impact of BES obtained and thus the BES will
2 depth of discharge on its lifetime is need to be replaced before the
2
ignored. end of the desired project
1 lifetime.
• The obtained result is robust • High microgrid total
0
10 20 30 40 50 60 70 80 90 100
against renewable generation and expansion cost.
3 load demand uncertainties. • The optimality of the obtained
Depth of Discharge (%) solution might be impacted.
Fig. 9. Microgrid total expansion cost for different depth of discharge value (10-
year BES lifetime case).
General algebraic modeling system (GAMS) is used to solve
errors are assumed to happen for 1000 hours/year. Increasing or the optimization problem in the studied cases. The problem is
decreasing the number of hours per year at which the implemented on a 2.4-GHz personal computer using CPLEX
uncertainties are considered leads to more conservative or 11.0. The obtained solution is found within a 0.05% gap of the
aggressive solution against data uncertainties. In the optimal solution; hence it provides a near-optimal solution. The
conservative solution, the obtained results are more robust gap is adjusted using the built-in functionalities of CPLEX in
against uncertainties but at the same time higher microgrid which in each iteration an upper bound and a lower bound of
expansion cost is expected. On the other hand, the aggressive the current solution are calculated and the relative difference is
solution yields less robust results against uncertainties with considered as an optimality gap. It is worth noting that in the
lower microgrid expansion total cost compared to the long-term planning problem it is not always possible to achieve
conservative solution. The 1000 hours/year used in this the optimal solution due to the complexity of the problem and
simulation can be considered as a moderate solution. The 10- the large number of binary and continuous variables. The
year BES lifetime case is resolved here using the proposed computation time, however, depends on the considered case,
model with the consideration of uncertainties. From the the number of islanding scenarios, and the optimality gap
numerical simulation results, it is found that when the among other factors. For the first case (i.e., the microgrid
uncertainties associated with renewable DG generation and scheduling problem without the BES installation) the problem
load demand are taken into consideration, the microgrid total is solved within seconds. When the BES installation is included
expansion cost increases to become $3,368,200/year. to the problem, the problem is solved within multiple hours.
Moreover, expensive BES technologies, which are The highest computational time is slightly less than 18 hours.
characterized with high lifecycle such as NaS battery become However, as the problem in hand is an expansion planning
economically feasible. The optimally determined parameters of problem, it is solved offline where the computation time is not
the installed BES units are given in Table VIII. The reason as important as in operation problems.
behind installing NiCd and NaS batteries instead of lead acid
battery, which is found to be the optimal BES technology in V. DISCUSSION
Case 1, is that considering the uncertainties in the microgrid
expansion problem requires the installed BES to be used more Based on the reviewed literature and the obtained results
frequently in order to overcome the rapid change in the from the conducted numerical simulations, the following is
renewable DGs generation and the load demand, especially concluded:
during islanding operation. Thus, BES technology with high • Installing the BES into the studied microgrid reduced
lifecycle is needed in such case. A summary of the studied the operation cost of the microgrid while at the same
cases’ advantages and disadvantages are shown in Table IX. time improved the system reliability by reducing the
amount of unserved energy during islanded operation.
TABLE VIII
NUMERICAL SIMULATION RESULTS FOR CASE 3. • The proposed model in this paper was able to
accurately determine the optimal BES size,

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
11

technology, units’ quantity, and maximum depth of to (A2), y would be equal to β. Therefore, the equation is
discharge that minimizes the total microgrid linearized, and the results of the constraints defined in (A2)-
expansion planning cost while considering the (A3) conform to the original equation in (A1).
microgrid islanded operation and the BES
degradation. APPENDIX B
• The obtained results showed that ignoring the impact A commonly used approach to solve MIP problems, such as
of BES operation on its lifetime results in an the one presented in this paper, is the branch and bound
unrealistic solution in which the installed BES might approach. Before explaining how this approach works, a
end up costing much more than what was determined
concept of MIP relaxation must be introduced. A relaxed MIP
before installing the BES.
problem can be defined based on the following two
• Considering the renewable generation and load
characteristics [54]:
demand uncertainties impacted the obtained optimal
1) Any solution to the original MIP problem is also a feasible
solution as larger BES size and higher number of
cycles are required. Thus, robust solution against solution to the relaxed problem.
parameter uncertainties can be achieved with the 2) The objective function value associated with the original
expense of higher BES investment cost. MIP solution is larger than or equal to the objective function
value associated with the relaxed problem solution.
VI. CONCLUSION A typical relaxed MIP problem is its corresponding LP
problem, which can be found by removing any integrality
A comprehensive BES optimal sizing solution for microgrid
constraints in the original MIP problem. To this end, solving the
applications was proposed in this paper. Critical factors such as
corresponding LP problem will yield one of three possible
the wide range of characteristics for different technologies, the
cases: infeasible solution, feasible solution that satisfies the
distributed deployment, the impact of depth of discharge and
original MIP integrality constraints, or feasible solution that
the number of charging/discharging cycles on the BES
does not satisfy the original MIP problem integrality
degradation, and the coordination of microgrid operation modes
constraints. If there is no solution to the LP problem, then the
are taken into consideration in the proposed model. A BES
problem is said to be infeasible and some of the constraints must
degradation model that is based on the relationship between the
be relaxed or the problem should be reformulated. In case of a
BES depth of discharge and lifecycle was implemented in the
feasible solution, if the obtained LP solution happens to satisfy
proposed BES sizing model to accurately estimate the impact
the original MIP integrality constraints, then the LP solution is
of BES operation on its lifetime. Moreover, a set of islanded
the optimal solution for the original MIP problem. However,
scenarios was included in the model to determine the microgrid
such optimistic case does not happen often and the LP solution
reliability when it operates in the islanded mode. Numerical
normally tends not to comply with the MIP integrality
simulations performed on a test microgrid validated the
constraints. In this case, the LP problem is divided into two sub-
effectiveness of the proposed model. Robust optimization
problems. This process is known as branching as the LP
approach was employed to address the uncertainties associated
problem is branched into sub-problems. These sub-problems
with renewable DGs generation and load demand.
are solved and the obtained solutions are compared with each
A follow-on work of this research will focus on studying the
other. If the solutions of both sub-problems satisfy the
impact of using the BES for multiple applications in the BES
optimal sizing. In order to study the ability of the BES to integrality conditions, they must be compared and the sub-
perform some applications such as voltage support, congestion problem solution that is associated with smaller objective
relief, and system upgrade deferral, a linear power flow model function value for minimization problem or larger objective
that can be used to calculate the voltage magnitude and angle at function value for maximization problem is selected as the
each bus and the active and reactive power flow must be optimal solution. If only one sub-problem solution satisfies the
developed. MIP integrality conditions, then this solution is saved as
incumbent solution (i.e., the optimal solution if no better
APPENDIX A solution is found) while the branching process is continued on
Linearization of bilinear terms: if variable y is equal to the the second sub-problem searching for a better solution that
multiplication of continuous variable β and k binary variables satisfies the MIP integrality conditions. The branching and
δ 1 , δ 2 , δ 3 , …, δ k such as illustrated in (A1), it can be described bounding steps are shown in the following figure.
by 2(k+1) constraints as shown in (A2)-(A3). M is a large
positive constant.
y = β δ 1δ 2δ 3 ...δ k (A1)
k k
β− ∑ M (1 − δ ) ≤ y ≤ β + ∑ M (1 − δ )
j =1
j
j =1
j
(A2)

− Mδ j ≤ y ≤ Mδ j ∀j ∈ (1,2,3,..., k ) (A3)
If at least one binary variable is zero, according to (A3), y would
be zero, and (A2) would be relaxed. If all binary variables are
one, all k constraints in (A3) would be relaxed, and according

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
12

[14] M. Koller, T. Borsche, A. Ulbig, and G. Andersson, “Defining a

degradation cost function for optimal control of a battery energy storage
system,” in PowerTech (POWERTECH), 2013 IEEE Grenoble, 2013,
pp. 1–6.
[15] C. Ju and P. Wang, “Energy management system for microgrids
including batteries with degradation costs,” in Power System
Technology (POWERCON), 2016 IEEE International Conference on,
2016, pp. 1–6.
[16] Y. Zhang and M.-Y. Chow, “Microgrid cooperative distributed energy
scheduling (CoDES) considering battery degradation cost,” in
Industrial Electronics (ISIE), 2016 IEEE 25th International Symposium
on, 2016, pp. 720–725.
[17] K. Abdulla et al., “Optimal Operation of Energy Storage Systems
Considering Forecasts and Battery Degradation,” IEEE Trans. Smart
Grid, pp. 1–1, 2016.
[18] E. Hajipour, M. Bozorg, and M. Fotuhi-Firuzabad, “Stochastic
Capacity Expansion Planning of Remote Microgrids With Wind Farms
and Energy Storage,” IEEE Trans. Sustain. Energy, vol. 6, no. 2, pp.
491–498, Apr. 2015.
[19] J. P. Fossati, A. Galarza, A. Martín-Villate, and L. Fontán, “A method
for optimal sizing energy storage systems for microgrids,” Renew.
Energy, vol. 77, pp. 539–549, May 2015.
[20] A. Toliyat and A. Kwasinski, “Energy storage sizing for effective
primary and secondary control of low-inertia microgrids,” in 2015
IEEE 6th International Symposium on Power Electronics for
REFERENCES Distributed Generation Systems (PEDG), 2015, pp. 1–7.
[21] T. Kerdphol, Y. Qudaih, and Y. Mitani, “Battery energy storage system
[1] K. G. Boroojeni, M. H. Amini, A. Nejadpak, S. S. Iyengar, B.
size optimization in microgrid using particle swarm optimization,” in
Hoseinzadeh, and C. L. Bak, “A theoretical bilevel control scheme for
IEEE PES Innovative Smart Grid Technologies, Europe, 2014, pp. 1–
power networks with large-scale penetration of distributed renewable
6.
resources,” in 2016 IEEE International Conference on Electro
[22] M. R. Aghamohammadi and H. Abdolahinia, “A new approach for
Information Technology (EIT), 2016, pp. 0510–0515.
optimal sizing of battery energy storage system for primary frequency
[2] Department of Energy Office of Electricity Delivery and Energy
control of islanded Microgrid,” Int. J. Electr. Power Energy Syst., vol.
Reliability. (2012). Summary Report: 2012 DOE Microgrid
54, pp. 325–333, Jan. 2014.
Workshop.[Online].Available:http://energy.gov/sites/prod/files/2012
[23] S. Bahramirad and H. Daneshi, “Optimal sizing of smart grid storage
%20Microgrid%20Workshop%20Report%2009102012.pdf. accessed
management system in a microgrid,” in 2012 IEEE PES Innovative
Nov. 1, 2015.
Smart Grid Technologies (ISGT), 2012, pp. 1–7.
[3] P. C. Loh, D. Li, Y. K. Chai, and F. Blaabjerg, “Autonomous Control
[24] S. Bahramirad, W. Reder, and A. Khodaei, “Reliability-Constrained
of Interlinking Converter With Energy Storage in Hybrid AC-DC
Optimal Sizing of Energy Storage System in a Microgrid,” IEEE Trans.
Microgrid,” IEEE Trans. Ind. Appl., vol. 49, no. 3, pp. 1374–1382, May
Smart Grid, vol. 3, no. 4, pp. 2056–2062, Dec. 2012.
2013.
[25] M. Ross, R. Hidalgo, C. Abbey, and G. Joós, “Analysis of Energy
[4] R. H. Lasseter and P. Paigi, “Microgrid: A conceptual solution,” in
Storage sizing and technologies,” in 2010 IEEE Electric Power and
Power Electronics Specialists Conference, 2004. PESC 04. 2004 IEEE
Energy Conference (EPEC), 2010, pp. 1–6.
35th Annual, 2004, vol. 6, pp. 4285–4290.
[26] M. Asensio, P. M. de Quevedo, G. Munoz-Delgado, and J. Contreras,
[5] C. Gouveia, J. Moreira, C. L. Moreira, and J. A. P. Lopes,
“Joint Distribution Network and Renewable Energy Expansion
“Coordinating Storage and Demand Response for Microgrid
Planning considering Demand Response and Energy Storage ?? Part I:
Emergency Operation,” IEEE Trans. Smart Grid, vol. 4, no. 4, pp.
Stochastic Programming Model,” IEEE Trans. Smart Grid, vol. PP, no.
1898–1908, Dec. 2013.
99, pp. 1–1, 2016.
[6] W. A. Omran, M. Kazerani, and M. M. A. Salama, “Investigation of
[27] M. Asensio, P. M. de Quevedo, G. Munoz-Delgado, and J. Contreras,
Methods for Reduction of Power Fluctuations Generated From Large
“Joint Distribution Network and Renewable Energy Expansion
Grid-Connected Photovoltaic Systems,” IEEE Trans. Energy Convers.,
Planning considering Demand Response and Energy Storage ?? Part II:
vol. 26, no. 1, pp. 318–327, Mar. 2011.
Numerical Results and considered Metrics,” IEEE Trans. Smart Grid,
[7] K. Boroojeni, M. H. Amini, A. Nejadpak, T. Dragicevic, S. S. Iyengar,
vol. PP, no. 99, pp. 1–1, 2016.
and F. Blaabjerg, “A Novel Cloud-Based Platform for Implementation
[28] T. A. Nguyen, M. L. Crow, and A. C. Elmore, “Optimal Sizing of a
of Oblivious Power Routing for Clusters of Microgrids,” IEEE Access,
Vanadium Redox Battery System for Microgrid Systems,” IEEE Trans.
vol. 5, pp. 607–619, 2017.
Sustain. Energy, vol. 6, no. 3, pp. 729–737, Jul. 2015.
[8] A. Khodaei, “Provisional Microgrids,” IEEE Trans. Smart Grid, vol. 6,
[29] X. Luo, J. Wang, M. Dooner, and J. Clarke, “Overview of current
no. 3, pp. 1107–1115, May 2015.
development in electrical energy storage technologies and the
[9] S. Gill, E. Barbour, I. A. G. Wilson, and D. Infield, “Maximising
application potential in power system operation,” Appl. Energy, vol.
revenue for non-firm distributed wind generation with energy storage
137, pp. 511–536, Jan. 2015.
in an active management scheme,” IET Renew. Power Gener., vol. 7,
[30] H. Chen, T. N. Cong, W. Yang, C. Tan, Y. Li, and Y. Ding, “Progress
no. 5, pp. 421–430, Sep. 2013.
in electrical energy storage system: A critical review,” Prog. Nat. Sci.,
[10] E. Sortomme and M. A. El-Sharkawi, “Optimal Power Flow for a
vol. 19, no. 3, pp. 291–312, Mar. 2009.
System of Microgrids with Controllable Loads and Battery Storage,” in
[31] I. Alsaidan, A. Khodaei, and W. Gao, “Determination of battery energy
2009 IEEE/PES Power Systems Conference and Exposition, 2009, pp.
storage technology and size for standalone microgrids,” in 2016 IEEE
1–5.
Power and Energy Society General Meeting (PESGM), 2016, pp. 1–5.
[11] S. Parhizi, H. Lotfi, A. Khodaei, and S. Bahramirad, “State of the Art
[32] X. Tan, Q. Li, and H. Wang, “Advances and trends of energy storage
in Research on Microgrids: A Review,” IEEE Access, vol. 3, pp. 890–
technology in Microgrid,” Int. J. Electr. Power Energy Syst., vol. 44,
925, 2015.
no. 1, pp. 179–191, Jan. 2013.
[12] J. Eyer and G. Corey, “Energy storage for the electricity grid: Benefits
[33] I. Alsaidan, A. Khodaei, and W. Gao, “Distributed energy storage
and market potential assessment guide,” Sandia Natl. Lab., vol. 20, no.
sizing for microgrid applications,” in 2016 IEEE/PES Transmission
10, p. 5, 2010.
and Distribution Conference and Exposition (T D), 2016, pp. 1–5.
[13] Q. Fu, A. Hamidi, A. Nasiri, V. Bhavaraju, S. B. Krstic, and P. Theisen,
[34] N. L. Diaz, T. Dragičević, J. C. Vasquez, and J. M. Guerrero, “Fuzzy-
“The Role of Energy Storage in a Microgrid Concept: Examining the
logic-based gain-scheduling control for state-of-charge balance of
opportunities and promise of microgrids.,” IEEE Electrification Mag.,
distributed energy storage systems for DC microgrids,” in 2014 IEEE
vol. 1, no. 2, pp. 21–29, Dec. 2013.

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2769639, IEEE
Transactions on Power Systems
13

Applied Power Electronics Conference and Exposition - APEC 2014, [57] [Online]. Available: http://industrial.alpha.com/index.php/products.
2014, pp. 2171–2176. [58] [Online]. Available: http://www.cdtechno.com/.
[35] W. Li and G. Joos, “Performance Comparison of Aggregated and [59] [Online]. Available: http://www.odysseybattery.com/documents/us-
Distributed Energy Storage Systems in a Wind Farm For Wind Power ody-tm-001_0411_000.pdf
Fluctuation Suppression,” in 2007 IEEE Power Engineering Society [60] N. Lu, M. R. Weimar, Y. V. Makarov, and C. Loutan, “An evaluation
General Meeting, 2007, pp. 1–6. of the NaS battery storage potential for providing regulation service in
[36] J. Cui, K. Li, Y. Sun, Z. Zou, and Y. Ma, “Distributed energy storage California,” in 2011 IEEE/PES Power Systems Conference and
system in wind power generation,” in 2011 4th International Exposition, 2011, pp. 1–9.
Conference on Electric Utility Deregulation and Restructuring and
Power Technologies (DRPT), 2011, pp. 1535–1540.
[37] E. Karanasios, M. Ampatzis, P. H. Nguyen, W. L. Kling, and A. van
Zwam, “A model for the estimation of the cost of use of Li-Ion batteries Ibrahim Alsaidan (S’15) received the B.S degree from Qassim
in residential storage applications integrated with PV panels,” in 2014 University, Saudi Arabia in 2008, and the M.S. degree from
49th International Universities Power Engineering Conference
(UPEC), 2014, pp. 1–6. University of Denver in 2012, where he is currently pursuing
[38] A. Bocca, A. Sassone, D. Shin, A. Macii, E. Macii, and M. Poncino, the Ph.D. degree. His research interests include modeling,
“An equation-based battery cycle life model for various battery sizing, and operating energy storage systems in distribution
chemistries,” in Very Large Scale Integration (VLSI-SoC), 2015 network.
IFIP/IEEE International Conference on, 2015, pp. 57–62.
[39] H. N. Seiger, “Effect of depth of discharge on cycle life of near-term
batteries,” presented at the 16th Intersociety Energy Conversion Amin Khodaei (SM’14) received his Ph.D. degree in electrical
Engineering Conference, 1981, pp. 102–110. engineering from the Illinois Institute of Technology, in 2010.
[40] A. F. Burke, “Cycle Life Considerations for Batteries in Electric and He was a visiting faculty (2010–2012) in the Robert W. Galvin
Hybrid Vehicles,” SAE Technical Paper, Warrendale, PA, SAE
Technical Paper 951951, Aug. 1995. Center for Electricity Innovation at IIT. He joined the
[41] L. H. Thaller, “Expected Cycle Life vs. Depth of Discharge University of Denver, Denver, CO, USA, in 2013 where he is
Relationships of Well-Behaved Single Cells and Cell Strings,” J. currently an Associate Professor and Chair of the Electrical and
Electrochem. Soc., vol. 130, no. 5, pp. 986–990, 1983. Computer Engineering Department. His research interests
[42] G. Muñoz-Delgado, J. Contreras, and J. M. Arroyo, “Joint Expansion
Planning of Distributed Generation and Distribution Networks,” IEEE include power system operation, planning, computational
Trans. Power Syst., vol. 30, no. 5, pp. 2579–2590, Sep. 2015. economics, and smart electricity grids.
[43] S. de la Torre, A. J. Conejo, and J. Contreras, “Transmission Expansion
Planning in Electricity Markets,” IEEE Trans. Power Syst., vol. 23, no. Wenzhong Gao (SM’03) received the M.S. and Ph.D. degrees
1, pp. 238–248, Feb. 2008.
[44] A. Khodaei, M. Shahidehpour, and S. Kamalinia, “Transmission in electrical and computer engineering, specializing in electric
Switching in Expansion Planning,” IEEE Trans. Power Syst., vol. 25, power engineering, from the Georgia Institute of Technology,
no. 3, pp. 1722–1733, Aug. 2010. Atlanta, GA, USA, in 1999 and 2002, respectively. He is
[45] A. J. Conejo, Ed., Decomposition techniques in mathematical currently with the Department of Electrical and Computer
programming: engineering and science applications. Berlin ; New
York: Springer, 2006. Engineering, University of Denver, Denver, CO, USA. His
[46] LONDON ECONOMICS, 2013. “Estimating the Value of Lost Load”. current research interests include renewable energy and
Briefing paper prepared for the Electricity Reliability Council of Texas, distributed generation, microgrid, smart grid, power system
Inc. by London Economics International LLC. June 17th 2013. protection, power electronics applications in power systems,
[47] A. Gabash and P. Li, “Active-Reactive Optimal Power Flow in
Distribution Networks With Embedded Generation and Battery power system modeling and simulation, and hybrid electric
Storage,” IEEE Trans. Power Syst., vol. 27, no. 4, pp. 2026–2035, Nov. propulsion systems.
2012. Dr. Gao is an Editor of the IEEE TRANSACTIONS ON
[48] A. Gabash and P. Li, “Flexible Optimal Operation of Battery Storage SUSTAINABLE ENERGY. He is the General Chair of the 48th
Systems for Energy Supply Networks,” IEEE Trans. Power Syst., vol.
28, no. 3, pp. 2788–2797, Aug. 2013. North American Power Symposium in 2016 and the IEEE
[49] A. Gabash and P. Li, “Variable reverse power flow-Part I: AR-OPF Symposium on Power Electronics and Machines in Wind
with reactive power of wind stations,” in Environment and Electrical Applications in 2012.
Engineering (EEEIC), 2015 IEEE 15th International Conference on,
2015, pp. 21–26.
[50] A. Gabash and P. Li, “Variable reverse power flow-part II: Electricity
market model and results,” in Environment and Electrical Engineering
(EEEIC), 2015 IEEE 15th International Conference on, 2015, pp. 27–
32.
[51] A. Gabash and P. Li, “Active-reactive optimal power flow for low-
voltage networks with photovoltaic distributed generation,” in 2012
IEEE International Energy Conference and Exhibition
(ENERGYCON), 2012, pp. 381–386.
[52] A. Gabash, M. E. Alkal, and P. Li, “Impact of allowed reverse active
power flow on planning PVs and BSSs in distribution networks
considering demand and EVs growth,” Power Energy Stud. Summit
PESS 2013 IEEE Stud. Branch Bielef., pp. 11–16, 2013.
[53] A. Khodaei, S. Bahramirad, and M. Shahidehpour, “Microgrid Planning
Under Uncertainty,” IEEE Trans. Power Syst., vol. 30, no. 5, pp. 2417–
2425, Sep. 2015.
[54] A. Thiele, T. Terry, and M. Epelman, “Robust linear optimization with
recourse,” Rapp. Tech., pp. 4–37, 2009.
[55] “Microgrid at Illinois Institute of Technology.” [Online]. Available:
http://iitmicrogrid.net/microgrid.aspx. [Accessed: 21-Mar-2015].
[56] “Home | ComEd’s Hourly Pricing Program.” [Online]. Available:
https://hourlypricing.comed.com/. [Accessed: 14-Apr-2015].

0885-8950 (c) 2017 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.