Received December 27, 2020, accepted January 17, 2021, date of publication January 27, 2021, date of current

version February 4, 2021.

Digital Object Identifier 10.1109/ACCESS.2021.3054860

Cycle-Life-Aware Optimal Sizing of Grid-Side

Battery Energy Storage
YUNFAN ZHANG 1 , YIFAN SU1 , ZHAOJIAN WANG 1 , (Member, IEEE),
FENG LIU 1 , (Senior Member, IEEE), AND CHENGHAO LI2
1 State Key Laboratory of Power Systems, Department of Electrical Engineering, Tsinghua University, Beijing 100084, China
2 State Grid Henan Electric Power Research Institute, Zhengzhou 450052, China
Corresponding author: Feng Liu ([email protected])
This work was supported by the HQ sponsored S&T Project of State Grid Corporation of China 5419-201924207A-0-0-00.

ABSTRACT Grid-side electrochemical battery energy storage systems (BESS) have been increasingly
deployed as a fast and flexible solution to promoting renewable energy resources penetration. However,
high investment cost and revenue risk greatly restrict its grid-scale applications. As one of the key factors
that affect investment cost, the cycle life of battery heavily depends on its charging/discharging actions during
the operation, particularly in the presence of uncertain renewable generation. In this context, it is necessary
to consider the operation-dependent cycle life of batteries in optimal BESS sizing, which imposes great
challenges to the modeling and solving of the planning problems. In this paper, we propose a novel two-level
optimal sizing model for grid-scale BESS, considering its operation under uncertainties induced by volatile
wind generation. In the lower level, a long-term chronological operation simulation of BESS is processed
with an accurate cycle life model of batteries; in the upper level, marginal economic utility analysis and BESS
size reforming are conducted to approach the optimal size of BESS. An iterative algorithm is designed to
solve the model effectively. The proposed method is verified on a modified IEEE RTS-24 system and a real
provincial power grid of China.

INDEX TERMS Battery energy storage, optimal sizing, cycle life, marginal utility analysis.

I. INTRODUCTION deployment of grid-scale batteries [3]. In this context, optimal

A. BACKGROUND planning, particularly the sizing of BESS, is key for invest-
Recent years have witnessed the proliferation of wind energy, ment viability. Since BESSs are relatively expensive for their
the most popular renewable energy resources. However, limited life span, battery lifetime assessment is necessary
the intrinsic stochastic nature of wind power imposes great when determining BESS size. Nevertheless, battery cycle life
challenges to power balancing and system secure opera- is affected by a set of stress factors, some of which are directly
tion [1]. Distinguished from other technologies, energy stor- determined by the way a grid-side battery is operated, such
age systems (ESS) can provide a fast and flexible solution to as state of charge (SoC) and the depth of discharge (DoD).
depressing power fluctuations, mitigating load peak-to-valley Thus the cycle life of BESS cannot be determined ex-ante
difference, and flattening load profile, hence facilitating bet- and is operation-dependent. Furthermore, the volatility of
ter integration of renewable generation. wind generation brings further challenges to the prediction
Among the current energy storage technologies, elec- of battery lifetime over the BESS project planning horizon.
trochemical batteries energy storage systems (BESS) have In this paper, we propose a novel sizing model to incorpo-
shown advantages in power density, energy density, dynamic rate precise non-linear BESS lifetime assessment during the
capability, ramp rate and cycle efficiency. With the recent planning horizon and develop a two-level iterative algorithm
advancements [2], grid-scale battery systems has been to solve this model effectively.
drawing constantly increasing attention. However, so far,
high-investment cost and revenue risk greatly restrict the B. LITERATURE REVIEW
To cope with uncertainties induced by volatile renew-
The associate editor coordinating the review of this manuscript and able generation, optimal BESS planning under uncertain-
approving it for publication was Behnam Mohammadi-Ivatloo . ties has been addressed in the literature of joint expansion
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
VOLUME 9, 2021 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 20179
 Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

planning of distributed generation and distribution net- are built on BESS with a given configuration, i.e., a fixed
works by scenario-based stochastic programming (SBSP) energy capacity and power rating, and focus on the battery
[4]–[6], chance-constrained stochastic programming (CCSP) system operation strategies.
[7], robust optimization [8], and distributionally robust opti- It is necessary to equally apply models accounting for
mization [9], where the uncertain demand and renewable battery cycle life and consider the effectiveness of service
generation is captured by a number of discrete representative lifetime when planning decisions are made, so that the battery
scenarios, a specific probability distribution, a continuous set can be planned and operated more economically [15]. How-
of all possible scenarios and an ambiguity set of probability ever, due to the intrinsic high nonlinearity of Peukert’s law
distributions, respectively. and the discrete logical judgments in the rainflow counting
In [4], a mixed binary integer linear programming model is algorithm [16], it is difficult to obtain an analytical expression
proposed for the optimal BESS allocation for power system for the cycle life assessment model and embed it into the
integrated with uncertain wind power generation. Benders planning optimization problem. To this end, several sim-
decomposition is applied to reduce the computational burden plified models are adopted [15], [17], [18]. Reference [17]
resulting from the scenario tree of the wind power output and considers sizing and allocation of mobile energy storage
the involving nonanticipative constraints. References [5], [6] for multi-services, where a fixed given project life in years
propose a nonlinear scenario-based stochastic programming is assumed and linear lifetime constraints are imposed on
framework for the joint planning of distribution network the battery charge exchanged with the grid. In [15] where
and energy storage, considering the impact of BESS and co-optimization of active distribution network with energy
price-dependent demand response programs. The associated storage is addressed, piecewise linearized battery lifetime
scenario-based deterministic equivalence is formulated into model based on weighted throughput method is utilized to
a mixed-integer linear program through linearization tech- formulate lifetime constraints for Lead-acid battery. In [18],
niques. Reference [7] considers the integrated planning of a linear empirical approach based on the cycling ampere-hour
electric power generation, natural gas network and storage. that the battery undergoes is adopted to estimate the battery
A two-stage iterative solution algorithm is provided to solve cycle life. Then the cost of energy storage is annualized
the chance-constrained mixed integer nonlinear program- by the yearly expected life, and a gradient-based search-
ming problem. In [8] and [9], robust optimization models ing algorithm is applied to determine the optimal sizing
are applied for BESS planning in transmission networks. The of energy storage for coping with wind power fluctuations.
existence of a feasible solution is warranted for any load and Reference [19] involves an accurate non-linear cycle life
renewable energy scenario or distribution in a pre-specified model for both first and second life segment of battery.
uncertainty set. A differential evolutionary algorithm is used to solve the
The planning results of both robust optimization and mixed-inter non-linear programming problem, thus a tradeoff
stochastic programming rely strongly on the model deployed between maximizing the profitability of energy storage with
for the uncertain parameters. Robust optimization based plan- multiple functionalities and extending its lifetime duration is
ning has to construct a proper uncertainty set, striking a achieved.
balance between robustness and economic efficiency. As for To conclude, most existing works assume a fixed service
the stochastic programming based methods, the selection of life of BESS, or a flat capacity degradation rate, or a sim-
probabilistic scenarios is nontrivial [10], with there being a plified cycle life model where battery cycle life is expressed
trade-off between computational complexity and representa- in terms of the charge-discharge power. However, estimation
tiveness of the scenario set. To get around this, techniques for of battery lifetime could be inaccurate and consequently
scenario reduction and decomposition, Monte Carlo simula- place the viability of BESS under question. To cope with
tion, and heuristic algorithms are applied to make the stochas- the nonlinearity and discrete logical judgments invoked by
tic programming model computationally tractable. But for Peukert’s law and rainflow counting algorithm in sizing
practically large systems, computational burden remains a problem, heuristic optimization techniques become candidate
challenge. solutions. Moreover, due to the difference in time scales of
Regarding battery cycle life, Peukert’s law [11] captures long-term planning and short-term operation of the system,
the non-linearity relationship between cycle life and DoD of dependence of battery cycle life on its operation strategies
battery, and has been widely used in battery storage involved imposes heavy computational burden on the planning deci-
system designs [3], [12]–[14]. Reference [3] offers mathe- sions, especially when available wind generation uncertain-
matical models of battery lifetime assessment by applying ties exist over the planning horizon.
Peukert’s Law and provides co-optimization bidding strate-
gies balancing battery degradation and profitability in power
market. In [12] and [13], energy management and economic C. CONTRIBUTION AND ORGANIZATION
scheduling of micro-grid with BESS are presented, respec- 1) CONTRIBUTION
tively, with the consideration of battery lifetime. In [14], To the best of our knowledge, consideration for dependence
a droop control based cycling strategy of BESS is proposed to of BESS lifetime on the operation under uncertainties over
reduce battery degradation costs. The aforementioned works the planning horizon has not been included in the existing

20180 VOLUME 9, 2021

Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

literature on grid-side BESS sizing. In this paper, we propose on dual shadow prices. It follows a size reforming of BESS
a novel optimal BESS sizing model with accurate cycle life to enhance economic efficiency and improve the planning
assessment under wind generation uncertainties, and an effec- decision. The proposed solution approach can achieve a sat-
tive algorithm to solve it. Specifically, we consider optimal isfactory balance between the accuracy of the results and the
sizing of both energy capacity and power rating of BESS. computing time as demonstrated in the numerical tests using
To deal with the lifetime of BESS, we iteratively conduct the data of IEEE system and a real provincial power grid of
chronological simulations throughout the life-cycle of BESS China.
and implement size reforming on the energy capacity and
power rating. The main contributions of this paper are as 2) ORGANIZATION
follows: The rest of the paper is organized as follows. The opti-
(1) Modeling. We propose a two-level optimal BESS siz- mal sizing model of grid-side BESS is stated in Section II.
ing model to bridge the yearly planning and hourly operation Section III introduces the chronological operation simulation
of BESS, as shown in Fig.1. The upper-level problem is to on the BESS-integrated power system, where an accurate
decide the optimal power and energy capacity of the BESS cycle life model of energy storage battery is considered.
with a given lifespan of BESS. The lower-level is to simu- Section IV presents marginal utility analysis on the BESS
late the yearly chronological operation of the power system, allocation plan and conducts size reforming. Case studies
including hourly BESS regulation under multiple scenar- on IEEE system and real-scale provincial power grid are
ios of renewable generation. Compared with existing works introduced in Section V. Finally, Section VI concludes the
[4]–[9], the proposed model does not rely on specific paper.
modeling deployed for uncertain wind generations and
accommodates for wide-ranging applications. Compared II. OPTIMAL SIZING MODEL OF GRID-SIDE BESS
with existing works [15], [17], [18], the proposed model eval- In this paper, grid-side BESS is allocated to mitigate wind
uates cycle-life degradation more accurately since a precise power fluctuations and reduce wind power spillage. Grid-side
Peukert’s Law based cycle life model is utilized for battery investors are to determine the optimal BESS sizing (i.e. the
lifetime estimation. energy capacity and power rating of BESS) with the goal of
(2) Solution Approach. We propose an iterative size economical optimality of the entire grid system. Considering
reforming procedure to coordinate the yearly planning and the high uncertainty and volatility of wind power, a scenario-
hourly operation of BESS. Specifically, we evaluate the cycle based stochastic programming method is employed to derive
life of BESS after each-round of hourly chronological opera- an initial BESS allocation strategy.
tion simulation, and then correct the cycle life of BESS. After-
ward, with the corrected cycle life, marginal utility analysis A. OBJECTIVE FUNCTION
on power and energy capacity of BESS is conducted based The objective function of the optimization problem is:
 
min π BES + π wc + π ls + π g (1)
PBES ,EBES

where the decision variables, PBES and EBES , denote the

power rating and energy capacity of BESS, respectively. The
objective in (1) consists of investment costs of BESS π BES ,
production costs of conventional generators π g , penalty of
curtailments on wind power π wc , as well as the penalty of
load shedding π ls . The aforementioned cost items will be
explained as follows:

1) INVESTMENT COST OF BESS

π BES denotes the total investment cost of BESS that being
divided into each year, with a specific formation of

π BES = AF C ep PBES + C ee EBES

(2)
ep
I (1 + I )Y
AF = ep . (3)
(1 + I )Y − 1
In (2), C ep and C ee are the unit investment cost of the power
rating and energy capacity of BESS, respectively. The BESS
total fixed investment cost (C ep PBES +C ee EBES ) is converted
to an annual value by multiplying the annuity factor (AF) AF.
FIGURE 1. Holistic structure of the model. This factor is defined in (3) where I denotes the discount rate

VOLUME 9, 2021 20181

 Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

g,min g,max
of grid-side BESS equipment, and Y ep denotes the anticipated where pi and pi represent the minimal and maximal
battery shelf life in years. active output of conventional generator i, respectively; and
gr
pi denotes the ramping rate of conventional generator i.
2) COST OF WIND CURTAILMENT
To promote renewable energy accommodation, penalties on 2) RESTRAINT CONDITIONS OF WIND OUTPUT
the curtailment of wind are considered. π wc denotes the The wind power consumption should be limited by its avail-
annual cost of wind curtailment, with a specific formation of able maximal wind output:
wg wg,max
, ∀s ∈ S, t ∈ T , i ∈ W.
 
wg,max wg 0 ≤ ps,t,i ≤ ps,t,i (9)
XX X
π wc = 365 ωs Ciwc ps,t,i − ps,t,i (4)
s∈S t∈T i∈W
3) RESTRAINT CONDITIONS OF LOAD SHEDDING
where Ciwc
is the penalty coefficient for unit curtailed wind The actual active power consumption of bus i at time slot t
wg,max wg
power of wind farm i. ps,t,i and ps,t,i denote the hourly under scenario s should be limited by its active power
available wind power and the hourly consumed wind power demand:
of wind farm i at time slot t in scenario s. ωs is the discrete
probability of scenario s. S represents the set of selected s,t,i ,
0 ≤ pls,t,i ≤ ple ∀s ∈ S, t ∈ T , i ∈ N . (10)
scenarios, T denotes the set of time slots in a scenario, and
W denotes the set of wind farms. 4) RESTRAINT CONDITIONS OF BATTERY ENERGY STORAGE
The charging and discharging of BESS should fulfill the
3) COST OF LOAD SHEDDING following conditions:
π ls denotes the annual cost of load shedding, with a specific s,t ≤ PBES ,
0 ≤ pch ∀s ∈ S, t ∈ T (11)
formation of
XXX   0≤ pdc
s,t ≤ PBES , ∀s ∈ S, t ∈ T (12)
π ls = 365 ωs Cils ple
s,t,i − p l
s,t,i (5) pch · pdc
= 0, ∀s ∈ S, t ∈ T (13)
s,t s,t
s∈S t∈T i∈N
µ · EBES ≤ Es,t ≤ µ · EBES , ∀s ∈ S, t ∈ T (14)
where Cils denotes the loss of unit load shedding at bus i, pdc
s,t
ple l
s,t,i and ps,t,i denote the active power demand and the active
Es,t = Es,t−1 + ηch · pch
s,t − , ∀s ∈ S, t ∈ T
ηdc
power consumption of bus i at time slot t in scenario s,
(15)
respectively. N denotes the set of buses in the grid.
E0 = E|T | (16)
4) COST OF CONVENTIONAL GENERATORS where pch dc
s,t , ps,t and Es,t are the charge power, discharge
πg denotes the annual generation cost of conventional gener- power and the energy of BESS at time slot t in scenario s,
ators, with a specific formation of respectively. Constraints (11)-(12) imply that the charging
XXX  g g
 and discharging power of BESS are limited by its power
π g = 365 ωs Ci2 (ps,t,i )2 + Ci1 ps,t,i + Ci0 (6) rating. Constraint (13) is the complementarity constraint of
s∈S t∈T i∈G
BESS device. ηch and ηdc are the charging and discharging
where
g
denotes the active power output of conventional
ps,t,i efficiencies of BESS. µ and µ are the upper and lower limits
generator i at time slot t under scenario s. G is the set of all of the SoC of BESS, respectively. Constraint (14) is the
conventional generators. Ci0 , Ci1 , Ci2 are the cost coefficients restraint condition of energy storage capacity. Constraints
of conventional generator i. (15) and (16) guarantee the energy balance of BESS where
|T | denotes the total number of time slots.
B. RESTRAINT CONDITIONS
This model is based on a direct current (DC) power flow [20] 5) RESTRAINT CONDITIONS OF POWER NETWORK
and leaves out the reactive power of BESS and network. This Power balance constraint (17) ensures the sum of all types of
simplification is adopted in many existing studies [3]–[9], generation is equal to total power consumption at each time
[15], [18], [19], since BESS’s support on real power balance interval.
X g X wg
is the main focus. All constraints are listed as follows. ps,t,i + ps,t,i + pdc ch
s,t = ps,t
i∈G i∈W
1) RESTRAINT CONDITIONS OF CONVENTIONAL X
+ pls,t,i , ∀s ∈ S, t ∈ T (17)
GENERATORS
i∈N
The active output of conventional generators should fulfill the
following operation conditions: Let pline
s,t,l denote the power flow on transmission line l at time
g,min g g,max
t in scenario s. For any scenario s ∈ S, time t ∈ T and any
pi ≤ ps,t,i ≤ pi ,
∀s ∈ S, t ∈ T , i ∈ G transmission line l ∈ L:
(7) pline
X g g
GSFl,i · ps,t,i +
X wg wg
GSFl,i · ps,t,i
g g gr s,t,l =
|ps,t,i − ps,t−1,i | ≤ pi , ∀s ∈ S, t ∈ T , i ∈ G (8) i∈G i∈W

20182 VOLUME 9, 2021

Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

X  
− GSFl,i · pls,t,i + GSFBES
l · pdc ch
s,t − ps,t (18) A. CYCLE-LIFE MODEL OF BESS
i∈N Electrochemical batteries have limited cycle life because of
the fading of active materials caused by the charging and
where GSFl,i is the generation shift factor (GSF) to line l
g wg discharging cycles. In this paper, we focus on lithium-ion
from bus i; GSFl,i the GSF to line l from generator i; GSFl,i
batteries, for which the DoD is the most important operational
the GSF to line l from wind farm i; GSFBES
l the GSF to line l
factor in battery degradation and cycle life assessment. The
from the BESS; L is the set of transmission lines in the power
lithium-ion BESS can perform a certain number of cycles at
grid. Constraint (19) enforces the bidirectional power flow
a specific DoD [3]:
limits of the transmission lines.
Ndfail = N100
fail
· d −kP (21)
−pline,max
l ≤ pline line,max
s,t,l ≤ pl , ∀s ∈ S, t ∈ T , l ∈ L. (19)
where Ndfail and fail
N100 are the maximum number of
C. FINDING INITIAL BESS ALLOCATION STRATEGY charge-discharge cycles at a specific DoD of d and a DoD
To determine an initial allocation strategy of grid-side BESS of 100% before the battery’s failure, respectively. kP and
fail are positive inherent parameters that can be provided by
N100
considering the uncertainty of wind output, first, we select
eight typical scenarios consisting of four daily wind curves the battery manufacturer. By keeping the loss of cycle life a
eq
with the most significant reverse peak regulation characteris- constant, the equivalent 100%-DoD cycle number n100 of nd
tics, of spring, summer, autumn and winter, respectively, and cycles at DoD of d is derived as:
eq
two daily load curves of working day and non-working day, n100 = nd · d kP , (22)
respectively. Then a BESS-concerned SBSP is established
as (20), where the typical scenes are intuitively assigned implying that a deeper DoD, a smaller kP or more frequent
equal probabilities. cycles (i.e., a larger nd ) gives rise to a shorter service life of
  the battery.
min π BES + π wc + π ls + π g A full cycle consists of one charging half cycle and one
PBES ,EBES discharging half cycle with the same DoD. In reality, however,
subject to (7)-(19) (20) two contiguous charge/discharge sessions might be asym-
metric. In this regard, the rainflow counting method [16] is
Note that the storage operational cost is neglected in (20)
applied to an SoC profile to identify cycles. Consider a daily
as the grid-side BESS is owned by the power grid com-
SoC profile SoCday , the rainflow algorithm outputs all the
pany itself. The bi-linear constraint (13) makes the prob-
cycle depth as:
lem (20) strongly non-convex and difficult to solve. Usually,    
the complementarity constraint (13) can be precisely lin- d half , d full = Rainflow SoCday (23)
earized by introducing auxiliary binary variables [21]. More-
over, the (epigraph of) quadratic generation cost function π g where d half is the vector of half cycles and d full is the vector
can be piecewise linearized. Thus (20) is converted into a of full cycles. For the specific steps of rainflow algorithm,
mixed integer linear programming (MILP). Then the optimal see [16]. Thus a battery’s daily equivalent 100%-DoD cycle
solution to problem (20), (P0BES , EBES
0 ), is set as the initial number is derived:
BESS allocation strategy. eq,day
X  kP X  kP
n100 = 0.5 djhalf + djfull . (24)
j j
III. CHRONOLOGICAL OPERATION SIMULATION OF THE
Then the realistic yearly service life of BESS is
BESS-INTEGRATED POWER SYSTEM
By solving the multi-scenario stochastic programming (20), N fail
Y rl = P365 100eq,day (25)
an initial grid-side BESS allocation strategy (P0BES , EBES
0 ) is
day=1 n100
obtained. However, this configuration scheme is based on
a limited number of typical scenes, not fully considering which may be different from the expected one Y ep . This
the impact of wind uncertainties. Moreover, this planning discordance may affect the economical optimality of BESS
assumes that the BESS has a fixed lifetime Y ep , which may allocation strategy. Next, we will show how realistic service
be not valid if the BESS has more or less frequent cycling life is obtained through all-year chronological operation sim-
than expected. As a matter of fact, the real lifetime of battery ulation.
is sensitive to operation. In this regard, we run a life-cycle
chronological operation simulation (COS) on the given con- B. CHRONOLOGICAL OPERATION SIMULATION
figuration scheme to estimate the realistic service life of 1) STEP 1: DATA PREPARATION
BESS, and then adjust the allocation plan through marginal First, we prepare data for the long-term BESS-concerned
benefit analysis that will be introduced in Section IV. This chronological operation simulation: the wind output profiles,
design is economically optimal while addressing the uncer- load forecasts, parameters of conventional generators and
tainties of wind power and the unique degradation character- states of critical tie-lines are needed. The annual chronologi-
istics of BESS. cal wind output profiles can be obtained by utilizing historical

VOLUME 9, 2021 20183

 Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

measured data, or generated according to meteorological data imizes system utility, especially when the MR and MC are
and the output characteristics of wind turbine generators. evaluated under the realistic cycle life of BESS.
Note that possible future expansion on system load during
the simulated period can be taken into account. A. MARGINAL REVENUE OF BESS
An annualized marginal grid-side benefit provided by the
2) STEP 2: DAILY OPERATION MODELING AND EVOLVING BESS is calculated based on the annual COS results, as in
Given the BESS allocation plan, year around operations are (29) and (30).
simulated by consecutively running daily simulations to ana- X365 X  
ch,day dc,day
lyze the interactions between the renewable generations and MRPBES = φt + φt (29)
day=1 t∈T
grid-side BESS during the whole-life-cycle of BESS. To this X365 X 
day

day
end, a daily COS model is formulated to find the optimal MREBES = µ · λt − µ · λt (30)
day=1 t∈T
hourly operation of BESS which is characterized by the daily
For the BESS with current sizing which is PkBES , EBES k


SoC profile SoCday , and that of the conventional generators. ,
Note that simulations on days can be carried out in parallel to MRPBES is the annualized MR of its power capacity and
accelerate the annual COS. MREBES is the annualized MR of its energy capacity. Namely,
By reducing (20) into one scenario (i.e., one specific additional per unit of capacity of power (energy) can reduce
day) and substituting
PBES and EBES by the given allocation the yearly costs of system by MRPBES (MREBES ) when addi-
scheme PkBES , EBESk (when k = 0 it denotes the initial tional BESS investment cost is not considered. Note that this
allocation plan), a BESS-integrated DC optimal power flow marginal contribution of BESS to power system is evaluated
problem is formulated as the daily operation model of COS. fully considering the fluctuations of wind power throughout
Note that in the daily operation model, the complementarity the year.
constraint of BESS, equality (13), is relaxed to simplify the
problem into a linear program to accelerate the computation. B. MARGINAL COST OF BESS
Though this may give rise to the situation that charging and The MC of BESS is calculated based on its life span as in (31)
discharging occur simultaneously, its influence on BESS’s and (32)
lifetime is quantified by recording the daily SoC profile.
Moreover, the duality of the linear BESS-integrated DC I (1 + I )Y
P
MCBES (Y ) = C ep (31)
optimal power flow problem is solved, for the dual variables (1 + I )Y − 1
of constraint (11), (12) and (14) are needed to calculate the I (1 + I )Y
marginal revenue (MR) of the BESS, as further explained in
E
MCBES (Y ) = C ee (32)
(1 + I )Y − 1
the Subsection IV. Notations are taken as follows:
where C ep and C ee are investment cost for per-unit of the
day day day
µ · EBES
k
≤ Et ≤ µ · EBES
k
(dual: λt , λt ) (26) capacity of power and energy of BESS, respectively, and are
ch,day ch,day converted into annual values. If Y takes the value of a fixed
pt ≤ PkBES (dual: φt ) (27)
dc,day dc,day
expected cycle life Y ep , then the capacity of power or energy
pt ≤ PBES (dual: φt
k
) (28) has a fixed MC. If Y takes the value of a realistic cycle life
Y rl , then the MC of BESS configuration can be varying and
3) STEP 3: RESULT OUTPUT operation-dependent. Note that only one-time investment cost
After the modeling and solving step, results from all parallel is considered here. Since the grid-side BESS is controlled
daily processes are combined as an annual outturn. First, by operators, operation costs caused by unmet discharged or
a realistic service life of BESS in years, Y rl , is obtained charged energy does not exist. Additional operational cost,
through (23)-(25). Moreover, values of dual multipliers λ, λ, e.g. damage to the state of health of BESS caused by over-
φ ch and φ dc are recorded. charge/-discharge, can be indirectly considered through the
operation-dependent variable life span Y rl .
IV. MARGINAL UTILITY ANALYSIS AND SIZE
REFORMING OF BESS C. SIZE REFORMING OF BESS
In this section, given the results from the chronological The normalized MU of the capacity of power and energy of
operation simulation, the optimality of the BESS allocation BESS can be calculated by (33) and (34), respectively. If the
is evaluated from the perspective of marginal utility (MU), corresponding MU is greater than zero, additional capacity
which is defined as the difference between the MR and the allocation to the current scheme is economically beneficial to
marginal cost (MC). In economic theory, MU analysis is an the system, till the point MU decreasing to zero. Otherwise,
effective tool to help companies determine the optimal yield a reduction on the current configuration of BESS is better, till
of goods. A firm should expand production until the point the point MU increasing to 0.
where MC is equal to MR. Motivated by this, the
relationship between the MR and MC of the power/energy P MRPBES − MCBES
P (Y rl )
MUBES = (33)
max MRPBES , MCBES
P (Y rl )


capacity of BESS helps to find the optimal sizing that max-

20184 VOLUME 9, 2021

Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

E
E (Y rl )
MREBES − MCBES
MUBES = (34)
max MREBES , MCBES
E (Y rl )

Therefore, if |MUBES P | and |MU E | are both less than a

BES
pre-set precision  > 0, we regard the current allocation plan
PkBES , EBES
k


as optimal. Otherwise,

we do size reforming on
the current scheme PkBES , EBESk as in (35) and (36).
 
BES = PBES · 1 + α · MUBES
Pk+1 k P
(35)
 
k+1
EBES k
= EBES · 1 + α · MUBES
E
(36)
k here denotes the iteration round of size reforming and
α > 0 is the step size which can be a fixed value or vary
with the step k. The adjusted configuration plan
 is denoted 
by Pk+1 BES , E k+1
BES . Then annual COS on the P k+1
BES , E k+1
BES -
integrated power system is needed, for further marginal utility
analysis on this configuration.
The block diagram for the proposed optimal BESS sizing
method based on simulation-based marginal utility analysis
is illustrated in Fig.2.

V. CASE STUDY
In this section, we test the proposed annual COS and MU
analysis based grid-side BESS allocation strategy. Simula-
tions are carried out on a modified IEEE RTS-24 buses system
and a provincial system of China, with MATLAB on a laptop
with Intel(R) Core(TM) i5-5200U 2.20GHz CPU and 4GB of
RAM.

A. SETUP
We do some modifications to the IEEE RTS-24 system:
1) A wind farm with a capacity of 300MW is located at bus
1. The BESS candidate position is also bus 1.
2) The active power load of the system is modified as
in Table.1.
3) The power limits of transmission lines (1,2), (1,3) and
(1,5) are set as 80MW, 50MW, 80MW, respectively.

TABLE 1. The proportion of each load.

FIGURE 2. The specific steps of optimal BESS sizing based on

simulation-based marginal utility analysis.

Limited by the capacity of sent-out gateway as well as

the reverse peak regulation characteristics of wind power,
the wind farm at bus 1 suffers from wind curtailment. To miti-
gate wind power fluctuations and reduce wind power spillage,
grid owner plans to allocate a battery energy storage system
The modified IEEE RTS-24 system is shown in Fig.3. nearby, i.e., at bus 1.
Parameters about line impedance, line capacity and conven- Two typical daily load profiles are considered: working day
tional generators can be found in [22]. For cost coefficients of and non-working day, as shown in Fig.4(a). The peak load
conventional generators, see Matpower CASE24_IEEE_RTS of the system on the working day is 3165.6MW, and that
data file [23]. of the non-working day is 2648MW. The wind power data

VOLUME 9, 2021 20185

 Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

FIGURE 4. (a) Two typical daily total load profiles. (b) Typical wind output
curves of four seasons.

TABLE 3. Cost of modified IEEE RTS-24 system without BESS (uniform

annual value, unit: million USD per year).
FIGURE 3. Modified IEEE RTS-24 system.

come from 2012 NREL position (116.6W◦ ,36.9N◦ ) and we

scale it to fit the capacity of 300MW. Fig.4(b) shows four
representative daily wind curves with the most significant
reverse peak regulation characteristics of spring, summer,
autumn and winter, respectively. Wind generation is assumed
to have zero marginal cost and is hence free.
The BESS-related parameters are given in Table.2.

TABLE 2. The BESS-related parameters.

TABLE 4. Cost of modified IEEE RTS-24 system with initial BESS plan
(uniform annual value, unit: million USD per year).

B. SIMULATION RESULTS modified IEEE
RTS-24 system with the BESS allocation of
1) INITIAL BESS ALLOCATION BASED ON SBSP P0BES , EBES
0 is shown in Table.4. With this BESS configu-
First, we solve the multi-period DC-OPF model for the cost ration, the annual comprehensive cost is reduced by 2.84%,
of the modified IEEE RTS-24 system without grid-side BESS from 481.55 million USD to 467.87 million USD. This is
devices under 8 typical scenarios. The cost of system without because the integration of BESS increases short-term system
BESS is shown in Table.3. Wind spillage and curtailment operation flexibility, permitting greater participation of wind
occur in the scene of spring and summer, during midnight. and reducing the generation cost of conventional generators
Consequently, the system pays for an annual penalty cost and penalty on wind curtailment. Fig.5 shows the wind con-
of 24.74 million USD, contributing to 5.14% of the annual sumption curve with and without BESS adjacent to the wind
total costs. farm, as well as the charge/discharge profiles of the BESS.
Then, we solve the scenario-based stochastic program- It can be seen from Fig.5 that when the expected output of
ming (20) and obtain an initial BESS configuration: P0BES a wind farm is large, it is often accompanied by charging
is 97.87MW and EBES 0 is 519.35MW.h. Cost of the of energy storage to reduce wind abandonment. When the

20186 VOLUME 9, 2021

Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

FIGURE 5. Wind power without and with BESS. (a) Working day of Spring.
(b) Working day of summer.

expected output of wind farm is small, BESS will release

energy to reduce the fluctuation of the joint wind farm.

2) MARGINAL UTILITY ANALYSIS AND SIZE REFORMING

Given the initial BESS allocation plan, we analyze the
marginal utility of power rating P0BES which is 97.87MW and
energy capacity EBES0 which is 519.35MW.h, through annual
chronological operation simulation. The annual COS and size
reforming process is iteratively done to find the configuration
that has both zero MUBESP and zero MU E . The final config-
BES
uration result would be optimal concerning wind fluctuations
throughout the year and operation-dependent battery cycle
life. FIGURE 6. (a) Power rating reforming of BESS. (b) Energy capacity
Size reforming on power rating and energy capacity are reforming of BESS. (c) Realistic and expected life of BESS.
shown in Fig.6 (a) and (b), respectively. Note that results in
these two subfigures come from the same iterative process,
since EBES and PBES are two configuration parameters cou-
pled in one system and consequently cannot be analyzed sep- 3) VERIFICATION OF OPTIMALITY
arately. After 14 rounds of COS and size reforming, the final Theoretically, the simulation-based marginal utility analysis
BESS configuration plan is as follows: P14 BES is 123.27MW, cannot characterize the sensitivity relationship between all
with MCBESP (Y rl ) being 7762.15$, MRP
BES being 7939.32$, values of continuous decision variables PBES , EBES and the
and the normalized marginal utility of power rating MUBES P
objective function. In this subsection, we verify that the pro-
being 2.23% which is smaller than the pre-set tolerance posed method achieves an error-bounded configuration plan
 = 5%; EBES 14 is 465.11MW.h, with MC E (Y rl ) being
BES within 14 rounds of annual COS and size reforming iterations.
31048.62$, MREBES beging 30124.35$, and the normalized First, we discretize the continuous configuration parameters
marginal utility of energy capacity being 2.98%. Realistic PBES , EBES into a two-dimensional grid with adaptive pixels.
BESS life derived through an accurate cycle life model is Then we run annual COS on each pixel and map the normal-
presented in Fig.6(c), indicating that the anticipated service ized MU on this grid.
life of BES may deviate from its real life a lot. The range of PBES is [97,128]MW and the range of EBES
Fig.6(a) also shows that the MR of the BESS power rat- is [425,522]MW.h. In order to alleviate computation burden
ing decreases with increasing power rating configuration. as well as ensure the accuracy of the contour map, the size of
Similarly, Fig.6(b) indicates that the MR of BESS energy pixels is varying and adaptive to local steepness. The contour
capacity increases with decreasing battery capacity. These are map of MUBES P is shown in Fig.7(a) and that of MUBES E is
coincident with the theoretical result in [24] that the marginal depicted in Fig.7(b). In addition, the trajectory of size forming
value from storage (without considering investment cost) is is given, showing how the initial configuration is adjusted to
decreasing with storage size. the point within pre-set tolerance  = 5%.

VOLUME 9, 2021 20187

 Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

FIGURE 9. Typical output curves of wind farms in four seasons:

(a) summer, (b) winter, (c) spring, and (d) autumn.

P . (b) Contour map of MU E .

FIGURE 7. (a) Contour map of MUBES BES

FIGURE 10. (a) Power rating reforming of BESS. (b) Energy capacity
reforming of BESS.

FIGURE 8. A provincial power grid of China employed as case study. The operation cost includes generation cost and penalty on wind
black circles are buses of 220 kV and over, the green squares are the wind curtailment and load shedding. When applying SBSP to
farms and the red triangle denotes the candidate location of BESS.
BESS sizing problem, there may be a big deviation between
the expected life of BESS and the realistic one. Even if the
anticipated life is accurate, such as 8 years in this case,
Moreover, we compare the proposed method with the the configuration plan derived through SBSP leads to higher
most frequently used scenario-based stochastic programming system costs, since it fails to evaluate the revenue of BESS
in Table.5, from the perspective of service life and total under hourly fluctuations of renewable generation during the
system cost during the entire cycle life of BESS. Here the whole life cycle.

20188 VOLUME 9, 2021

Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

TABLE 5. Cost of modified IEEE RTS-24 system with different BESS configurations (uniform annual value, unit: million USD per year).

TABLE 6. Normalized typical daily total load profiles. non-working day of autumn (Scen 8). Normalized typical
daily load profiles of the system are given in Table.6. The
base value is selected as the maximum total power load of
system in 2019, which is 6.4097 × 104 MW. Typical curves of
the 82 wind farms are illustrated in Fig.9.
The initial BES allocation plan is as follows: P0BES is
113.34.87MW and EBES 0 is 222.58.35MW.h, while the nor-
malized marginal utility of this configuration is −40.86%
and −9.23%, respectively. This indicates that the initial
allocation through SBSP leads to an over investment on
BESS, especially the power rating. After 11 rounds of iter-
ations, the reformed BES size is: P11 BES is 55.60MW, EBES
11
P E
is 128.64MW.h with MUBES is −2.93%, MUBES is −4.36%
and the realistic service life is Y rl = 10.96. The iterative size
reforming process is shown in Fig.10.

VI. CONCLUSION
In this paper, we have proposed an optimal BESS sizing
method considering the operation-dependent cycle life of
batteries. It is showed that our two-level model can more
accurately estimate the cycle-life deregulation of batteries
and the revenue of BESS under hourly fluctuation of uncer-
tain renewable generation, hence provides more reasonable
sizing strategies. The tests on IEEE benchmark system and a
real power grid demonstrate the effectiveness and practicality
of the proposed method.
The scope of this work was limited to the sizing of battery
energy storage in consideration of battery cycle-life. Installa-
tion location is not optimized. Therefore, it is recommended
that future studies related to BESS allocation planning should
C. VERIFICATION ON A PROVINCIAL SYSTEM OF CHINA consider the decision on installation location.
To demonstrate the scalability and efficiency of the proposed
framework on practical size system, a provincial power sys- REFERENCES
tem of China is adopted for case study, as shown in Fig.8. [1] A. Sturt and G. Strbac, ‘‘Efficient stochastic scheduling for simulation of
There are 398 buses of 220kV and over, 594 branches, wind-integrated power systems,’’ IEEE Trans. Power Syst., vol. 27, no. 1,
147 conventional generators and 82 wind farms in the system. pp. 323–334, Feb. 2012.
[2] R. H. Byrne, T. A. Nguyen, D. A. Copp, B. R. Chalamala, and I. Gyuk,
Power flows through 220kV and over transmission lines are ‘‘Energy management and optimization methods for grid energy storage
considered. Locations of wind farms are illustrated in Fig.8 systems,’’ IEEE Access, vol. 6, pp. 13231–13260, 2018.
with green square symbols. One centralized battery energy [3] G. He, Q. Chen, C. Kang, P. Pinson, and Q. Xia, ‘‘Optimal bidding
strategy of battery storage in power markets considering performance-
storage system is to be built with a connection point labeled based regulation and battery cycle life,’’ IEEE Trans. Smart Grid, vol. 7,
by the red triangle symbol in Fig.8. We run yearly COS no. 5, pp. 2359–2367, Sep. 2016.
based on real load and wind power data of the grid from 2019. [4] P. Xiong and C. Singh, ‘‘Optimal planning of storage in power systems
integrated with wind power generation,’’ IEEE Trans. Sustain. Energy,
We select 8 representative scenarios, which are work- vol. 7, no. 1, pp. 232–240, Jan. 2016.
ing day of summer (Scen 1), non-working day of summer [5] M. Asensio, P. Meneses de Quevedo, G. Munoz-Delgado, and J. Contreras,
(Scen 2), working day of winter (Scen 3), non-working day of ‘‘Joint distribution network and renewable energy expansion planning
considering demand response and energy storage—Part I: Stochastic pro-
winter (Scen 4), working day of spring (Scen 5), non-working gramming model,’’ IEEE Trans. Smart Grid, vol. 9, no. 2, pp. 655–666,
day of spring (Scen 6), working day of autumn (Scen 7) and Mar. 2018.

VOLUME 9, 2021 20189

 Y. Zhang et al.: Cycle-Life-Aware Optimal Sizing of Grid-Side Battery Energy Storage

[6] M. Asensio, P. Meneses de Quevedo, G. Mu noz-Delgado, and J. Contreras, YUNFAN ZHANG received the B.Sc. degree in
‘‘Joint distribution network and renewable energy expansion planning con- electrical engineering from Tsinghua University,
sidering demand response and energy storage—Part II: Numerical results,’’ Beijing, China, in 2018, where she is currently
IEEE Trans. Smart Grid, vol. 9, no. 2, pp. 667–675, Mar. 2018. pursuing the Ph.D. degree. Her research interests
[7] B. Odetayo, M. Kazemi, J. MacCormack, W. D. Rosehart, H. Zareipour, include applied optimization, energy economics,
and A. Reza Seifi, ‘‘A chance constrained programming approach to the game theory, and their applications.
integrated planning of electric power generation, natural gas network
and storage,’’ IEEE Trans. Power Syst., vol. 33, no. 6, pp. 6883–6893,
Nov. 2018.
[8] R. A. Jabr, I. Dzafic, and B. C. Pal, ‘‘Robust optimization of storage
investment on transmission networks,’’ IEEE Trans. Power Syst., vol. 30,
no. 1, pp. 531–539, Jan. 2015. YIFAN SU received the B.Sc. degree in electri-
[9] L. Yang, R. Xei, W. Wei, C. Sun, and S. Mei, ‘‘Coordinated planning of
cal engineering from Tsinghua University, Beijing,
storage unit in a remote wind farm and grid connection line: A distribu-
China, in 2019, where he is currently pursuing the
tionally robust optimization approach,’’ in Proc. IEEE Innov. Smart Grid
Technol. Asia (ISGT Asia), May 2019, pp. 424–428. Ph.D. degree. His research interests include dis-
[10] I. J. Scott, P. M. S. Carvalho, A. Botterud, and C. A. Silva, ‘‘Clustering tributed optimization and robust dispatch in smart
representative days for power systems generation expansion planning: grid.
Capturing the effects of variable renewables and energy storage,’’ Appl.
Energy, vol. 253, Nov. 2019, Art. no. 113603.
[11] R. Rao, S. Vrudhula, and D. N. Rakhmatov, ‘‘Battery modeling for energy-
aware system design,’’ Computer, vol. 36, no. 12, pp. 77–87, Dec. 2003.
[12] D. Tran and A. M. Khambadkone, ‘‘Energy management for lifetime
extension of energy storage system in micro-grid applications,’’ IEEE ZHAOJIAN WANG (Member, IEEE) received
Trans. Smart Grid, vol. 4, no. 3, pp. 1289–1296, Sep. 2013. the B.Sc. degree from Tianjin University, Tian-
[13] C. Liu, X. Wang, X. Wu, and J. Guo, ‘‘Economic scheduling model of jin, China, in 2013, and the Ph.D. degree from
microgrid considering the lifetime of batteries,’’ IET Gener., Transmiss. Tsinghua University, Beijing, China, in 2018.
Distrib., vol. 11, no. 3, pp. 759–767, Feb. 2017. From 2016 to 2017, he was a Visiting Ph.D. Stu-
[14] J.-O. Lee, Y.-S. Kim, T.-H. Kim, and S.-I. Moon, ‘‘Novel droop control dent at California Institute of Technology, CA,
of battery energy storage systems based on battery degradation cost in USA. From 2018 to 2020, he was a Postdoctoral
islanded DC microgrids,’’ IEEE Access, vol. 8, pp. 119337–119345, 2020. Scholar with Tsinghua University. His research
[15] X. Dong, Z. Yuying, and J. Tong, ‘‘Planning-operation co-optimization interests include power system stability analysis,
model of active distribution network with energy storage considering the distributed control, and microgrid planning.
lifetime of batteries,’’ IEEE Access, vol. 6, pp. 59822–59832, 2018.
[16] S. D. Downing and D. F. Socie, ‘‘Simple rainflow counting algorithms,’’
Int. J. Fatigue, vol. 1, no. 1, pp. 31–40, Jan. 1982. FENG LIU (Senior Member, IEEE) received the
[17] H. Abdeltawab and Y. A.-R.-I. Mohamed, ‘‘Mobile energy storage sizing B.Sc. and Ph.D. degrees in electrical engineer-
and allocation for multi-services in power distribution systems,’’ IEEE ing from Tsinghua University, Beijing, China,
Access, vol. 7, pp. 176613–176623, 2019. in 1999 and 2004, respectively. From 2015 to
[18] Y. Liu, W. Du, L. Xiao, H. Wang, and S. Bu, ‘‘Sizing energy stor- 2016, he was a Visiting Associate with the Cal-
age based on a life-cycle saving dispatch strategy to support frequency ifornia Institute of Technology, CA, USA. He is
stability of an isolated system with wind farms,’’ IEEE Access, vol. 7,
currently an Associate Professor with Tsinghua
pp. 166329–166336, 2019.
University. He is the author or coauthor of more
[19] Y. Zhang, Y. Xu, H. Yang, Z. Y. Dong, and R. Zhang, ‘‘Optimal whole-
life-cycle planning of battery energy storage for multi-functional ser- than 200 peer-reviewed technical articles and two
vices in power systems,’’ IEEE Trans. Sustain. Energy, vol. 11, no. 4, books, and holds more than 20 issued or pending
pp. 2077–2086, Oct. 2020. patents. His research interests include stability analysis, optimal control,
[20] Z. Li, Q. Guo, H. Sun, and J. Wang, ‘‘Sufficient conditions for exact relax- robust dispatch, and game theory based decision making in energy and power
ation of complementarity constraints for storage-concerned economic dis- systems. He is an Associate Editor of several international journals, including
patch,’’ IEEE Trans. Power Syst., vol. 31, no. 2, pp. 1653–1654, Mar. 2016. IEEE TRANSACTIONS ON SMART GRID and IEEE POWER AND ENERGY SOCIETY
[21] J. Fortuny-Amat and B. McCarl, ‘‘A representation and economic interpre- LETTER. He also served as a Guest Editor for IEEE TRANSACTIONS ON ENERGY
tation of a two-level programming problem,’’ J. Oper. Res. Soc., vol. 32, CONVERSION.
no. 9, pp. 783–792, Sep. 1981. CHENGHAO LI received the B.S. and Ph.D.
[22] P. Subcommittee, ‘‘IEEE reliability test system,’’ IEEE Trans. Power App.
degrees from the Huazhong University of Science
Syst., vols. PAS-98, no. 6, pp. 2047–2054, Nov. 1979.
and Technology, Wuhan, China, in 2009 and 2015,
[23] R. D. Zimmerman, C. E. Murillo-Sanchez, and R. J. Thomas, ‘‘MAT-
POWER: Steady-state operations, planning, and analysis tools for power respectively. He is currently a Senior Engineer of
systems research and education,’’ IEEE Trans. Power Syst., vol. 26, no. 1, State Grid Henan Electric Power Research Insti-
pp. 12–19, Feb. 2011. tute, China. His research interests include power
[24] P. Harsha and M. Dahleh, ‘‘Optimal management and sizing of energy systems, renewable energy integration, and energy
storage under dynamic pricing for the efficient integration of renew- storage systems.
able energy,’’ IEEE Trans. Power Syst., vol. 30, no. 3, pp. 1164–1181,
May 2015.

20190 VOLUME 9, 2021