This article has been accepted for publication in IEEE Access.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2019.Doi Number

Optimal allocation of BESS to maximize

efficiency under constrained SOC in
Parallel Inverter-Based Scaled-Up Slack
Bus
Soo Hyoung Lee, Member, IEEE
Department of Electrical and Control Engineering, Mokpo National University, Jeollanam-do 58554, Korea
Corresponding author: Soo Hyoung Lee (e-mail: [email protected]).
This work was supported in part by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT)
(No. 2022R1A2C2006688) and in part by the Korea Institute of Energy Technology Evaluation and Planning (KETEP) grant funded by
the Korean government (MOTIE) (20223A10100030, Development of Synchronous Condenser Model and Power System Inertia
Operating Technology)

ABSTRACT The parallel inverter-based scaled-up slack bus includes multiple battery energy storage
systems (BESSs) to support the slack operation without short of real power. The multiple BESSs require a
well-designed state of the charge (SOC) balancing algorithm for maximally effective usage. Until now,
however, the SOC balancing algorithms only focus on strong SOC balancing without careful consideration
of BESS efficiency during its charging and discharging. This paper proposes a soft SOC balancing algorithm
based on the Levenberg-Marquardt (L-M) algorithm considering BESS's charging and discharging efficiency.
The charges or discharges of the BESSs are primarily determined by their efficiencies that indirectly reflect
the SOCs. As a result, the proposed method provides some margins to improve efficiency rather than tightly
focusing on only the SOCs balancing between multiple BESSs. The optimizations and balancing are
conducted by MATLAB software. Their related converter interactions are simulated using a power system
computer-aided design/electromagnetic transient including DC (PSCAD/EMTDCTM), to measure data for
optimization.

INDEX TERMS Efficiency-based SOC balancing, Levenberg-Marquardt algorithm, optimal SOC

balancing, parallel inverter-based slack bus, scaled-up slack bus.

I. INTRODUCTION open questions to be discussed. Representatively, there is a

voltage stability problem. Despite the excellent stability of the
A. MOTIVATION AND INCITEMENT current Korean power system, it will not be guaranteed to keep
Recently, the environment has gotten lots of attention the degree of stability in the near future when massive
worldwide due to the advent of COVID-19. Especially renewable energy sources are connected to the power grid via
regarding global warming issues, the temperature increase power converters [3]-[4]. From the viewpoint of energy
must be limited to under 1.5oC. The CO2 emissions must be balance, renewable energies require much energy storage. The
reduced to below 45% of the 2010 levels by 2030 [1]. In line battery energy storage system (BESS) is currently preferred to
with this global trend, the Korean government announced be installed combined with inverters to cover the intermittency
“renewable energy 3020”, increasing the country’s of renewable energy sources.
dependency on renewable energy sources by up to 20% by The generation bus voltage is kept constant by the excitation
2030 [2]. Unfortunately, most renewable generators are system of the conventional synchronous generator. However,
formed as variable renewable energy (VRE), which fluctuates it does not apply to a DG based on the converter. In addition,
in nature, such as solar and wind. the size of the converter is limited by the ratings of switch
Despite the positive aspects of global movements into devices, so it is hard for the converter-based generator controls
renewable energy expansion, many technical issues are still the voltage as the synchronous generator. The BESS system is

VOLUME XX, 2019 1

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

generally installed as modules to build up to the required size. method has also been proposed [14]. Here, the multiple ESSs
Nevertheless, the SOC balancing between the modules differs are operated by a high-level management system based on a
from the most efficient operation of the entire BESS system. consensus algorithm rather than local device control. A
Decentralized SOC balancing control has been proposed for
B. LITERATURE REVIEW autonomous DC microgrid applications [15]. That eliminated
The synchronized GPS-time-based control for multiple slack communication delay is causing performance deterioration by
buses has been proposed [5] to improve the voltage stability adding a high pass filter algorithm.
of the inverter-driven power system. An essential prerequisite There have also been several optimization methods for
here is the sizes of the inverter and backbone energy source. BESS siting and sizing [16]-[17]. However, those methods
The inverter and its energy source cannot keep their role as a focus on the problems in the distribution system rather than
slack generator when they face huge events exceeding their the multiple BESSs to emulate a slack bus.
sizes. Unfortunately, the practical inverter has a much smaller
power rating than the conventional generator. To overcome C. CONTRIBUTION AND PAPER ORGANIZATION
the limit, the control and operation of multiple inverter groups Although there are other types of energy storage, the BESS is
are proposed for a slack operation [6]. Similarly, the backbone the preferred energy storage to be used with the inverter in
energy source must be composed of plural energy sources current. This paper proposes an optimal SOC balancing of
unless a unit source is large enough to cover the maximum parallel BESS to maximize the efficiency of the inverter-based
output. Therefore, the multiple sources require a balancing scaled-up slack bus. For the SOC balancing, both efficiencies
control between their SOCs if they have limited energy, such of the inverters and BESSs are included in the objective
as BESSs. There is an I-SOC droop-based balancing method function of the optimization instead of the simple yet
for modular multilevel supercapacitor ESS [7]. The current straightforward comparison of relative SOC levels. Although
reference for each module is determined based on the droop the main control target is maximum efficiency, not the SOCs
curve depending on the SOC. From a macro point of view, it balancing, the SOC levels of the multiple BESSs maintain
is merely an improvement from direct on-and-off control close to each other based on the efficiency versus SOC
based on the order of SOCs. A method includes voltage characteristics derived from electrical battery model
control in its SOC control loop [8]. The voltage affects the estimations in [18]-[23].
droop ratio that determines each cell’s current concerning the It has been challenging to accurately balance the energy
SOC of the individual battery cell. It is an improved form of using conventional power-sharing methods such as droop and
[7] in the viewpoint of automatic selection of droop ratio. VMS in [24]-[27]. The proposed method allows the multiple
There is a unique SOC balancing method for modular ESS [9]. inverter-based scaled-up slack generator [28] to operate its
Here the ESS is composed of several serial-connected backbone power source efficiently so that the group of small-
modules. The current references for each module are size inverters can stably operate a relatively large inertia-free
determined considering SOC and the required output current stand-alone (IFSA) microgrid [29].
of the ESS. A study has been conducted on SOC balancing This paper is organized as follows. First, the problem
between multiple ESS with different capacities [10]. Its formulation is described in Section II. Then, Section III
primary purpose is mitigating SOC unbalances caused by describes the efficiency corresponding to SOC, power ratio,
applying the conventional SOC balancing method. There is a and elapsed time. After that, the SOC balancing is given in
study of consensus-based stochastic cell balancing control Section IV with maximum efficiency based on Levenberg-
[11]. This has given a model-free accurate SOC estimation Marquardt (L-M) algorithm. After that, computational
that could only be achieved using model-based control. A. B. analysis and case studies are shown in Sections V-VII. Finally,
Ahmad et al. proposed an ESS topology and its related SOC the conclusion is given in Section VIII.
balancing method for grid-scale three-phase BESS [12]. Here,
the BESS has been applied as a single-arm to eliminate the II. PROBLEM FORMULATION
need to balance between arms. The SOC balancing between The scaled-up slack generator is one of the core techniques to
multiple modules has been conducted based on the insertion implement the IFSA microgrid with large load consumption.
or bypass of each BESS. P. Wang et al. proposed a three-port Although it can solve the instantaneous power mismatch
converter-based hybrid energy storage system, including a between generation and load, mitigating a huge energy
battery and supercapacitor and its balancing control [13]. The imbalance with a small BESS is still challenging. In other
battery is balanced by voltage control based on SOC-related words, the generation-load balancing is up to the cooperation
droop and supercapacitor voltage. The supercapacitor operates of multiple BESSs, including SOC balancing. So far, although
based on its physical topology without the direct application only SOC balancing has been used as the objective function of
of active control. Still, it is indirectly affected by the battery the conventional BESS operation, it has been considered
SOC balancing control via applying its voltage to the battery separately from other objectives, such as achieving maximum
balancing control. In addition to the previously described efficiency. Maintaining the same SOC levels between multiple
device-level SOC balancing, a microgrid-level SOC sharing BESSs requires lots of charging and discharging operations.

VOLUME XX, 2019 2

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

BESS with high/low SOC must discharge/charge even if the where Vt,crg and Vt,dsc are charging and discharging terminal
estimated deviation is insignificant. In the worst case, several voltages of the battery. In rough, the efficiency reduces
BESSs can discharge during the other BESSs' charging. according to the power increase or SOC decrease. To
maximize the efficiency of the multiple BESS system, it is
III. EFFICIENCY CORRESPONDING TO SOC, POWER advantageous to share power between all BESS to minimize
RATIO, AND ELAPSED TIME the power of individual BESS. Also, the battery shows high
There have been numerous studies to estimate the SOC and efficiency if the elapsed time is short from its last
SOH (state of the health) [30]-[31], which are the essential charge/discharge. This is because the charges across the
parameters to implement the equivalent electrical model of the equivalent capacitors discharge through the parallel resistors
battery shown in Figure 1. The data for electrical motive force during the elapsed time.
(Voc), the series resistance (Rseries), and resistive/capacitive In practice, the BESS includes not only a battery but also a
components (Rsec/Csec, Rmin/Cmin, and Rhour/Chour) of time converter whose efficiency vary respect to the power ratio.
constants vary according to SOC levels. The time constants Due to the most efficient power range being up to the design,
are determined by the efficiency curve is assumed as the polynomial of (3) by
τ = RC (1) considering general converter designs.

η conv= 0.5 ( ( P + 0.01) + (1.01 − P ) )

0.05 0.01
The battery model makes it possible to acquire much data, (3)
including efficiency, correctly without directly measuring the
actual battery in operation. where P is power in pu. The efficiency, η BESS ( = η batη conv ) is
shown in Figure 3 concerning the elapsed time, power, and
SOC.

FIGURE 1. The electrical battery model has three time constants: second,
minute, and hour.

The electrical battery model with parameters in [18] shows

the charging-discharging efficiency (without converter)
versus power curves in Figure 2. The battery model does not
consider leak of charges, so the efficiency is calculated by
η=
bat Vt ,crg − Vt ,dsc (2)

FIGURE 3. BESS (i.e., battery + converter) efficiency VS power curves

concerning SOC (t: elapsed time from last charge/discharge).

The scaled-up slack bus is implemented in Figure 4 based

on multiple BESSs. The real and reactive powers of the slack
are measured and used to generate power references for each
BESS. The reactive power references are calculated by
dividing the measured one by the number of BESSs in the PQ
operation. The real power references are computed by the
optimization process described in Section IV to maximize
efficiency. The efficiency vs. power and SOC curves are
updated according to the SOH for accurate estimation of
optimal sizes of BESSs.
According to the curves in Figure 2, it is advantageous to
operate all batteries to maximize efficiency by minimizing
FIGURE 2. Battery (without converter) efficiency VS power curves power in each BESSs. By contrast, minimum power does not
concerning SOC (t: elapsed time from last charge/discharge). guarantee the maximum efficiency of the individual BESS,
including the converter, according to the curves in Figure 3.

VOLUME XX, 2019 3

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

The correlation between efficiency and power is reduced μ make the L-M algorithm similar to the Gauss-Newton and
because of the inverter. In other words, it is not easy to the gradient descent methods, respectively. Then, the
intuitively select the optimal operating points without huge associated parameters required in the L-M algorithm are
computational effort to compare efficiencies resulting from determined by (5).
the conditions of multiple BESSs. It becomes more difficult
β = [ PBESS 1 , PBESS i , PBESS n ] (5)
considering the curves of Figure 3 vary according to the SOH.
where PBESS i and n are the power of BESSi and the number of
BESS, respectively. The Jacobian matrix, J, is composed of
J i = ε i Ai (6)
where
∂ηBESS
Ai = (7)
∂PBESS i
To determine εi, a temporary update is performed by
( A T A + µ I )δ = A T [ y − f (β)] (8)
Then, temporary deviations of powers are calculated by
( noi +1)
∆PBESS
tmp
i = PBESS i − PBESS i
( noi )
(9)
where noi is the number of iterations. After that, εi is
determined by

FIGURE 4. Scaled-up slack bus based on multiple BESSs.

IV. SOC BALANCING BASED ON LEVENBERG-

MARQUARDT ALGORITHM
It looks in Figure 3 that the maximally efficient powers lie on
a virtual line regardless of the elapsed time or SOC. Also, the
efficiency curves are always convex in an arbitrary SOC.
Therefore, local optimization based on Jacobian can find the
maximum efficiency point.
To deal with the issue, the Levenberg-Marquardt (L-M)
optimization algorithm [32] is applied to select the optimal
sizes of multiple BESSs. This L-M algorithm is the combined
type of gradient descent and Gauss-Newton algorithms. Then,
during the optimization process, it updates the parameters
adaptively by (4).
(J T J + µ I )δ = J T [ y − f (β)] (4)
where the matrices J and I are the Jacobian and identity
matrices, respectively. The vector, β, includes the parameters
determining the objective function’s value. The parameter
deviation vector updates it, δ, at every iteration to minimize FIGURE 5. Optimization process to determine sizes of BESSs during
the objective function. The relatively small and large values of discharging.

VOLUME XX, 2019 4

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

FIGURE 6. Discharging efficiency of BESS VS SOC (t: elapsed time from FIGURE 7. Charging efficiency of BESS VS SOC (t: elapsed time from the
the last charge). last discharge).

maximum efficiency discharging operation. By contrast, it is

 max 
 ( PBESS i − PBESS i)
high
unclear if the BESS with a low SOC firstly charges during the
, PBESS i ≥ PBESS
high
i & ∆PBESS i
tmp
≥ 0
 ( P max − PBESS i )  maximum efficiency charging operation. Therefore, this study
 BESS i  applies the optimization procedure only to discharging. All
ε i =  ( PBESS i i)
 low − PBESS
min  BESSs charge power evenly.
 (10)
, PBESS i ≤ PBESS i & ∆PBESS i ≤ 0
 ( BESS i i)
low tmp
 P − PBESS
min
 V. COMPUTATIONAL ANALYSIS
  Several computational analyses have been conducted with a
 1  scaled-up slack bus based on multiple BESSs to verify the
 , otherwise 
proposed optimization algorithm. The elapsed times are
The εi makes the power reference not exceed the maximum assumed as 5 minutes, considering the general requirements
capacity of BESS. Finally, the optimal powers of BESSs are for the long-term energy storage devices in the power grid.
determined by the optimization procedure shown in Figure 5. The system in Figure 4 is considered for the following case
Initially, their values are set as equal. Then, the optimization studies. The size of each BESS is assumed as
process is carried out by applying the L-M algorithm of (4) 0.1MW/0.3MWh considering the actual islanded power
with its associated parameters in (5) and (6). These selected system in Korea of the previous study [6]. The main target of
optimal sizes are then used to update the BESS power this study is the green slack control box in Figure 4. First, the
allocation. After that, the optimization process monitors three efficiency data are collected by simulating the electrical
values, measured power, SOC, and SOH, to determine a repeat battery model in Figure 1 by considering various conditions of
of the previous procedure. The critical values, εP, εC, and εH, charging/discharging and SOC using the PSCAD/EMTDCTM.
are selected considering accuracy, computational effort, etc. Those data are used to verify the performance of the
Finally, the corresponding BESS sizes are determined by optimization. Then, the optimization is conducted based on
referring to the BESS power allocation table. MATLAB simulation to update the BESS power allocation
The optimization process cannot be fast enough to carry table, which could be used for the output reference generation
every instant elapsed time out. This is because the optimal size during power system operation.
must be determined before the operation of the BESSs, but the In the case of a single BESS, there is no option to select the
elapsed time can be checked only after the operation of the best combination of BESS outputs. In this case, the BESS
BESSs. The efficiency curves in Figure 3 can be redrawn output must be determined by following the power output
separately as discharge and charge efficiency contour maps, as from the common slack bus. Therefore, at least two BESSs are
shown in Figs. 6 and 7, respectively. A higher SOC always required for valid optimization applications.
results in a higher discharging efficiency regardless of power
or elapsed time, as shown in Figure 6. In contrast, higher SOC A. OPTIMIZATION OF TWO BESS POWERS
only sometimes leads to higher charging efficiency. This Two BESSs are connected to the scaled-up slack bus. The
stands out when the power is medium-level, as shown in SOCs are assumed as 0.7 and 0.4, respectively, for BESS 1
Figure 7(b). Therefore, it is most likely that a BESS with a and 2. As mentioned in Figure 6, the BESS shows higher
high SOC will first discharge among multiple BESSs during efficiency when its SOC is higher. Therefore, BESS 1 should

VOLUME XX, 2019 5

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

FIGURE 8. Optimization trajectories (solid-line) concerning various sums FIGURE 9. Efficiency improvements during optimization in Figure 8.
of BESS powers (dots display optimal points).

release more power than BESS 2 for high efficiency. However, B. OPTIMIZATION OF THREE BESS POWERS
both BESSs should discharge if the required power is Three BESSs are connected to the scaled-up slack bus. The
sufficiently large. This is because the extremely high SOCs are assumed as 0.7, 0.6, and 0.4, respectively, for BESS
discharging of one BESS results in low total efficiency. 1, 2, and 3. As previously described, the BESS shows higher
Initial outputs of the BESSs are allocated to make the BESS efficiency when its SOC is higher. Therefore, the released
2, which has smaller SOC, not operate when the total output is power from the BESSs should decrease by order of BESS 1,
smaller or equal to 0.8pu. In case of the total output is between 2, and 3 for high efficiency.
0.8pu and 1.6pu, the value subtracts BESS 1 from the total The optimal combinations of powers are shown in Figure
output allocated to BESS 2. Identical outputs are allocated to 10 when the total power is 1pu. The initial combinations of
both BESSs when the total output is larger than 1.6pu. The BESSs outputs are selected randomly. The 1pu total power
optimal combinations of powers are shown in Figure 8 region is illustrated by the yellow plane in the left-upper three-
according to total power between 0 to 2 pu. The optimization dimensional graph. The solid line and dots are optimization
trajectories are shown in solid-line, and the optimal points are trajectories and optimal points, respectively. The right-upper
shown in dots. The trajectories are parallel to a virtual diagonal graph shows the floor projection according to the direction of
line, which starts from (zero PESS1, one PESS2) and ends at (one the PBESS3 axis. Similarly, the left-lower and right-lower
PESS1, zero PESS2). Those mean the entire dis-charges are graphs show the floor projections according to the directions
constant through every trajectory. The BESS 2 does not of the PBESS1 and PBESS2 axes, respectively. In most cases,
discharge when the total power is smaller than 0.86pu. The 0.5122pu, 0.4878pu, and 0 are determined for the BESS 1, 2,
BESS 1 and 2 discharge 0.5326pu and 0.3274pu, respectively, and 3, respectively, as the optimal combination. In several
when the entire power is 0.86pu. The changes in the BESSs cases, 0.5332pu, 0, and 0.4668pu are determined for BESS 1,
are related to the efficiency curves in Figure 6. Meanwhile, the 2, and 3, respectively. This means that BESS 1 always
BESS 2 discharges smaller than the BESS 1 as long as the total discharges, and BESS 2 has much more opportunity to release
power is smaller than about 1.9pu. This is because the SOC of its power than BESS 3. Considering their SOCs, BESS with
the BESS 1 is larger than that of BESS 2. The difference high SOC tends to discharge more, so it converges to a
between discharging results in SOC balance. Figure 8 shows balanced SOC.
that the optimization trajectories are relatively long when the The optimization is also applied to the other cases where
total power is smaller than 0.78pu. In other words, the power the total power is not 1pu. For clear comparison in the one- or
difference between BESSs is relatively large here. Also, the two-BESS operating range, the SOCs are allocated 0.7, 0.4,
efficiency improvements are remarkable in that region, as and 0.35. Initial outputs of the BESSs are allocated to make
shown in Figure 9. The effectiveness is pronounced well when the BESS 2 and 3, which have small SOC, not operate when
the total power is much smaller, which is the general status of the total output is smaller or equal to 0.8pu. In case of the total
the power grid, than the maximum capacity. The efficiency output is between 0.8pu and 1.6pu, the value subtracts BESS
improves by up to about 0.2% in this representative case. The 1 from the total output allocated to BESS 2. BESS 3 is still
improvement differs according to the values selected as the zero. In case of the total output is between 1.6pu and 2.4pu,
initial BESS outputs. the value subtracts BESS 1 and 2 from the total output
allocated to BESS 3. Identical outputs are allocated to all

VOLUME XX, 2019 6

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

FIGURE 10. Optimization trajectories (solid-line) when the sum of BESS FIGURE 11. Efficiencies of two- and three-BESS-connection.
powers is 1pu (dots display optimal points).

FIGURE 12. Optimal discharging power of BESS 2 concerning its SOC FIGURE 13. Optimal discharging power of BESS 7 concerning its SOC
and total power of two BESSs. and total power of seven BESSs.

BESSs when the total output exceeds 2.4pu. As a result, the while its SOC is smaller than 0.7pu (i.e., the same as the SOC
three-BESS-connected case shows higher efficiency than the of BESS 1), as shown in Figure 12. This is the same as the
two-BESS-connected case when the total power is relatively result of Figure 8. By contrast, BESS 1 does not discharge
large, as shown in Figure 11. when the PBESSs is smaller than about 0.75pu and the SOC of
BESS 2, SOC2, is larger than 0.7pu. The slope of PBESS2
VI. POWER ALLOCATION CONCERNING SOC concerning PBESSs is smaller than 1 when the PBESSs is larger
The previous computational analyses showed the optimization than about 0.75pu. It means that BESS 1 and BESS 2
with several representative SOCs. This section analyzes the discharge together. The PBESS2 slightly increases according to
amount of discharging power allocation of the BESS the SOC2 increase near 0.75pu of PBESSs. In brief, the proposed
concerning the various SOCs and the various total power of algorithm makes the BESS more discharge when its SOC is
BESSs. relatively high, regardless of the absolute level of the SOC. As
a result, the SOC is balanced while the BESS maintains
A. OPTIMAL ALLOCATION BETWEEN TWO BESS maximum efficiency.
POWERS
Two BESSs are connected to the scaled-up slack bus. The B. OPTIMAL ALLOCATION BETWEEN SEVEN BESS
SOC of BESS 1 is assumed as 0.7 pu, and the SOC of BESS POWERS
2 varies from 0.3pu to 0.8pu, while the total power of the The previous analysis is applied to seven BESSs connected to
BESSs varies from 0 to 2pu. The BESS 2 does not discharge a scaled-up slack bus. The SOC of BESS 1 to 6 is assumed as
when the total power, PBESSs, is smaller than about 0.75pu 0.38pu, 0.45pu, 0.52pu, 0.59pu, 0.66pu, and 0.73pu,

VOLUME XX, 2019 7

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

FIGURE 14. Optimal discharging powers of seven BESSs with SOCs in FIGURE 15. Optimal discharging powers of seven BESSs with random
order of [0.31pu, 0.32pu, 0.325pu, 0.33pu, 0.335pu, 0.34pu, and 0.35pu], SOCs (SOC1 < SOC2 < SOC3 < SOC4 < SOC5 < SOC6 < SOC7) and the
and the related total efficiency. related total efficiency.

respectively. The SOC of BESS 7 varies from 0.3pu to 0.8pu, smaller than those when PBESSs is 2.87pu, such as 0.4539pu,
while the total power of the BESSs varies from 0 to 7pu. The 0.4629pu, 0.4723pu, 0.4818pu, and 0.4911pu, their relative
BESS 7 does not discharge when the total power, PBESSs, is size follows the relative levels of SOCs regardless of their
smaller than about 2.64pu while its SOC is smaller than absolute values. As a result, more points of discharging power
0.38pu (i.e., the same as the SOC of BESS 1), as shown in are above the reference diagonal solid line when the SOC is
Figure 13. The minimum PBESSs for the discharge of BESS 7 higher. In other words, the proposed method successfully
reduces according to the increase of SOC7. This means the balances the SOCs of BESSs even if the SOCs are mutually
BESS 7 tends to discharge more when the SOC7 is high. In very close. The total efficiency improves as a result of the
brief, the proposed algorithm makes the target BESS more optimal allocation. It is remarkable when PBESSs is relatively
discharged when its SOC is relatively high, regardless of the small (i.e., enough space exists for optimization), as shown in
absolute level of the SOC and the total number of BESSs. As the bottom right graph in Figure 14.
a result, the SOC is balanced while the BESS maintains
maximum efficiency. B. BALANCING IN CASE OF RANDOM SOCS
The proposed algorithm is applied to seven BESSs with
VII. BESS BALANCING IN CASES OF VARIOUS SOCS random SOCs for much closer investigation. Seven random
The proposed algorithm is verified by applying more cases, SOCs are sorted in rise from BESS 1 to 7 in each case. More
such as BESSs with similar SOCs or random SOCs. discharging power points are above the reference diagonal
solid line when the SOC is higher, as shown in Figure 15. Also,
A. BALANCING IN CASE OF SIMILAR SOCS the BESS with the lower SOC has more chance to release zero
To verify the performance in harsh conditions, seven BESSs power. In other words, the number of zero points decreases
with similar SOCs are considered. The SOC of BESS 1 to 7 is according to BESS 1 to 5, as shown in Figure 15. In conclusion,
assumed as 0.31pu, 0.32pu, 0.325pu, 0.33pu, 0.335pu, 0.34pu, the proposed method successfully balances the SOCs of
and 0.35pu, respectively. PBESS1 is zero when PBESSs is at most BESSs in arbitrary conditions of SOCs. The total efficiency
2.87pu. It is not zero when PBESSs is at least 2.94pu. Also, improves as a result of the optimal allocation. It is remarkable
PBESS2 is zero when PBESSs is at most 0.35pu. It is not zero when when PBESSs is relatively small, as shown in the bottom right
PBESSs is at least 0.42pu. The power increases according to the graph in Figure 15.
order of BESS 1 to 7 as 0pu, 0.06583pu, 0.06757pu,
0.06921pu, 0.07074pu, 0.07214pu, and 0.07451pu when the VIII. CONCLUSION
PBESSs is 0.42pu. Although the powers of BESS 3 to 5 are This paper proposed an optimal SOC balancing of BESS
smaller than those when PBESSs is 0.35pu, such as 0.06939pu based on the Levenberg-Marquardt (L-M) algorithm to
and 0.06968pu, their relative size follows the relative levels of maximize efficiency for the parallel inverter-based scaled-up
SOCs regardless of their absolute values. The power increases slack bus. The proposed method improves the BESS
according to the order of BESS 1 to 7 as 0.2836pu, 0.3707pu, efficiency and properly balances their SOCs. In relation, there
0.4007pu, 0.4286pu, 0.4549pu, 0.4795pu, and 0.522pu when have been many state estimation and modeling methods for
the PBESSs is 2.94pu. Although the powers of BESS 2 to 6 are

VOLUME XX, 2019 8

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

batteries that help the accuracy of the proposed method, as Converters in DC Hybrid Energy Storage Systems,” IEEE Access, Vol.
7, pp. 157173-157184, Oct. 2019.
described in the introduction. [14] W. Kang, M. Chen, B. Li, F. Chen, W. Lai, H. Lin, and B. Zhao,
To verify the performance of the proposed method, “Distributed Reactive Power Control and SOC Sharing Method for
parallel inverters in the DG models were simulated by the Battery Energy Storage System in Microgrids,” IEEE Access, Vol. 7,
electromagnetic transient program (EMTP). The proposed pp. 2169-3536, April 2019.
[15] X. Lin, R. Zamora, and C. A. Baguley, “A Fully Filter-Based
method successfully played a key role in the scaled-up slack Decentralized Control With State of Charge Balancing Strategy for
generator of the IFSA microgrid. As a result, a multiple- Battery Energy Storage Systems in Autonomous DC Microgrid
inverter slack generator could be implemented with multiple Applications,” IEEE Access, Vol. 9, pp. 15028-15040, Jan. 2021.
small BESSs even if there is no single BESS as large as the [16] S. Khunkitti, P. Boonluk, and A. Siritaratiwat, “Optimal Location and
Sizing of BESS for Performance Improvement of Distribution
conventional synchronous machine. Systems with High DG Penetration,” Wiley Hindawi International
This study verified the performance of the single scaled- Transactions on Electrical Energy Systems, Vol. 2022, pp. 1-16, Jun.
up slack generator. However, it is expected that further 2022.
studies are required to apply the proposed method to a [17] P. Boonluk, A. Siritaratiwat, P. Fuangfoo, and S. Khunkitti, “Optimal
Siting and Sizing of Battery Energy Storage Systems for Distribution
practical system, including many types of individual BESS. Network of Distribution System Operators,” MDPI Batteries, Vol. 6,
The proposed method would encourage the expansion of No. 4, pp. 1-16, Nov. 2020.
stand-alone microgrids by enabling the grid-forming inverter [18] Y. Cao, R. C. Kroeze, and P. T. Krein, “Multi-timescale Parametric
with the currently commercialized BESSs. Electrical Battery Model for Use in Dynamic Electric Vehicle
Simulations,” IEEE Trans. on Transportation Electrification, Vol. 2,
No. 4, pp. 432-442, Dec. 2016.
REFERENCES [19] D. M. Rosewater, D. A. Copp, T. A. Nguyen, R. H. Byrne, and S.
[1] V. M-. Delmotte, P. Zhai, H-. O. Pӧrtner, D. Roberts, J. Skea, P. R. Santoso, “Battery Energy Storage Models for Optimal Control,” IEEE
Shukla, A. Pirani, W. M-. Okia, C. Péan, R. Pidcock, S. Connors, J. B. Access, Vol. 7, pp. 178357-178391, Dec. 2019.
R. Matthews, Y. Chen, X, Zhou, M. I. Gomis, E. Lonnoy, T. Maycock, [20] G. Giordano, V. Klass, M. Behm, G. Lindbergh, and J. Sjöberg,
M. Tignor, and T. Waterfield, “Global warming of 1.5oC,” IPCC “Model-Based Lithium-Ion Battery Resistance Estimation From
Special Report. Oct. 2018. Electric Vehicle Operating Data,” IEEE Trans. on Vehicular
[2] “Renewable Energy 3020,” Ministry of Trade, Industry and Energy, Technology, Vol. 67, No. 5, pp. 3720-3728, May 2018.
Sejong, Chung-nam, Republic of Korea, February 2017. [21] M. K. S. Verma, S. Basu, R. S. Patil, K. S. Hariharan, S. P. Adiga, S.
[3] L. G. Meegahapola, S. Bu, D. P. Wadduwage, C. Y. Chung, and X. M. Kolake, D. Oh, T. Song, Y. Sung, “On-Board State Estimation in
Yu, “Review on Oscillatory Stability in Power Grids With Renewable Electrical Vehicles: Achieving Accuracy and Computational
Energy Sources: Monitoring, Analysis, and Control Using Efficiency Through an Electrochemical Model,” IEEE Trans. on
Synchrophasor Technology,” IEEE Trans. on Industrial Electronics, Vehicular Technology, Vol. 69, No. 3, pp. 2563-2575, Mar. 2020.
Vol. 68, No. 1, pp. 519-531, Jan. 2021. [22] G. Saldaña, J. I. S. Martín, I. Zamora, F. J. Asensio, and O. Oñederra,
[4] S. Maihemuti, W. Wang, H. Wang, J. Wu, and X. Zhang, “Dynamic “Analysis of the Current Electric Battery Models for Electric Vehicle
Security and Stability Region Under Different Renewable Energy Simulation,” MDPI Energies, Vol. 12, pp. 1-27, July 2019.
Permeability in IENGS System,” IEEE Access, Vol. 9, pp. 19800- [23] Y. Yu, N. Narayan, V. V-. Garita, J. P-. Gerber, Z. Qin, M.
19817, Jan. 2021. Wagemaker, P. Bauer, and M. Zeman, “Constructing Accurate
[5] S. H. Lee, D. Choi, and J-. W. Park, “Inertia-Free Stand-Alone Equivalent Electrical Circuit Models of Lithium Iron Phosphate and
Microgrid-Part I: Analysis on Synchronized GPS Time-Based Control Lead–Acid Battery Cells for Solar Home System Applications,”
and Operation,” IEEE Trans. on Industry Applications, Vol. 54, No. 5, MDPI Energies, Vol. 11, pp. 1-20, Sept. 2018.
pp. 4048-4059, Sept.-Oct. 2018. [24] M. M. Bijaieh, W. W. Weaver, and R. D. Robinett, “Energy Storage
[6] S. H. Lee, “Scaled-Up Slack Generator Based on Parallel Inverters for Requirements for Inverter-Based Microgrids Under Droop Control in
a Reliable IFSA Microgrid,” IEEE Access, Vol. 9, pp. 133204-133214, d-q Coordinates,” IEEE Trans. on Energy Conversion, Vol. 35, No. 2,
Aug. 2021. pp. 611-620, June 2020.
[7] K. Bi, L. Sun, Q. An, and J. Duan, “Active SOC Balancing Control [25] A. G. Paspatis, G. C. Konstantopoulos, and J. M. Guerrero, “Enhanced
Strategy for Modular Multilevel Super Capacitor Energy Storage Current-Limiting Droop Controller for Grid-Connected Inverters to
System,” IEEE Trans. on Power Electronics, Vol. 34, No. 5, pp. 4981- Guarantee Stability and Maximize Power Injection Under Grid Faults,”
4992, May 2019. IEEE Trans. on Control Systems Technology, Vol. 29, No. 2, pp. 841-
[8] Y. Cao and J. A. A. Qahouq, “Hierarchical SOC Balancing Control 849, Mar. 2021.
for Battery Energy Storage System,” IEEE Trans. on Industrial [26] X. Meng, J. Liu, and Z. Liu, “A Generalized Droop Control for Grid-
Electronics, Vol. 68, No. 10, pp. 9386-9397, Oct. 2021. Supporting Inverter Based on Comparison Between Traditional Droop
[9] K. Bi, W. Yang, D. Xu, and W. yan, “Dynamic SOC Balance Strategy Control and Virtual Synchronous Generator Control,” IEEE Trans. on
for Modular Energy Storage System Based on Adaptive Droop Power Electronics, Vol. 34, No. 6, pp. 5416-5438, June. 2019.
Control,” IEEE Access, Vol. 8, pp. 41418-41431, Feb. 2020. [27] D. Choi, J-. W. Park, and S. H. Lee, “Virtual Multi-Slack Droop
[10] Q. Wu, R. Guan, X. Sun, Y. Wang, and X. Li, “SoC Balancing Control of Stand-Alone Microgrid With High Renewable Penetration
Strategy for Multiple Energy Storage Units with Different Capacities Based on Power Sensitivity Analysis,” IEEE Trans. on Power Systems,
in Islanded Microgrids Based on Droop Control,” IEEE Journal of Vol. 33, No. 3, pp. 3408-3417, May 2018.
Emerging and Selected Topics in Power Electronics, Vol. 6, No. 4, pp. [28] S. H. Lee, “Scaled-Up Slack Generator Based on Parallel Inverters for
1932-1941, Dec. 2018. a Reliable IFSA Microgrid,” IEEE Access, Vol. 9, pp. 133204-133214,
[11] I. Bistritz and N. Bambos, “Consensus-Based Stochastic Control for Aug. 2021.
Model-Free Cell Balancing,” IEEE Trans. on Control of Network [29] S. H. Lee, D. Choi, and J-. W. Park, “Inertia-Free Stand-Alone
Systems, Vol. 8, No. 3, pp. 1139-1150, Sept. 2021. Microgrid-Part I: Analysis on Synchronized GPS Time-Based Control
[12] A. B. Ahmad, C. A. Ooi, and D. Ishak, “State-of-Charge Balancing and Operation,” IEEE Trans. on Industry Applications, Vol. 54, No. 5,
Control for Optimal Cell Utilisation of a Grid-Scale Three-Phase pp. 4048-4059, Sept./Oct. 2018.
Battery Energy Storage System Using Hybrid Modular Multilevel [30] M. Galeotti, L. Cina, C. Giammanco, S. Cordiner, and A. D. Carlo,
Converter Topology Without Redundant Cells,” IEEE Access, Vol. 9, “Performance analysis and SOH (state of health) evaluation of lithium
pp. 53920-53935, April 2021. polymer batteries through electrochemical impedance spectroscopy,”
[13] P. Wang, P. Ren, X. Lu, W. Wang, and D. Xu, “A Distributed State- Elsevier Energy, Vol. 89, pp. 678-686, July 2015.
of-Charge Balancing Control Scheme for Three-Port Output-Series

VOLUME XX, 2019 9

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4
 This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2023.3237773

[31] J. Hoog, J. Jaguemont, M. A-. Monem, P. V. D. Bossche, J. V. Mierlo,

and N. Omar, “Combining an Electrothermal and Impedance Aging
Model to Investigate Thermal Degradation Caused by Fast Charging,”
MDPI Energies, Vol. 11, pp. 1-15, Mar. 2018.
[32] B. M. Wilamowski and H. Yu, “Improved Computation for
Levenberg-Marquardt Training,” IEEE Trans. on Neural Networks,
Vol. 21, No. 6, pp. 930-937, June 2010.

Soo Hyoung Lee (S’08–M’12) received B.S. and Ph.D.

in electrical engineering from the School of Electrical
and Electronic Engineering, Yonsei University, Seoul,
South Korea, in 2008 and 2012, respectively. During
2012–2014, he was a Postdoctoral
Research Associate with the School of Electrical and
Computer Engineering, Georgia Institute of
Technology, Atlanta, GA, USA. During 2014–2018,
he was a Senior Researcher in the Advanced Power Grid Research Division
at Korea Electrotechnology Research Institute, Uiwang, South Korea. He is
an Associated Professor at the Department of Electrical and Control
Engineering at Mokpo National University, Mokpo, South Korea. His
research interests include converter-based microgrids, optimal coordination
of distributed generation systems, converter control for distributed
generation systems, implementation of multi-level converters for low-
voltage AC systems, and non-isolated DC-DC converters for high-voltage
applications.

VOLUME XX, 2019 10

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4