Received 5 October 2022, accepted 27 October 2022, date of publication 1 November 2022, date of current version 8 November 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3218683

An Intelligent BMS With Probabilistic MO-GSA Based

CDMAS Integrating Edge Controller Analytics
C. R. SARIN, (Member, IEEE), AND GEETHA MANI , (Member, IEEE)
School of Electrical Engineering, VIT, Vellore 632014, India
Corresponding author: Geetha Mani ([email protected])

ABSTRACT One of the most challenging facets of any Battery Management System is scheduling the
charging and discharging cycles of each battery without compromising the uninterruptible power supplies to
meet demand. A battery-powered system may include several batteries of varying sorts, models, makes,
sizes, and lifespans, etc which employs a diversity of charging techniques. Subsequently, each battery
would have its own charging-discharging path. Even when a battery of the same capacity and make is
used with the same load profile, the charging curves differ. This is because batteries have differential
electrochemical characteristics and deteriorate to some extent with each use. Accordingly, the Battery
Management System should schedule the batteries so that degradation and usage are kept to a minimum.
This paper developed an adaptive Intelligent Battery Management System that can schedule batteries with
minimal power loss, increased battery life, and higher financial benefit, even when batteries of various sizes,
capacities, production types, lifespans, charge cycle models etc are incorporated. Multi-zonal approaches
are used, combining the benefits of edge analytics and Centralized Data Management and Analytics System.
Every millisecond, the battery parameters will be monitored using an energy monitoring circuit integrated
with an edge controller. However, the edge controller alone will not be able to process such a vast volume of
data on its own, the data will be divided into two categories and analyzed in two phases. All of the big data is
delivered to a Centralized Data Management and Analytics System using Low-power wide-area network, and
this data is labeled as primary data. A second set of data is extracted from big data within the edge controller
via metadata processing before it is transmitted to Centralized Data Management and Analytics System.
This secondary data is processed against safety standards stored in read only memory and rapid judgments
are performed using edge analytics if necessary. The Centralized Data Management and Analytics System
employs a number of analytics techniques. An Auto-Regressive Integrated Moving Average method will be
used to forecast the State of Charge of batteries. With the help of this forecasted data, the Multi-Objective
Gravitational Search Algorithm is then used to schedule the best battery allocation based on a number
of objectives such as battery temperature runaway, unit cost of consumption including span of service,
last used time period, State of Charge (%), State of Charge (WH), and so on. Between Auto-Regressive
Integrated Moving Average and Gravitational Search Algorithm, a Naive Bayes probabilistic estimator is
encased to identify the best general population for Gravitational Search Algorithm, avoiding repeated battery
swapping and improving power efficiency. The whole device is evaluated in Hardware in a Loop model.
When comparing the performance of the developed model to that of other optimization models, it is evident
that Gravitational Search Algorithm outperforms other methods when population is constrained.

INDEX TERMS ARIMA, BMS, edge controller, microgrid, MOGSA, Naive Bayes estimator.
NOMENCLATURE
Symbols ε Moving average operator.
k Bias constant. M1 , M2 Mass particles.
θ Autoregressive operator. F Magnitude of the gravitational
The associate editor coordinating the review of this manuscript and force.
approving it for publication was Bin Zhou . G Gravitational constant.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
115802 VOLUME 10, 2022
C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

n Iteration identifier. conjunction with BSS to meet the AC load. However, there
a Acceleration. are other limitations and shortcomings with battery-powered
P(c|x) Probability of linked sources. systems that must be addressed. To begin with, the capacity
P(c) Prior probability of selection. of the battery is limited [3], [4]. BSS must be charged on a
P(x|c) Likelihood probability of the regular basis and cannot be utilized to satisfy demand while
sources. being charged [5]. Thereupon the battery will be charged,
Bi Identifier of Battery. discharged, or left idle in between load cycles [6]. Battery
Bt SOC of Battery for time t. Management System (BMS) will schedule the battery for
Abbreviations both charging and discharging, ensuring that the battery is
OCV Open Circuit Voltage. maximum exploited and charged ahead of time for immediate
SCC Short Circuit Current. use. Another concern that should be resolved is battery aging,
LI Lithium-ion. which causes the battery to degrade with each use [7], [8].
LP Lithium Polymer. The overall longevity and number of charge cycles of the
NC Nickel Cadmium. battery are lowered proportionally [9]. To address this issue,
NM Nickel Metal Hydrides. it is preferable to host a variety of battery management
ZC Zinc–carbon. methods and preventive initiatives [10]. As a result, there has
LA Lead Acid. been a great deal of research into battery-operated system
ARIMA Auto-Regressive Integrated Moving monitoring, analysis, regulation, and management.
Average.
WH Watt Hour. B. LITERATURE REVIEW
SOC State of Charge. Yuanyuan Li et al. have performed a comprehensive inves-
DOD Depth of Discharge. tigation into feature selection techniques for estimating
BMS Battery Management System. battery condition using charging–discharging profiles [11].
BSS Battery Storage Systems. They offer extensive information on factors that impact
CDMAS Centralized Data Management and battery health as well as evaluating parameters for battery
Analytics System. analysis. Wang, Yujie, et al. undertake a thorough evaluation
GA Genetic algorithm. of the most widely used battery modeling and condition
HC Hill Climbing Algorithm. estimation methods for BMS [12]. Dai, Haifeng, et al.
HIL Hardware in a Loop. proposed a multi-layer framework for a BMS architecture
IBMS Intelligent Battery Management that solves a variety of issues such as overloading, external
System. heat, over current, temperature, excessive fast charging,
i,j Iterative constants. internal/external short circuit, battery life, thermal runaway,
LPWAN Low-power wide-area network. and so on [10]. The preceding papers assists in identifying
MO Multi-Objective methods. the factors and parameters that influence battery scheduling.
MOGSA Multi-Objective Gravitational The available methods for battery scheduling must now be
Search Algorithm. established.
NBPE Naive Bayes Probability Estimator The design of a fuzzy logic controller based on Particle
Algorithm. Swarm Optimization (PSO) for charging–discharging and
MG Microgrid. scheduling of BSS in Microgrid (MG) applications is
PSO Particle Swarm Optimization Algo- discussed in a paper by Faisal, M., et al. [13]. Liu, Chunyang.,
rithm. et al. proposes a method for scheduling microgrids in real
RP Reserve Power. time which takes into account the isolation of distributed
SA Simulated annealing Algorithm. generators and BSS over the operating cycle [14]. Rolling
‘S’ / ‘NS’ Selected / Not Selected. scheduling models, including daytime and hourly schedules,
are developed with the latter taking into account potential
irregularities in the status of charges. Luo, Liang, et al.
I. INTRODUCTION proposes a hybrid energy BMS with uncertainties around
A. BACKGROUND energy production, load demand and variations in grid
Many microgrids are now capable of functioning inde- bidding [3]. Wang, Shuoqi, et al. assess a comparative study
pendently of the main utility grid because of expanding on battery aging paradigm effect on the microgrid with four
installations of distributed energy sources such as solar and standard semi-empirical models. This paper serves as the
wind [1]. Battery Storage Systems (BSS) are employed primary reference paper for this study [15]. The majority
in such systems, enabling energy to be stored and used of existing techniques rely on centralized control of battery
as needed. [2]. Even if such models cost more up front, scheduling, yet such models are inefficient when it comes
they have a more stable and efficient power supply with to making rapid decisions. Decentralized approaches lack
higher reliability over time. Inverters are sometimes used in coordination between sources and are inefficient in battery

VOLUME 10, 2022 115803

 C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

scheduling when aging is taken into account. Yi, Zonggen, only a few batteries will be required at any given time to
et al. present a highly efficient receding horizon management satisfy demand, and some will be left idle for potential
technique for large plug-in electric vehicle populations demand. The battery schedule management system will
that uses a combination of centralised and decentralised extend battery charge cycles over varying time spans so
technologies to enable dynamic charging coordination [16]. that the system still satisfies demand but gives backup
By utilising the energy sharing and storage capabilities of batteries higher priority for charging. The BMS will also
electricity batteries in both buildings and electric vehicles, take into account certain other factors in order to reduce the
Huang, Pei, et al. suggest a coordinated control of building number of charge cycles on the battery and hence reduce
prosumers for enhancing cluster-level performance [17]. deterioration.
Dini, Anoosh, et al. represents unknown parameters with a This paper endeavors to establish an Intelligent Battery
hybrid stochastic/robust optimization that blends bounded Management System (IBMS) hardware that are scalable,
uncertainty-based robust optimization with scenario-based adaptive and dynamic, with following objectives.
stochastic programming [18]. The three articles mentioned 1) A two-part hardware architecture is adapted, with an
above are intended to establish coordinated control among edge controller and a Centralized Data Management
distributed energy sources, and they contributed to the and Analytics System (CDMAS). The model employs a
development of the control strategies employed in this multi-zonal analysis, which feature metadata analysis,
study. Edge-controlled design cases are being evaluated optimization, forecasting, and probabilistic estimation,
now. among other things. A Low Power Wide Area Network
Preetha, V and Vavilapalli demonstrates how a microcon- (LPWAN) facilitates data transmission between the
troller is utilised to construct a battery management system Edge controller and the CDMAS.
that uses the Hardware in a Loop approach to monitor and 2) Each battery system is coupled to a decentralized
secure the battery pack for safe operation [19]. Feng, Cheng, edge node controller, which also acts as an energy
et al present a thorough assessment of multidisciplinary monitoring circuit. Every millisecond, they records
research on Edge computing applications in smart grid, open circuit voltage and short circuit current of battery,
covering definitions, architectures, characteristics, and key load voltage, load current, and temperature of cells. The
enabling technologies, as well as application scenarios [20]. data is segmented into two divisions at this level based
Crocioni, Giulia, et al. investigate a number of machine on a metadata analysis since the edge controller alone
learning techniques for projecting Li-Ion battery perfor- cannot process such a huge volume of data. All of the
mance, with a particular focus on prognostic Li-Ion batteries data is transmitted to the CDMAS through LPWAN,
[21]. Brännvall, Rickard, et al. investigate the hardware and is analyzed by the CDMAS as primary data. A new
and software components utilized in renewable energy edge set of data is extracted from the primary data and is
controller applications for data collecting and control [22]. used as secondary data. It has the same essence as the
The articles described above can assist in developing edge- original data but is sampled over a shorter period of
controlled frameworks, data management, and hardware time. This data is analyzed using an edge analytics,
modelling. which is frequently used for decentralized control.
3) Now the present Watt Hour (WH) capacity and State of
C. MOTIVATION AND CONTRIBUTION Charge (SOC) or Depth of Discharge (DOD) of each
The prime objective of any BMS is to ensure that demand battery will be computed. Using the above data as well
is satisfied without interruption [23]. To achieve this aim, as historical data, Auto-Regressive Integrated Moving
BMS must accomplish two principal tasks [24]. The first Average (ARIMA) based time series analysis is used
stage entails appropriate scheduling and management of to forecast the battery charge cycle for the next time
battery charging and discharging. Other metrics, such as stamp.
temperature regulation and age degradation concerns, should 4) The best sources from all available sources are
be explored as a secondary goal. The rate of degradation evaluated using a Multi-Objective Gravitational Search
is a vital consideration when scheduling the battery. Each Algorithm (MO - GSA) based on these forecast results.
usage causes the battery to deteriorate to a certain extent Unit cost adjusted to Capex and Opex, temperature,
due to parasitic and electrolyte reactions. As such, the State SOC (Percentile), SOC(WH), availability of charging,
of Health and longevity of the battery is also reduced to time period after the battery was last used, charge
some extent after each use. Consequently, life expectancy cycle count, and aging factor are parameters taken into
and maximum achievable charging cycles are also reduced. account when choosing sources.
By using selective scheduling, the BMS can mitigate this 5) The general population for GSA is defined using
issue. a Naive Bayes Probability Estimator (NBPE). This
The cumulative capacity of all batteries in a microgrid avoids integration time delay, swapping transients,
system might be greater than the maximum demand. helps to achieve optimum battery utilization and
Nonetheless, average demand will be 50 to 70 % of maximum increase power quality by avoiding repeated switching
capacity, with peak demand lasting only a few hours. Hence between batteries.

115804 VOLUME 10, 2022

C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

FIGURE 1. Block diagram of two-stage IMBS application framework.

D. PAPER ORGANIZATION temperature sensor module with a 4051 multiplexer. For

There are three sections of the paper. The first section each element, every millisecond, module would monitor
contains an introduction to the problem, an analysis of the temperature, OCV, SCC, battery voltage and current, load
problem, context, and a brief literature review of the current voltage and current. This is a massive amount of data, and
quandary states. The second section moves into a deeper the edge controller cannot analyze it on its own.
level about the methods that have been established. The third Zone II - Communication: As a result, the data will
section takes step by step through the method, execution, with be divided into two processing groups at this stage. All of
results and discussions. The experiment is carried out in a the data will be transferred to a CDMAS through LPWAN,
laboratory setting using a scalable Hardware in a Loop (HIL) and the CDMAS will analyze all of the big data. However,
model. making decisions based on such a massive quantities of data
takes time, and there should be a redundancy framework in
II. PROBLEM FORMULATION - INTELLIGENT BATTERY place for making fast decisions in crisis situations. Thusly,
MANAGEMENT SYSTEM(IBMS) an edge analytic architecture is being established, with the
The IBMS is established using a two-stage multi-zonal mul- edge controller serving as a supplementary processor. Unlike
tiprocessing data analytics HIL framework. Figure 1 depicts CDMAS, the edge controller is incapable of handling massive
the block diagram of a two-stage IBMS architecture that quantities of data. Consequently, a sampling of a big dataset
incorporates an edge controller and a CDMAS. Decentralized would be performed within the edge controller itself.
administration is greatly aided by the edge node, while Zone III – Edge Sampling and Metadata analytics:
centralized management is facilitated through CDMAS. In this zone of activity, data is organized into samples
CDMAS is based on a single board controller, whereas edge either by time or by input from metadata analysis based
node is based on a WIFI micro-controller. The edge node on past data. Often metadata analyses are conducted on
is directly coupled to the power system elements. Internally, these newly sampled data in order to identify the nature
both controllers have a two-layer design. The first layer is of variations. These analyses include determining maximum
the hardware layer, which consists of the physical elements and minimum demand, temperature, voltage, and current
of the system. The second layer is the firmware layer, which boundaries based on each battery, safe operating limits, and
contains instructions for implementing analytics routines. rate of change in SOC/DOD etc. Each parameter would have
A flow chart of the entire computational process is depicted certain standard values that will be stored in the EEPROM of
in the figure 2. the edge controller. If any vector floats above the estimated
Zone I - Edge Monitoring: This zone encompasses the limit, the particular battery/load would shut down safely and
measurement and monitoring layer within the edge node. automatically.
A voltage sensor module based on a resistive divider circuit Zone IV - Edge analytics: Within the Edge node, interim
and a current sensor with an INA219 IC are incorporated analyses are performed to ensure safe operation and rapid
within the edge controller for battery output monitoring as decision making. In most cases, the battery deviations are
shown in figure 3. Sensors are connected to a WEMOS smooth. It can also differ beyond limits in certain circum-
D1 ESP8266/160MHz/4M Flash WIFI board, which will stances. As a result, anomalies in the curves are regularly
communicate via LPWAN. The voltage sensor is directly tested at regular intervals, and any deviation is immediately
attached to the analogue pin, while the current sensor is identified for further investigation. Over the local data, local
connected to the I2C model. The temperature is measured ordinary least squares method assumptions are applied to
in parallel with the voltage sensor using an LM35-based investigate if the battery is charging or discharging. The rate

VOLUME 10, 2022 115805

 C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

FIGURE 2. Flow chart of entire computational model.

of change of the curve is often established to distinguish Zone V – CDMAS Database Management: The orga-
local variations in SOC/DOD. Using local data, ARIMA nization of data in CDMAS obtained through LPWAN falls
is used to estimate future variations in the SOC/DOC if under the next operational zones. When the data is sent,
the edge controller has sufficient processing power. This it will be stored in a MariaDB at CDMAS. A Raspberry
information aids in the diagnosis of overloading, overheating, Pi 3 B+ configured with a Broadcom 2711 64-bit quad-core
excessively rapid charging, internal/external short circuits, Cortex-A72 processor, 2GB RAM, and 2.4GHz and 5GHz.
thermal runaway, and other issues. The wireless LPWAN model serves as the communication

115806 VOLUME 10, 2022

C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

damaging batteries may be prioritized. GSA is a weighted

population based optimization tool for identifying the right
solutions under technical and economic constraints. The
selection factors for the batteries should be well established
before applying the same. Based on the type of battery and
the operating conditions, each battery is assigned a specific
weight. Precision of measurements will be improved by
using dynamical weight distribution. Table 1 displays eight
weighted parameters for deciding the selection fitness of
batteries. The table explains the significance and expectations
of each variable. Now GSA is used to select the most
favorable sources based on these weights.
FIGURE 3. Edge measurement and analytics controller. The GSA is a modern meta-heuristic search optimization
algorithm that is based on gravity and mass interactions.
The term ‘‘gravitational force’’ refers to the tendency of
network. A separate table is created for each battery with a masses to accelerate toward each other. The gravitational
grand table that contains each battery, SOC/DOD and overall force between two particles is proportional to the product of
demand as the last column. their masses and inversely proportional to their distance as
Zone VI – CDMAS Metadata Analysis: This section shown in equation 2. F is the magnitude of the gravitational
examines the metadata in the CDMAS data and applies force, G is gravitational constant, M1 and M2 are the mass
it to evaluate the nature of the curve. Correlogram, auto- of the first and second particles respectively, and R is the
correlation, and partial correlation data aids in determining distance between the two particles.
the randomness and potency of forecasting. To extract the M1 ∗ M2
characteristics of the SOC curve, exploratory data analysis is F =G∗ (2)
R2
used. These analyses also provide inference on the maximum
Acceleration, a depends only on the force and its mass M as
and minimum demand, temperature, voltage, and current
described in equation 3.
boundaries for each battery, as well as the rate of change in
SOC/DOD. F
a= (3)
Zone VII – ARIMA: The next step entails forecasting M
the SOC (percentage and WH) using ARIMA, which will To use the GSA in this case, first create a general population
help with potential battery scheduling. After evaluating by labeling SOC of each battery Bi as mass where i is the
SOC for WH capacity, the SOC percentile is estimated. battery index. Now, for iteration n, the force of interaction
The auto-regressive term (p), the number of nonseasonal between masses is determined using equation 4, and the better
deviations required for stationary (d), and the number and worst models are defined using equations 5, 6 and 7 for
of lagged forecast errors (q) are the three constitutional population of N .
parameters used to calculate ARIMA. For a battery Bi whose
n Gn ∗ Mi (n) ∗ Mj (n) ∗ (Mj (n) − Mi (n))
SOC is Bt and k is a constant, the equation 1 represents a Fi,j = (4)
mathematical model for predicting SOC using ARIMA where R2i,j ∗ 
θ and ε are autoregressive and moving average operators. best(n) = min fitj (n) (5)
∀j∈1,..,n

Bt = k + ϕt ∗ BIt + . . . . . . . . . + ϕp ∗ BIt−p worst(n) = max fitj (n) (6)

∀j∈1,..,n
+ θt εtI + . . . . . . . . . + θq ∗ εt−q
I
+ εt (1) fit i (n) − worst(n)
mi (n) = (7)
best(n) − worst(n)
Zone VIII - Multi-objective GSA: After SOC of each
After calculating the individual weights and forces, equation
sources has been forecasted, the next step is to schedule
8 is used to assess the overall force.
the battery. As previously explained, the cumulative capacity
n
of batteries will usually surpass the actual demand. Only X
Fi (n) = randj ∗ Fi,j (n) (8)
a few of the available resources are used to satisfy the
jk,bestj 6 =1
demand, while the remainder are turned off or left idle.
Selecting the right sources from all available sources to Zone IX - Probabilistic GSA: Using the methodology
satisfy demand is a challenging task. Each battery has its own described above, GSA can identify the best sources. However,
set of benefits and drawbacks. The best batteries with the a typical GSA for battery distribution can cause power quality
least amount of power loss and the longest length of supply issues. The method would attempt to select the best samples
should be chosen, with particular attention paid to the cell from all available sources. As previously said, sources with
aging and deterioration. Because each battery has a specific the shorter time period in usage and sources that have been
make, Capex, Opex, and aging price diminution, the least used the most historically will be given more weight. As a

VOLUME 10, 2022 115807

 C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

TABLE 1. Objectives of the MO-GSA.

result, each iteration chooses new sources. This is a problem TABLE 2. Capacity of battery system.
because repeated battery integration and disintegration can
degrade battery quality and decrease power quality due
to the transient effect. Selective filtering and probabilistic
evaluation of the population of GSA can solve this problem.
In general, GSA models generate a random population of
sources. Rather than using a random sample, existing sources
are blended in with the general population and assessed
for fitness. However, combining all of the sources would
reduce the diversity and heterogeneity of the population.
As a result, with the help of a Naive Bayes probabilistic
estimator, an intermediate sampling is generated, which aids
in the formation of a sample population from currently linked be linked to embedded C for analysis and Python for the
sources. P(c|x) represents the probability of current linked dashboard, as C++ is quicker for data analysis than Python.
sources, P(c) represents the prior probability of selection,
P(x|c) represents the likelihood, which is the probability of A. DATA MONITORING AND PREPROCESSING
the indicator given population, and P(x) represents the last
The first phase of the designed method integrates the
linked probability, equation 9 is being used to select the
measurement and monitoring system. An energy monitoring
sample population.
module is built into the edge controller to track each battery
and charge activities continually. Modules individually mea-
P(c ∨ x) = P(x1 ∨ c) ∗ P(x2 ∨ c) ∗ . . . . . . . . . . . . ∗ P(xn ∨ c) sure each battery’s OCV and SCC, load voltage and current,
(9) instantaneous temperature and so on every millisecond. This
information is utilized for the assessment of battery SOC
III. EXPERIMENTAL RESULTS AND DISCUSSION (WH and percent) and demand. Processing all of the data
As previously said, the operations are carried out in Hardware using only an edge controller is very difficult. So, the
in a Loop model. The experimental power system is built data is split into two classes. The measured values were
up as a DC microgrid architecture with a maximum storage immediately moved to the CDMAS for primary analysis.
capacity of 8140 WH and differential resistive loads. In this The second step of the process would retrieve valuable
experimental arrangement, as shown in table 2, forty two specimen data from all of the big data and process it within
different batteries of six different make are used. The batteries the edge computing model. The last comprehensive power
ranging only from 5 WH to 1000 WH are used for research capacity, last SOC (% and WH) of each battery, last measured
rather than the use of high-capacity batteries. These great current and voltage, last demand, and other data will be
differences in the make and capacity of batteries are intended stored as historical data in the edge controller’s EEPROM.
to give versatility and heterogeneity in the analysis. The This information will be updated as new data becomes
demand is also varied even from minimum capacity to accessible. This is useful for quickly detecting faults and
maximum capacity for ensuring reliability. The Node RED abnormal situations, as well as performing fault dissolution
modules may create an IoT link at the server end. This may by protective relaying. The figure 4 displays sample data in

115808 VOLUME 10, 2022

C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

FIGURE 4. SOC data stored in SQL DB for ten batteries at five stamps.

TABLE 3. Sample data for all the battery samples.

FIGURE 6. Charge cycle deviations of nine batteries.

FIGURE 7. Charge cycle for single charge – discharge.

FIGURE 5. Real-time SOC dashboard with Node-Red for one battery at

CDMAS.

the CDMAS SQL database which provides sample SOC data FIGURE 8. Charge cycle for multiple charge – discharge.
from 10 batteries over a span of five different time stamps.
Table 3 displays sample data from all battery sources for
a particular time instant. Continuous curve data is more The charge cycle with single charge/discharge is depicted
significant than instantaneous data and it can be used to in figure 7. A battery does not necessarily need to be charged
predict battery parameters and make decisions. The figure 5 and discharged in a single cycle. At times, a battery will
displays real-time SOC data from one individual battery on begin charging even though it is not completely discharged.
the dashboard supported by python plotting. As previously This is particularly true when a battery isn’t loaded for
said, each battery has its own set of characteristics and a certain period of time but a higher demand is expected
parameters. As a consequence, each battery’s charging and in the near future. A multi-charge/discharge battery cycle
discharging characteristics can differ. The diagram 6 depicts is visualized in figure 8. Within those circumstances, each
a typical battery charge cycle curve of nine different batteries charge and discharge cycle must be forecasted separately. The
over time. Even if all of the batteries began charging at charging status and strategy are calculated at first. Charging
the same time, they charge and discharge in various ways is usually performed in a smooth and continuous process.
depending on their type, capacity, and attached load. As a Pulse charging is a rapid charging method for Lithium ion
consequence, battery scheduling is a time-dependent process batteries. A charging loop over a millisecond time interval
in which estimates must be redefined based on the state of is demonstrated in figure 9 and the charging pulse can be
charge of each battery at the time. distinguished. These charge pulses may be observed as an

VOLUME 10, 2022 115809

 C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

FIGURE 9. Changes in SOC during battery pulse charging. FIGURE 12. Four discharging paths.

FIGURE 13. SOC (%) - multiple charging cycles.

FIGURE 10. Changes in SOC – three charging methods.

The discharge curve of the same battery for four different

SOCs and four different loads could be seen in the figure 12.
The same battery is used with four different loads and
four different initial SOCs in this experiment. It has been
discovered that the discharging curve differs depending on
SOC and load parameters. Since the charging and discharging
curves differ depending on current SOC state, charging type,
load, battery type, load type etc, there will be several SOC
curves that overlap, as seen in figure 13. Figure shows
the cases of SOCs of four different batteries during a one
hour time-frame. As a result, a dynamic method is used to
FIGURE 11. SOC (WH) - four discharging path. calculate the SOC and other battery parameters. The values
of instantaneous battery parameters, as well as temporal data,
anomaly during analysis. The curve is smoothed in the next are considered for analysis simultaneously. The analysis is
stage to avoid this problem. Since the pulses are measured in done independently for each curve, and then the data is used
mWH, the forecasting accuracy is unaffected. to make an estimate.
The charging curves of three different batteries of the same
capacity with three different charging cases have been seen in B. METADATA ANALYSIS AND FORECASTING
diagram 9. It’s important to keep in mind that as the charging A Correlogram may be used to evaluate the correlation
methodology changes, so does the charging time. Similarly, between two or more curves, as well as the randomness
figure 11 shows the discharge curves of four separate batteries between the curves. Figure 14 shows a Correlogram of nine
for a sample load. The discharge curves for four distinct sample battery SOCs, with only a few displaying a higher
batteries, each with a different initial SOC, are shown. correlation. This data reveals that the curves have a high level
The charging curves of three different batteries of the same of randomness, necessitating a complex parallel analysis for
capacity with three different charging cases have been seen in such a model. In time series analysis, the auto-correlation
diagram 10. It’s important to keep in mind that as the charging curve aids in identifying fluctuations and randomness in the
methodology changes, so does the charging time. Similarly, results over time.
figure 11 shows the discharge curves of four separate batteries Auto-correlation and partial correlation of a sample curve
for a sample load. The discharge curves for four distinct are depicted in figures 15 and 16. The data indicates a
batteries, each with a different initial SOC, are shown. moderately smooth transition of the curve. This illustrates

115810 VOLUME 10, 2022

C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

FIGURE 17. Forecasting the charging SOC (%).

FIGURE 14. Correlogram of nine sample battery SOCs.

FIGURE 18. Forecasting the discharging SOC (WH).

FIGURE 15. Auto- correlation of SOC curve.

FIGURE 19. Forecasting of the SOC for a complete cycle.

charging SOC of a battery is shown in the diagram 17 under

two conditions. In the first example, the battery is fully
FIGURE 16. Partial correlation of SOC curve. charged and turned off. In the second example, the battery
is filled to quarter capacity and then left uncharged or
unused. The figure 18 illustrates the real and expected
that individual curve variances are modest and consistent. discharge SOC values of a battery for a uniform load.
However, when it comes to multiple batteries, there is a When individual charging and discharging SOC estimates
significant disparity in the variability of individual SOC are completed, the SOC over the complete battery cycle
curves. If the system has a limited degree of randomness, may be calculated. SOC forecasting for a battery cycle
a Naive Bayes projection is used to approximate the curve. with multiple charging - discharging curves is shown in the
A model like this is incredibly fast as compared to ARIMA. diagram 19.
If the curves have a higher randomness, forecasting of each
charging and discharging curve is performed individually C. PROBABILISTIC MO - GSA
using ARIMA. The cumulative SOC capacity of batteries may be much
The charging condition of batteries is forecasted initially, greater than demand at any given time, and only a few
followed by discharge forecasting. The actual and forecasted batteries will be required to meet real demand. Queuing,

VOLUME 10, 2022 115811

 C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

TABLE 4. MO-GSA populations generated from linked sources using NBPE.

TABLE 5. MO-GSA implementation and optimal source scheduling.

TABLE 6. MO-GSA implementation and optimal source scheduling for multiple loads.

on the other hand, would make all accessible outlets available, the population’s diversity and competitiveness. To assess
regardless of their utility. As a result, selecting the best selection parameters for adding existing linked sources to
suitable batteries from a wide range of options is critical. the population, the Naive Bayes probability estimator is
It is a complex process to calculate the optimum battery used. A variety of factors influence selection, including unit
package from all available batteries to ensure that the cost, temperature, SOC (percentage), SOC(WH), charging
cumulative WH exceeds or meets the demand with the availability, last used time, apparent lifespan, number of
slightest conceivable difference, smaller power losses and a charge cycles, and so on. Each battery is assigned a
larger financial bufferment. A multi-objective optimization specific priority based on these data. An example of the
algorithm can help to solve these problems. The best selection status of ten random batteries is shown in the
battery combination is computed by the MO-GSA from table 4.
all available sources. Heuristically, GSA looks for optimal If the population has been determined, the GSA algorithm
sources between all available sources in all possible angles. is used to find the best sources. GSA will select all optimum
Eight constraints as mentioned in table 1 are applicable to sources that meet all of the constraints and parameters.
the set of selected population of sources in line with socio- A sample data set for optimum battery allocation for
technological constraints. As already stated, the development a 756 WH requirement is shown in the table 5. While there
of a random population for the GSA often results in are seventeen batteries in all, only a few will be enough
frequent swapping between batteries due to constraints that to satisfy demand. It is worth noting that 35 percent of the
focus on charge cycle count, charging time, SOC, life sources are already attached and are measured using NBPE.
expectancy etc. The GSA will assess the battery distribution based on the
As a result, the population is constructed in such a manner weight targets allocated to each battery, as seen in the table
that a significant portion of all batteries that are now linked 5. The final distribution is seen in the last row of the table.
to the utility is still included in the overall population. The sample selection status of the batteries is seen in the
If production is less than 50 % of all usable power, this could table 6 for various loads of the same general population and
account for 50 % of the total population. As the percentage varying battery periods. Table depicts the fitness estimate
share of demand over total battery capacity rises, it will after multiple iterations. Within iterations, fitness will change
rise as well. However, choosing all connected sources limits before arriving at the final solution.

115812 VOLUME 10, 2022

C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

TABLE 7. Comparison of mathematical models. The benefits of edge analytics blended with streamlined mon-
itoring of multi-zonal processes. The developed hardware
monitored the battery parameters, and the data was classified
into two groups using metadata analysis. All big data was
delivered to a central server over an LPWAN as the initial
block of information. The second batch of data used safe
TABLE 8. Comparison of GSA with NPBE - GSA. boundary interpretation within the edge controller, where
data was processed in the short term and decisions were
taken quickly. A variety of numerical tools were used by
the central processor. ARIMA was first used to forecast the
SOC. The MO-GSA was then utilized to plan the appropriate
battery distribution based on a variety of objectives, including
battery temperature runaway, unit consumables such as duty
The performance of the deployed framework is now being period, last time of use, SOC (percent and WH), and so
evaluated using equivalent mathematical approaches. The on. A Naive Bayes Probabilistic Estimator was sandwiched
evaluation is performed in two stages. First, the performance between ARIMA and GSA to select the optimum general
of MOGSA is compared to that of other optimization methods population for GSA, avoiding repetitive battery switching
such as MO-GA, MO-SA, and MO-HC. Each model receives and maximizing power usage. In the Hardware in a Loop
the identical inputs, including the same load demand, SOC, concept, the entire system was tested. The time cycle and
Battery population, and constraint matrix. Two parameters fitness of conventional GSA and Probabilistic MO-GSA
are used to evaluate the performance of a model. Consider the were discovered, and it was determined that the revised
following three load levels: 1/3 load, 1/2 load, and 3/4 load. model was capable of identifying the best outcomes with the
These three load levels are successfully applied to a wide fewest number of iterations and time. The performance of the
population of all sources, and the best sources are chosen constructed system is evaluated by comparing the results to
using all four approaches. Two aspects of the result are those of several similar models such as MO-GA, MO-SA, and
chosen for evaluation. As previously stated, the total source MO-HC. The advantage of NPBE is evaluated by comparing
capacity chosen is always greater than the load demand. The the outcomes of NPBE - GSA with conventional GSA.
best optimal strategy should keep the demand-supply gap as A follow-up study focusing on load scheduling and shedding
small as feasible. A table 7 shows the Reserve Power (RP) using bio-inspired methods for BMS will be undertaken in
for each source, with MO-GSA having the lowest reserve the near future.
allocation of all sources. Another factor to consider while
assessing is the number of battery sources to be employed. REFERENCES
Because the battery life decreases with each use, selecting
[1] F. Mohammadi, B. Mohammadi-Ivatloo, G. B. Gharehpetian, M. H. Ali,
the configuration with the fewest battery allocations is always W. Wei, O. Erdinc, and M. Shirkhani, ‘‘Robust control strategies for
the best choice. In virtually all situations, MO-GSA employs microgrids: A review,’’ IEEE Syst. J., vol. 16, no. 2, pp. 2401–2412,
Jun. 2022.
the fewest number of battery sources feasible.
[2] K. Komala, K. P. Kumar, and S. H. C. Cherukuri, ‘‘Storage and
The performance of MO-GSA is compared to that of non-storage methods of power balancing to counter uncertainty in
NBPE-MOGSA using the percentage of exclusion approach. hybrid microgrids—A review,’’ J. Energy Storage, vol. 36, Apr. 2021,
The customary GSA will seek to engage a wide set of Art. no. 102348.
[3] L. Luo, S. S. Abdulkareem, A. Rezvani, M. R. Miveh, S. Samad,
general battery population, independent of their prior usage. N. Aljojo, and M. Pazhoohesh, ‘‘Optimal scheduling of a renewable based
However, because the objective function is always looking for microgrid considering photovoltaic system and battery energy storage
the least-used batteries, frequent switching between power under uncertainty,’’ J. Energy Storage, vol. 28, Apr. 2020, Art. no. 101306.
sources will happen. By integrating some elements from [4] X. Lin and R. Zamora, ‘‘Controls of hybrid energy storage systems in
microgrids: Critical review, case study and future trends,’’ J. Energy
previously connected sources, NPBE - GSA is used to Storage, vol. 47, Mar. 2022, Art. no. 103884.
segregate particular elements from the rest of the population. [5] B. Yang, J. Wang, Y. Chen, D. Li, C. Zeng, Y. Chen, Z. Guo, H. Shu,
As a result, the best model is one that has the fewest variants X. Zhang, T. Yu, and L. Sun, ‘‘Optimal sizing and placement of energy
storage system in power grids: A state-of-the-art one-stop handbook,’’
while still fulfilling the primary objective. The table 8 shows J. Energy Storage, vol. 32, Dec. 2020, Art. no. 101814.
the proportion of deviations in traditional GSA and NPBE - [6] G. Kaur, A. Prakash, and K. U. Rao, ‘‘A critical review of microgrid
GSA. It’s worth mentioning that in the general population, adaptive protection techniques with distributed generation,’’ Renew.
NPBE- GSA has lower variances than GSA. Energy Focus, vol. 39, pp. 99–109, Dec. 2021.
[7] M. Woody, M. Arbabzadeh, G. M. Lewis, G. A. Keoleian, and A.
Stefanopoulou, ‘‘Strategies to limit degradation and maximize li-ion
IV. CONCLUSION battery service lifetime—Critical review and guidance for stakeholders,’’
This paper established an adaptive framework for battery J. Energy Storage, vol. 28, Apr. 2020, Art. no. 101231.
management to schedule batteries of different sizes, capacity, [8] S. Liu, J. Wang, H. Liu, Q. Liu, J. Tang, and Z. Li, ‘‘Battery degradation
model and multiple-indicators based lifetime estimator for energy storage
performance, life and charging models with lower power system design and operation: Experimental analyses of cycling-induced
loss, improved battery life, and higher financial benefits. aging,’’ Electrochimica Acta, vol. 384, Jul. 2021, Art. no. 138294.

VOLUME 10, 2022 115813

 C. R. Sarin, G. Mani: Intelligent BMS With Probabilistic MO-GSA

[9] X. Li, C. Yuan, X. Li, and Z. Wang, ‘‘State of health estimation for [23] Z. Wang, D. T. Gladwin, M. J. Smith, and S. Haass, ‘‘Practical state
li-ion battery using incremental capacity analysis and Gaussian process estimation using Kalman filter methods for large-scale battery systems,’’
regression,’’ Energy, vol. 190, Jan. 2020, Art. no. 116467. Appl. Energy, vol. 294, Jul. 2021, Art. no. 117022.
[10] H. Dai, B. Jiang, X. Hu, X. Lin, X. Wei, and M. Pecht, ‘‘Advanced battery [24] M. S. H. Lipu, M. A. Hannan, T. F. Karim, A. Hussain, M. H. M. Saad,
management strategies for a sustainable energy future: Multilayer design A. Ayob, M. S. Miah, and T. M. I. Mahlia, ‘‘Intelligent algorithms and
concepts and research trends,’’ Renew. Sustain. Energy Rev., vol. 138, control strategies for battery management system in electric vehicles:
Mar. 2021, Art. no. 110480. Progress, challenges and future outlook,’’ J. Cleaner Prod., vol. 292,
[11] Y. Li, D.-I. Stroe, Y. Cheng, H. Sheng, X. Sui, and R. Teodorescu, Apr. 2021, Art. no. 126044.
‘‘On the feature selection for battery state of health estimation based on
charging–discharging profiles,’’ J. Energy Storage, vol. 33, Jan. 2021,
Art. no. 102122.
[12] Y. Wang, J. Tian, Z. Sun, L. Wang, R. Xu, M. Li, and Z. Chen,
‘‘A comprehensive review of battery modeling and state estimation
approaches for advanced battery management systems,’’ Renew. Sustain.
C. R. SARIN (Member, IEEE) received the B.E.
Energy Rev., vol. 131, Oct. 2020, Art. no. 110015. degree in electrical and electronics engineering
[13] M. Faisal, M. A. Hannan, P. J. Ker, M. S. A. Rahman, R. A. Begum, and the M.E. degree in mechatronics engineer-
and T. M. I. Mahlia, ‘‘Particle swarm optimised fuzzy controller for ing from Anna University, in 2010 and 2012,
charging–discharging and scheduling of battery energy storage system in respectively, and the M.B.A. degree in human
MG applications,’’ Energy Rep., vol. 6, pp. 215–228, Dec. 2020. resource management from Annamalai University,
[14] C. Liu, Y. Qin, and H. Zhang, ‘‘Real-time scheduling strategy for in 2015. He is currently pursuing the Ph.D. degree
microgrids considering operation interval division of DGs and batteries,’’ in applications of machine learning in the energy
Global Energy Interconnection, vol. 3, no. 5, pp. 442–452, Oct. 2020. sector with the Vellore Institute of Technology,
[15] S. Wang, D. Guo, X. Han, L. Lu, K. Sun, W. Li, D. U. Sauer, and M. Vellore. He has been an Assistant Professor with
Ouyang, ‘‘Impact of battery degradation models on energy management the Department of Electrical and Electronics, Vimal Jyothi Engineering
of a grid-connected DC microgrid,’’ Energy, vol. 207, Sep. 2020, College, Kannur, Kerala, since June 2012. His research interests include
Art. no. 118228. intelligent energy management systems, embedded system and applications
[16] Z. Yi, D. Scoffield, J. Smart, A. Meintz, M. Jun, M. Mohanpurkar, of swarm intelligence in robots, and electric vehicles.
and A. Medam, ‘‘A highly efficient control framework for centralized
residential charging coordination of large electric vehicle populations,’’ Int.
J. Electr. Power Energy Syst., vol. 117, May 2020, Art. no. 105661.
[17] P. Huang, M. Lovati, X. Zhang, and C. Bales, ‘‘A coordinated control
to improve performance for a building cluster with energy storage,
electric vehicles, and energy sharing considered,’’ Appl. Energy, vol. 268, GEETHA MANI (Member, IEEE) received the
Jun. 2020, Art. no. 114983. B.E. degree in electronic and instrumentation
[18] A. Dini, S. Pirouzi, M. Norouzi, and M. Lehtonen, ‘‘Hybrid stochas- engineering from Madurai Kamaraj University, the
tic/robust scheduling of the grid-connected microgrid based on the linear M.E. degree in instrumentation engineering from
coordinated power management strategy,’’ Sustain. Energy, Grids Netw., MIT Campus, Anna University, India, and the
vol. 24, Dec. 2020, Art. no. 100400. Doctor of Philosophy degree in advanced process
[19] V. Preetha, K. Vavilapalli, N. K. Jeevan, and P. Abilash, ‘‘Model- control from Anna University, Chennai. She has
based hardware-in-the loop testing of battery management system,’’ SAE, been teaching for 14 years at reputed institutions.
Warrendale, PA, USA, Tech. Paper 2022-28-0388, 2022. She is currently an Associate Professor with the
[20] C. Feng, Y. Wang, Q. Chen, Y. Ding, G. Strbac, and C. Kang, ‘‘Smart grid School of Electrical Engineering, Vellore Institute
encounters edge computing: Opportunities and applications,’’ Adv. Appl. of Technology, Vellore, India. She was a recipient of the IEI Young Engineers
Energy, vol. 1, Feb. 2021, Art. no. 100006.
Award to recognize her contribution in the field of electrical engineering
[21] G. Crocioni, D. Pau, J.-M. Delorme, and G. Gruosso, ‘‘Li-ion batteries
parameter estimation with tiny neural networks embedded on intelligent
from the Institution of Engineers, India, in 2016. She also authored and
IoT microcontrollers,’’ IEEE Access, vol. 8, pp. 122135–122146, 2020. published several research articles in peer-reviewed international journals
[22] R. Brännvall, M. Siltala, J. Gustafsson, J. Sarkinen, M. Vesterlund, and and book chapters. Her research interests include process control, soft
J. Summers, ‘‘EDGE: Microgrid data center with mixed energy storage,’’ sensing, industrial automation, and the Internet of Things.
in Proc. 11th ACM Int. Conf. Future Energy Syst., Jun. 2020.

115814 VOLUME 10, 2022