Proceedings of the Second International Conference on Edge Computing and Applications (ICECAA 2023)

IEEE Xplore Part Number: CFP23BV8-ART; ISBN: 979-8-3503-4757-9

Li-ion Battery Life Estimation using K-Nearest

Neighbor Algorithm
Allirani S Pooja K Soundarya E
Associate Professor , Department of Electrical and Department of Electrical and
Department of Electrical and Electronics Engineering Electronics Engineering
2023 2nd International Conference on Edge Computing and Applications (ICECAA) | 979-8-3503-4757-9/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICECAA58104.2023.10212304

Electronics Engineering Sri Ramakrishna Engineering College, Sri Ramakrishna Engineering College,
Sri Ramakrishna Engineering College, Coimbatore Coimbatore
Coimbatore [email protected] [email protected]
[email protected]
Sreeram S Nair
Department of Electrical and
Electronics Engineering
Sri Ramakrishna Engineering College,
Coimbatore
[email protected]
Abstract- Accurate Li -Ion battery life cycle prediction is As battery voltage cannot be utilised as the sole
crucial for the Battery Management System (BMS). In measurement for predicting capacity, lithium
comparison to straight methodological approaches and batteries are more challenging to estimate than lead-
prototype strategies, statistics approaches have been garnering acid batteries. Compared to alkaline or lead-acid
a lot of interest in digital State of Health (SO H) estimation batteries, lithium-ion batteries have a significantly
be cause of its defined structure, scalability for online process, flatter discharge profile, which results in much less
and inde pe ndent from proposed converter. O nce after voltage loss for a given amount of capacity depleted.
e xtracting a robust health indicator to gauge the SO H, a When battery capacity is determined solely by
supe rvised learning algorithm is built utilizing data-driven
voltage, this leads to erroneous evaluations of the
te chniques to map the relationship between them. A partly
charge and discharge current cycle is suggested in this concept, battery's capacity. For instance, over 80% of their
which is constructed using insights of battery deterioration SOC, lithium-ion, lithium iron phosphate cells
me trics. The input is then sent into the KNN algorithm, which virtually show no voltage change. The amount of
has a strong capacity for generalization and rapid learning. load on the battery also affects the cell voltage
The KNN algorithm, which has quick l earning and high because to Ohmic and other losses. Many portable
ge neralization capabilities, is then given the data. The impact lithium battery systems solely estimate capacity using
on e stimating accuracy is thoroughly assessed when charging voltage, which might result in inaccurate readings
and discharging capacities are chosen. A publicly available and confusion. This frequently happens when a
dataset is used to test the proposed strategy, and the outcomes lithium battery's high current drain leads in an
show that it works as expected. The primary battery SO H
apparent low voltage situation and a falsely low
e stimation techniques are reviewed in this research, together
with their key benefits and drawbacks for real-time automobile
capacity measurement. For instance, drone operators
compatibility, particularly in the case of hybrid electric are instructed to disregard alerts about low battery
syste ms. capacity when their machines do challenging
manoeuvres because the high current drain appears to
Ke ywords- Battery Management System (BMS), Health result in low voltage on the battery. Lithium batteries
Indicator (HI), State of Health (SO H), K-Nearest Neighbor require a more intricate algorithm to produce precise
Algorithm (KNN).
capacity readings while accounting for situations like
I. INTRODUCTION these. A lithium battery-powered system is unreliable
unless the BMS can accurately read the capacity. The
Li-ion batteries are widely used in various SOC sensor in the BMS mentioned in this study is
electronic devices and applications due to their high substantially more accurate than the SOC sensor in
energy density and long cycle life. Accurate comparable batteries because it uses numerous
estimation of the remaining battery life is crucial for sensors and detection techniques to measure the
optimizing device performance, preventing capacity of a lithium battery while accounting for
unexpected shutdowns, and enhancing user Ohmic losses and other polarisation effects.
experience. In recent years, the k-nearest neighbor
(KNN) algorithm has emerged as a promising The KNN algorithm is a non-parametric machine
approach for Li-ion battery life estimation due to its learning algorithm that relies on the similarity
ability to capture complex non-linear relationships between instances to make predictions. In the context
and adapt to varying battery conditions. of Li-ion battery life estimation, the KNN algorithm
identifies the k most similar historical battery

979-8-3503-4757-9/23/$31.00 ©2023 IEEE 1606

Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:31:49 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Edge Computing and Applications (ICECAA 2023)
IEEE Xplore Part Number: CFP23BV8-ART; ISBN: 979-8-3503-4757-9

instances from a training dataset and uses their reactive power from the DG, but not from the ESS, is
known battery life values to estimate the remaining also considered.
battery life of a test instance. By leveraging the local
In the reference [14] a convex FPO was proposed,
context and considering neighboring instances, the
KNN algorithm can provide accurate estimations for using the second-order conical model, to perform the
allocation of ESS in SD with the objective of
the remaining battery life.
reducing daily operating costs. Then, the generalized
This research contributes to the field of Li-ion Benders decomposition is used to speed up the model
battery management by providing a reliable and solution. Finally, the solution of the problem is
efficient method for estimating battery life. The further accelerated by relaxing integrality constraints
developed model can be integrated into battery
Finally, [15] proposed a joint optimization model
management systems and enable better decision-
making regarding battery us age and optimization. for ESS and substation allocations, in addition to
network reconfiguration with the aim of reducing
The insights gained from this study can also be
applied to other domains requiring accurate investment costs, active losses, cutting generation of
PVDG and WDG and load shedding. The model is
estimation of remaining life or degradation.
formulated using an exact convex relaxation of the
The proposed system offers a practical and FPO problem, and takes into account: use of OLTC
effective solution for optimizing battery performance controls, capacitor banks, static reactive
and enhancing user experience in various compensators and reactive dispatch of the DG, load
applications. uncertainties, PVDG and WDG, as well as the
possibility of choosing the type of ESS to be
II. LITERATURE REVIEW allocated.
The majority of current works transmit raw data
III. PROPOSED SYSYEM
to the cloud for health prediction, increasing the
transmitter's power requirements. A special iThing The machine learning-based charge and discharge
architecture was developed that could execute the analysis with the k-nearest neighbor technique is
SOH estimation and RUL prediction on-board what we're applying in this proposed Li-ion battery
utilising peak extraction approach with little compute life estimation system (KNN algorithm). The most
load and storage demand because standard health basic machine literacy algorithm is KNN, which is
prediction processes do not work for IoT devices based on the supervised literacy paradigm. The
deployed in the wild. With the same data sets, which suggested system is applicable to all Li-Ion battery
have a lower precision, the life and health of every types as well as other battery types. A dataset of each
battery was compared and identified. And previous bracket is obtained and trained to predict the
research primarily focused on predicting the health of outcomes accurately. An ATMEGA 328P
lead acid batteries; it was incompatible with Li-ion microprocessor serves as the main regulator for this
batteries and other lithium kinds. system that monitors the battery's charging and
discharging status. An ESP8266 Wi-Fi module is
The problem of optimal allocation of storage
used to record the values and send them to the server.
systems is addressed in [12] using an exact convex Through a webpage, the Li-ion battery's current state
FPO model recently proposed by the authors
is shown. both the charging and the discharge rate is
themselves [11]. In addition, network reconfiguration kept in amp per hour form. Additionally, the solution
was considered to obtain seasonal configurations of
uses the KNN algorithm, that either categorises a
the distribution system topology based on security
battery's count of used cycles by contrasting the
constraints and with the aim of reducing technical tackle records out from pall of the being test data.
losses. Uncertainties related to load and renewable
Out from total count of utilized cycles, the leftover
generation were considered from a scenario creation cycles and life are calculated, and the result is then
and subsequent reduction approach. Benders
displayed on the user's mobile device. The design
decomposition was used to divide the model into a was developed with the assumption that electric
master problem and some independent subproblems
vehicles would play a significant role in
that can be solved in parallel.
transportation in the future and might benefit from
Literature [13] proposed a linearized FPO model this design. The tackle factors gather the input
with an operational scheme based on model battery's data and store it in the pall when a battery is
predictive control to perform the allocation of ESSB supplied as an input to the system. The battery's
distributed in the SD. In this representation, the information is provided as an input to the machine
ESSB are allocated in the system with the objective literacy system that has been trained; additionally, the
of performing energy price arbitration and reducing system analyses the data using the learned data sets
the PVDG cut. The proposed problem also takes into and produces a result. Li-ion battery’s State of
account a battery life calculation model according to Charge is the main piece of information used to
its operation. Finally, the possibility of dispatching calculate the battery's lifetime. The vaticination

979-8-3503-4757-9/23/$31.00 ©2023 IEEE 1607

Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:31:49 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Edge Computing and Applications (ICECAA 2023)
IEEE Xplore Part Number: CFP23BV8-ART; ISBN: 979-8-3503-4757-9

model also takes into account the batteries'

temperature, current, and voltage. These are the
battery-related facts that were discussed before in the
environment. As a result, following system analysis,
it shows how long the battery lasts, which also
enables to determine when to get a new battery in
place of the old one. The design was developed with
the assumption that electric vehicles will play a
significant role in transportation in the future, and
that this design canbe practical for that. Also, this
design will assist the diligent in determining the
demand for batteries for E-Vehicles based on
information on when the client will change their
battery.
IV. PROPOSED BLOCK DIAGRAM
The proposed Block Diagram of Li-ion life Fig. 3. Hardware Simulation
estimation using KNN Algorithm is shown in Fig. 1.
The system has an integrated Atmega328p
microcontroller, which interfaces current, voltage,
and temperature sensors to A0, A1, and A2
respectively and has built-in ADC features. Once the
system starts, the charging process begins, which is
controlled using a Darlington transistor using the D2
pin in the microcontroller. Then, the discharging part
is controlled using a similar darlington transistor
from D3 pins in the microcontroller. All the collected
data is deduced.
By considering Voltage Measurement Accuracy,
Sampling Rate, Power Efficiency, Compatibility and
Integration, Noise Filtering and Stability, Size and
Form Factor, Cost, and Reliability and Durability and
Fig.1. Block Diagram of Li-ion life estimation using KNN evaluating various sensor options based on their
Algorithm performance, compatibility, power efficiency, and
cost, the most suitable sens or can be selected for
The circuit Diagram of Li-ion life estimation
estimating Li-ion battery life using the KNN
using KNN Algorithm is shown in Fig. 2. algorithm. It is important to thoroughly test and
validate the chosen sensor in real-world scenarios to
ensure accurate and reliable battery life predictions.
V. GETTING ARDUINO UNO READY FOR
PROGRAMMING ESP8266
ESP8266 module must be linked to a computer in
order to programme it. USB is preferred to Serial
Adapter like an FTDI because the Module ESP8266
ESP-01 most effectively helps in serial connection.
An on-board USB to Serial Adapter on the Arduino
UNO is utilised to programme the Arduino. This is
utilized in this work to programme the ESP8266. To
make sure Arduino isn't using the TX and RX pins
that is used to connect to the ESP8266 Module, bare
minimum sketch can be uploaded to Arduino.

Fig. 2. Circuit Diagram

The ESP8266 serves as a data acquisition and
communication module in the Li-ion battery life
The hardware simulation of Li-ion life estimation estimation system. It acquires battery parameters
using KNN Algorithm is shown in Fig. 3. through integrated or external sensors, preprocesses
the data, wirelessly transmits it to a cloud server, and
collaborates with the server to execute the KNN

979-8-3503-4757-9/23/$31.00 ©2023 IEEE 1608

Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:31:49 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Edge Computing and Applications (ICECAA 2023)
IEEE Xplore Part Number: CFP23BV8-ART; ISBN: 979-8-3503-4757-9

algorithm for battery life estimation. Its power

management capabilities contribute to overall energy
efficiency and extended battery life.
NOTE: The setup and loop routines are the only
ones in the bare minimum sketch; all other functions
lack data. In this situation, a spare Arduino UNO
Board with a broken ATmega328p IC is available. So
the Arduino UNO's microcontroller was taken out
and was started to use as a USB to Serial Converter
VI. WORKING OF ESP8266 ARDUINO
INTERFACE
Ensure that all of the connections indicated above
are made correctly. To the system connect the
Auduino after setting up the ESP8266 in
Programming Mode (GPIO0 is linked to GND). Press
the RST button and open the Arduino IDE once the Fig. 4. Hardware prototype
ESP8266 module has been powered on. Choose the
"Generic ESP8266" Board from the Board options
(Tools -> Board). In the IDE, choose the correct port
number. Open the Blink Sketch now, and switch the
LED Pin to position 2. In this case, 2 refers to the
ESP8266 module's GPIO2 pin. It is verified that
GPIO0 is attached to GND before pressing the
upload button, and then the RST button is pushed.
While the add button is clicked, the compilation and
upload of the code will take some time. The status
bar is shown at the end of the IDE. GPIO0 van be
disconnected from GND once the application has
been successfully uploaded. The LED that is attached
to GPIO2 will flash. Fig. 5. Hardware result
VII. KNN ALGORITHM TRAINED CLASS By applying the strategies such as Feature
K-Nearest Neighbor comprises one of the basic Selection, Data Preprocessing, Optimal K-Value
overseen machine learning algorithms. Selection, Distance Metric Selection, Data
Augmentation, Cross-Validation and Model
The K-NN algorithm places the new data point in Evaluation, Ensemble Methods, Model Updating and
the category that matches the present categories the Adaptation, Domain Expertise and Domain-Specific
most, assuming that the new model and the previous Adjustments and continuously refining the Li-ion
reports are related. battery life estimation process using the KNN
After storing all of the past information, a new algorithm, the performance and accuracy of the
data point is categorised using the K-NN algorithm predictions can be improved, leading to more reliable
based on similarity. This indicates that new data can and useful battery life estimates for various
be reliably and quickly categorised using the K-NN applications.
approach. Achieving high accuracy involves several
The K-NN technique can be used for regression strategies and considerations such as High-Quality
even though classification problems are where it is Training Data, Relevant Features Selection, Data
most typically applied. Preprocessing, Optimal K-Value Selection, Distance
Metric Selection, Weighted Voting, Cross -Validation
The hardware prototype is shown in Fig 4. The and Model Evaluation, Ensemble Methods, and
hardware result displayed in webpage is shown in Fig Continuous Model Refinement.
5.
The research constraints that are faced during the
process of experimentation are careful designing
experiments, leveraging available resources, and
continuously refining their battery life estimation.

979-8-3503-4757-9/23/$31.00 ©2023 IEEE 1609

Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:31:49 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Edge Computing and Applications (ICECAA 2023)
IEEE Xplore Part Number: CFP23BV8-ART; ISBN: 979-8-3503-4757-9

VIII. CONCLUSION AND FUTURE WORK performance and enhancing user experience in a wide
range of applications.
The main obstacle to achieving sustainable
battery performance is Li-ion battery longevity. The
degradation path is dominated by application-specific
REFERENCES
usage, and an accurate ageing prognosis is currently
difficult to make. A wide-ranging effect of accurate [1] 1.B. Bole, C. Kulkarni, and M. Daigle (2014) ‘Adaptation of
battery life predictions is that it can be used to an electrochemistry-based li-ion battery model to account for
deterioration observed under randomized use’ page 9.
diagnose problems by understanding how the battery [2] G.-W. You, S. Park, and D. Oh (2016) ‘Real-time state-of-
behaves in specific situations. In this research work, health estimation for electric vehicle batteries: A dat a-driven
various strategies are created using various processes, approach,’Appl.Energy,vol.176,pp. 92–103.
[3] R. R. Richardson, M.A. Osborne and D.A. Howey, (2016)
and model performances are contrasted. Using a ‘Gaussian process regression for forecasting battery state of
similar training dataset, the evaluation is tried on health’ J. Power Sources ,vol. 357, pp. 209–219.
brand-new cells, then simulated. The method gives a [4] W. Dron, S. Duquennoy, T . Voigt, K. Hachicha, and P. Gard
(2014) ‘ An emulation-based method for lifetime estimation
crystal-clear indicator of when the battery in an of wireless sensor networks’ In Proceedings of the 2014
electric car needs to be changed, and the model's IEEE International Conference on Distributed Computing in
output is correct. Based on the data sets supplied into Sensor Systems, DCOSS ’14, pages 241–248, Washington,
DC, USA. IEEE Computer Society.
the system and the input battery, the model predicts [5] X.Zheng and X. Deng, (2018) ‘State-of-health prediction for
the outcome. As a result, the calculation of the car's lithium-ion batteries with multiple Gaussian process
regression model,’ IEEE Access, vol. 384, pp. 387 –395.
Li-ion battery charge has been replaced with an [6] Y. Wang, X. Zhang and Z. Chen (2016) ‘A novel state of
estimation of the battery life at the charging station health estimation method of Li-ion battery using group
for electric vehicles. By expanding the data set and method of data handling,’ J.Power Sources, vol. 327, pp.457–
464.
improving the algorithm, the system can be used in [7] J. Yu, ‘‘State of health prediction of lithium-ion batteries:
the future for all types of batteries and provide Multiscale logic regression and Gaussian process regression
accuracy of 99.99%. ensemble,’’ Rel. Eng. Syst. Saf.,vol. 174, pp. 82–95, Jun.
2018.
[8] B. Saha and K. Goebel. (2007). Battery data set. NASA
In conclusion, the use of the k-nearest neighbor Ames prognostics data repository. NASA Ames, Moffett
(KNN) algorithm for Li-ion battery life estimation Field, CA, USA. [Online]. Available:
has shown promising results in accurately predicting http://ti.arc.nasa.gov/project/prognostic-data-repository
[9] J. Liu, A. Saxena, K. Goebel, B. Saha, and W. Wang, ‘‘An
the remaining battery life. Through the utilization of adaptive recurrent neural network for remaining useful life
a diverse and representative dataset, preprocessing prediction of lithiumion batteries,’’ Dept. Mech. Aerosp.
techniques, and appropriate feature selection, the Eng., Carleton Univ., Ottawa, ON,Canada, T ech. Rep.
ADA562707, 2010.
KNN algorithm has demonstrated its ability to [10] 10.Y. Gao, X. Zhang, B. Guo, C. Zhu, J. Wiedemann, L.
capture the complex non-linear relationships between Wang, andJ. Cao, ‘‘Health-aware multiobjective optimal
battery parameters and remaining life. charging strategy with coupled electrochemical-thermal-
aging model for lithium-ion battery,’’ IEEET rans. Ind.
Informat., vol. 16, no. 5, pp. 3417–3429, May 2020.
By leveraging the local context and considering [11] Nick, M., Cherkaoui, R., Boudec, J. L. & Paolone, M.,
the similarities between instances, the KNN 2018a. An Exact Convex Formulation of the Optimal Power
algorithm provides accurate estimations of battery Flow in Radial Distribution Networks Including T ransverse
Components. IEEE T ransactions on Automatic Control,
life. It offers advantages such as adaptability to 63(3), pp. 682-697.
different battery chemistries, interpretability of the [12] Nick, M., Cherkaoui, R., Boudec, J. L. & Paolone, M.,
estimation process, and computational efficiency, 2018a. An Exact Convex Formulation of the Optimal Power
Flow in Radial Distribution Networks Including T ransverse
making it suitable for real-time applications and Components. IEEE T ransactions on Automatic Control,
resource-constrained environments. 63(3), pp. 682-697.
[13] Fortenbacher, P., Ulbig, A. & Andersson, G., 2018. Optimal
Future research directions may include further Placement and Sizing of Distributed Battery Storage in Low
Voltage Grids Using Receding Horizon Control Strategies.
refinement of the KNN algorithm by exploring IEEE T ransactions on Power Systems, 33(3), pp. 2383-2394.
different distance metrics, incorporating weighted [14] Lin, Z., Hu, Z., Zhang, H. & Song, Y., 2019. Optimal ESS
voting schemes, or applying ensemble methods to Allocation in Distribution Network Using Accelerated
Generalised Benders Decomposition. IET Generation,
improve the estimation accuracy. Additionally, T ransmission & Distribution, 13(13), pp. 2738 -2746
investigating the integration of real-time data [15] Xie, S., Hu, Z., Yang, L. & Wang, J., 2020. Expansion
acquisition and online learning techniques can Planning of Active Distribution System Considering Mult iple
Active Network Managements and the Optimal Load-
enhance the adaptability of the model to changing Shedding Direction. International Journal on Electrical Power
battery conditions. and Energy Systems, Volume 115, pp. 1- 20.

In conclusion, the Li-ion battery life estimation

using the KNN algorithm presents a reliable,
efficient, and interpretable approach for estimating
remaining battery life. The research contributes to the
advancement of battery management systems and
offers valuable insights into optimizing battery

979-8-3503-4757-9/23/$31.00 ©2023 IEEE 1610

Authorized licensed use limited to: ULAKBIM UASL - Okan Universitesi. Downloaded on January 09,2024 at 14:31:49 UTC from IEEE Xplore. Restrictions apply.