STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS
Accurate State of Health (SOH) estimation is critically important for the safe and long-term operation of lithium-ion batteries. However, the implementation of estimation algorithms in embedded Battery Management Systems (BMS) is significantly limited by available computational resources. This paper...
Gespeichert in:
| Datum: | 2026 |
|---|---|
| Hauptverfasser: | , , , , |
| Format: | Artikel |
| Sprache: | Englisch |
| Veröffentlicht: |
Institute of Renewable Energy National Academy of Sciences of Ukraine
2026
|
| Schlagworte: | |
| Online Zugang: | https://ve.org.ua/index.php/journal/article/view/599 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | Vidnovluvana energetika |
| Завантажити файл: |  |
Institution
Vidnovluvana energetika| _version_ | 1871104079684435968 |
|---|---|
| author | Sorochynskyi , Ya. Bosak , Al. Dubovyk , V. Kulakovskyi , L. Bosak , An. |
| author_facet | Sorochynskyi , Ya. Bosak , Al. Dubovyk , V. Kulakovskyi , L. Bosak , An. |
| author_institution_txt_mv | [
{
"author": "Ya. Sorochynskyi ",
"institution": "National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv, Ukraine."
},
{
"author": "Al. Bosak ",
"institution": "National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv, Ukraine."
},
{
"author": "V. Dubovyk ",
"institution": "National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv, Ukraine."
},
{
"author": "L. Kulakovskyi ",
"institution": "National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv, Ukraine."
},
{
"author": "An. Bosak ",
"institution": "General Energy Institute of the National Academy of Sciences of Ukraine."
}
] |
| author_sort | Sorochynskyi , Ya. |
| baseUrl_str | https://ve.org.ua/index.php/journal/oai |
| collection | OJS |
| datestamp_date | 2026-07-18T06:32:24Z |
| description | Accurate State of Health (SOH) estimation is critically important for the safe and long-term operation of lithium-ion batteries. However, the implementation of estimation algorithms in embedded Battery Management Systems (BMS) is significantly limited by available computational resources. This paper presents a scientifically grounded comparative analysis of two computationally efficient machine learning methods – Support Vector Regression (SVR) and Random Forest – to determine the optimal algorithm for practical implementation, with a dual focus on prediction accuracy and computational efficiency. Based on an open dataset, systematic hyperparameter tuning with GridSearchCV and cross-validation was conducted for both models. The performance of the final, optimized models was evaluated by accuracy metrics (RMSE, MAE) and indicators of suitability for embedded systems (forecasting time, model size). The results showed that even after thorough optimization, the SVR model demonstrated higher prediction accuracy (RMSE 2.67% vs. 3.08% for Random Forest). An even more significant advantage was found in the efficiency analysis: SVR was found to be almost 170 times faster (forecast time of 0.119 ms vs 20.570 ms) and 200 times more compact (model size 8.6 kB vs 1718.6 kB). It is concluded that, for the problem of estimating SOH based on cyclic data, the SVR model is the optimal candidate for practical implementation in embedded BMS. It offers the best balance of high accuracy and minimal hardware resource requirements, outperforming Random Forest across all key criteria for engineering practice. |
| doi_str_mv | 10.36296/1819-8058.2026.1(84).109-115 |
| first_indexed | 2026-03-31T01:00:06Z |
| format | Article |
| fulltext |
109
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
UDC 621.3:620.96 https://doi.org/10.36296/1819-8058.2026.1(84).109-115
STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS:
A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS
Received Oct. 31, 2025; accepted Mar. 23, 2026
Available online Mar. 31, 2026
Sorochynskyi Ya.1, Bosak Al.2, Dubovyk V.3,
Kulakovskyi L.4, Bosak An.5
Author for correspondence: Bosak Alla
e-mail: allabosak@lll.kpi.ua
Abstract. Accurate State of Health (SOH) estimation is critically
important for the safe and long-term operation of lithium-ion
batteries. However, the implementation of estimation algo-
rithms in embedded Battery Management Systems (BMS) is
significantly limited by available computational resources. This
paper presents a scientifically grounded comparative analysis
of two computationally efficient machine learning methods –
Support Vector Regression (SVR) and Random Forest – to de-
termine the optimal algorithm for practical implementation,
with a dual focus on prediction accuracy and computational efficiency. Based on an open dataset, systematic hy-
perparameter tuning with GridSearchCV and cross-validation was conducted for both models. The performance of
the final, optimized models was evaluated by accuracy metrics (RMSE, MAE) and indicators of suitability for embed-
ded systems (forecasting time, model size). The results showed that even after thorough optimization, the SVR model
demonstrated higher prediction accuracy (RMSE 2.67% vs. 3.08% for Random Forest). An even more significant ad-
vantage was found in the efficiency analysis: SVR was found to be almost 170 times faster (forecast time of 0.119
ms vs 20.570 ms) and 200 times more compact (model size 8.6 kB vs 1718.6 kB). It is concluded that, for the problem
of estimating SOH based on cyclic data, the SVR model is the optimal candidate for practical implementation in em-
bedded BMS. It offers the best balance of high accuracy and minimal hardware resource requirements, outperform-
ing Random Forest across all key criteria for engineering practice.
Key words: Keywords: State of Health, SOH, battery management system, BMS, Battery Energy Storage Systems
(BESS), renewable energy, distributed generation, microgrids, lithium-ion batteries, machine learning, embed-
ded systems, Support Vector Regression, Random Forest, computational efficiency.
ОЦІНКА СТАНУ ПРАЦЕЗДАТНОСТІ (SOH) ДЛЯ ВБУДОВАНИХ СИСТЕМ КЕРУВАННЯ БАТАРЕЯМИ:
ПОРІВНЯЛЬНИЙ АНАЛІЗ ОБЧИСЛЮВАЛЬНО ЕФЕКТИВНИХ МЕТОДІВ
Отримано 31 жов. 2025 р.; рекомендовано до публікації 23 бер. 2026 р.
Доступно онлайн 31 бер. 2026 р.
Сорочинський Я. З.1, Босак А. В.2, Дубовик В. Г.3,
Кулаковський Л. Я.4, Босак А. В.5
Автор для кореспонденції: Босак Алла,
e-mail: allabosak@lll.kpi.ua
Анотація. Точна оцінка стану працездатності (State of
Health, SOH) є критично важливою для безпечної та довгові-
чної експлуатації літій-іонних акумуляторів. Однак реаліза-
ція алгоритмів оцінки у вбудованих системах керування ба-
тареями (BMS) суттєво обмежується доступними
обчислювальними ресурсами. В роботі проведено науково
обґрунтований порівняльний аналіз двох обчислювально
ефективних методів машинного навчання – методу опор-
них векторів для регресії (SVR) та випадкового лісу – а для
визначення оптимального алгоритму для практичної реалізації з подвійним фокусом на точності
1 аспірант
https://orcid.org/0009-0008-8056-520X
2 канд. техн. наук, доц.
https://orcid.org/0000-0003-0545-9980
3 старший викладач
https://orcid.org/0000-0001-8884-8222
4 канд. техн. наук, доц.
https://orcid.org/0000-0003-1273-6894
5 д-р. філософії, мол. наук. співроб.
https://orcid.org/0000-0002-4667-9720
1,2,3,4 Національний технічний університет
України «Київський політехнічний інститут
імені Ігоря Сікорського», Київ, Україна.
5 Інститут загальної енергетики НАН України.
1 Postgraduate Student
https://orcid.org/0009-0008-8056-520X
2 Cand. of Tech. Sciences, Assoc. Prof.
https://orcid.org/0000-0003-0545-9980
3 Senior Lecturer
https://orcid.org/0000-0001-8884-8222
4 Cand. of Tech. Sciences, Assoc. Prof.
https://orcid.org/0000-0003-1273-6894
5 PhD, Junior Researcher
https://orcid.org/0000-0002-4667-9720
1, 2, 3, 4 National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv,
Ukraine.
5 General Energy Institute of the National
Academy of Sciences of Ukraine.
110
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
прогнозування та обчислювальній ефективності. На основі відкритого набору даних, для обох моделей
було проведено систематичний підбір гіперпараметрів за допомогою GridSearchCV з перехресною валі-
дацією. Ефективність фінальних, оптимізованих моделей оцінювалася за метриками точності (RMSE,
MAE) та показниками придатності для вбудованих систем (час прогнозування, розмір моделі). Резуль-
тати показали, що навіть після ретельної оптимізації модель SVR продемонструвала вищу точність
прогнозування (RMSE 2.67% проти 3.08% у Випадковий Ліс). Ще більш значуща перевага була виявлена в
аналізі ефективності: SVR виявився майже у 170 разів швидшим (час прогнозу 0.119 мс проти 20.570 мс)
та у 200 разів компактнішим (розмір моделі 8.6 кБ проти 1718.6 кБ). Зроблено висновок, що для задачі
оцінки SOH на основі циклічних даних модель SVR є оптимальним кандидатом для практичної реалізації
у вбудованих BMS. Вона забезпечує найкращий баланс високої точності та мінімальних вимог до апара-
тних ресурсів, перевершуючи Випадковий Ліс за всіма ключовими для інженерної практики критеріями.
Ключові слова: стан працездатності, SOH, система керування батареями, BMS, системи накопичення
енергії, відновлювана енергетика, розподілена генерація, мікромережі, літій-іонні акумулятори, ма-
шинне навчання, вбудовані системи, метод опорних векторів, випадковий ліс, обчислювальна ефекти-
вність.
List of abbreviations and symbols
BMS – Battery Management System
EV – Electric Vehicle
MAE – Mean Absolute Error
RMSE – RMS Error
SOH – State of Health
SVR – Support Vector Method for Regression
RES – Renewable Energy Sources
BESS – Battery Energy Storage Systems
RUL – Remaining Useful Life
Introduction. The rapid transition of the global economy to-
wards carbon-neutral models is predicated on the massive
deployment of renewable energy sources (RES). However,
the integration of stochastic generation sources, such as so-
lar (PV) and wind power plants, into the general power grid
is impossible without high-efficiency buffer systems. In this
context, lithium-ion batteries have become a key technolog-
ical element for ensuring frequency stabilization and peak
shaving. According to the IEA's annual report "Renewables
2025. Analysis and forecasts to 2030", the share of RES in the
global energy mix is critically dependent on the pace of Bat-
tery Energy Storage Systems (BESS) deployment [1].
The efficiency of such systems in the energy sector depends
directly on the accuracy of estimating the storage units' re-
maining useful life. For renewable energy facilities - ranging
from residential solar installations to industrial wind farms
- errors in SOH estimation can lead to incorrect generation
planning and financial losses. Given that the modern Smart
Grid paradigm implies decentralization, diagnostic algo-
rithms are required to operate autonomously at each grid
node, ensuring the reliability of computations of green en-
ergy supply and its analysis without imposing excessive
computational resources.
Lithium-ion batteries have become the dominant technol-
ogy in this field, finding wide application in electric vehicles
(EVs), consumer electronics, and stationary battery energy
storage systems (BESS) used for grid stabilization [2, 3].
However, despite their advantages, lithium-ion batteries
are complex electrochemical systems prone to degradation
processes that reduce their capacity and power over time
and across operating cycles [4-6].
To ensure the safe, reliable, and efficient operation of these
systems, the Battery Management System plays a key role
[7-9]. One of the most critical functions of the BMS is the
accurate estimation of the battery's internal states, among
which the State of Health (SOH) is a fundamental indicator.
SOH, typically defined as the ratio of the current maximum
capacity to the initial capacity, serves as an indicator of the
battery's "aging" level. Accurate SOH estimation is critical
for predicting the Remaining Useful Life (RUL), optimizing
charging and discharging strategies, and preventing hazard-
ous failures associated with excessive degradation [10, 11].
The engineering challenge is that SOH estimation must be
performed in real-time directly on board the device where
the BMS operates. In recent years, many methods have been
proposed for estimating SOH based on data, in particular, us-
ing complex deep learning architectures that demonstrate
high accuracy in laboratory conditions. However, their high
computational complexity and significant memory consump-
tion make their practical implementation on resource-lim-
ited BMS platforms non-trivial, and often impossible.
Thus, there is a gap in research associated with the need for
a systematic analysis of methods that would provide an op-
timal compromise between estimation accuracy and com-
putational efficiency. The purpose of this work is to conduct
a comparative analysis of two classic but powerful machine
learning methods – the method of support vectors for re-
gression (SVR) and random forest (Random Forest) – for the
SOH estimation problem. Unlike studies focused solely on
accuracy, the proposed analysis is conducted considering
the key engineering constraints of an embedded system.
The main contribution of this article is to provide a quanti-
tative assessment of not only accuracy but also computa-
tional costs (forecasting time and model size), thereby en-
abling practical recommendations for BMS developers to
select the most appropriate algorithm for implementation.
111
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
2. Materials and methods.
2.1. Description of the data set. The study is based on an
open dataset [12]. This dataset contains long-term cycling
data for eight pouch-type lithium-ion battery cells manu-
factured by Kokam with Nickel-Manganese-Cobalt (NMC)
oxide-based cathodes. The experiment was conducted un-
der controlled laboratory conditions at a temperature of 40
°C. Key characteristics of each battery cell are provided in
Table 1. This data includes initial and final capacity, demon-
strating the degree of degradation, as well as the total num-
ber of cycles performed.
Table 1. Summary of the battery cell characteristics in the dataset
Element
Initial
Capacity (Ah)
Final capacity (Ah) Final SOH (%) Number of cycles
Cell1 724.121 530.596 73.3 8200
Cell2 718.741 505.717 70.4 7700
Cell3 718.763 535.989 74.6 8100
Cell4 721.116 555.575 77 5100
Cell5 720.406 430.906 59.8 5000
Cell6 716.658 560.977 78.3 5000
Cell7 713.381 553.081 77.5 8100
Cell8 711.957 528.389 74.2 8100
For this study, cell capacity data obtained during periodic
characterization tests are used. Specifically, the capacity
parameter during a 1C charge (denoted as C1ch in the
original dataset) is analyzed, as it is a reliable indicator of
degradation. Fig. 1 visualizes the raw capacity-degradation
data for all eight cells.
Fig. 1. Charge capacity degradation trajectories for all 8 cells from the Oxford dataset
As seen in Fig. 1, although all cells exhibit a similar general
trend of capacity reduction, there is significant cell-to-cell
variation in the degradation rate. This variability under-
scores the need to develop models capable of adapting to
the individual behavior of each cell. The advantage of this
dataset is the availability of complete degradation trajecto-
ries for several cells, which enables training and testing
models on independent samples, thereby simulating real-
world conditions in which a trained model is applied to a
new, previously unknown battery cell.
2.2. SOH calculation and data preparation. The State of
Health for each element was calculated as the ratio of the
current maximum capacity to the initial capacity, expressed
as a percentage, according to formula [13]:
( ) ( )n initial
SOH n = Q / Q 100%, (1)
where: ( )SOH n - the state of health at the n-th measure-
ment cycle; n
Q - is the maximum charge capacity at the n-
th measurement cycle; initial
Q is the initial capacity, defined
112
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
as the maximum charge capacity at the first available meas-
urement cycle.
A data-based approach was chosen to train the models. The
cycle number was used as the single input feature for the
models, as it is a direct indicator of the battery's operating
time. The calculated SOH was the target variable. This min-
imalist approach to feature selection is intentional, as it re-
duces the requirements for sensory equipment and BMS
computational resources.
To ensure an objective assessment of the models' general-
ization ability, the data were split according to a "cross-cell
validation" principle. Cells Cell1 through Cell6 were used to
form the training set, while cells Cell7 and Cell8 were com-
pletely excluded from the training process and reserved for
the final model testing.
2.3. Description of machine learning models.
Support Vector Regression (SVR). SVR is an extension of
the Support Vector Method (SVM) for regression tasks [14].
The main idea of the method is to construct a hyperplane
that best approximates the data, while ignoring points that
fall within a certain "tube" around the hyperplane. This al-
lows the model to be less sensitive to noise in the data. To
model non-linear dependencies, such as SOH degradation,
SVR uses the "kernel trick". In this work, the Radial Basis
Function (RBF) kernel was used, which is versatile and ef-
fective for a wide range of tasks.
Random Forest is an ensemble machine learning method
that averages predictions of a large number of independent
decision trees for regression tasks [15]. Each tree in the
"forest" is trained on a random subsample from a training
dataset. Furthermore, when constructing each node of a
tree, only a random subset of features is considered. This
dual element of randomness makes the model robust
against overfitting and capable of capturing complex non-
linear relationships in the data without requiring complex
hyperparameter tuning.
2.4. Evaluation criteria. The performance of the proposed
models was assessed in two main areas: accuracy and com-
putational efficiency.
1. Estimation accuracy – to quantify the accuracy of SOH pre-
diction, standard metrics for regression tasks were used
[16]:
• Root Mean Squared Error (RMSE):
n
2
i i
i=1
1
RMSE= (y -y )
n
ˆ . (2)
• Mean Absolute Error (MAE):
n
i i
i=1
1
MAE = y - y
n
,ˆ (3)
where n is the number of points in the test sample, i
y is
the real SOH value, and is i
ŷ the predicted SOH value.
2. Computational efficiency – to assess the models' suitability
for implementation in embedded systems, the following
parameters were analyzed:
• Forecasting time measurements were taken on a stand-
ardized computing platform to ensure the comparability of
results.
• Model size – the amount of memory (in kilobytes) re-
quired to store the trained model. This parameter directly
correlates with the requirements for the BMS microcon-
troller's Flash memory.
3. Research results. This section presents the results of ap-
plying the trained SVR and Random Forest models to the
independent test dataset (cells Cell7 and Cell8). The mod-
els' performance is evaluated based on the two key criteria
defined in the previous section: prediction accuracy and
computational efficiency.
3.1. Accuracy of SOH estimation. Quantitative indicators of
forecasting accuracy for both models in the test sample are
summarized in Table 2. The RMSE and MAE metrics were
calculated by comparing the predicted SOH values with the
actual values obtained from experimental data.
Table 2. Accuracy comparison of SVR and Random Forest
models on the test sample
Model Metric Value (%)
SVR RMSE 2.65
SVR MAE 2.30
Random Forest RMSE 3.08
Random Forest MAE 2.68
The results turned out to be partially counter-intuitive. De-
spite Random Forest often being considered a more pow-
erful algorithm, in this task, the simpler SVR model demon-
strated higher prediction accuracy (RMSE 2.68% vs. 3.08%).
For a visual assessment of the prediction quality, Fig. 2 pre-
sents a graph of the actual SOH degradation for cell Cell7
(which was not used in training) and the predictions gener-
ated by both models.
Analysis of the forecast graph (Fig. 2) helps explain this phe-
nomenon. The Random Forest model, trying to reproduce
stochastic (random) fluctuations and anomalous capacity
drops, generated a prediction with high variance. This re-
sulted in significant local errors, which adversely affected
the overall accuracy metric.
Instead, the SVR with the RBF core built a smooth, general-
ized degradation model. By ignoring sharp outliers, which are
difficult to predict, SVR provided a more stable and robust
forecast that, on average, was closer to the actual values.
To ensure an objective comparison and achieve maximum
performance of each algorithm, a systematic hyperparam-
eter tuning procedure was conducted. For this purpose,
the GridSearchCV tool from the scikit-learn library was
used, which implements an exhaustive search over a given
113
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
grid of parameters using cross-validation. A 3-fold cross-
validation strategy (cv=3) was applied to the training set
(Cell1-Cell6) for a reliable assessment of each parameter
combination's quality. Negative mean squared error
(neg_mean_squared_error) was chosen as the optimiza-
tion metric.
Fig. 2. Graphical comparison of SOH predictions from SVR and Random Forest models with actual data for test Cell7.
(Blue line - actual data, orange - Random Forest prediction, green - SVR prediction)
Quantitative indicators of forecasting accuracy for opti-
mized models on the test sample are summarized in Ta-
ble 3. The RMSE and MAE metrics were calculated by com-
paring the predicted SOH values with the actual ones
obtained from experimental data that was not used during
the training.
Table 3. Accuracy comparison of optimized SVR and Ran-
dom Forest models on the test sample
Model Metric Value (%)
SVR (optimal) RMSE 2.67
SVR (optimal) MAE 2.33
Random Forest (optimal) RMSE 3.08
Random Forest (optimal) MAE 2.68
The results obtained, even after systematic hyperparame-
ter optimization, were partly counter-intuitive. Despite
Random Forest often being considered a more powerful al-
gorithm, in this task, the simpler SVR model demonstrated
higher prediction accuracy (RMSE 2.67% vs. 3.08%). This in-
dicates that SVR's advantage is not accidental but is due to
the fundamental properties of the algorithm and the na-
ture of the data.
For a visual assessment of the quality of forecasting, Fig. 3
presents a graph of the real degradation of SOH for the
Cell7 element and the predictions generated by both opti-
mized models.
This demonstrates a key trade-off between model flexibility
and robustness. For systems where prediction stability and
minimizing maximum error are important, a smoother
model like SVR may be a better choice, even if it is not able
to reproduce all the nuances of the process. For diagnostics
tasks, where, conversely, it is important to capture an
anomaly, the behavior of Random Forest might be more in-
formative (but this is beyond the scope of simple SOH pre-
diction).
3.2. Analysis of computational efficiency. To assess the
suitability of the models for implementation in embedded
BMS, an analysis of their computational requirements was
conducted. The results, including the average time for one
prediction and the size of the trained optimized model, are
presented in Table 4.
Table 4. Comparison of the computational efficiency of
the models
Model
Forecast Time
(ms)
Model Size (kB)
SVR 0.12 8.9
Random Forest 23.568 968.5
The quantitative assessment of the models' computational
efficiency, presented in Table 4, demonstrates the signifi-
cant advantages of the SVR model in both parameters. Its
size is 8.9 kB, which is more than 100 times smaller than
that of the Random Forest model (968.5 kB). A similar dif-
ference is observed in performance: the inference time for
SVR was 0.121 ms, while for Random Forest it was 23.5 ms,
making SVR nearly 200 times faster.
114
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
Fig. 3. Graphical comparison of SOH predictions from SVR and Random Forest models with actual data (optimized mod-
els) for test cell Cell7
4. Discussion of results. The results of the computational
efficiency analysis (Table 4) are important for the practical
implementation of the algorithm in embedded systems and
definitively confirm the advantages of SVR. From an engi-
neering perspective, the observed differences are critical.
The SVR model size of just 8.9 kB allows it to be easily inte-
grated into the memory of most low-cost microcontrollers
used in BMS, which have limited Flash memory (often 64-
256 kB). In contrast, the Random Forest model, with a size
of almost 1 MB, requires a significantly more powerful and
expensive hardware platform, which is unacceptable for
many commercial applications.
Even more significant is the difference in prediction time.
The SVR speed (0.121 ms) means that the SOH estimate
places virtually zero load on the microcontroller's CPU. This
allows the processor to perform other critical real-time
tasks, such as voltage and current monitoring, cell balanc-
ing, and executing safety algorithms. The slower perfor-
mance of Random Forest (23.568 ms) creates a significantly
higher load, which can complicate the guaranteed execu-
tion of other BMS functions within strict time frames.
Thus, combining the accuracy analysis (Table 3) and the ef-
ficiency analysis (Table 4) yields an unambiguous conclu-
sion. Despite SVR being a simpler model, it not only proved
to be more accurate but also demonstrated orders of mag-
nitude higher computational efficiency. This makes it not
just a compromise, but the optimal choice for practical im-
plementation in embedded battery management systems.
5. Conclusions. This paper conducted a comparative analy-
sis of two machine learning methods, SVR and Random For-
est, for the task of State of Health (SOH) estimation of
lithium-ion batteries, with a focus on their suitability for im-
plementation in embedded BMS.
Based on experimental data, it was established that the SVR
model outperforms the Random Forest model across all cri-
teria critical to engineering practice. It achieved higher pre-
diction accuracy (RMSE 2.67% vs. 3.08%) and demonstrated
computational efficiency orders of magnitude higher, being
more than 100 times more compact and nearly 200 times
faster.
The proposed approach is instrumental in the continued
upscaling of renewable energy deployment. Energy storage
systems coupled with solar or wind installations are fre-
quently deployed in remote locations characterized by lim-
ited telemetry capabilities. The high computational effi-
ciency of the SVR method facilitates the integration of high-
precision diagnostics directly into the controllers of BESS
units. This enhances the overall reliability of power systems
with high RES penetration by enabling more accurate fore-
casting of available battery capacity for grid balancing and
by extending the operational lifespan of cost-intensive stor-
age assets.
Future research may focus on including additional parame-
ters, such as temperature, in the SVR model's input feature
vector to further enhance its accuracy under real-world op-
erating conditions.
115
Відновлювана енергетика. № 1/2026 | Комплексні проблеми енергетичних систем на основі НВДЕ
REFERENCES:
1. IEA (2025), Renewables 2025. Analysis and forecasts to
2030, IEA, Paris.
2. Tian, J., Fan, Y., Pan, T., Zhang, X., Yin, J., & Zhang, Q.
(2024). A critical review on inconsistency mechanism,
evaluation methods and improvement measures for
lithium-ion battery energy storage systems. Renewable
and Sustainable Energy Reviews, 189, 113978.
https://doi.org/10.1016/j.rser.2023.113978
3. Lv, Y., Geng, X., Luo, W., Chu, T., Li, H., Liu, D., ... & Li,
C. (2023). Review on influence factors and prevention
control technologies of lithium-ion battery energy
storage safety. Journal of Energy Storage, 72, 108389.
https://doi.org/10.1016/j.est.2023.108389
4. Rahman, T., & Alharbi, T. (2024). Exploring lithium-Ion
battery degradation: A concise review of critical
factors, impacts, data-driven degradation estimation
techniques, and sustainable directions for energy
storage systems. Batteries, 10(7), 220.
https://doi.org/10.3390/batteries10070220
5. Edge, J. S., O’Kane, S., Prosser, R., Kirkaldy, N. D., Patel,
A. N., Hales, A., ... & Offer, G. J. (2021). Lithium ion
battery degradation: what you need to know. Physical
Chemistry Chemical Physics, 23(14), 8200-8221.
https://doi.org/10.1039/D1CP00359C
6. Shchurov, N. I., Dedov, S. I., Malozyomov, B. V., Shtang,
A. A., Martyushev, N. V., Klyuev, R. V., & Andriashin, S.
N. (2021). Degradation of lithium-ion batteries in an
electric transport complex. Energies, 14(23), 8072.
https://doi.org/10.3390/en14238072
7. Habib, A. A., Hasan, M. K., Issa, G. F., Singh, D., Islam,
S., & Ghazal, T. M. (2023). Lithium-ion battery
management system for electric vehicles: constraints,
challenges, and recommendations. Batteries, 9(3), 152.
https://doi.org/10.3390/batteries9030152
8. S. Mishra, S. C. Swain and R. K. Samantaray, "A Review
on Battery Management system and its Application in
Electric vehicle," 2021 International Conference on
Advances in Computing and Communications (ICACC),
Kochi, Kakkanad, India, 2021, pp. 1-6, doi:
10.1109/ICACC-202152719.2021.9708114.
9. S. Thangavel, D. Mohanraj, T. Girijaprasanna, S. Raju, C.
Dhanamjayulu and S. M. Muyeen, "A Comprehensive
Review on Electric Vehicle: Battery Management
System, Charging Station, Traction Motors," in IEEE
Access, vol. 11, pp. 20994-21019, 2023, doi:
10.1109/ACCESS.2023.3250221
10. Cheng, M., Zhang, X., Ran, A., Wei, G., & Sun, H.
(2023). Optimal dispatch approach for second-life
batteries considering degradation with online SoH
estimation. Renewable and Sustainable Energy
Reviews, 173, 113053.
https://doi.org/10.1016/j.rser.2022.113053
11. Pang, B., Chen, L., & Dong, Z. (2022). Data-driven
degradation modeling and SOH prediction of Li-ion
batteries. Energies, 15(15), 5580.
https://doi.org/10.3390/en15155580
12. Christoph R. Birkl, "Diagnosis and Prognosis of
Degradation in Lithium-Ion Batteries", PhD thesis,
Department of Engineering Science, University of
Oxford, 2017.
https://doi.org/10.5287/bodleian:KO2kdmYGg
13. Chen, X., Hu, Y., Li, S., Wang, Y., Li, D., Luo, C., ... &
Yang, Y. (2021). State of health (SoH) estimation and
degradation modes analysis of pouch
NMC532/graphite Li-ion battery. Journal of Power
Sources, 498, 229884.
https://doi.org/10.1016/j.jpowsour.2021.229884
14. Montesinos López, O. A., Montesinos López, A., &
Crossa, J. (2022). Support vector machines and support
vector regression. In Multivariate statistical machine
learning methods for genomic prediction (pp. 337-
378). Cham: Springer International Publishing.
https://doi.org/10.1007/978-3-030-89010-0_9
15. Salman, H. A., Kalakech, A., & Steiti, A. (2024). Random
forest algorithm overview. Babylonian Journal of
Machine Learning, 2024, 69-79.
https://doi.org/10.58496/BJML/2024/007
16. Hodson, T. O. (2022). Root mean square error (RMSE) or
mean absolute error (MAE): When to use them or not.
Geoscientific Model Development Discussions, 2022, 1-
10. https://doi.org/10.5194/gmd-15-5481-2022
|
| id | veorgua-article-599 |
| institution | Vidnovluvana energetika |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2026-07-19T01:18:24Z |
| publishDate | 2026 |
| publisher | Institute of Renewable Energy National Academy of Sciences of Ukraine |
| record_format | ojs |
| resource_txt_mv | veorgua/19/f87f9f68b862df3a29eec9b8dc3c6a19.pdf |
| spelling | veorgua-article-5992026-07-18T06:32:24Z STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS ОЦІНКА СТАНУ ПРАЦЕЗДАТНОСТІ (SOH) ДЛЯ ВБУДОВАНИХ СИСТЕМ КЕРУВАННЯ БАТАРЕЯМИ: ПОРІВНЯЛЬНИЙ АНАЛІЗ ОБЧИСЛЮВАЛЬНО ЕФЕКТИВНИХ МЕТОДІВ Sorochynskyi , Ya. Bosak , Al. Dubovyk , V. Kulakovskyi , L. Bosak , An. Keywords: State of Health, SOH, battery management system, BMS, Battery Energy Storage Systems (BESS), renewable energy, distributed generation, microgrids, lithium-ion batteries, machine learning, embedded systems, Support Vector Regression, Random Forest, computational efficiency. стан працездатності, SOH, система керування батареями, BMS, системи накопичення енергії, відновлювана енергетика, розподілена генерація, мікромережі, літій-іонні акумулятори, машинне навчання, вбудовані системи, метод опорних векторів, випадковий ліс, обчислювальна ефективність. Accurate State of Health (SOH) estimation is critically important for the safe and long-term operation of lithium-ion batteries. However, the implementation of estimation algorithms in embedded Battery Management Systems (BMS) is significantly limited by available computational resources. This paper presents a scientifically grounded comparative analysis of two computationally efficient machine learning methods – Support Vector Regression (SVR) and Random Forest – to determine the optimal algorithm for practical implementation, with a dual focus on prediction accuracy and computational efficiency. Based on an open dataset, systematic hyperparameter tuning with GridSearchCV and cross-validation was conducted for both models. The performance of the final, optimized models was evaluated by accuracy metrics (RMSE, MAE) and indicators of suitability for embedded systems (forecasting time, model size). The results showed that even after thorough optimization, the SVR model demonstrated higher prediction accuracy (RMSE 2.67% vs. 3.08% for Random Forest). An even more significant advantage was found in the efficiency analysis: SVR was found to be almost 170 times faster (forecast time of 0.119 ms vs 20.570 ms) and 200 times more compact (model size 8.6 kB vs 1718.6 kB). It is concluded that, for the problem of estimating SOH based on cyclic data, the SVR model is the optimal candidate for practical implementation in embedded BMS. It offers the best balance of high accuracy and minimal hardware resource requirements, outperforming Random Forest across all key criteria for engineering practice. Точна оцінка стану працездатності (State of Health, SOH) є критично важливою для безпечної та довговічної експлуатації літій-іонних акумуляторів. Однак реалізація алгоритмів оцінки у вбудованих системах керування батареями (BMS) суттєво обмежується доступними обчислювальними ресурсами. В роботі проведено науково обґрунтований порівняльний аналіз двох обчислювально ефективних методів машинного навчання – методу опорних векторів для регресії (SVR) та випадкового лісу – а  для визначення оптимального алгоритму для практичної реалізації з подвійним фокусом на точності прогнозування та обчислювальній ефективності. На основі відкритого набору даних, для обох моделей було проведено систематичний підбір гіперпараметрів за допомогою GridSearchCV з перехресною валідацією. Ефективність фінальних, оптимізованих моделей оцінювалася за метриками точності (RMSE, MAE) та показниками придатності для вбудованих систем (час прогнозування, розмір моделі). Результати показали, що навіть після ретельної оптимізації модель SVR продемонструвала вищу точність прогнозування (RMSE 2.67% проти 3.08% у Випадковий Ліс). Ще більш значуща перевага була виявлена в аналізі ефективності: SVR виявився майже у 170 разів швидшим (час прогнозу 0.119 мс проти 20.570 мс) та у 200 разів компактнішим (розмір моделі 8.6 кБ проти 1718.6 кБ). Зроблено висновок, що для задачі оцінки SOH на основі циклічних даних модель SVR є оптимальним кандидатом для практичної реалізації у вбудованих BMS. Вона забезпечує найкращий баланс високої точності та мінімальних вимог до апаратних ресурсів, перевершуючи Випадковий Ліс за всіма ключовими для інженерної практики критеріями. Institute of Renewable Energy National Academy of Sciences of Ukraine 2026-03-28 Article Article application/pdf https://ve.org.ua/index.php/journal/article/view/599 10.36296/1819-8058.2026.1(84).109-115 Vidnovluvana energetika ; No. 1(84) (2026): Scientific and applied Journal renewable energy ; 109-115 Возобновляемая энергетика; ##issue.no## 1(84) (2026): Scientific and applied Journal renewable energy ; 109-115 Відновлювана енергетика; № 1(84) (2026): Науково-прикладний журнал Відновлювана енергетика; 109-115 2664-8172 1819-8058 10.36296/1819-8058.2026.1(84) en https://ve.org.ua/index.php/journal/article/view/599/510 Copyright (c) 2026 Ya. Sorochynskyi , Al. Bosak , V. Dubovyk , L. Kulakovskyi , An. Bosak https://creativecommons.org/licenses/by-nc-nd/4.0 |
| spellingShingle | Keywords: State of Health SOH battery management system BMS Battery Energy Storage Systems (BESS) renewable energy distributed generation microgrids lithium-ion batteries machine learning embedded systems Support Vector Regression Random Forest computational efficiency. Sorochynskyi , Ya. Bosak , Al. Dubovyk , V. Kulakovskyi , L. Bosak , An. STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS |
| title | STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS |
| title_alt | ОЦІНКА СТАНУ ПРАЦЕЗДАТНОСТІ (SOH) ДЛЯ ВБУДОВАНИХ СИСТЕМ КЕРУВАННЯ БАТАРЕЯМИ: ПОРІВНЯЛЬНИЙ АНАЛІЗ ОБЧИСЛЮВАЛЬНО ЕФЕКТИВНИХ МЕТОДІВ |
| title_full | STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS |
| title_fullStr | STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS |
| title_full_unstemmed | STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS |
| title_short | STATE OF HEALTH (SOH) ESTIMATION FOR EMBEDDED BATTERY MANAGEMENT SYSTEMS: A COMPARATIVE ANALYSIS OF COMPUTATIONALLY EFFICIENT METHODS |
| title_sort | state of health (soh) estimation for embedded battery management systems: a comparative analysis of computationally efficient methods |
| topic | Keywords: State of Health SOH battery management system BMS Battery Energy Storage Systems (BESS) renewable energy distributed generation microgrids lithium-ion batteries machine learning embedded systems Support Vector Regression Random Forest computational efficiency. |
| topic_facet | Keywords: State of Health SOH battery management system BMS Battery Energy Storage Systems (BESS) renewable energy distributed generation microgrids lithium-ion batteries machine learning embedded systems Support Vector Regression Random Forest computational efficiency. стан працездатності SOH система керування батареями BMS системи накопичення енергії відновлювана енергетика розподілена генерація мікромережі літій-іонні акумулятори машинне навчання вбудовані системи метод опорних векторів випадковий ліс обчислювальна ефективність. |
| url | https://ve.org.ua/index.php/journal/article/view/599 |
| work_keys_str_mv | AT sorochynskyiya stateofhealthsohestimationforembeddedbatterymanagementsystemsacomparativeanalysisofcomputationallyefficientmethods AT bosakal stateofhealthsohestimationforembeddedbatterymanagementsystemsacomparativeanalysisofcomputationallyefficientmethods AT dubovykv stateofhealthsohestimationforembeddedbatterymanagementsystemsacomparativeanalysisofcomputationallyefficientmethods AT kulakovskyil stateofhealthsohestimationforembeddedbatterymanagementsystemsacomparativeanalysisofcomputationallyefficientmethods AT bosakan stateofhealthsohestimationforembeddedbatterymanagementsystemsacomparativeanalysisofcomputationallyefficientmethods AT sorochynskyiya ocínkastanupracezdatnostísohdlâvbudovanihsistemkeruvannâbatareâmiporívnâlʹnijanalízobčislûvalʹnoefektivnihmetodív AT bosakal ocínkastanupracezdatnostísohdlâvbudovanihsistemkeruvannâbatareâmiporívnâlʹnijanalízobčislûvalʹnoefektivnihmetodív AT dubovykv ocínkastanupracezdatnostísohdlâvbudovanihsistemkeruvannâbatareâmiporívnâlʹnijanalízobčislûvalʹnoefektivnihmetodív AT kulakovskyil ocínkastanupracezdatnostísohdlâvbudovanihsistemkeruvannâbatareâmiporívnâlʹnijanalízobčislûvalʹnoefektivnihmetodív AT bosakan ocínkastanupracezdatnostísohdlâvbudovanihsistemkeruvannâbatareâmiporívnâlʹnijanalízobčislûvalʹnoefektivnihmetodív |