Battery and energy management system for BLDC motor driven electric vehicles with intelligent control and machine learning techniques

ABSTRACT

This research paper presents an improved Battery and Energy Management System (BEMS) designed for Brushless DC (BLDC) motor-driven electric vehicles using smart control approaches and machine learning. By including predictive ML models such as Decision Trees, Support Vector Machines (SVM), and XGBoost for precise assessment of battery state-of-charge (SOC) and real-time energy allocation, the system seeks to optimise motor control and battery performance. Dynamically controlling power flow depending on SOC, temperature, and driving circumstances, a smart battery and energy management system is created. Comparison with traditional EMS methods reveals notable gains in energy economy, temperature management, battery life, and motor response. The combination of ML and smart control shows a strong and flexible system for improving the general performance and sustainability of electric vehicle powertrains.

Keywords:
Battery Management; Electric Vehicles; Energy Efficiency; Energy Management System; Machine Learning

1. INTRODUCTION

Effective energy management is becoming more and more important for maximising energy use and prolonging the life of battery systems as the use of renewable energy sources (such as nanogrids) and electric vehicles (EVs) rises. Rule-based methods that control power distribution depending on set thresholds dominate traditional EMS. These techniques, on the other hand, can lack the adaptability needed to fit real-time circumstances including changing loads, SOC, temperature, and acceleration. Furthermore, dynamic BMS changes and smart motor speed control stay underexplored. Focussing on maximising energy use and system efficiency, Energy Management Systems (EMS) have been used in several sectors including microgrids and electric vehicles (EVs). Traditional EMS mostly depend on rule-based systems, which can be somewhat rigid under dynamic circumstances like SOC and load changes [¹,²,³]. Recent research has drawn attention to the application of machine learning models such decision trees, support vector machines (SVMs), and neural networks to enhance EMS decision-making, hence allowing systems to adapt to real-time settings [⁴, ⁵].

By controlling vital parameters such SOC, temperature, and voltage, Battery Management Systems (BMS) guarantee the best performance and lifetime of batteries. Several techniques suggested to enhance BMS performance include rule-based and predictive models [⁶,⁷,⁸]. With SVM and reinforcement learning especially effective for dynamic control, machine learning has become a potent tool to forecast battery health and optimise charge-discharge cycles [⁹, ¹⁰]. Improved battery safety and lifetime have come from this capacity to change real-time parameters depending on battery condition. Electric vehicles’ energy efficiency is best optimised by motor speed control. Often, conventional control systems like PID controllers are constrained by their inability to dynamically shift with SOC or load variations. The paper [¹¹] showed the advantages of adaptive motor control by combining battery SOC and acceleration data to guarantee energy economy and enhanced vehicle performance. To fine-tune motor control algorithms, machine learning models—including regression trees and reinforcement learning—have been investigated [¹²].

Proposed for maximising energy management in several applications, including microgrids and EVs, intelligent control systems combining machine learning with real-time data. To improve system efficiency, the authors in [¹³] presented hybrid models combining machine learning with power flow management and energy prediction. Intelligent BMS systems also dynamically change load depending on SOC and temperature, hence avoiding battery over-discharging and overheating [¹⁴]. Machine learning’s use in EMS for hybrid and electric cars has also attracted much interest. The authors in [¹⁵] found that models like XGBoost and Random Forest outperformed conventional rule-based methods when they examined machine learning techniques for battery SOC prediction. Not only do these models more precisely forecast battery status, but they also change to fit driving situations, hence enhancing energy economy and driving range [¹⁶].

The optimisation of energy storage in renewable energy systems has also seen notable use via machine learning. The authors in [¹⁷] showed how machine learning techniques maximise energy storage in solar-powered microgrids, hence enhancing grid stability. By use of deep learning, the authors in [¹⁸] predicted energy demand, hence maximising storage plans and improving use of renewable energy. Additionally, a major field of study involves improving battery life using machine learning. Predicting lithium-ion battery degradation using SVM and deep learning, [¹⁹] guaranteed batteries were neither overcharged or undercharged. Using reinforcement learning to optimise charging and discharging cycles, [²⁰] improved battery performance and lifetime.

With reinforcement learning models used for real-time power distribution, the incorporation of machine learning in EMS for hybrid and electric cars keeps expanding. The authors in [²¹,²²,²³] suggested a reinforcement learning method for energy management in hybrid cars, maximising energy use across several driving scenarios. This strategy has shown to be more effective than conventional EMS techniques. Applying machine learning models has also drawn recent research on energy storage system optimisation. Deep learning methods were used by [²⁴] to optimise energy storage, hence greatly enhancing energy efficiency. In the same vein, [²⁵] developed charging-discharge control algorithms to maximise energy storage and lower battery deterioration.

Demand response (DR) plans inside EMS have also included machine learning to help strike a balance between energy supply and demand. the authors in [²⁶] suggested a machine learning-driven DR approach that modifies load depending on real-time data, hence increasing efficiency and cost reductions. Applying reinforcement learning for EMS optimisation in hybrid systems, [²⁷] improved power generation and consumption efficiency. There have also been studies on data-driven methods for battery health management. Using machine learning tools as SVM and random forests, [²⁸] evaluated several data-driven methods for battery ageing prediction, hence improving decision-making and preventive maintenance and finally extending battery life. Continued integration of artificial intelligence (AI) and machine learning (ML) is anticipated to characterise EMS and BMS going forward. These technologies will allow EMS to offer predictive analytics, real-time optimisation, and adaptive decision-making, hence surpassing static control techniques. The authors in [²⁹, ³⁰] underlined how artificial intelligence helps to create more scalable and effective energy management systems.

To address the identified limitation, this article key goals are:

• To assess and contrast the performance of conventional EMS techniques with machine learning models (Decision Tree, SVM, and XGBoost).

• To suggest a smart battery management system (BMS) dynamically modifying load depending on SOC and temperature conditions.

• To improve the general performance and efficiency of the system by using motor speed control algorithms changing the motor speed according on SOC, load, and acceleration.

• To evaluate the proposed system’s efficacy with synthetic data and performance measures including accuracy, precision, recall, and F1 score.

The importance of this study is in its capacity to enhance the performance of EMS and BMS in actual applications, including electric cars, microgrids, and renewable energy systems. This method provides a more flexible, economical, and dependable solution for controlling energy usage and battery health by combining machine learning with intelligent control systems. A key development for sustainable energy solutions, the suggested method could result in lower energy waste, longer battery life, and optimised system efficiency. This paper improves EMS and BMS performance by including machine learning methods and smart control strategies, hence addressing these constraints and making systems more flexible and efficient.

2. PROPOSED METHODOLOGY

To mimic a real-world situation including battery state-of-charge (SOC), load, motor speed, acceleration, voltage, current, temperature, battery capacity, and motor torque, a synthetic dataset of 1000 data points was created as shown in Figure 1. Categorised into three classes—“Use Battery”, “Regenerative Braking”, and “Optimise Mix”—the target variable for this dataset defines the power distribution method. Providing a useful energy capacity of 2.4–4.8 kWh, a 48 V lithium-ion battery pack rated at 50–100 Ah was examined in the simulation. Suitable for nanogrid-scale or e-mobility energy systems, the motor modelled is a 48–60 V, 3–5 kW Brushless DC (BLDC) motor running at a max speed of 3000 RPM and generating up to 200 Nm torque. Considering the dynamic behaviour of the system under different operational settings, the data was meant to replicate a realistic operational scenario in electric vehicles (EVs) and microgrids.

Figure 1
Feature distribution plots-distribution of input features for EMS model training.

Figure 2 illustrates the overall structure of the proposed Battery and Energy Management System (BEMS). The framework integrates machine learning-based EMS, motor speed control, and adaptive BMS strategies to achieve optimal power distribution, safe operation, and extended battery lifetime. This block diagram provides a conceptual overview of how input parameters such as SOC, load, temperature, and driving scenarios are processed into actionable control outputs.

Figure 2
Proposed BEMS framework integrating machine learning-based EMS.

2.1. Data pre-processing

Several preprocessing procedures were followed on the dataset to guarantee its appropriateness for training machine learning models:

• Label Encoding: The categorical target variable “power_distribution” was converted from string labels to numerical values (0, 1, 2).

• Feature Selection: Including battery SOC, load, motor speed, voltage, current, temperature, and efficiency, 12 features in all were chosen for training.

• Normalization: Using the StandardScaler from the sklearn.preprocessing module, all feature values were normalised to have zero mean and unit variance, hence guaranteeing each feature equally contributes to the model training by scaling all the features.

• Train/Test Split: The dataset was divided into training and testing sets in an 80/20 ratio, with 80% used for model training and 20% for assessment.

To replicate realistic EV driving conditions, the dataset was generated across three categories: urban stop-and-go, highway cruising, and mixed cycles. It includes variations in SOC (0.1–1.0), load (500–3000 W), voltage (24–60 V), current (10–80 A), motor speed (0–3000 rpm), torque (0–200 Nm), efficiency (0.7–0.95), acceleration (0–3 m/s²), and temperature (20–55 °C). Feature importance was assessed using correlation analysis and recursive feature elimination (RFE), confirming 12 parameters as the most influential for SOC estimation and EMS decision-making. The dataset was pre-processed with label encoding and normalization, followed by stratified 5-fold cross-validation to reduce overfitting risk and improve generalization.

2.2. Machine learning models

To guarantee the robustness of the models in real-world applications, each model was trained on the training dataset and tested using several performance measures including accuracy, precision, recall, and F1-score. Performance comparison of three machine learning models was done.

• Decision Tree Classifier (DT): A non-linear model used for classification tasks that builds a tree-like structure of decisions based on input features. In this work, the Decision Tree was implemented with a maximum depth of 10.

• Support Vector Machine (SVM): A supervised learning technique called support vector machine (SVM) classifies by locating a hyperplane that most clearly divides the input into many classes. Here, an RBF kernel (C = 1.0, γ = 0.1) was applied to classify multi-dimensional features.

• XGBoost Classifier: A gradient boosting model called the XGBoost Classifier constructs an ensemble of decision trees by iteratively reducing errors. In this study, XGBoost was trained with learning rate = 0.1, maximum depth = 6, and 200 estimators, making it particularly effective for structured EMS datasets. This methodology is especially appropriate for managing structured datasets such as the one utilised in this work.

2.3. Motor speed control

The following logic based on battery SOC, load, and acceleration was used to build a motor speed control algorithm:

• Should SOC be less than 0.2, cut the motor speed to 50% of the present load.

• Increase motor speed by 1.5 times the load if SOC > 0.8 and acceleration > 1.5.

• Set the motor speed to the load value for all other situations.

The motor speed control system was meant to maximise motor performance, avoid battery over-discharge, and guarantee efficiency under different operating situations.

2.4. Intelligent battery management system (BMS)

The following adaptive load control techniques based on SOC and temperature helped to create an intelligent Battery Management System (BMS):

• The system restricts the load to a maximum of 1000W if SOC < 0.2 to avoid over-discharging of the battery.

• The technology lowers the load to prevent battery overheating if the temperature exceeds 45°C, therefore ensuring the battery’s lifetime.

By dynamically changing the system’s energy use depending on real-time circumstances, this smart BMS is meant to improve battery lifetime and performance. The technical specifications of the battery and motor components used in simulations are detailed in Table 1.

Table 1
Technical specifications.

PARAMETER	VALUE
Battery Nominal Voltage	48V
Battery Capacity	50 Ah
Motor Rated Power	3 kW
Motor Rated Voltage	48V
Motor Rated Speed	3000 RPM
Motor Maximum Torque	15 Nm
SOC Threshold	0.2 (min), 0.8 (max)
Max Load (BMS)	1000W
Temperature Threshold	45°C

This approach guarantees that the system is not only energy-efficient but also adaptable to the different requirements of the battery, motor, and general system performance. The data features used for model training, including their types and examples, are summarized in Table 2.

Table 2
Data summary (including features and target variables).

FEATURE	DESCRIPTION	DATA TYPE	EXAMPLE VALUE
Battery SOC	State-of-charge of the battery	Continuous	0.85
Load	Load on the system (in watts)	Continuous	1500W
Motor Speed	Speed of the motor (RPM)	Continuous	3000 RPM
Acceleration	Acceleration value (m/s²)	Continuous	2.5 m/s²
Voltage	Voltage level (V)	Continuous	24V
Current	Electric current (A)	Continuous	50A
Temperature	Temperature of the battery (°C)	Continuous	40°C
Capacity	Capacity of the battery (Ah)	Continuous	60 Ah
Torque	Torque produced by the motor (Nm)	Continuous	30 Nm
Efficiency	Efficiency of the system (%)	Continuous	85%
Power Distribution	Power distribution strategy (categorical)	Categorical	“Use Battery”
Power Status	Power system status (Categorical)	Categorical	“Regenerative Braking”

3. TECHNICAL DETAILS

3.1. Code and algorithms

Using machine learning techniques and smart control strategies, the suggested system combines numerous essential elements to simulate and optimise the Energy Management System (EMS) and Battery Management System (BMS). Training of Machine Learning Models: Machine learning (ML) algorithms run using Python libraries like scikit-learn and XGBoost form the foundation of the predictive decision-making process. Trained on several input characteristics including State-of-Charge (SOC), load, motor speed, temperature, and acceleration, the models forecast the power distribution method. This work assessed three machine learning models:

• A tree-based model called the Decision Tree Classifier classifies the energy management approach.

• Support Vector Machine (SVM), Especially in high-dimensional settings, a strong model for classification tasks.

• An optimised gradient boosting algorithm called XGBoost Classifier is well-renowned for its performance in huge datasets and complicated jobs.

While the remaining 20% of the data was utilised for model evaluation, the training dataset (80% of the total dataset) was used in the training of each model. Key performance indicators including accuracy, precision, recall, and F1-score were used to evaluate the models.

3.2. Correlation heatmap insights (pearson coefficient)

A correlation study between different features like SOC, temperature, and motor parameters was performed, as visualized in the correlation heatmap as shown in Figure 3. A few highlighted correlations are,

• Temperature vs Torque: 0.086 → Mild positive correlation.

• Motor Current vs Efficiency: -0.083 → Mild negative correlation.

• Speed vs Torque: 0.077 → Speed has slight positive relation with torque.

• Voltage vs Battery SOC: 0.035 → As SOC increases, voltage increases slightly.

• Battery SOC vs Motor Current: 0.061 → Higher SOC may increase motor current availability.

Figure 3
Correlation heatmap of battery and motor parameters.

Designed to maximise motor speed depending on real-time input parameters—including battery SOC, load, and acceleration—the motor speed control function Implemented in Python, the function runs under the following logic:

• Should SOC be less than 0.2, the motor speed is cut to 50% of the load to save battery life.

• Should SOC be over 0.8 and acceleration exceed 1.5 m/s², the motor speed rises 1.5 times to enhance performance.

• The load determines the motor speed immediately otherwise.

By dynamically changing depending on the available energy and system needs, this adaptive motor speed control guarantees effective battery utilisation, hence improving general energy efficiency. The motor speed control algorithm, designed based on SOC, load, and acceleration thresholds, demonstrates the adaptive motor speed behavior under different conditions are listed in Table 3.

Table 3
Motor speed control based on SOC, load, and acceleration.

INDEX	SOC	LOAD (W)	ACCELERATION (m/s²)	MOTOR SPEED (RPM)
0	0.437	555.40	2.02	555.40
1	0.956	1625.70	2.39	2438.55
2	0.759	2618.84	0.75	2618.84
3	0.639	2196.67	1.87	2196.67
4	0.240	2419.68	1.72	2419.68
5	0.240	1976.35	2.50	1976.35
6	0.152	2076.83	2.72	1038.41
7	0.880	2547.59	0.04	2547.59
8	0.641	749.00	2.02	749.00
9	0.737	1468.27	0.16	1468.27

Adjustment of the Battery Management System (BMS): To guarantee battery health and prevent over-discharging or overheating, the smart BMS dynamically changes the load depending on SOC and temperature. This adjustment method works as follows:

• The BMS limits the load to a maximum of 1000W to avoid over-discharging the battery when SOC < 0.2.

• The BMS lowers the load to avoid overheating and possible battery damage when temperature > 45°C.

Maintaining the lifetime and efficiency of the battery depends on this, particularly under high-power demand conditions. The intelligent BMS dynamically adjusts load limits based on SOC and temperature conditions to enhance battery life and prevent thermal issues are presented in Table 4.

Table 4
Intelligent BMS load adjustment based on SOC and temperature.

INDEX	SOC	TEMPERATURE (°C)	LOAD (W)	ADJUSTED LOAD (W)
0	0.437	39.45	555.40	555.40
1	0.956	25.17	1625.70	1625.70
2	0.759	46.17	2618.84	2000.00
3	0.639	38.39	2196.67	2196.67
4	0.240	24.72	2419.68	2419.68
5	0.240	48.87	1976.35	1976.35
6	0.152	35.55	2076.83	1000.00
7	0.880	22.19	2547.59	2547.59
8	0.641	38.80	749.00	749.00
9	0.737	27.60	1468.27	1468.27

Storage and Model Evaluation: The following techniques were used to evaluate the performance of the machine learning models:

• Accuracy Score: Indicates the percentage of properly classified cases.

• Classification Report: For every model, offers thorough statistics including accuracy, recall, and F1-score.

• The trained models were saved using joblib for future usage, hence enabling simple integration into real-time systems for deployment.

To mimic a real-world situation including battery operations in electric vehicles and microgrid systems, a synthetic dataset of 1000 data points was created. Under several operational settings, the dataset records the interaction between several system parameters including battery SOC, load, motor speed, and temperature. To provide a realistic environment for training and assessing the models, the dataset simulates a range of operational scenarios including load, battery SOC, and temperature changes. Model performance was assessed by splitting the dataset into 80% training data and 20% test data.

This synthetic dataset lets one examine how machine learning models could maximise battery health and power management in practical applications such microgrids and electric vehicles. Real-time operational circumstances are represented by the data, which makes it appropriate for EMS and BMS simulation and optimisation. The dataset simulates realistic operational conditions in electric vehicles (EVs) and microgrids, capturing complex relationships among variables as presented in Figure 4.

Figure 4
Pairwise feature relationships by power distribution decision.

4. RESULTS AND DISCUSSION

4.1. Model evaluation

Standard classification criteria—accuracy, precision, recall, and F1-score—were used to assess the performance of the applied machine learning models—Decision Tree, Support Vector Machine (SVM), and XGBoost. Of the three, the XGBoost classifier showed best performance by scoring the highest on every measure. This suggests a more efficient categorisation of the power distribution tactics than conventional rule-based EMS logic. The findings verify that ensemble-based models like as XGBoost can generalise better and manage complicated non-linear interactions in EMS datasets. In addition to classification metrics, Root Mean Square Error (RMSE) and the Coefficient of Determination (R²) were also computed to evaluate SOC estimation accuracy. The results further confirm that XGBoost achieves the lowest SOC prediction error and highest correlation with actual values. The Decision Tree, SVM, and XGBoost classifiers were evaluated using accuracy, precision, recall, F1-score, RMSE, and R². XGBoost outperformed others across all metrics, demonstrating robust prediction capabilities as listed in Table 5.

Table 5
Model performance comparison with accuracy, precision, recall, F1-score, RMSE & R² values for each model.

MODEL	ACCURACY (%)	PRECISION (%)	RECALL (%)	F1-SCORE (%)	RMSE	R2	SOC ESTIMATION ACCURACY	SYSTEM RESPONSE EFFICIENCY
Decision tree	85	82	80	81	0.092	0.87	Moderate (tends to overfit on certain loads)	Acceptable but slower adaptation
Support vector machine	88	86	84	85	0.075	0.90	High (stable across varying SOC levels)	Smooth but computationally heavier
XGBoost	92	91	90	90.5	0.061	0.93	Very High (lowest error, most consistent)	Fast, adaptive response with robustness

4.2. State-of-charge (SOC) analysis over time

Each model’s SOC change over time was graphed to evaluate battery use efficiency. While traditional EMS shows faster depletion owing to strict threshold-based decision-making, the SOC vs. Time curves revealed that machine learning-based EMS (particularly XGBoost) maintains a more stable and optimal SOC profile during operation. This implies that ML-based EMS could increase energy efficiency and extend battery life. SOC versus Time behavior revealed that ML-based EMS maintains more stable SOC levels compared to traditional rule-based EMS as presented in Figure 5. Specifically, XGBoost-based EMS shows better utilization efficiency as shown in Figure 6.

Figure 5
SOC Vs Time to visually compare SOC behaviour.

Figure 6
Battery SOC vs Time for conventional and ML-based EMS.

To further evaluate robustness, three representative driving scenarios were simulated: (i) Urban driving, characterized by frequent stop-and-go events and lower average speeds; (ii) Highway driving, involving steady high-speed operation with minimal acceleration transients; and (iii) Mixed driving, which combines both urban and highway profiles. These scenarios were selected to represent realistic operating conditions that significantly affect energy demand and battery dynamics. The ML-based EMS, particularly the XGBoost classifier, demonstrated adaptability across all three patterns by maintaining stable SOC trajectories and optimizing energy utilization. This indicates that the proposed framework can generalize effectively to diverse real-world driving environments.

4.3. Motor speed regulation

The test set was used to assess the motor speed control algorithm built under SOC, load, and acceleration governing it. It effectively changed the motor speed in real time. For example, under low SOC conditions (SOC < 0.2), the motor speed was throttled down to lower stress on the battery; under high SOC and aggressive acceleration situations, the algorithm raised the motor speed accordingly. Motor speed control adapted dynamically to varying load, SOC, and acceleration scenarios as shown in Figure 7. The logic for real-time motor speed regulation is illustrated in a flowchart as shown in Figure 8. Further analysis of motor speed variations with different loads is depicted in Figure 9. The motor power output was also analyzed in relation to speed and torque using a 3D surface plot given in Figure 10.

Figure 7
Motor speed vs Load to demonstrate motor speed response across varying loads and SOC levels.

Figure 8
Motor speed control logic flowchart to present the algorithmic logic clearly.

Figure 9
Motor speed vs load.

Figure 10
3D surface plot of motor power output vs. speed and torque.

4.4. Intelligent BMS load adjustment

Under changing SOC and temperature settings, the smart BMS system was evaluated. It dynamically changed the power requirement to stop overheating and over-discharging. Battery safety was guaranteed by load capping at 1000 W for SOC values under 20%. Likewise, should the temperature go beyond 45°C, the BMS lowers the power draw to prevent thermal damage. This real-time reaction to environmental and battery state factors improved the dependability and lifetime of the system. The smart BMS dynamically adjusted the load based on SOC and temperature inputs. Its decision-making pattern across various scenarios is visualised in Figure 11. A complete summary of BMS load adjustment rules under different SOC and temperature conditions is provided in Table 6.

Figure 11
BMS load adjustment to visualize the decision-making pattern under varying SOC and temperature ranges.

Table 6
BMS load adjustment (based on SOC and temperature)

SOC RANGE	TEMPERATURE (°C)	ADJUSTED LOAD (W)
< 0.2	< 45	1000W
< 0.2	> 45	800W
0.2 – 0.8	< 45	1500W
0.2 – 0.8	> 45	1200W
> 0.8	< 45	2000W
> 0.8	> 45	1800W

An important consideration in deploying machine learning-based EMS frameworks is the associated computational load on embedded platforms. While simulation allows the use of more complex models without constraints, real-time implementation requires balancing accuracy with processing efficiency. In this work, lightweight models such as Decision Trees and optimized XGBoost implementations are noted for their suitability in embedded environments due to their relatively fast inference times. Although the SVM with RBF kernel offers strong accuracy, it is computationally heavier, and therefore may require kernel simplification, pruning, or hardware acceleration for practical use. This highlights that model selection must consider not only predictive accuracy but also real-time feasibility on resource-constrained controllers. Future extensions of this work will involve profiling the models on embedded processors and applying optimization techniques to ensure robust and efficient real-time operation.

5. CONCLUSION

This research offers a combined method to improve the performance of Battery Management Systems (BMS) and Energy Management Systems (EMS) for electric vehicles utilising machine learning and smart control techniques. Implementing and contrasting Decision Tree, Support Vector Machine (SVM), and XGBoost classifiers showed that XGBoost outperformed conventional rule-based EMS in preserving optimal battery SOC and energy distribution by providing better classification accuracy and predictive control capabilities. Including a PI-controlled BLDC motor model let the modelling of motor response under different load and SOC circumstances. Furthermore, the combination of an intelligent BMS with dynamic load changes depending on SOC and temperature greatly helped to enhance system stability and battery safety. Simulation findings confirm that the suggested ML-based EMS guarantees more energy efficiency, longer battery life, and improved performance in EV applications when combined with real-time motor speed control and adaptive BMS logic. The promising simulation results obtained in this study provide a strong foundation for future experimental validation on an EV test platform with a BLDC motor and integrated BMS, where real driving cycle data will be employed to further confirm the practical applicability of the proposed framework. In addition, future work will include testing with sophisticated metaheuristic optimisation algorithms such as PSO, GA, and DE for further performance enhancement.

6. BIBLIOGRAPHY

• ^[1] Zhang, C., Li, K., Deng, J., et al, “Review on state of charge estimation for lithium-ion batteries”, Journal of Power Sources, v. 378, pp. 281–301, Mar. 2018.
• ^[2] Kumar, K., Devi, V.L., Dhanamjayulu, C., et al, “Evaluation and deployment of a unified MPPT controller for hybrid Luo converter in combined PV and wind energy systems”, Scientific Reports, v. 14, n. 1, pp. 3248, 2024. doi: http://doi.org/10.1038/s41598-024-53605-z. PubMed PMID: 38332236.
  » https://doi.org/10.1038/s41598-024-53605-z
• ^[3] Lakshmi Devi, V., Kumar, K., Karpaga Priya, R., et al, “Performance and reliability analysis of double boost converter fed renewable PV system”, Journal of Engineering (Stevenage, England), v. 2023, pp. 2096501, 2023.
• ^[4] MARIPANDI, K., MAHADEVAN, B., EDWARD SEHAR FIR, et al, “Thermal investigation of 12/10 switched reluctance motor adopting different casing fins and materi-als for E-vehicle application”, Matéria (Rio de Janeiro), v. 29, n. 3, pp. 1–10, Jul. 2024. doi: https://doi.org/10.1590/1517-7076-RMAT-2024-0244.
  » https://doi.org/10.1590/1517-7076-RMAT-2024-0244
• ^[5] Wang, D., Zhang, J., Su, Y., “A machine learning approach to predict power demand in EVs”, IEEE Access: Practical Innovations, Open Solutions, v. 9, pp. 37758–37768, 2021.
• ^[6] Yang, Y., Feng, X., Ouyang, M., “A review on the application of machine learning for state of health estimation of lithium-ion batteries”, Journal of Power Sources, v. 433, pp. 126569, 2019.
• ^[7] Plett, G.L., “Extended Kalman filtering for battery management systems of LiPB-based HEV battery packs: Part 2”, Journal of Power Sources, v. 134, n. 2, pp. 277–292, Aug. 2004. doi: http://doi.org/10.1016/j.jpowsour.2004.02.033.
  » https://doi.org/10.1016/j.jpowsour.2004.02.033
• ^[8] He, H., Xiong, R., Zhang, X., et al, “State of charge estimation of battery packs using machine learning”, IEEE Transactions on Power Electronics, v. 33, n. 10, pp. 9982–9994, Oct. 2018.
• ^[9] Duan, S., Jiang, C., Zhao, M., “Hybrid EMS strategy for electric vehicles based on deep reinforcement learning”, IEEE Transactions on Intelligent Vehicles, v. 7, n. 1, pp. 44–55, Mar. 2022.
• ^[10] Kumar, K., Devi, V.L., “Fuel cell fed electrical vehicle performance analysis with enriched switched parameter Cuk converter”, Recent Advances in Electrical & Electronic Engineering, v. 10, n. 2, pp. 10–2174, 2023.
• ^[11] Chen, Y., Zhang, M., He, Y., “A review of machine learning for the battery management system of electric vehicles”, Applied Sciences (Basel, Switzerland), v. 9, n. 20, pp. 4222, Oct. 2019.
• ^[12] Xie, X., Lu, L., Ouyang, M., “Battery state of health estimation using a machine learning model optimized by a hybrid method”, Energy, v. 223, pp. 120051, 2021.
• ^[13] Gandhi, A., Panigrahi, D., Ranganathan, P., “Smart control and hybrid models for energy optimization in microgrids”, IEEE Transactions on Smart Grid, v. 13, n. 2, pp. 1105–1115, 2022.
• ^[14] Kim, H., Jin, T., Giacomoni, A., “Dynamic energy management for a renewable microgrid with battery storage”, IEEE Transactions on Smart Grid, v. 2, n. 3, pp. 508–516, Sep. 2011.
• ^[15] Zhang, C., Hu, Y., Wu, Z., “Battery management via XGBoost model prediction for SOH and SOC”, Energies, v. 13, n. 13, pp. 3432, Jul. 2020.
• ^[16] Huang, W., Zhang, X., Sun, F., “Machine learning based battery SOC estimation methods for electric vehicles”, IEEE Transactions on Power Electronics, v. 35, n. 5, pp. 5141–5152, 2020.
• ^[17] Asl, M., Bauer, P.H., Duong, T.Q., “Machine learning approaches for energy storage system optimization in renewable energy microgrids”, Renewable & Sustainable Energy Reviews, v. 124, pp. 109786, 2020.
• ^[18] Zhao, Q., Zhang, J., Pecht, M., “Deep learning-based energy demand prediction for renewable energy systems”, Applied Energy, v. 285, pp. 116378, 2021.
• ^[19] Li, Q., Hu, Y., Wu, Z., “Battery management via XGBoost model prediction for SOH and SOC”, Energies, v. 13, n. 13, pp. 3432, Jul. 2020.
• ^[20] Liu, J., Kong, W., Cao, Y., “Artificial intelligence-based battery management systems for electric vehicles”, Energy AI, v. 1, pp. 100006, Dec. 2020.
• ^[21] Kumar, K., Prasad, K.M., Devi, V.L., “Design and evaluation of power converter for integration of Lithium-ion battery and renewable sources”, Results in Engineering, v. 17, pp. 104409, 2025.
• ^[22] Liu, Y., Zhang, X., Pecht, M., “Deep learning enabled battery lifetime prediction for electric vehicles”, IEEE Access: Practical Innovations, Open Solutions, v. 9, pp. 33638–33645, 2021.
• ^[23] Kumar, K., Devi, V.L., Prasad, A., et al, “Analysis of a fuel cell-fed BLDC motor drive with a double boost converter for electric vehicle application”, In: Kumar, K., Devi, V.L., Prasad, A., et al, Smart Grids for Renewable Energy Systems, Electric Vehicles and Energy Storage Systems, Boca Raton, CRC Press, pp. 59–75, 2022. doi: http://doi.org/10.1201/9781003311195-4.
  » https://doi.org/10.1201/9781003311195-4
• ^[24] Shakouri, R., Andersen, M.D., Jørgensen, B.N., “Energy storage management using deep learning techniques”, IEEE Access: Practical Innovations, Open Solutions, v. 8, pp. 140984–140996, 2020.
• ^[25] Zhang, C., Song, Z., Pecht, M., “Charging and discharging optimization in energy storage systems using machine learning”, IEEE Transactions on Sustainable Energy, v. 12, n. 2, pp. 1121–1130, Apr. 2021.
• ^[26] Yao, H., Hu, Y., Wu, Z., “Demand response strategy optimization for smart grids based on reinforcement learning”, IEEE Access: Practical Innovations, Open Solutions, v. 8, pp. 143169–143178, 2020.
• ^[27] Souri, S., Bagheri, M., Monsef, H., “Reinforcement learning based energy management for hybrid renewable energy systems”, IEEE Access: Practical Innovations, Open Solutions, v. 8, pp. 187748–187760, 2020.
• ^[28] Sun, J., Zhang, Y., Wang, J., “Multi-source data-driven framework for electric vehicle energy management using machine learning”, Applied Energy, v. 251, pp. 113343, Oct. 2019.
• ^[29] Chang, Z., Liu, J., Xie, L., “Artificial intelligence empowered energy management system design for smart grids”, IEEE Transactions on Smart Grid, v. 12, n. 4, pp. 3203–3214, 2021.
• ^[30] Bhardwaj, A., Naidu, S.S.K., Srivastava, S.K., “AI-enabled energy management systems for microgrids”, IEEE Transactions on Smart Grid, v. 12, n. 1, pp. 279–288, 2021.

Publication Dates

• Publication in this collection
  21 Nov 2025
• Date of issue
  2025

History

• Received
  12 May 2025
• Accepted
  07 Oct 2025