Rout Smitanjali, Samal Sudhansu Kumar, Gelmecha Demissie Jobir, Mishra Satyasis
Department of Electrical Engineering Centurion, University of Technology & Management, Bhubaneswar, Odisha, India.
Department of Electronics and Communication Engineering, Adama Science and Technology University (ASTU), Adama, Ethiopia.
Sci Rep. 2025 Aug 19;15(1):30438. doi: 10.1038/s41598-025-93775-y.
Accurate estimation of the State of Health (SOH) is crucial for ensuring the performance, safety, and longevity of lithium-ion batteries in electric vehicles. Traditional methods, such as Coulomb Counting and the Extended Kalman Filter, often lack the accuracy and computational efficiency required for modern applications. This study proposes an advanced framework that leverages machine learning models to model the nonlinear degradation patterns of lithium-ion batteries by focusing on key features such as voltage, current, internal resistance, and temperature. The proposed framework incorporates optimized pre-processing techniques, including normalization, to improve data quality and ensure consistency across varying battery conditions. Advanced machine learning models, including Adaboost, Xgboost, Ridge Regression, Decision Trees, Random Forests, Artificial Neural Networks, and Long Short-Term Memory Networks (LSTM), are employed to analyze battery performance. Among these, the LSTM network demonstrates outstanding capability in capturing long-term dependencies in sequential battery data, achieving a mean squared error of 0.000115 and an R2 score of 0.9982. It also accurately predicts the remaining life cycle of the battery using temporal patterns derived from MATLAB model datasheets, significantly reducing estimation errors. A comprehensive comparison using performance metrics such as root mean squared error, mean absolute error, and R2 scores highlights the LSTM model's superiority while evaluating the suitability of other approaches. The proposed method not only improves estimation accuracy but also reduces computational demands through optimized feature selection and model training strategies, making it highly suitable for real-time applications in lightweight electric vehicles with limited computational resources. This research bridges the gap between theoretical advancements in data-driven techniques and their practical deployment in real-world battery management systems.
准确估计健康状态（SOH）对于确保电动汽车中锂离子电池的性能、安全性和使用寿命至关重要。传统方法，如库仑计数法和扩展卡尔曼滤波器，往往缺乏现代应用所需的准确性和计算效率。本研究提出了一个先进的框架，该框架利用机器学习模型，通过关注电压、电流、内阻和温度等关键特征，对锂离子电池的非线性退化模式进行建模。所提出的框架采用了优化的预处理技术，包括归一化，以提高数据质量并确保在不同电池条件下的一致性。采用了先进的机器学习模型，包括Adaboost、Xgboost、岭回归、决策树、随机森林、人工神经网络和长短期记忆网络（LSTM）来分析电池性能。其中，LSTM网络在捕捉连续电池数据中的长期依赖性方面表现出卓越能力，均方误差达到0.000115，R2分数达到0.9982。它还利用从MATLAB模型数据表得出的时间模式准确预测电池的剩余生命周期，显著减少估计误差。使用均方根误差、平均绝对误差和R2分数等性能指标进行的全面比较突出了LSTM模型的优越性，同时评估了其他方法的适用性。所提出的方法不仅提高了估计准确性，还通过优化特征选择和模型训练策略降低了计算需求，使其非常适合计算资源有限的轻型电动汽车的实时应用。这项研究弥合了数据驱动技术的理论进展与其在实际电池管理系统中的实际部署之间 的差距。
