Integrating Multilayer Perceptron and Support Vector Regression for Enhanced State of Health Estimation in Lithium-Ion Batteries

• Published in: IEEE access Vol. 13; pp. 11463 - 11478
• Main Authors: Jafari, Sadiqa, Kim, Jisoo, Choi, Wonil, Byun, Yung-Cheol
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Batteries; battery performance; Electric vehicles; Ensemble learning; Evaluation; Lithium-ion batteries; Lithium-ion battery; Machine learning; Measurement; Multilayer perceptrons; optimization algorithms; Prediction algorithms; Predictive maintenance; Predictive models; Radio frequency; Random forests; Root-mean-square errors; SOH; Support vector machines; Temperature distribution; Temperature measurement
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3497656

Accurately evaluating the State of Health (SOH) of batteries is crucial for guaranteeing the secure and dependable functioning of Electric Vehicles (EVs). This paper presents a novel strategy for tackling the difficulties associated with intricate preprocessing and the demand for extensive data in conventional approaches to SOH measurement. Using sophisticated machine learning algorithms, we suggest an all-encompassing methodology for predicting the SOH. Our approach includes meticulous data preparation, which includes analyzing crucial operating elements such as voltage, current, and temperature. We utilized Support Vector Regression (SVR) and Multilayer Perceptron (MLP) models, which were fine-tuned using hyperparameter optimization. The models were assessed using evaluation metrics such as Root Mean Squared Error (RMSE), Mean Squared Error (MSE), and R-squared <inline-formula> <tex-math notation="LaTeX">R^{2} </tex-math></inline-formula>. In order to improve the accuracy of our predictions, we combined these models into a stacked ensemble using a Random Forest (RF) meta-model. This resulted in an <inline-formula> <tex-math notation="LaTeX">R^{2} </tex-math></inline-formula> value of 0.987, MAE of 0.02559, MSE of 0.0013, and RMSE of 0.00624. The results indicate that the ensemble outperforms individual models in predicting SOH. This research highlights the capacity of ensemble learning in predictive maintenance and battery management.