Work

An Integrated Explainable Machine Learning Framework for Cell-to-Pack Imbalance Prediction in Electric Vehicle Battery Systems

Year: 2026
Type: article
Abstract: This study presents a machine learning framework for predicting cell-to-pack imbalance in electric vehicle batteries. A set of battery attributes that affect battery performance, including State of Charge (SOC), State of Health (SOH), temperature (including ambient temperature and humidity), and cabin temperature, were used to make this prediction. The collected battery data were pre-processed to improve prediction performance. Three preprocessing methods were used: (i) Variational Mode Decomposition (VMD); (ii) deep denoising autoencoders, and (iii) feature selection using minimum Redundancy Maximum Relevance (mRMR) mutual information. This preprocessing step removes noise from the dataset and identifies useful information for prediction. Cell-to-pack imbalance prediction was performed using three interpretable machine learning models: (i) an Explainable Boosting Machine (EBM); (ii) fuzzy regression; and (iii) RuleFit. These models provide transparent predictions and help in understanding the factors responsible for battery imbalance. Models were evaluated on an independent 20-sample test set held out from the 100-record dataset. Among the three interpretable models, Fuzzy Regression achieved the strongest predictive performance (R² = 0.998), followed by RuleFit (R² = 0.976), while EBM achieved a lower R² of 0.201 on this independent test set despite providing detailed feature-level explanations. These results indicate that combining signal preprocessing, feature selection, and interpretable regression can provide useful and explainable information for cell-to-pack imbalance prediction, while also showing that performance can vary substantially between interpretable models on a small, two-pack dataset; validation on larger and more varied battery datasets is needed to confirm generalizability.
Language: English
Open Access status: bronze
OpenAlex ID: W7226019377
FWCI:
Cites:
Cited by:
Related to: 10