A Fault Prediction Method for Lithium-ion Batteries by Fusing Internal and External Features with Stacked Integration Models

Battery status is influenced by various internal and external factors, and battery management systems (BMSs) still face many challenges in fault prediction. To address issues such as neglecting external features, insufficient feature selection dimensions, and limited BMS hardware computation in real-time environments, this study proposes a lithium-ion battery fault prediction method that integrates internal and external features with a stacking ensemble model. By combining internal sensor data with external features, a sliding time window-Lempel-Ziv-Welch method is proposed. Feature importance is calculated using the Gini coefficient, and multiple discriminant correlation analysis is used for feature fusion across different domains. Finally, hyperparameters are optimized using Newton-Raphson-based optimizer algorithm, and the stacking model is trained, with recall, precision, accuracy, and area under the curve (AUC) as evaluation metrics. Experimental results show that compared to individual models, the proposed method improves precision, recall, and accuracy by 11.2%, 9.88%, and 9.1%, respectively, with an AUC of 0.897. This research aims to enhance battery fault prediction accuracy and reliability, providing new technical support for the safe and reliable operation of new energy vehicles.