Global–Local Feature Fusion for Accurate Long-Term Battery State-of-Health Forecast

Le résumé fourni par la source

To ensure the safety and reliability of lithium-ion battery (LIB) operation, accurate prediction of the state of health (SOH) of the batteries is crucial. However, long-term SOH prediction is difficult due to the complexity of LIBs’ internal mechanisms and the influence of the external environment. In this study, we propose a novel convolutional neural network (CNN) architecture that combines Transformer with entropy-driven attention and patch embedding. First, different battery health indicators (HIs) are fully extracted, and then the Pearson correlation coefficient is used to select the most relevant HIs to SOH from them. Next, the patch embedding module is utilized to enhance the feature characterization. Through entropy-driven attention, the Transformer encoder module effectively and precisely captures long-range dependencies between data and grasps the long-term trend of battery aging as a whole. CNN concentrates on sequence variations and fluctuations that take place within short-term time windows by extracting localized characteristics through convolutional operations. Lastly, the two complementary features, local and global, are fully integrated, providing rich information for prediction. Numerous experiments indicate that the model achieves excellent results on different battery datasets, showing good accuracy and robustness.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.