A hybrid neural network based on variational mode decomposition denoising for predicting state-of-health of lithium-ion batteries.

Accurately predicting the State of Health (SOH) of lithium-ion batteries is essential for ensuring their safe and reliable operation, and reducing maintenance and service costs for associated equipment. Nevertheless, the aging data of lithium-ion batteries displays pronounced nonlinearity and is plagued by issues such as capacity regeneration. To address this issue, this study proposes a framework for SOH prediction of lithium-ion batteries based on Variational Mode Decomposition (VMD) and CNN-Transformer. First, the original data undergoes a VMD smoothing process to eliminate capacity regeneration and a portion of the noise signals. Subsequently, Convolutional Neural Networks (CNN) is utilized for feature extraction. Then, a modified Transformer model is employed to capture the inherent correlations in the time series and map the features to future SOH values. An iterative strategy is adopted to predict SOH for each charge-discharge cycle. The experimental results on the CALCE dataset demonstrate that the proposed method can accurately predict the SOH of lithium-ion batteries using just 5 %–6 % of the complete cycle's aging data. Additionally, comparative results on the NASA dataset show that, compared to the latest relevant literature, the proposed method achieves high prediction accuracy while maintaining exceptional generalization. • Apply VMD to remove the effect of irregular noise on model performance. • Use CNN-transformer model for feature extraction and long dependency learning. • Iterative prediction strategies require only a small amount of data. [ABSTRACT FROM AUTHOR]