A method for estimating the SOH of lithium-ion batteries under complex charging conditions using dilated residual temporal encoding

In practical applications, various charging protocols and complex operational conditions present significant challenges for accurately estimating the state of health (SOH) of lithium-ion batteries. Currently available methods, which rely on single charging scenarios and manual feature extraction, exhibit limited generalizability and cannot be easily adapted to diverse charging requirements. Therefore, an end-to-end learning framework that integrates multisource data is proposed in this paper. First, a comprehensive dataset is constructed by incorporating negative pulse fast charging conditions, realistic fast charging simulations, and publicly available data, providing the model with a rich array of learning samples and overcoming the limitations associated with single charging scenarios. Second, a deep neural network based on dilated residual temporal encoding is developed. This network effectively extracts multidimensional features of battery degradation through multiscale convolution techniques, addresses the vanishing gradient problem in deep network training via residual gating mechanisms, and accurately captures long- and short-term temporal dependencies through adaptive temporal encoding. Finally, the experimental results demonstrate that the proposed method exhibits stable performance across various scenarios, including negative pulse fast charging and dynamic load conditions; thus, this method provides a reliable solution for the adaptation of battery management systems to diverse charging requirements.