Accurate estimation of state of health for lithium-ion batteries based on Pyraformer and TimeGAN data augmentation

Accurate state of health estimation of lithium batteries is crucial to ensure the safe and reliable operation of new energy vehicles. However, it remains a challenge to adequately develop highly accurate models for estimating state of health based on limited data. In this paper, an SOH estimation method integrating data augmentation techniques and data-driven models is proposed. Specifically, features that are strongly correlated with battery aging are first obtained by integrating the collaboration of multiple features. The time series generative adversarial network is used to expand the training data, and the SOH estimation model with the ability to capture multivariate dynamic relationships is constructed with the transformer network as the core, which improves the accuracy of predicting the capacity degradation of lithium-ion batteries. Through the validation on battery data with different material systems, the results show that the SOH estimation model for Li-ion batteries proposed in this paper can achieve high-precision SOH estimation. The RMSE and MAE of the estimation results are around 0.25 % and 0.2 %, respectively. This approach has a wide range of application potentials and is expected to realize high-precision and high-real-time battery full life cycle health management based on the end-cloud collaborative architecture and cloud BMS.