State of health estimation for Lithium-ion batteries based on novel feature extraction and BiGRU-Attention model

Accurate battery state of health (SOH) estimation is crucial to ensure the reliable and safe use of lithium batteries. Effective extraction of health features (HFs) and data-driven model selection are essential for achieving accurate SOH estimation. This paper proposes lithium-ion battery SOH estimation based on the novel HFs extraction and bidirectional gated recurrent unit Attention (BiGRU-Attention). The peak position, height, valley position, and height are derived from the rate of incremental capacity (ROIC) curve during the constant current charging stage to characterize battery deterioration. Spearman approach is employed to analysis the relationship between the extracted HFs and SOH. Three different data sets and seven comparison algorithms are utilized to achieve battery SOH estimation. The comparison results demonstrate that the maximum ΔSOH error of the BiGRU-Attention method is less than 4%, with RMSE and MAE values both being less than 1.2%. The cross-validation method is utilized to assess the generalizability of the model, The findings indicate that the RMSE and MAE values of the BiGRU-Attention method are both below 1.9% and outperformed other methods. The HFs proposed in this study effectively demonstrate battery performance degradation, and the selected method demonstrates high precision, robustness, and generalizability.