Lithium-Ion Battery State of Health Prediction Based on RL-VC and BiTCN-BiGRU-SA Model

Accurate estimation of the state of health (SOH) in lithium-ion batteries serves as a critical technical prerequisite for optimizing the performance of battery management systems (BMS), ensuring system safety, and enhancing economic benefits. To address the challenges of insufficient feature extraction quality and complex nonlinear characteristics in lithium-ion battery SOH estimation, this paper proposes an SOH estimation method based on revised Lorentzian-voltage capacity (RL-VC) and bidirectional temporal convolutional network-bidirectional gated recurrent unit-self attention mechanism (BiTCN-BiGRU-SA). First, an RL-VC model is constructed to obtain the battery incremental capacity (IC) curve and extract health indicators that characterize battery aging. Traditional incremental capacity analysis (ICA) depends on the selection of sampling intervals and filtering parameters, and is easily affected by noise, resulting in blurred features, whereas the RL-VC model, based on Lorentzian function fitting, can generate a smooth and clearly featured IC curve without complex filtering, thereby enhancing the linear correlation between characteristic parameters and SOH. Second, a deep learning framework integrating BiTCN, BiGRU, and SA mechanism is constructed. The BiTCN layer extracts temporal critical features of battery operational data through bidirectional learning, the BiGRU layer captures long-term dependencies in the data, and the SA layer calculates attention scores based on BiGRU outputs to further enhance SOH estimation accuracy. Finally, the proposed method is validated using both an experimental dataset and the publicly available NASA dataset. The results demonstrate that the method exhibits high accuracy and robustness under different battery types and aging conditions, with a maximum mean absolute error (MAE) of 0.77% and a maximum root mean square error (RMSE) of 1%, outperforming comparison models such as TCN-GRU and BiTCN-BiGRU.