Simultaneous Estimation of State of Health and Remaining Useful Life for Lithium-ion Batteries Using a Transfer-Learning-Based Fusion Model

Accurate battery-state estimation is considered essential for safe and stable lithium-ion battery operation. A novel joint framework is proposed in this study, in which only capacity data are employed as input, and the learning rate, hidden-layer node count and regularization coefficient of the Transformer-BiLSTM model are optimized by the Newton-Raphson-based optimizer algorithm (NRBO). Capacity-rebound patterns and long-term degradation are captured more accurately by NRBO algorithm, while local optima are avoided. Transfer learning is introduced, and positional embedding is fused with self-attention, so aging trajectories across chemistries are predicted accurately. The model is shown to support not only intra-type battery transfer but also inter-type transfer across different chemistries. The developed hybrid model is validated not only for cells tested at 4 °C, 24 °C and 43 °C, but also for data collected under both dynamic and constant operating conditions. Using only the first 10 % of the data, the proposed model keeps both mean absolute error (MAE) and root means square error (RMSE) below 1 % across all three battery types.