Abstract

Reliable state-of-health (SOH) estimation is important for the safe operation and management of lithium-ion batteries. This study proposes an SOH estimation framework that combines six health factors extracted from incremental-capacity curves, a transformer encoder, a state space model (SSM) decoder, and the Rime Optimization Algorithm (RIME). The transformer extracts relationships across different cycle positions, while the SSM describes the continuous change in battery health. RIME jointly selects the attention-head number and SSM state dimension. The model was evaluated using four NASA batteries through cross-battery four-fold validation and prediction experiments starting at 30%, 50%, and 70% of the cycle sequence. Ten models were compared under the same data partitions and preprocessing procedure. In four-fold validation, RIME–SSM–Transformer achieved an average MAE of 0.537% and an average RMSE of 0.740%, giving the lowest errors among all models. At the 30%, 50%, and 70% prediction starting points, its average MAE values were 1.194%, 0.810%, and 0.576%, while the corresponding RMSE values were 1.428%, 0.987%, and 0.654%. Parameter sensitivity analysis showed that six attention heads and an SSM state dimension of 64 gave the lowest validation error within the tested range. External validation on four additional batteries produced an average MAE of 0.629% and an average RMSE of 0.827%. These results show that the proposed framework provides accurate and stable SOH estimates under different degradation paths and amounts of available battery history.

IPC Classification

Keywords