Answers are grounded in this paper and cite their sources.
Accurate state-of-health (SOH) estimation of lithium-ion batteries remains challenging under limited labeled cells, heterogeneous degradation trajectories, and variable operating conditions. To address this issue, this study proposes temporal convolution–LightMamba HealthNet, combining physically motivated cycle-level health features with short-window temporal modeling for SOH estimation under limited training-cell budgets. Specifically, each charge–discharge cycle is segmented into charging and discharging stages, and 12 physically interpretable health indicators are constructed from charging dynamics, discharge response, and reference voltage-curve deviation. The compact descriptors summarize measurable responses to battery aging and operating conditions. Based on these compact health features, temporal convolution–LightMamba HealthNet is developed to learn degradation evolution from historical feature sequences. The temporal convolutional module captures local cycle-to-cycle variations, while LightMambaProxy combines depthwise convolution, a feedforward bottleneck and gating to transform short-window features. The model is evaluated on HUST and XJTU, with training-cell-budget comparisons showing lower mean test errors than the evaluated TCN and LSTM configurations. Ablation studies and parameter-matched module comparisons support the predictive contributions of linear projection, temporal convolution and LightMambaProxy. Matched supplementary comparisons with TCN, LSTM, Mamba and Transformer show the lowest mean test errors for the evaluated HealthNet configuration, with a competitive measured accuracy–cost trade-off.
No takes yet. Share an insight, caveat, or question.
Lv et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: